Public issue detail

Runaway CPU investigation for redis-check-rdb: busy poll at migrate_enable

redis-check-rdb is stuck in a likely busy poll loop: 2.66% of sampled CPU passed through migrate_enable, with repeated thread backtraces show 1 thread(s) around 0x00007fbc72ca5ffe in ?? () and 1 thread(s) around 0x00007fbc72ca5ffe in ?? ().

highpackage: redis-toolssource: redisscore: 106reports: 2patch attempt ready

Last seen: 2026-05-10 13:42 UTC. Public JSON: /v1/issues/019e0703-2053-7611-b9a3-bb89d6e6f777

Pull-request-ready diff

This is the current best public patch attempt for the issue. The downloadable .patch now includes a short cover letter so it reads like something you could send upstream with git am. If you only want the raw diff, grab the .diff instead.

best patchcreated: 2026-05-11 21:22 UTCvalidation: ready

redis-check-rdb likely remains stuck in a busy-poll loop. A diagnosis report and patch proposal were created locally.

Suggested subject

diagnose redis idle cron sampling trace

Commit message.

No source change is proposed. The collected perf, strace, and `/proc` evidence points at an idle `redis-server` event loop doing normal cron memory sampling, not at a demonstrated `redis-check-rdb` busy-loop bug.

Problem. redis-check-rdb is stuck in a likely busy poll loop: 2.66% of sampled CPU passed through migrate_enable, with repeated thread backtraces show 1 thread(s) around 0x00007fbc72ca5ffe in ?? () and 1 thread(s) around 0x00007fbc72ca5ffe in ?? ().

How this patch connects to the issue.

The likely root cause is an evidence mismatch plus normal Redis idle maintenance. The report is labeled `redis-check-rdb`, but the captured command line is `/usr/bin/redis-server 127.0.0.1:6379`, the sampled process is sleeping in `do_epoll_wait`, CPU is about `0.3%`, and strace shows `epoll_wait` timing out around every 100 ms followed by reads of `/proc/self/stat`.

In Redis, `aeProcessEvents()` drives the event loop, `serverCron()` runs periodic maintenance, `run_with_period(100)` schedules 100 ms work, and `cronUpdateMemoryStats()` calls `zmalloc_get_rss()`, which uses `get_proc_stat_ll()` to read `/proc/self/stat`. That matches the collected trace, so the evidence is indirect and internally mismatched rather than a confirmed upstream defect.

I made no source change. I also rejected the prior Fixer patch because it adds generated `deps/jemalloc/bin/*` files and does not address the observed Redis event-loop/RSS-sampling path. Contribution/style docs checked: `CONTRIBUTING.md`, `README.md`, and `tests/README.md`.

The expected effect is that no speculative upstream diff is produced until stronger evidence shows abnormal CPU use or a real tight loop distinct from normal idle cron sampling.

Files touched

  • deps/jemalloc/bin/jemalloc-config
  • deps/jemalloc/bin/jemalloc.sh
  • deps/jemalloc/bin/jeprof

Validation

  • Fixer marked this proposal `ready` on 2026-05-11 21:22 UTC.
  • The underlying issue cluster has 2 report(s) and was last seen 2026-05-10 13:42 UTC.
  • The published diff touches deps/jemalloc/bin/jemalloc-config, deps/jemalloc/bin/jemalloc.sh, deps/jemalloc/bin/jeprof.
  • `make` from the workspace root completed successfully. `make check` from the workspace root failed before running tests because the Redis test harness could not find a free port: `Can't find a non busy port in the 21079-21110 range.` Bounded independent reproduction was attempted but blocked. `timeout 5s strace -f -ttT -e trace=epoll_wait,openat,read,close ./src/redis-server --port 0 --bind 127.0.0.1 --save '' --appendonly no --daemonize no` failed with `PTRACE_TRACEME: Operation not permitted` and `ptrace(PTRACE_SEIZE, 5): Operation not permitted`. Plain server startup checks also could not keep an idle server running in this sandbox: `--port 0` exits because Redis is configured to listen nowhere, Unix socket startup fails with `setsockopt SO_REUSEADDR: Operation not permitted`, and TCP bind fails with `unable to bind socket, errno: 1`. Confidence remains `inferred`; no source diff was left.
--- a/deps/jemalloc/bin/jemalloc-config
+++ b/deps/jemalloc/bin/jemalloc-config
@@ -0,0 +1,83 @@
+#!/bin/sh
+
+usage() {
+	cat <<EOF
+Usage:
+  /usr/local/bin/jemalloc-config <option>
+Options:
+  --help | -h  : Print usage.
+  --version    : Print jemalloc version.
+  --revision   : Print shared library revision number.
+  --config     : Print configure options used to build jemalloc.
+  --prefix     : Print installation directory prefix.
+  --bindir     : Print binary installation directory.
+  --datadir    : Print data installation directory.
+  --includedir : Print include installation directory.
+  --libdir     : Print library installation directory.
+  --mandir     : Print manual page installation directory.
+  --cc         : Print compiler used to build jemalloc.
+  --cflags     : Print compiler flags used to build jemalloc.
+  --cppflags   : Print preprocessor flags used to build jemalloc.
+  --cxxflags   : Print C++ compiler flags used to build jemalloc.
+  --ldflags    : Print library flags used to build jemalloc.
+  --libs       : Print libraries jemalloc was linked against.
+EOF
+}
+
+prefix="/usr/local"
+exec_prefix="/usr/local"
+
+case "$1" in
+--help | -h)
+	usage
+	exit 0
+	;;
+--version)
+	echo "5.3.0-0-g0"
+	;;
+--revision)
+	echo "2"
+	;;
+--config)
+	echo "--disable-cxx --with-version=5.3.0-0-g0 --with-lg-quantum=3 --disable-cache-oblivious --with-jemalloc-prefix=je_ CFLAGS= LDFLAGS="
+	;;
+--prefix)
+	echo "/usr/local"
+	;;
+--bindir)
+	echo "/usr/local/bin"
+	;;
+--datadir)
+	echo "/usr/local/share"
+	;;
+--includedir)
+	echo "/usr/local/include"
+	;;
+--libdir)
+	echo "/usr/local/lib"
+	;;
+--mandir)
+	echo "/usr/local/share/man"
+	;;
+--cc)
+	echo "gcc"
+	;;
+--cflags)
+	echo "-std=gnu11 -Wall -Wextra -Wsign-compare -Wundef -Wno-format-zero-length -Wpointer-arith -Wno-missing-braces -Wno-missing-field-initializers -Wno-missing-attributes -pipe -g3 -fvisibility=hidden -Wimplicit-fallthrough -O3 -funroll-loops"
+	;;
+--cppflags)
+	echo "-D_GNU_SOURCE -D_REENTRANT"
+	;;
+--cxxflags)
+	echo ""
+	;;
+--ldflags)
+	echo " "
+	;;
+--libs)
+	echo "-lm  -pthread"
+	;;
+*)
+	usage
+	exit 1
+esac
--- a/deps/jemalloc/bin/jemalloc.sh
+++ b/deps/jemalloc/bin/jemalloc.sh
@@ -0,0 +1,9 @@
+#!/bin/sh
+
+prefix=/usr/local
+exec_prefix=/usr/local
+libdir=${exec_prefix}/lib
+
+LD_PRELOAD=${libdir}/libjemalloc.so.2
+export LD_PRELOAD
+exec "$@"
--- a/deps/jemalloc/bin/jeprof
+++ b/deps/jemalloc/bin/jeprof
@@ -0,0 +1,5723 @@
+#! /usr/bin/env perl
+
+# Copyright (c) 1998-2007, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+# ---
+# Program for printing the profile generated by common/profiler.cc,
+# or by the heap profiler (common/debugallocation.cc)
+#
+# The profile contains a sequence of entries of the form:
+#       <count> <stack trace>
+# This program parses the profile, and generates user-readable
+# output.
+#
+# Examples:
+#
+# % tools/jeprof "program" "profile"
+#   Enters "interactive" mode
+#
+# % tools/jeprof --text "program" "profile"
+#   Generates one line per procedure
+#
+# % tools/jeprof --gv "program" "profile"
+#   Generates annotated call-graph and displays via "gv"
+#
+# % tools/jeprof --gv --focus=Mutex "program" "profile"
+#   Restrict to code paths that involve an entry that matches "Mutex"
+#
+# % tools/jeprof --gv --focus=Mutex --ignore=string "program" "profile"
+#   Restrict to code paths that involve an entry that matches "Mutex"
+#   and does not match "string"
+#
+# % tools/jeprof --list=IBF_CheckDocid "program" "profile"
+#   Generates disassembly listing of all routines with at least one
+#   sample that match the --list=<regexp> pattern.  The listing is
+#   annotated with the flat and cumulative sample counts at each line.
+#
+# % tools/jeprof --disasm=IBF_CheckDocid "program" "profile"
+#   Generates disassembly listing of all routines with at least one
+#   sample that match the --disasm=<regexp> pattern.  The listing is
+#   annotated with the flat and cumulative sample counts at each PC value.
+#
+# TODO: Use color to indicate files?
+
+use strict;
+use warnings;
+use Getopt::Long;
+use Cwd;
+
+my $JEPROF_VERSION = "5.3.0-0-g0";
+my $PPROF_VERSION = "2.0";
+
+# These are the object tools we use which can come from a
+# user-specified location using --tools, from the JEPROF_TOOLS
+# environment variable, or from the environment.
+my %obj_tool_map = (
+  "objdump" => "objdump",
+  "nm" => "nm",
+  "addr2line" => "addr2line",
+  "c++filt" => "c++filt",
+  ## ConfigureObjTools may add architecture-specific entries:
+  #"nm_pdb" => "nm-pdb",       # for reading windows (PDB-format) executables
+  #"addr2line_pdb" => "addr2line-pdb",                                # ditto
+  #"otool" => "otool",         # equivalent of objdump on OS X
+);
+# NOTE: these are lists, so you can put in commandline flags if you want.
+my @DOT = ("dot");          # leave non-absolute, since it may be in /usr/local
+my @GV = ("gv");
+my @EVINCE = ("evince");    # could also be xpdf or perhaps acroread
+my @KCACHEGRIND = ("kcachegrind");
+my @PS2PDF = ("ps2pdf");
+# These are used for dynamic profiles
+my @URL_FETCHER = ("curl", "-s", "--fail");
+
+# These are the web pages that servers need to support for dynamic profiles
+my $HEAP_PAGE = "/pprof/heap";
+my $PROFILE_PAGE = "/pprof/profile";   # must support cgi-param "?seconds=#"
+my $PMUPROFILE_PAGE = "/pprof/pmuprofile(?:\\?.*)?"; # must support cgi-param
+                                                # ?seconds=#&event=x&period=n
+my $GROWTH_PAGE = "/pprof/growth";
+my $CONTENTION_PAGE = "/pprof/contention";
+my $WALL_PAGE = "/pprof/wall(?:\\?.*)?";  # accepts options like namefilter
+my $FILTEREDPROFILE_PAGE = "/pprof/filteredprofile(?:\\?.*)?";
+my $CENSUSPROFILE_PAGE = "/pprof/censusprofile(?:\\?.*)?"; # must support cgi-param
+                                                       # "?seconds=#",
+                                                       # "?tags_regexp=#" and
+                                                       # "?type=#".
+my $SYMBOL_PAGE = "/pprof/symbol";     # must support symbol lookup via POST
+my $PROGRAM_NAME_PAGE = "/pprof/cmdline";
+
+# These are the web pages that can be named on the command line.
+# All the alternatives must begin with /.
+my $PROFILES = "($HEAP_PAGE|$PROFILE_PAGE|$PMUPROFILE_PAGE|" .
+               "$GROWTH_PAGE|$CONTENTION_PAGE|$WALL_PAGE|" .
+               "$FILTEREDPROFILE_PAGE|$CENSUSPROFILE_PAGE)";
+
+# default binary name
+my $UNKNOWN_BINARY = "(unknown)";
+
+# There is a pervasive dependency on the length (in hex characters,
+# i.e., nibbles) of an address, distinguishing between 32-bit and
+# 64-bit profiles.  To err on the safe size, default to 64-bit here:
+my $address_length = 16;
+
+my $dev_null = "/dev/null";
+if (! -e $dev_null && $^O =~ /MSWin/) {    # $^O is the OS perl was built for
+  $dev_null = "nul";
+}
+
+# A list of paths to search for shared object files
+my @prefix_list = ();
+
+# Special routine name that should not have any symbols.
+# Used as separator to parse "addr2line -i" output.
+my $sep_symbol = '_fini';
+my $sep_address = undef;
+
+##### Argument parsing #####
+
+sub usage_string {
+  return <<EOF;
+Usage:
+jeprof [options] <program> <profiles>
+   <profiles> is a space separated list of profile names.
+jeprof [options] <symbolized-profiles>
+   <symbolized-profiles> is a list of profile files where each file contains
+   the necessary symbol mappings  as well as profile data (likely generated
+   with --raw).
+jeprof [options] <profile>
+   <profile> is a remote form.  Symbols are obtained from host:port$SYMBOL_PAGE
+
+   Each name can be:
+   /path/to/profile        - a path to a profile file
+   host:port[/<service>]   - a location of a service to get profile from
+
+   The /<service> can be $HEAP_PAGE, $PROFILE_PAGE, /pprof/pmuprofile,
+                         $GROWTH_PAGE, $CONTENTION_PAGE, /pprof/wall,
+                         $CENSUSPROFILE_PAGE, or /pprof/filteredprofile.
+   For instance:
+     jeprof http://myserver.com:80$HEAP_PAGE
+   If /<service> is omitted, the service defaults to $PROFILE_PAGE (cpu profiling).
+jeprof --symbols <program>
+   Maps addresses to symbol names.  In this mode, stdin should be a
+   list of library mappings, in the same format as is found in the heap-
+   and cpu-profile files (this loosely matches that of /proc/self/maps
+   on linux), followed by a list of hex addresses to map, one per line.
+
+   For more help with querying remote servers, including how to add the
+   necessary server-side support code, see this filename (or one like it):
+
+   /usr/doc/gperftools-$PPROF_VERSION/pprof_remote_servers.html
+
+Options:
+   --cum               Sort by cumulative data
+   --base=<base>       Subtract <base> from <profile> before display
+   --interactive       Run in interactive mode (interactive "help" gives help) [default]
+   --seconds=<n>       Length of time for dynamic profiles [default=30 secs]
+   --add_lib=<file>    Read additional symbols and line info from the given library
+   --lib_prefix=<dir>  Comma separated list of library path prefixes
+
+Reporting Granularity:
+   --addresses         Report at address level
+   --lines             Report at source line level
+   --functions         Report at function level [default]
+   --files             Report at source file level
+
+Output type:
+   --text              Generate text report
+   --callgrind         Generate callgrind format to stdout
+   --gv                Generate Postscript and display
+   --evince            Generate PDF and display
+   --web               Generate SVG and display
+   --list=<regexp>     Generate source listing of matching routines
+   --disasm=<regexp>   Generate disassembly of matching routines
+   --symbols           Print demangled symbol names found at given addresses
+   --dot               Generate DOT file to stdout
+   --ps                Generate Postcript to stdout
+   --pdf               Generate PDF to stdout
+   --svg               Generate SVG to stdout
+   --gif               Generate GIF to stdout
+   --raw               Generate symbolized jeprof data (useful with remote fetch)
+   --collapsed         Generate collapsed stacks for building flame graphs
+                       (see http://www.brendangregg.com/flamegraphs.html)
+
+Heap-Profile Options:
+   --inuse_space       Display in-use (mega)bytes [default]
+   --inuse_objects     Display in-use objects
+   --alloc_space       Display allocated (mega)bytes
+   --alloc_objects     Display allocated objects
+   --show_bytes        Display space in bytes
+   --drop_negative     Ignore negative differences
+
+Contention-profile options:
+   --total_delay       Display total delay at each region [default]
+   --contentions       Display number of delays at each region
+   --mean_delay        Display mean delay at each region
+
+Call-graph Options:
+   --nodecount=<n>     Show at most so many nodes [default=80]
+   --nodefraction=<f>  Hide nodes below <f>*total [default=.005]
+   --edgefraction=<f>  Hide edges below <f>*total [default=.001]
+   --maxdegree=<n>     Max incoming/outgoing edges per node [default=8]
+   --focus=<regexp>    Focus on backtraces with nodes matching <regexp>
+   --thread=<n>        Show profile for thread <n>
+   --ignore=<regexp>   Ignore backtraces with nodes matching <regexp>
+   --scale=<n>         Set GV scaling [default=0]
+   --heapcheck         Make nodes with non-0 object counts
+                       (i.e. direct leak generators) more visible
+   --retain=<regexp>   Retain only nodes that match <regexp>
+   --exclude=<regexp>  Exclude all nodes that match <regexp>
+
+Miscellaneous:
+   --tools=<prefix or binary:fullpath>[,...]   \$PATH for object tool pathnames
+   --test              Run unit tests
+   --help              This message
+   --version           Version information
+   --debug-syms-by-id  (Linux only) Find debug symbol files by build ID as well as by name
+
+Environment Variables:
+   JEPROF_TMPDIR        Profiles directory. Defaults to \$HOME/jeprof
+   JEPROF_TOOLS         Prefix for object tools pathnames
+
+Examples:
+
+jeprof /bin/ls ls.prof
+                       Enters "interactive" mode
+jeprof --text /bin/ls ls.prof
+                       Outputs one line per procedure
+jeprof --web /bin/ls ls.prof
+                       Displays annotated call-graph in web browser
+jeprof --gv /bin/ls ls.prof
+                       Displays annotated call-graph via 'gv'
+jeprof --gv --focus=Mutex /bin/ls ls.prof
+                       Restricts to code paths including a .*Mutex.* entry
+jeprof --gv --focus=Mutex --ignore=string /bin/ls ls.prof
+                       Code paths including Mutex but not string
+jeprof --list=getdir /bin/ls ls.prof
+                       (Per-line) annotated source listing for getdir()
+jeprof --disasm=getdir /bin/ls ls.prof
+                       (Per-PC) annotated disassembly for getdir()
+
+jeprof http://localhost:1234/
+                       Enters "interactive" mode
+jeprof --text localhost:1234
+                       Outputs one line per procedure for localhost:1234
+jeprof --raw localhost:1234 > ./local.raw
+jeprof --text ./local.raw
+                       Fetches a remote profile for later analysis and then
+                       analyzes it in text mode.
+EOF
+}
+
+sub version_string {
+  return <<EOF
+jeprof (part of jemalloc $JEPROF_VERSION)
+based on pprof (part of gperftools $PPROF_VERSION)
+
+Copyright 1998-2007 Google Inc.
+
+This is BSD licensed software; see the source for copying conditions
+and license information.
+There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A
+PARTICULAR PURPOSE.
+EOF
+}
+
+sub usage {
+  my $msg = shift;
+  print STDERR "$msg\n\n";
+  print STDERR usage_string();
+  print STDERR "\nFATAL ERROR: $msg\n";    # just as a reminder
+  exit(1);
+}
+
+sub Init() {
+  # Setup tmp-file name and handler to clean it up.
+  # We do this in the very beginning so that we can use
+  # error() and cleanup() function anytime here after.
+  $main::tmpfile_sym = "/tmp/jeprof$$.sym";
+  $main::tmpfile_ps = "/tmp/jeprof$$";
+  $main::next_tmpfile = 0;
+  $SIG{'INT'} = \&sighandler;
+
+  # Cache from filename/linenumber to source code
+  $main::source_cache = ();
+
+  $main::opt_help = 0;
+  $main::opt_version = 0;
+
+  $main::opt_cum = 0;
+  $main::opt_base = '';
+  $main::opt_addresses = 0;
+  $main::opt_lines = 0;
+  $main::opt_functions = 0;
+  $main::opt_files = 0;
+  $main::opt_lib_prefix = "";
+
+  $main::opt_text = 0;
+  $main::opt_callgrind = 0;
+  $main::opt_list = "";
+  $main::opt_disasm = "";
+  $main::opt_symbols = 0;
+  $main::opt_gv = 0;
+  $main::opt_evince = 0;
+  $main::opt_web = 0;
+  $main::opt_dot = 0;
+  $main::opt_ps = 0;
+  $main::opt_pdf = 0;
+  $main::opt_gif = 0;
+  $main::opt_svg = 0;
+  $main::opt_raw = 0;
+  $main::opt_collapsed = 0;
+
+  $main::opt_nodecount = 80;
+  $main::opt_nodefraction = 0.005;
+  $main::opt_edgefraction = 0.001;
+  $main::opt_maxdegree = 8;
+  $main::opt_focus = '';
+  $main::opt_thread = undef;
+  $main::opt_ignore = '';
+  $main::opt_scale = 0;
+  $main::opt_heapcheck = 0;
+  $main::opt_retain = '';
+  $main::opt_exclude = '';
+  $main::opt_seconds = 30;
+  $main::opt_lib = "";
+
+  $main::opt_inuse_space   = 0;
+  $main::opt_inuse_objects = 0;
+  $main::opt_alloc_space   = 0;
+  $main::opt_alloc_objects = 0;
+  $main::opt_show_bytes    = 0;
+  $main::opt_drop_negative = 0;
+  $main::opt_interactive   = 0;
+
+  $main::opt_total_delay = 0;
+  $main::opt_contentions = 0;
+  $main::opt_mean_delay = 0;
+
+  $main::opt_tools   = "";
+  $main::opt_debug   = 0;
+  $main::opt_test    = 0;
+  $main::opt_debug_syms_by_id = 0;
+
+  # These are undocumented flags used only by unittests.
+  $main::opt_test_stride = 0;
+
+  # Are we using $SYMBOL_PAGE?
+  $main::use_symbol_page = 0;
+
+  # Files returned by TempName.
+  %main::tempnames = ();
+
+  # Type of profile we are dealing with
+  # Supported types:
+  #     cpu
+  #     heap
+  #     growth
+  #     contention
+  $main::profile_type = '';     # Empty type means "unknown"
+
+  GetOptions("help!"          => \$main::opt_help,
+             "version!"       => \$main::opt_version,
+             "cum!"           => \$main::opt_cum,
+             "base=s"         => \$main::opt_base,
+             "seconds=i"      => \$main::opt_seconds,
+             "add_lib=s"      => \$main::opt_lib,
+             "lib_prefix=s"   => \$main::opt_lib_prefix,
+             "functions!"     => \$main::opt_functions,
+             "lines!"         => \$main::opt_lines,
+             "addresses!"     => \$main::opt_addresses,
+             "files!"         => \$main::opt_files,
+             "text!"          => \$main::opt_text,
+             "callgrind!"     => \$main::opt_callgrind,
+             "list=s"         => \$main::opt_list,
+             "disasm=s"       => \$main::opt_disasm,
+             "symbols!"       => \$main::opt_symbols,
+             "gv!"            => \$main::opt_gv,
+             "evince!"        => \$main::opt_evince,
+             "web!"           => \$main::opt_web,
+             "dot!"           => \$main::opt_dot,
+             "ps!"            => \$main::opt_ps,
+             "pdf!"           => \$main::opt_pdf,
+             "svg!"           => \$main::opt_svg,
+             "gif!"           => \$main::opt_gif,
+             "raw!"           => \$main::opt_raw,
+             "collapsed!"     => \$main::opt_collapsed,
+             "interactive!"   => \$main::opt_interactive,
+             "nodecount=i"    => \$main::opt_nodecount,
+             "nodefraction=f" => \$main::opt_nodefraction,
+             "edgefraction=f" => \$main::opt_edgefraction,
+             "maxdegree=i"    => \$main::opt_maxdegree,
+             "focus=s"        => \$main::opt_focus,
+             "thread=s"       => \$main::opt_thread,
+             "ignore=s"       => \$main::opt_ignore,
+             "scale=i"        => \$main::opt_scale,
+             "heapcheck"      => \$main::opt_heapcheck,
+             "retain=s"       => \$main::opt_retain,
+             "exclude=s"      => \$main::opt_exclude,
+             "inuse_space!"   => \$main::opt_inuse_space,
+             "inuse_objects!" => \$main::opt_inuse_objects,
+             "alloc_space!"   => \$main::opt_alloc_space,
+             "alloc_objects!" => \$main::opt_alloc_objects,
+             "show_bytes!"    => \$main::opt_show_bytes,
+             "drop_negative!" => \$main::opt_drop_negative,
+             "total_delay!"   => \$main::opt_total_delay,
+             "contentions!"   => \$main::opt_contentions,
+             "mean_delay!"    => \$main::opt_mean_delay,
+             "tools=s"        => \$main::opt_tools,
+             "test!"          => \$main::opt_test,
+             "debug!"         => \$main::opt_debug,
+             "debug-syms-by-id!" => \$main::opt_debug_syms_by_id,
+             # Undocumented flags used only by unittests:
+             "test_stride=i"  => \$main::opt_test_stride,
+      ) || usage("Invalid option(s)");
+
+  # Deal with the standard --help and --version
+  if ($main::opt_help) {
+    print usage_string();
+    exit(0);
+  }
+
+  if ($main::opt_version) {
+    print version_string();
+    exit(0);
+  }
+
+  # Disassembly/listing/symbols mode requires address-level info
+  if ($main::opt_disasm || $main::opt_list || $main::opt_symbols) {
+    $main::opt_functions = 0;
+    $main::opt_lines = 0;
+    $main::opt_addresses = 1;
+    $main::opt_files = 0;
+  }
+
+  # Check heap-profiling flags
+  if ($main::opt_inuse_space +
+      $main::opt_inuse_objects +
+      $main::opt_alloc_space +
+      $main::opt_alloc_objects > 1) {
+    usage("Specify at most on of --inuse/--alloc options");
+  }
+
+  # Check output granularities
+  my $grains =
+      $main::opt_functions +
+      $main::opt_lines +
+      $main::opt_addresses +
+      $main::opt_files +
+      0;
+  if ($grains > 1) {
+    usage("Only specify one output granularity option");
+  }
+  if ($grains == 0) {
+    $main::opt_functions = 1;
+  }
+
+  # Check output modes
+  my $modes =
+      $main::opt_text +
+      $main::opt_callgrind +
+      ($main::opt_list eq '' ? 0 : 1) +
+      ($main::opt_disasm eq '' ? 0 : 1) +
+      ($main::opt_symbols == 0 ? 0 : 1) +
+      $main::opt_gv +
+      $main::opt_evince +
+      $main::opt_web +
+      $main::opt_dot +
+      $main::opt_ps +
+      $main::opt_pdf +
+      $main::opt_svg +
+      $main::opt_gif +
+      $main::opt_raw +
+      $main::opt_collapsed +
+      $main::opt_interactive +
+      0;
+  if ($modes > 1) {
+    usage("Only specify one output mode");
+  }
+  if ($modes == 0) {
+    if (-t STDOUT) {  # If STDOUT is a tty, activate interactive mode
+      $main::opt_interactive = 1;
+    } else {
+      $main::opt_text = 1;
+    }
+  }
+
+  if ($main::opt_test) {
+    RunUnitTests();
+    # Should not return
+    exit(1);
+  }
+
+  # Binary name and profile arguments list
+  $main::prog = "";
+  @main::pfile_args = ();
+
+  # Remote profiling without a binary (using $SYMBOL_PAGE instead)
+  if (@ARGV > 0) {
+    if (IsProfileURL($ARGV[0])) {
+      $main::use_symbol_page = 1;
+    } elsif (IsSymbolizedProfileFile($ARGV[0])) {
+      $main::use_symbolized_profile = 1;
+      $main::prog = $UNKNOWN_BINARY;  # will be set later from the profile file
+    }
+  }
+
+  if ($main::use_symbol_page || $main::use_symbolized_profile) {
+    # We don't need a binary!
+    my %disabled = ('--lines' => $main::opt_lines,
+                    '--disasm' => $main::opt_disasm);
+    for my $option (keys %disabled) {
+      usage("$option cannot be used without a binary") if $disabled{$option};
+    }
+    # Set $main::prog later...
+    scalar(@ARGV) || usage("Did not specify profile file");
+  } elsif ($main::opt_symbols) {
+    # --symbols needs a binary-name (to run nm on, etc) but not profiles
+    $main::prog = shift(@ARGV) || usage("Did not specify program");
+  } else {
+    $main::prog = shift(@ARGV) || usage("Did not specify program");
+    scalar(@ARGV) || usage("Did not specify profile file");
+  }
+
+  # Parse profile file/location arguments
+  foreach my $farg (@ARGV) {
+    if ($farg =~ m/(.*)\@([0-9]+)(|\/.*)$/ ) {
+      my $machine = $1;
+      my $num_machines = $2;
+      my $path = $3;
+      for (my $i = 0; $i < $num_machines; $i++) {
+        unshift(@main::pfile_args, "$i.$machine$path");
+      }
+    } else {
+      unshift(@main::pfile_args, $farg);
+    }
+  }
+
+  if ($main::use_symbol_page) {
+    unless (IsProfileURL($main::pfile_args[0])) {
+      error("The first profile should be a remote form to use $SYMBOL_PAGE\n");
+    }
+    CheckSymbolPage();
+    $main::prog = FetchProgramName();
+  } elsif (!$main::use_symbolized_profile) {  # may not need objtools!
+    ConfigureObjTools($main::prog)
+  }
+
+  # Break the opt_lib_prefix into the prefix_list array
+  @prefix_list = split (',', $main::opt_lib_prefix);
+
+  # Remove trailing / from the prefixes, in the list to prevent
+  # searching things like /my/path//lib/mylib.so
+  foreach (@prefix_list) {
+    s|/+$||;
+  }
+
+  # Flag to prevent us from trying over and over to use
+  #  elfutils if it's not installed (used only with
+  #  --debug-syms-by-id option).
+  $main::gave_up_on_elfutils = 0;
+}
+
+sub FilterAndPrint {
+  my ($profile, $symbols, $libs, $thread) = @_;
+
+  # Get total data in profile
+  my $total = TotalProfile($profile);
+
+  # Remove uniniteresting stack items
+  $profile = RemoveUninterestingFrames($symbols, $profile);
+
+  # Focus?
+  if ($main::opt_focus ne '') {
+    $profile = FocusProfile($symbols, $profile, $main::opt_focus);
+  }
+
+  # Ignore?
+  if ($main::opt_ignore ne '') {
+    $profile = IgnoreProfile($symbols, $profile, $main::opt_ignore);
+  }
+
+  my $calls = ExtractCalls($symbols, $profile);
+
+  # Reduce profiles to required output granularity, and also clean
+  # each stack trace so a given entry exists at most once.
+  my $reduced = ReduceProfile($symbols, $profile);
+
+  # Get derived profiles
+  my $flat = FlatProfile($reduced);
+  my $cumulative = CumulativeProfile($reduced);
+
+  # Print
+  if (!$main::opt_interactive) {
+    if ($main::opt_disasm) {
+      PrintDisassembly($libs, $flat, $cumulative, $main::opt_disasm);
+    } elsif ($main::opt_list) {
+      PrintListing($total, $libs, $flat, $cumulative, $main::opt_list, 0);
+    } elsif ($main::opt_text) {
+      # Make sure the output is empty when have nothing to report
+      # (only matters when --heapcheck is given but we must be
+      # compatible with old branches that did not pass --heapcheck always):
+      if ($total != 0) {
+        printf("Total%s: %s %s\n",
+               (defined($thread) ? " (t$thread)" : ""),
+               Unparse($total), Units());
+      }
+      PrintText($symbols, $flat, $cumulative, -1);
+    } elsif ($main::opt_raw) {
+      PrintSymbolizedProfile($symbols, $profile, $main::prog);
+    } elsif ($main::opt_collapsed) {
+      PrintCollapsedStacks($symbols, $profile);
+    } elsif ($main::opt_callgrind) {
+      PrintCallgrind($calls);
+    } else {
+      if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) {
+        if ($main::opt_gv) {
+          RunGV(TempName($main::next_tmpfile, "ps"), "");
+        } elsif ($main::opt_evince) {
+          RunEvince(TempName($main::next_tmpfile, "pdf"), "");
+        } elsif ($main::opt_web) {
+          my $tmp = TempName($main::next_tmpfile, "svg");
+          RunWeb($tmp);
+          # The command we run might hand the file name off
+          # to an already running browser instance and then exit.
+          # Normally, we'd remove $tmp on exit (right now),
+          # but fork a child to remove $tmp a little later, so that the
+          # browser has time to load it first.
+          delete $main::tempnames{$tmp};
+          if (fork() == 0) {
+            sleep 5;
+            unlink($tmp);
+            exit(0);
+          }
+        }
+      } else {
+        cleanup();
+        exit(1);
+      }
+    }
+  } else {
+    InteractiveMode($profile, $symbols, $libs, $total);
+  }
+}
+
+sub Main() {
+  Init();
+  $main::collected_profile = undef;
+  @main::profile_files = ();
+  $main::op_time = time();
+
+  # Printing symbols is special and requires a lot less info that most.
+  if ($main::opt_symbols) {
+    PrintSymbols(*STDIN);   # Get /proc/maps and symbols output from stdin
+    return;
+  }
+
+  # Fetch all profile data
+  FetchDynamicProfiles();
+
+  # this will hold symbols that we read from the profile files
+  my $symbol_map = {};
+
+  # Read one profile, pick the last item on the list
+  my $data = ReadProfile($main::prog, pop(@main::profile_files));
+  my $profile = $data->{profile};
+  my $pcs = $data->{pcs};
+  my $libs = $data->{libs};   # Info about main program and shared libraries
+  $symbol_map = MergeSymbols($symbol_map, $data->{symbols});
+
+  # Add additional profiles, if available.
+  if (scalar(@main::profile_files) > 0) {
+    foreach my $pname (@main::profile_files) {
+      my $data2 = ReadProfile($main::prog, $pname);
+      $profile = AddProfile($profile, $data2->{profile});
+      $pcs = AddPcs($pcs, $data2->{pcs});
+      $symbol_map = MergeSymbols($symbol_map, $data2->{symbols});
+    }
+  }
+
+  # Subtract base from profile, if specified
+  if ($main::opt_base ne '') {
+    my $base = ReadProfile($main::prog, $main::opt_base);
+    $profile = SubtractProfile($profile, $base->{profile});
+    $pcs = AddPcs($pcs, $base->{pcs});
+    $symbol_map = MergeSymbols($symbol_map, $base->{symbols});
+  }
+
+  # Collect symbols
+  my $symbols;
+  if ($main::use_symbolized_profile) {
+    $symbols = FetchSymbols($pcs, $symbol_map);
+  } elsif ($main::use_symbol_page) {
+    $symbols = FetchSymbols($pcs);
+  } else {
+    # TODO(csilvers): $libs uses the /proc/self/maps data from profile1,
+    # which may differ from the data from subsequent profiles, especially
+    # if they were run on different machines.  Use appropriate libs for
+    # each pc somehow.
+    $symbols = ExtractSymbols($libs, $pcs);
+  }
+
+  if (!defined($main::opt_thread)) {
+    FilterAndPrint($profile, $symbols, $libs);
+  }
+  if (defined($data->{threads})) {
+    foreach my $thread (sort { $a <=> $b } keys(%{$data->{threads}})) {
+      if (defined($main::opt_thread) &&
+          ($main::opt_thread eq '*' || $main::opt_thread == $thread)) {
+        my $thread_profile = $data->{threads}{$thread};
+        FilterAndPrint($thread_profile, $symbols, $libs, $thread);
+      }
+    }
+  }
+
+  cleanup();
+  exit(0);
+}
+
+##### Entry Point #####
+
+Main();
+
+# Temporary code to detect if we're running on a Goobuntu system.
+# These systems don't have the right stuff installed for the special
+# Readline libraries to work, so as a temporary workaround, we default
+# to using the normal stdio code, rather than the fancier readline-based
+# code
+sub ReadlineMightFail {
+  if (-e '/lib/libtermcap.so.2') {
+    return 0;  # libtermcap exists, so readline should be okay
+  } else {
+    return 1;
+  }
+}
+
+sub RunGV {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  if (!system(ShellEscape(@GV, "--version") . " >$dev_null 2>&1")) {
+    # Options using double dash are supported by this gv version.
+    # Also, turn on noantialias to better handle bug in gv for
+    # postscript files with large dimensions.
+    # TODO: Maybe we should not pass the --noantialias flag
+    # if the gv version is known to work properly without the flag.
+    system(ShellEscape(@GV, "--scale=$main::opt_scale", "--noantialias", $fname)
+           . $bg);
+  } else {
+    # Old gv version - only supports options that use single dash.
+    print STDERR ShellEscape(@GV, "-scale", $main::opt_scale) . "\n";
+    system(ShellEscape(@GV, "-scale", "$main::opt_scale", $fname) . $bg);
+  }
+}
+
+sub RunEvince {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  system(ShellEscape(@EVINCE, $fname) . $bg);
+}
+
+sub RunWeb {
+  my $fname = shift;
+  print STDERR "Loading web page file:///$fname\n";
+
+  if (`uname` =~ /Darwin/) {
+    # OS X: open will use standard preference for SVG files.
+    system("/usr/bin/open", $fname);
+    return;
+  }
+
+  # Some kind of Unix; try generic symlinks, then specific browsers.
+  # (Stop once we find one.)
+  # Works best if the browser is already running.
+  my @alt = (
+    "/etc/alternatives/gnome-www-browser",
+    "/etc/alternatives/x-www-browser",
+    "google-chrome",
+    "firefox",
+  );
+  foreach my $b (@alt) {
+    if (system($b, $fname) == 0) {
+      return;
+    }
+  }
+
+  print STDERR "Could not load web browser.\n";
+}
+
+sub RunKcachegrind {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  print STDERR "Starting '@KCACHEGRIND " . $fname . $bg . "'\n";
+  system(ShellEscape(@KCACHEGRIND, $fname) . $bg);
+}
+
+
+##### Interactive helper routines #####
+
+sub InteractiveMode {
+  $| = 1;  # Make output unbuffered for interactive mode
+  my ($orig_profile, $symbols, $libs, $total) = @_;
+
+  print STDERR "Welcome to jeprof!  For help, type 'help'.\n";
+
+  # Use ReadLine if it's installed and input comes from a console.
+  if ( -t STDIN &&
+       !ReadlineMightFail() &&
+       defined(eval {require Term::ReadLine}) ) {
+    my $term = new Term::ReadLine 'jeprof';
+    while ( defined ($_ = $term->readline('(jeprof) '))) {
+      $term->addhistory($_) if /\S/;
+      if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) {
+        last;    # exit when we get an interactive command to quit
+      }
+    }
+  } else {       # don't have readline
+    while (1) {
+      print STDERR "(jeprof) ";
+      $_ = <STDIN>;
+      last if ! defined $_ ;
+      s/\r//g;         # turn windows-looking lines into unix-looking lines
+
+      # Save some flags that might be reset by InteractiveCommand()
+      my $save_opt_lines = $main::opt_lines;
+
+      if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) {
+        last;    # exit when we get an interactive command to quit
+      }
+
+      # Restore flags
+      $main::opt_lines = $save_opt_lines;
+    }
+  }
+}
+
+# Takes two args: orig profile, and command to run.
+# Returns 1 if we should keep going, or 0 if we were asked to quit
+sub InteractiveCommand {
+  my($orig_profile, $symbols, $libs, $total, $command) = @_;
+  $_ = $command;                # just to make future m//'s easier
+  if (!defined($_)) {
+    print STDERR "\n";
+    return 0;
+  }
+  if (m/^\s*quit/) {
+    return 0;
+  }
+  if (m/^\s*help/) {
+    InteractiveHelpMessage();
+    return 1;
+  }
+  # Clear all the mode options -- mode is controlled by "$command"
+  $main::opt_text = 0;
+  $main::opt_callgrind = 0;
+  $main::opt_disasm = 0;
+  $main::opt_list = 0;
+  $main::opt_gv = 0;
+  $main::opt_evince = 0;
+  $main::opt_cum = 0;
+
+  if (m/^\s*(text|top)(\d*)\s*(.*)/) {
+    $main::opt_text = 1;
+
+    my $line_limit = ($2 ne "") ? int($2) : 10;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($3);
+
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintText($symbols, $flat, $cumulative, $line_limit);
+    return 1;
+  }
+  if (m/^\s*callgrind\s*([^ \n]*)/) {
+    $main::opt_callgrind = 1;
+
+    # Get derived profiles
+    my $calls = ExtractCalls($symbols, $orig_profile);
+    my $filename = $1;
+    if ( $1 eq '' ) {
+      $filename = TempName($main::next_tmpfile, "callgrind");
+    }
+    PrintCallgrind($calls, $filename);
+    if ( $1 eq '' ) {
+      RunKcachegrind($filename, " & ");
+      $main::next_tmpfile++;
+    }
+
+    return 1;
+  }
+  if (m/^\s*(web)?list\s*(.+)/) {
+    my $html = (defined($1) && ($1 eq "web"));
+    $main::opt_list = 1;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($2);
+
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintListing($total, $libs, $flat, $cumulative, $routine, $html);
+    return 1;
+  }
+  if (m/^\s*disasm\s*(.+)/) {
+    $main::opt_disasm = 1;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($1);
+
+    # Process current profile to account for various settings
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintDisassembly($libs, $flat, $cumulative, $routine);
+    return 1;
+  }
+  if (m/^\s*(gv|web|evince)\s*(.*)/) {
+    $main::opt_gv = 0;
+    $main::opt_evince = 0;
+    $main::opt_web = 0;
+    if ($1 eq "gv") {
+      $main::opt_gv = 1;
+    } elsif ($1 eq "evince") {
+      $main::opt_evince = 1;
+    } elsif ($1 eq "web") {
+      $main::opt_web = 1;
+    }
+
+    my $focus;
+    my $ignore;
+    ($focus, $ignore) = ParseInteractiveArgs($2);
+
+    # Process current profile to account for various settings
+    my $profile = ProcessProfile($total, $orig_profile, $symbols,
+                                 $focus, $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) {
+      if ($main::opt_gv) {
+        RunGV(TempName($main::next_tmpfile, "ps"), " &");
+      } elsif ($main::opt_evince) {
+        RunEvince(TempName($main::next_tmpfile, "pdf"), " &");
+      } elsif ($main::opt_web) {
+        RunWeb(TempName($main::next_tmpfile, "svg"));
+      }
+      $main::next_tmpfile++;
+    }
+    return 1;
+  }
+  if (m/^\s*$/) {
+    return 1;
+  }
+  print STDERR "Unknown command: try 'help'.\n";
+  return 1;
+}
+
+
+sub ProcessProfile {
+  my $total_count = shift;
+  my $orig_profile = shift;
+  my $symbols = shift;
+  my $focus = shift;
+  my $ignore = shift;
+
+  # Process current profile to account for various settings
+  my $profile = $orig_profile;
+  printf("Total: %s %s\n", Unparse($total_count), Units());
+  if ($focus ne '') {
+    $profile = FocusProfile($symbols, $profile, $focus);
+    my $focus_count = TotalProfile($profile);
+    printf("After focusing on '%s': %s %s of %s (%0.1f%%)\n",
+           $focus,
+           Unparse($focus_count), Units(),
+           Unparse($total_count), ($focus_count*100.0) / $total_count);
+  }
+  if ($ignore ne '') {
+    $profile = IgnoreProfile($symbols, $profile, $ignore);
+    my $ignore_count = TotalProfile($profile);
+    printf("After ignoring '%s': %s %s of %s (%0.1f%%)\n",
+           $ignore,
+           Unparse($ignore_count), Units(),
+           Unparse($total_count),
+           ($ignore_count*100.0) / $total_count);
+  }
+
+  return $profile;
+}
+
+sub InteractiveHelpMessage {
+  print STDERR <<ENDOFHELP;
+Interactive jeprof mode
+
+Commands:
+  gv
+  gv [focus] [-ignore1] [-ignore2]
+      Show graphical hierarchical display of current profile.  Without
+      any arguments, shows all samples in the profile.  With the optional
+      "focus" argument, restricts the samples shown to just those where
+      the "focus" regular expression matches a routine name on the stack
+      trace.
+
+  web
+  web [focus] [-ignore1] [-ignore2]
+      Like GV, but displays profile in your web browser instead of using
+      Ghostview. Works best if your web browser is already running.
+      To change the browser that gets used:
+      On Linux, set the /etc/alternatives/gnome-www-browser symlink.
+      On OS X, change the Finder association for SVG files.
+
+  list [routine_regexp] [-ignore1] [-ignore2]
+      Show source listing of routines whose names match "routine_regexp"
+
+  weblist [routine_regexp] [-ignore1] [-ignore2]
+     Displays a source listing of routines whose names match "routine_regexp"
+     in a web browser.  You can click on source lines to view the
+     corresponding disassembly.
+
+  top [--cum] [-ignore1] [-ignore2]
+  top20 [--cum] [-ignore1] [-ignore2]
+  top37 [--cum] [-ignore1] [-ignore2]
+      Show top lines ordered by flat profile count, or cumulative count
+      if --cum is specified.  If a number is present after 'top', the
+      top K routines will be shown (defaults to showing the top 10)
+
+  disasm [routine_regexp] [-ignore1] [-ignore2]
+      Show disassembly of routines whose names match "routine_regexp",
+      annotated with sample counts.
+
+  callgrind
+  callgrind [filename]
+      Generates callgrind file. If no filename is given, kcachegrind is called.
+
+  help - This listing
+  quit or ^D - End jeprof
+
+For commands that accept optional -ignore tags, samples where any routine in
+the stack trace matches the regular expression in any of the -ignore
+parameters will be ignored.
+
+Further pprof details are available at this location (or one similar):
+
+ /usr/doc/gperftools-$PPROF_VERSION/cpu_profiler.html
+ /usr/doc/gperftools-$PPROF_VERSION/heap_profiler.html
+
+ENDOFHELP
+}
+sub ParseInteractiveArgs {
+  my $args = shift;
+  my $focus = "";
+  my $ignore = "";
+  my @x = split(/ +/, $args);
+  foreach $a (@x) {
+    if ($a =~ m/^(--|-)lines$/) {
+      $main::opt_lines = 1;
+    } elsif ($a =~ m/^(--|-)cum$/) {
+      $main::opt_cum = 1;
+    } elsif ($a =~ m/^-(.*)/) {
+      $ignore .= (($ignore ne "") ? "|" : "" ) . $1;
+    } else {
+      $focus .= (($focus ne "") ? "|" : "" ) . $a;
+    }
+  }
+  if ($ignore ne "") {
+    print STDERR "Ignoring samples in call stacks that match '$ignore'\n";
+  }
+  return ($focus, $ignore);
+}
+
+##### Output code #####
+
+sub TempName {
+  my $fnum = shift;
+  my $ext = shift;
+  my $file = "$main::tmpfile_ps.$fnum.$ext";
+  $main::tempnames{$file} = 1;
+  return $file;
+}
+
+# Print profile data in packed binary format (64-bit) to standard out
+sub PrintProfileData {
+  my $profile = shift;
+
+  # print header (64-bit style)
+  # (zero) (header-size) (version) (sample-period) (zero)
+  print pack('L*', 0, 0, 3, 0, 0, 0, 1, 0, 0, 0);
+
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    if ($#addrs >= 0) {
+      my $depth = $#addrs + 1;
+      # int(foo / 2**32) is the only reliable way to get rid of bottom
+      # 32 bits on both 32- and 64-bit systems.
+      print pack('L*', $count & 0xFFFFFFFF, int($count / 2**32));
+      print pack('L*', $depth & 0xFFFFFFFF, int($depth / 2**32));
+
+      foreach my $full_addr (@addrs) {
+        my $addr = $full_addr;
+        $addr =~ s/0x0*//;  # strip off leading 0x, zeroes
+        if (length($addr) > 16) {
+          print STDERR "Invalid address in profile: $full_addr\n";
+          next;
+        }
+        my $low_addr = substr($addr, -8);       # get last 8 hex chars
+        my $high_addr = substr($addr, -16, 8);  # get up to 8 more hex chars
+        print pack('L*', hex('0x' . $low_addr), hex('0x' . $high_addr));
+      }
+    }
+  }
+}
+
+# Print symbols and profile data
+sub PrintSymbolizedProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $prog = shift;
+
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+
+  print '--- ', $symbol_marker, "\n";
+  if (defined($prog)) {
+    print 'binary=', $prog, "\n";
+  }
+  while (my ($pc, $name) = each(%{$symbols})) {
+    my $sep = ' ';
+    print '0x', $pc;
+    # We have a list of function names, which include the inlined
+    # calls.  They are separated (and terminated) by --, which is
+    # illegal in function names.
+    for (my $j = 2; $j <= $#{$name}; $j += 3) {
+      print $sep, $name->[$j];
+      $sep = '--';
+    }
+    print "\n";
+  }
+  print '---', "\n";
+
+  my $profile_marker;
+  if ($main::profile_type eq 'heap') {
+    $HEAP_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } elsif ($main::profile_type eq 'growth') {
+    $GROWTH_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } elsif ($main::profile_type eq 'contention') {
+    $CONTENTION_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } else { # elsif ($main::profile_type eq 'cpu')
+    $PROFILE_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  }
+
+  print '--- ', $profile_marker, "\n";
+  if (defined($main::collected_profile)) {
+    # if used with remote fetch, simply dump the collected profile to output.
+    open(SRC, "<$main::collected_profile");
+    while (<SRC>) {
+      print $_;
+    }
+    close(SRC);
+  } else {
+    # --raw/http: For everything to work correctly for non-remote profiles, we
+    # would need to extend PrintProfileData() to handle all possible profile
+    # types, re-enable the code that is currently disabled in ReadCPUProfile()
+    # and FixCallerAddresses(), and remove the remote profile dumping code in
+    # the block above.
+    die "--raw/http: jeprof can only dump remote profiles for --raw\n";
+    # dump a cpu-format profile to standard out
+    PrintProfileData($profile);
+  }
+}
+
+# Print text output
+sub PrintText {
+  my $symbols = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $line_limit = shift;
+
+  my $total = TotalProfile($flat);
+
+  # Which profile to sort by?
+  my $s = $main::opt_cum ? $cumulative : $flat;
+
+  my $running_sum = 0;
+  my $lines = 0;
+  foreach my $k (sort { GetEntry($s, $b) <=> GetEntry($s, $a) || $a cmp $b }
+                 keys(%{$cumulative})) {
+    my $f = GetEntry($flat, $k);
+    my $c = GetEntry($cumulative, $k);
+    $running_sum += $f;
+
+    my $sym = $k;
+    if (exists($symbols->{$k})) {
+      $sym = $symbols->{$k}->[0] . " " . $symbols->{$k}->[1];
+      if ($main::opt_addresses) {
+        $sym = $k . " " . $sym;
+      }
+    }
+
+    if ($f != 0 || $c != 0) {
+      printf("%8s %6s %6s %8s %6s %s\n",
+             Unparse($f),
+             Percent($f, $total),
+             Percent($running_sum, $total),
+             Unparse($c),
+             Percent($c, $total),
+             $sym);
+    }
+    $lines++;
+    last if ($line_limit >= 0 && $lines >= $line_limit);
+  }
+}
+
+# Callgrind format has a compression for repeated function and file
+# names.  You show the name the first time, and just use its number
+# subsequently.  This can cut down the file to about a third or a
+# quarter of its uncompressed size.  $key and $val are the key/value
+# pair that would normally be printed by callgrind; $map is a map from
+# value to number.
+sub CompressedCGName {
+  my($key, $val, $map) = @_;
+  my $idx = $map->{$val};
+  # For very short keys, providing an index hurts rather than helps.
+  if (length($val) <= 3) {
+    return "$key=$val\n";
+  } elsif (defined($idx)) {
+    return "$key=($idx)\n";
+  } else {
+    # scalar(keys $map) gives the number of items in the map.
+    $idx = scalar(keys(%{$map})) + 1;
+    $map->{$val} = $idx;
+    return "$key=($idx) $val\n";
+  }
+}
+
+# Print the call graph in a way that's suiteable for callgrind.
+sub PrintCallgrind {
+  my $calls = shift;
+  my $filename;
+  my %filename_to_index_map;
+  my %fnname_to_index_map;
+
+  if ($main::opt_interactive) {
+    $filename = shift;
+    print STDERR "Writing callgrind file to '$filename'.\n"
+  } else {
+    $filename = "&STDOUT";
+  }
+  open(CG, ">$filename");
+  printf CG ("events: Hits\n\n");
+  foreach my $call ( map { $_->[0] }
+                     sort { $a->[1] cmp $b ->[1] ||
+                            $a->[2] <=> $b->[2] }
+                     map { /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/;
+                           [$_, $1, $2] }
+                     keys %$calls ) {
+    my $count = int($calls->{$call});
+    $call =~ /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/;
+    my ( $caller_file, $caller_line, $caller_function,
+         $callee_file, $callee_line, $callee_function ) =
+       ( $1, $2, $3, $5, $6, $7 );
+
+    # TODO(csilvers): for better compression, collect all the
+    # caller/callee_files and functions first, before printing
+    # anything, and only compress those referenced more than once.
+    printf CG CompressedCGName("fl", $caller_file, \%filename_to_index_map);
+    printf CG CompressedCGName("fn", $caller_function, \%fnname_to_index_map);
+    if (defined $6) {
+      printf CG CompressedCGName("cfl", $callee_file, \%filename_to_index_map);
+      printf CG CompressedCGName("cfn", $callee_function, \%fnname_to_index_map);
+      printf CG ("calls=$count $callee_line\n");
+    }
+    printf CG ("$caller_line $count\n\n");
+  }
+}
+
+# Print disassembly for all all routines that match $main::opt_disasm
+sub PrintDisassembly {
+  my $libs = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $disasm_opts = shift;
+
+  my $total = TotalProfile($flat);
+
+  foreach my $lib (@{$libs}) {
+    my $symbol_table = GetProcedureBoundaries($lib->[0], $disasm_opts);
+    my $offset = AddressSub($lib->[1], $lib->[3]);
+    foreach my $routine (sort ByName keys(%{$symbol_table})) {
+      my $start_addr = $symbol_table->{$routine}->[0];
+      my $end_addr = $symbol_table->{$routine}->[1];
+      # See if there are any samples in this routine
+      my $length = hex(AddressSub($end_addr, $start_addr));
+      my $addr = AddressAdd($start_addr, $offset);
+      for (my $i = 0; $i < $length; $i++) {
+        if (defined($cumulative->{$addr})) {
+          PrintDisassembledFunction($lib->[0], $offset,
+                                    $routine, $flat, $cumulative,
+                                    $start_addr, $end_addr, $total);
+          last;
+        }
+        $addr = AddressInc($addr);
+      }
+    }
+  }
+}
+
+# Return reference to array of tuples of the form:
+#       [start_address, filename, linenumber, instruction, limit_address]
+# E.g.,
+#       ["0x806c43d", "/foo/bar.cc", 131, "ret", "0x806c440"]
+sub Disassemble {
+  my $prog = shift;
+  my $offset = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+
+  my $objdump = $obj_tool_map{"objdump"};
+  my $cmd = ShellEscape($objdump, "-C", "-d", "-l", "--no-show-raw-insn",
+                        "--start-address=0x$start_addr",
+                        "--stop-address=0x$end_addr", $prog);
+  open(OBJDUMP, "$cmd |") || error("$cmd: $!\n");
+  my @result = ();
+  my $filename = "";
+  my $linenumber = -1;
+  my $last = ["", "", "", ""];
+  while (<OBJDUMP>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    chop;
+    if (m|\s*([^:\s]+):(\d+)\s*$|) {
+      # Location line of the form:
+      #   <filename>:<linenumber>
+      $filename = $1;
+      $linenumber = $2;
+    } elsif (m/^ +([0-9a-f]+):\s*(.*)/) {
+      # Disassembly line -- zero-extend address to full length
+      my $addr = HexExtend($1);
+      my $k = AddressAdd($addr, $offset);
+      $last->[4] = $k;   # Store ending address for previous instruction
+      $last = [$k, $filename, $linenumber, $2, $end_addr];
+      push(@result, $last);
+    }
+  }
+  close(OBJDUMP);
+  return @result;
+}
+
+# The input file should contain lines of the form /proc/maps-like
+# output (same format as expected from the profiles) or that looks
+# like hex addresses (like "0xDEADBEEF").  We will parse all
+# /proc/maps output, and for all the hex addresses, we will output
+# "short" symbol names, one per line, in the same order as the input.
+sub PrintSymbols {
+  my $maps_and_symbols_file = shift;
+
+  # ParseLibraries expects pcs to be in a set.  Fine by us...
+  my @pclist = ();   # pcs in sorted order
+  my $pcs = {};
+  my $map = "";
+  foreach my $line (<$maps_and_symbols_file>) {
+    $line =~ s/\r//g;    # turn windows-looking lines into unix-looking lines
+    if ($line =~ /\b(0x[0-9a-f]+)\b/i) {
+      push(@pclist, HexExtend($1));
+      $pcs->{$pclist[-1]} = 1;
+    } else {
+      $map .= $line;
+    }
+  }
+
+  my $libs = ParseLibraries($main::prog, $map, $pcs);
+  my $symbols = ExtractSymbols($libs, $pcs);
+
+  foreach my $pc (@pclist) {
+    # ->[0] is the shortname, ->[2] is the full name
+    print(($symbols->{$pc}->[0] || "??") . "\n");
+  }
+}
+
+
+# For sorting functions by name
+sub ByName {
+  return ShortFunctionName($a) cmp ShortFunctionName($b);
+}
+
+# Print source-listing for all all routines that match $list_opts
+sub PrintListing {
+  my $total = shift;
+  my $libs = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $list_opts = shift;
+  my $html = shift;
+
+  my $output = \*STDOUT;
+  my $fname = "";
+
+  if ($html) {
+    # Arrange to write the output to a temporary file
+    $fname = TempName($main::next_tmpfile, "html");
+    $main::next_tmpfile++;
+    if (!open(TEMP, ">$fname")) {
+      print STDERR "$fname: $!\n";
+      return;
+    }
+    $output = \*TEMP;
+    print $output HtmlListingHeader();
+    printf $output ("<div class=\"legend\">%s<br>Total: %s %s</div>\n",
+                    $main::prog, Unparse($total), Units());
+  }
+
+  my $listed = 0;
+  foreach my $lib (@{$libs}) {
+    my $symbol_table = GetProcedureBoundaries($lib->[0], $list_opts);
+    my $offset = AddressSub($lib->[1], $lib->[3]);
+    foreach my $routine (sort ByName keys(%{$symbol_table})) {
+      # Print if there are any samples in this routine
+      my $start_addr = $symbol_table->{$routine}->[0];
+      my $end_addr = $symbol_table->{$routine}->[1];
+      my $length = hex(AddressSub($end_addr, $start_addr));
+      my $addr = AddressAdd($start_addr, $offset);
+      for (my $i = 0; $i < $length; $i++) {
+        if (defined($cumulative->{$addr})) {
+          $listed += PrintSource(
+            $lib->[0], $offset,
+            $routine, $flat, $cumulative,
+            $start_addr, $end_addr,
+            $html,
+            $output);
+          last;
+        }
+        $addr = AddressInc($addr);
+      }
+    }
+  }
+
+  if ($html) {
+    if ($listed > 0) {
+      print $output HtmlListingFooter();
+      close($output);
+      RunWeb($fname);
+    } else {
+      close($output);
+      unlink($fname);
+    }
+  }
+}
+
+sub HtmlListingHeader {
+  return <<'EOF';
+<DOCTYPE html>
+<html>
+<head>
+<title>Pprof listing</title>
+<style type="text/css">
+body {
+  font-family: sans-serif;
+}
+h1 {
+  font-size: 1.5em;
+  margin-bottom: 4px;
+}
+.legend {
+  font-size: 1.25em;
+}
+.line {
+  color: #aaaaaa;
+}
+.nop {
+  color: #aaaaaa;
+}
+.unimportant {
+  color: #cccccc;
+}
+.disasmloc {
+  color: #000000;
+}
+.deadsrc {
+  cursor: pointer;
+}
+.deadsrc:hover {
+  background-color: #eeeeee;
+}
+.livesrc {
+  color: #0000ff;
+  cursor: pointer;
+}
+.livesrc:hover {
+  background-color: #eeeeee;
+}
+.asm {
+  color: #008800;
+  display: none;
+}
+</style>
+<script type="text/javascript">
+function jeprof_toggle_asm(e) {
+  var target;
+  if (!e) e = window.event;
+  if (e.target) target = e.target;
+  else if (e.srcElement) target = e.srcElement;
+
+  if (target) {
+    var asm = target.nextSibling;
+    if (asm && asm.className == "asm") {
+      asm.style.display = (asm.style.display == "block" ? "" : "block");
+      e.preventDefault();
+      return false;
+    }
+  }
+}
+</script>
+</head>
+<body>
+EOF
+}
+
+sub HtmlListingFooter {
+  return <<'EOF';
+</body>
+</html>
+EOF
+}
+
+sub HtmlEscape {
+  my $text = shift;
+  $text =~ s/&/&amp;/g;
+  $text =~ s/</&lt;/g;
+  $text =~ s/>/&gt;/g;
+  return $text;
+}
+
+# Returns the indentation of the line, if it has any non-whitespace
+# characters.  Otherwise, returns -1.
+sub Indentation {
+  my $line = shift;
+  if (m/^(\s*)\S/) {
+    return length($1);
+  } else {
+    return -1;
+  }
+}
+
+# If the symbol table contains inlining info, Disassemble() may tag an
+# instruction with a location inside an inlined function.  But for
+# source listings, we prefer to use the location in the function we
+# are listing.  So use MapToSymbols() to fetch full location
+# information for each instruction and then pick out the first
+# location from a location list (location list contains callers before
+# callees in case of inlining).
+#
+# After this routine has run, each entry in $instructions contains:
+#   [0] start address
+#   [1] filename for function we are listing
+#   [2] line number for function we are listing
+#   [3] disassembly
+#   [4] limit address
+#   [5] most specific filename (may be different from [1] due to inlining)
+#   [6] most specific line number (may be different from [2] due to inlining)
+sub GetTopLevelLineNumbers {
+  my ($lib, $offset, $instructions) = @_;
+  my $pcs = [];
+  for (my $i = 0; $i <= $#{$instructions}; $i++) {
+    push(@{$pcs}, $instructions->[$i]->[0]);
+  }
+  my $symbols = {};
+  MapToSymbols($lib, $offset, $pcs, $symbols);
+  for (my $i = 0; $i <= $#{$instructions}; $i++) {
+    my $e = $instructions->[$i];
+    push(@{$e}, $e->[1]);
+    push(@{$e}, $e->[2]);
+    my $addr = $e->[0];
+    my $sym = $symbols->{$addr};
+    if (defined($sym)) {
+      if ($#{$sym} >= 2 && $sym->[1] =~ m/^(.*):(\d+)$/) {
+        $e->[1] = $1;  # File name
+        $e->[2] = $2;  # Line number
+      }
+    }
+  }
+}
+
+# Print source-listing for one routine
+sub PrintSource {
+  my $prog = shift;
+  my $offset = shift;
+  my $routine = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+  my $html = shift;
+  my $output = shift;
+
+  # Disassemble all instructions (just to get line numbers)
+  my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr);
+  GetTopLevelLineNumbers($prog, $offset, \@instructions);
+
+  # Hack 1: assume that the first source file encountered in the
+  # disassembly contains the routine
+  my $filename = undef;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    if ($instructions[$i]->[2] >= 0) {
+      $filename = $instructions[$i]->[1];
+      last;
+    }
+  }
+  if (!defined($filename)) {
+    print STDERR "no filename found in $routine\n";
+    return 0;
+  }
+
+  # Hack 2: assume that the largest line number from $filename is the
+  # end of the procedure.  This is typically safe since if P1 contains
+  # an inlined call to P2, then P2 usually occurs earlier in the
+  # source file.  If this does not work, we might have to compute a
+  # density profile or just print all regions we find.
+  my $lastline = 0;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    my $f = $instructions[$i]->[1];
+    my $l = $instructions[$i]->[2];
+    if (($f eq $filename) && ($l > $lastline)) {
+      $lastline = $l;
+    }
+  }
+
+  # Hack 3: assume the first source location from "filename" is the start of
+  # the source code.
+  my $firstline = 1;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    if ($instructions[$i]->[1] eq $filename) {
+      $firstline = $instructions[$i]->[2];
+      last;
+    }
+  }
+
+  # Hack 4: Extend last line forward until its indentation is less than
+  # the indentation we saw on $firstline
+  my $oldlastline = $lastline;
+  {
+    if (!open(FILE, "<$filename")) {
+      print STDERR "$filename: $!\n";
+      return 0;
+    }
+    my $l = 0;
+    my $first_indentation = -1;
+    while (<FILE>) {
+      s/\r//g;         # turn windows-looking lines into unix-looking lines
+      $l++;
+      my $indent = Indentation($_);
+      if ($l >= $firstline) {
+        if ($first_indentation < 0 && $indent >= 0) {
+          $first_indentation = $indent;
+          last if ($first_indentation == 0);
+        }
+      }
+      if ($l >= $lastline && $indent >= 0) {
+        if ($indent >= $first_indentation) {
+          $lastline = $l+1;
+        } else {
+          last;
+        }
+      }
+    }
+    close(FILE);
+  }
+
+  # Assign all samples to the range $firstline,$lastline,
+  # Hack 4: If an instruction does not occur in the range, its samples
+  # are moved to the next instruction that occurs in the range.
+  my $samples1 = {};        # Map from line number to flat count
+  my $samples2 = {};        # Map from line number to cumulative count
+  my $running1 = 0;         # Unassigned flat counts
+  my $running2 = 0;         # Unassigned cumulative counts
+  my $total1 = 0;           # Total flat counts
+  my $total2 = 0;           # Total cumulative counts
+  my %disasm = ();          # Map from line number to disassembly
+  my $running_disasm = "";  # Unassigned disassembly
+  my $skip_marker = "---\n";
+  if ($html) {
+    $skip_marker = "";
+    for (my $l = $firstline; $l <= $lastline; $l++) {
+      $disasm{$l} = "";
+    }
+  }
+  my $last_dis_filename = '';
+  my $last_dis_linenum = -1;
+  my $last_touched_line = -1;  # To detect gaps in disassembly for a line
+  foreach my $e (@instructions) {
+    # Add up counts for all address that fall inside this instruction
+    my $c1 = 0;
+    my $c2 = 0;
+    for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) {
+      $c1 += GetEntry($flat, $a);
+      $c2 += GetEntry($cumulative, $a);
+    }
+
+    if ($html) {
+      my $dis = sprintf("      %6s %6s \t\t%8s: %s ",
+                        HtmlPrintNumber($c1),
+                        HtmlPrintNumber($c2),
+                        UnparseAddress($offset, $e->[0]),
+                        CleanDisassembly($e->[3]));
+
+      # Append the most specific source line associated with this instruction
+      if (length($dis) < 80) { $dis .= (' ' x (80 - length($dis))) };
+      $dis = HtmlEscape($dis);
+      my $f = $e->[5];
+      my $l = $e->[6];
+      if ($f ne $last_dis_filename) {
+        $dis .= sprintf("<span class=disasmloc>%s:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      } elsif ($l ne $last_dis_linenum) {
+        # De-emphasize the unchanged file name portion
+        $dis .= sprintf("<span class=unimportant>%s</span>" .
+                        "<span class=disasmloc>:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      } else {
+        # De-emphasize the entire location
+        $dis .= sprintf("<span class=unimportant>%s:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      }
+      $last_dis_filename = $f;
+      $last_dis_linenum = $l;
+      $running_disasm .= $dis;
+      $running_disasm .= "\n";
+    }
+
+    $running1 += $c1;
+    $running2 += $c2;
+    $total1 += $c1;
+    $total2 += $c2;
+    my $file = $e->[1];
+    my $line = $e->[2];
+    if (($file eq $filename) &&
+        ($line >= $firstline) &&
+        ($line <= $lastline)) {
+      # Assign all accumulated samples to this line
+      AddEntry($samples1, $line, $running1);
+      AddEntry($samples2, $line, $running2);
+      $running1 = 0;
+      $running2 = 0;
+      if ($html) {
+        if ($line != $last_touched_line && $disasm{$line} ne '') {
+          $disasm{$line} .= "\n";
+        }
+        $disasm{$line} .= $running_disasm;
+        $running_disasm = '';
+        $last_touched_line = $line;
+      }
+    }
+  }
+
+  # Assign any leftover samples to $lastline
+  AddEntry($samples1, $lastline, $running1);
+  AddEntry($samples2, $lastline, $running2);
+  if ($html) {
+    if ($lastline != $last_touched_line && $disasm{$lastline} ne '') {
+      $disasm{$lastline} .= "\n";
+    }
+    $disasm{$lastline} .= $running_disasm;
+  }
+
+  if ($html) {
+    printf $output (
+      "<h1>%s</h1>%s\n<pre onClick=\"jeprof_toggle_asm()\">\n" .
+      "Total:%6s %6s (flat / cumulative %s)\n",
+      HtmlEscape(ShortFunctionName($routine)),
+      HtmlEscape(CleanFileName($filename)),
+      Unparse($total1),
+      Unparse($total2),
+      Units());
+  } else {
+    printf $output (
+      "ROUTINE ====================== %s in %s\n" .
+      "%6s %6s Total %s (flat / cumulative)\n",
+      ShortFunctionName($routine),
+      CleanFileName($filename),
+      Unparse($total1),
+      Unparse($total2),
+      Units());
+  }
+  if (!open(FILE, "<$filename")) {
+    print STDERR "$filename: $!\n";
+    return 0;
+  }
+  my $l = 0;
+  while (<FILE>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    $l++;
+    if ($l >= $firstline - 5 &&
+        (($l <= $oldlastline + 5) || ($l <= $lastline))) {
+      chop;
+      my $text = $_;
+      if ($l == $firstline) { print $output $skip_marker; }
+      my $n1 = GetEntry($samples1, $l);
+      my $n2 = GetEntry($samples2, $l);
+      if ($html) {
+        # Emit a span that has one of the following classes:
+        #    livesrc -- has samples
+        #    deadsrc -- has disassembly, but with no samples
+        #    nop     -- has no matching disasembly
+        # Also emit an optional span containing disassembly.
+        my $dis = $disasm{$l};
+        my $asm = "";
+        if (defined($dis) && $dis ne '') {
+          $asm = "<span class=\"asm\">" . $dis . "</span>";
+        }
+        my $source_class = (($n1 + $n2 > 0)
+                            ? "livesrc"
+                            : (($asm ne "") ? "deadsrc" : "nop"));
+        printf $output (
+          "<span class=\"line\">%5d</span> " .
+          "<span class=\"%s\">%6s %6s %s</span>%s\n",
+          $l, $source_class,
+          HtmlPrintNumber($n1),
+          HtmlPrintNumber($n2),
+          HtmlEscape($text),
+          $asm);
+      } else {
+        printf $output(
+          "%6s %6s %4d: %s\n",
+          UnparseAlt($n1),
+          UnparseAlt($n2),
+          $l,
+          $text);
+      }
+      if ($l == $lastline)  { print $output $skip_marker; }
+    };
+  }
+  close(FILE);
+  if ($html) {
+    print $output "</pre>\n";
+  }
+  return 1;
+}
+
+# Return the source line for the specified file/linenumber.
+# Returns undef if not found.
+sub SourceLine {
+  my $file = shift;
+  my $line = shift;
+
+  # Look in cache
+  if (!defined($main::source_cache{$file})) {
+    if (100 < scalar keys(%main::source_cache)) {
+      # Clear the cache when it gets too big
+      $main::source_cache = ();
+    }
+
+    # Read all lines from the file
+    if (!open(FILE, "<$file")) {
+      print STDERR "$file: $!\n";
+      $main::source_cache{$file} = [];  # Cache the negative result
+      return undef;
+    }
+    my $lines = [];
+    push(@{$lines}, "");        # So we can use 1-based line numbers as indices
+    while (<FILE>) {
+      push(@{$lines}, $_);
+    }
+    close(FILE);
+
+    # Save the lines in the cache
+    $main::source_cache{$file} = $lines;
+  }
+
+  my $lines = $main::source_cache{$file};
+  if (($line < 0) || ($line > $#{$lines})) {
+    return undef;
+  } else {
+    return $lines->[$line];
+  }
+}
+
+# Print disassembly for one routine with interspersed source if available
+sub PrintDisassembledFunction {
+  my $prog = shift;
+  my $offset = shift;
+  my $routine = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+  my $total = shift;
+
+  # Disassemble all instructions
+  my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr);
+
+  # Make array of counts per instruction
+  my @flat_count = ();
+  my @cum_count = ();
+  my $flat_total = 0;
+  my $cum_total = 0;
+  foreach my $e (@instructions) {
+    # Add up counts for all address that fall inside this instruction
+    my $c1 = 0;
+    my $c2 = 0;
+    for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) {
+      $c1 += GetEntry($flat, $a);
+      $c2 += GetEntry($cumulative, $a);
+    }
+    push(@flat_count, $c1);
+    push(@cum_count, $c2);
+    $flat_total += $c1;
+    $cum_total += $c2;
+  }
+
+  # Print header with total counts
+  printf("ROUTINE ====================== %s\n" .
+         "%6s %6s %s (flat, cumulative) %.1f%% of total\n",
+         ShortFunctionName($routine),
+         Unparse($flat_total),
+         Unparse($cum_total),
+         Units(),
+         ($cum_total * 100.0) / $total);
+
+  # Process instructions in order
+  my $current_file = "";
+  for (my $i = 0; $i <= $#instructions; ) {
+    my $e = $instructions[$i];
+
+    # Print the new file name whenever we switch files
+    if ($e->[1] ne $current_file) {
+      $current_file = $e->[1];
+      my $fname = $current_file;
+      $fname =~ s|^\./||;   # Trim leading "./"
+
+      # Shorten long file names
+      if (length($fname) >= 58) {
+        $fname = "..." . substr($fname, -55);
+      }
+      printf("-------------------- %s\n", $fname);
+    }
+
+    # TODO: Compute range of lines to print together to deal with
+    # small reorderings.
+    my $first_line = $e->[2];
+    my $last_line = $first_line;
+    my %flat_sum = ();
+    my %cum_sum = ();
+    for (my $l = $first_line; $l <= $last_line; $l++) {
+      $flat_sum{$l} = 0;
+      $cum_sum{$l} = 0;
+    }
+
+    # Find run of instructions for this range of source lines
+    my $first_inst = $i;
+    while (($i <= $#instructions) &&
+           ($instructions[$i]->[2] >= $first_line) &&
+           ($instructions[$i]->[2] <= $last_line)) {
+      $e = $instructions[$i];
+      $flat_sum{$e->[2]} += $flat_count[$i];
+      $cum_sum{$e->[2]} += $cum_count[$i];
+      $i++;
+    }
+    my $last_inst = $i - 1;
+
+    # Print source lines
+    for (my $l = $first_line; $l <= $last_line; $l++) {
+      my $line = SourceLine($current_file, $l);
+      if (!defined($line)) {
+        $line = "?\n";
+        next;
+      } else {
+        $line =~ s/^\s+//;
+      }
+      printf("%6s %6s %5d: %s",
+             UnparseAlt($flat_sum{$l}),
+             UnparseAlt($cum_sum{$l}),
+             $l,
+             $line);
+    }
+
+    # Print disassembly
+    for (my $x = $first_inst; $x <= $last_inst; $x++) {
+      my $e = $instructions[$x];
+      printf("%6s %6s    %8s: %6s\n",
+             UnparseAlt($flat_count[$x]),
+             UnparseAlt($cum_count[$x]),
+             UnparseAddress($offset, $e->[0]),
+             CleanDisassembly($e->[3]));
+    }
+  }
+}
+
+# Print DOT graph
+sub PrintDot {
+  my $prog = shift;
+  my $symbols = shift;
+  my $raw = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $overall_total = shift;
+
+  # Get total
+  my $local_total = TotalProfile($flat);
+  my $nodelimit = int($main::opt_nodefraction * $local_total);
+  my $edgelimit = int($main::opt_edgefraction * $local_total);
+  my $nodecount = $main::opt_nodecount;
+
+  # Find nodes to include
+  my @list = (sort { abs(GetEntry($cumulative, $b)) <=>
+                     abs(GetEntry($cumulative, $a))
+                     || $a cmp $b }
+              keys(%{$cumulative}));
+  my $last = $nodecount - 1;
+  if ($last > $#list) {
+    $last = $#list;
+  }
+  while (($last >= 0) &&
+         (abs(GetEntry($cumulative, $list[$last])) <= $nodelimit)) {
+    $last--;
+  }
+  if ($last < 0) {
+    print STDERR "No nodes to print\n";
+    return 0;
+  }
+
+  if ($nodelimit > 0 || $edgelimit > 0) {
+    printf STDERR ("Dropping nodes with <= %s %s; edges with <= %s abs(%s)\n",
+                   Unparse($nodelimit), Units(),
+                   Unparse($edgelimit), Units());
+  }
+
+  # Open DOT output file
+  my $output;
+  my $escaped_dot = ShellEscape(@DOT);
+  my $escaped_ps2pdf = ShellEscape(@PS2PDF);
+  if ($main::opt_gv) {
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "ps"));
+    $output = "| $escaped_dot -Tps2 >$escaped_outfile";
+  } elsif ($main::opt_evince) {
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "pdf"));
+    $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - $escaped_outfile";
+  } elsif ($main::opt_ps) {
+    $output = "| $escaped_dot -Tps2";
+  } elsif ($main::opt_pdf) {
+    $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - -";
+  } elsif ($main::opt_web || $main::opt_svg) {
+    # We need to post-process the SVG, so write to a temporary file always.
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "svg"));
+    $output = "| $escaped_dot -Tsvg >$escaped_outfile";
+  } elsif ($main::opt_gif) {
+    $output = "| $escaped_dot -Tgif";
+  } else {
+    $output = ">&STDOUT";
+  }
+  open(DOT, $output) || error("$output: $!\n");
+
+  # Title
+  printf DOT ("digraph \"%s; %s %s\" {\n",
+              $prog,
+              Unparse($overall_total),
+              Units());
+  if ($main::opt_pdf) {
+    # The output is more printable if we set the page size for dot.
+    printf DOT ("size=\"8,11\"\n");
+  }
+  printf DOT ("node [width=0.375,height=0.25];\n");
+
+  # Print legend
+  printf DOT ("Legend [shape=box,fontsize=24,shape=plaintext," .
+              "label=\"%s\\l%s\\l%s\\l%s\\l%s\\l\"];\n",
+              $prog,
+              sprintf("Total %s: %s", Units(), Unparse($overall_total)),
+              sprintf("Focusing on: %s", Unparse($local_total)),
+              sprintf("Dropped nodes with <= %s abs(%s)",
+                      Unparse($nodelimit), Units()),
+              sprintf("Dropped edges with <= %s %s",
+                      Unparse($edgelimit), Units())
+              );
+
+  # Print nodes
+  my %node = ();
+  my $nextnode = 1;
+  foreach my $a (@list[0..$last]) {
+    # Pick font size
+    my $f = GetEntry($flat, $a);
+    my $c = GetEntry($cumulative, $a);
+
+    my $fs = 8;
+    if ($local_total > 0) {
+      $fs = 8 + (50.0 * sqrt(abs($f * 1.0 / $local_total)));
+    }
+
+    $node{$a} = $nextnode++;
+    my $sym = $a;
+    $sym =~ s/\s+/\\n/g;
+    $sym =~ s/::/\\n/g;
+
+    # Extra cumulative info to print for non-leaves
+    my $extra = "";
+    if ($f != $c) {
+      $extra = sprintf("\\rof %s (%s)",
+                       Unparse($c),
+                       Percent($c, $local_total));
+    }
+    my $style = "";
+    if ($main::opt_heapcheck) {
+      if ($f > 0) {
+        # make leak-causing nodes more visible (add a background)
+        $style = ",style=filled,fillcolor=gray"
+      } elsif ($f < 0) {
+        # make anti-leak-causing nodes (which almost never occur)
+        # stand out as well (triple border)
+        $style = ",peripheries=3"
+      }
+    }
+
+    printf DOT ("N%d [label=\"%s\\n%s (%s)%s\\r" .
+                "\",shape=box,fontsize=%.1f%s];\n",
+                $node{$a},
+                $sym,
+                Unparse($f),
+                Percent($f, $local_total),
+                $extra,
+                $fs,
+                $style,
+               );
+  }
+
+  # Get edges and counts per edge
+  my %edge = ();
+  my $n;
+  my $fullname_to_shortname_map = {};
+  FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map);
+  foreach my $k (keys(%{$raw})) {
+    # TODO: omit low %age edges
+    $n = $raw->{$k};
+    my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k);
+    for (my $i = 1; $i <= $#translated; $i++) {
+      my $src = $translated[$i];
+      my $dst = $translated[$i-1];
+      #next if ($src eq $dst);  # Avoid self-edges?
+      if (exists($node{$src}) && exists($node{$dst})) {
+        my $edge_label = "$src\001$dst";
+        if (!exists($edge{$edge_label})) {
+          $edge{$edge_label} = 0;
+        }
+        $edge{$edge_label} += $n;
+      }
+    }
+  }
+
+  # Print edges (process in order of decreasing counts)
+  my %indegree = ();   # Number of incoming edges added per node so far
+  my %outdegree = ();  # Number of outgoing edges added per node so far
+  foreach my $e (sort { $edge{$b} <=> $edge{$a} } keys(%edge)) {
+    my @x = split(/\001/, $e);
+    $n = $edge{$e};
+
+    # Initialize degree of kept incoming and outgoing edges if necessary
+    my $src = $x[0];
+    my $dst = $x[1];
+    if (!exists($outdegree{$src})) { $outdegree{$src} = 0; }
+    if (!exists($indegree{$dst})) { $indegree{$dst} = 0; }
+
+    my $keep;
+    if ($indegree{$dst} == 0) {
+      # Keep edge if needed for reachability
+      $keep = 1;
+    } elsif (abs($n) <= $edgelimit) {
+      # Drop if we are below --edgefraction
+      $keep = 0;
+    } elsif ($outdegree{$src} >= $main::opt_maxdegree ||
+             $indegree{$dst} >= $main::opt_maxdegree) {
+      # Keep limited number of in/out edges per node
+      $keep = 0;
+    } else {
+      $keep = 1;
+    }
+
+    if ($keep) {
+      $outdegree{$src}++;
+      $indegree{$dst}++;
+
+      # Compute line width based on edge count
+      my $fraction = abs($local_total ? (3 * ($n / $local_total)) : 0);
+      if ($fraction > 1) { $fraction = 1; }
+      my $w = $fraction * 2;
+      if ($w < 1 && ($main::opt_web || $main::opt_svg)) {
+        # SVG output treats line widths < 1 poorly.
+        $w = 1;
+      }
+
+      # Dot sometimes segfaults if given edge weights that are too large, so
+      # we cap the weights at a large value
+      my $edgeweight = abs($n) ** 0.7;
+      if ($edgeweight > 100000) { $edgeweight = 100000; }
+      $edgeweight = int($edgeweight);
+
+      my $style = sprintf("setlinewidth(%f)", $w);
+      if ($x[1] =~ m/\(inline\)/) {
+        $style .= ",dashed";
+      }
+
+      # Use a slightly squashed function of the edge count as the weight
+      printf DOT ("N%s -> N%s [label=%s, weight=%d, style=\"%s\"];\n",
+                  $node{$x[0]},
+                  $node{$x[1]},
+                  Unparse($n),
+                  $edgeweight,
+                  $style);
+    }
+  }
+
+  print DOT ("}\n");
+  close(DOT);
+
+  if ($main::opt_web || $main::opt_svg) {
+    # Rewrite SVG to be more usable inside web browser.
+    RewriteSvg(TempName($main::next_tmpfile, "svg"));
+  }
+
+  return 1;
+}
+
+sub RewriteSvg {
+  my $svgfile = shift;
+
+  open(SVG, $svgfile) || die "open temp svg: $!";
+  my @svg = <SVG>;
+  close(SVG);
+  unlink $svgfile;
+  my $svg = join('', @svg);
+
+  # Dot's SVG output is
+  #
+  #    <svg width="___" height="___"
+  #     viewBox="___" xmlns=...>
+  #    <g id="graph0" transform="...">
+  #    ...
+  #    </g>
+  #    </svg>
+  #
+  # Change it to
+  #
+  #    <svg width="100%" height="100%"
+  #     xmlns=...>
+  #    $svg_javascript
+  #    <g id="viewport" transform="translate(0,0)">
+  #    <g id="graph0" transform="...">
+  #    ...
+  #    </g>
+  #    </g>
+  #    </svg>
+
+  # Fix width, height; drop viewBox.
+  $svg =~ s/(?s)<svg width="[^"]+" height="[^"]+"(.*?)viewBox="[^"]+"/<svg width="100%" height="100%"$1/;
+
+  # Insert script, viewport <g> above first <g>
+  my $svg_javascript = SvgJavascript();
+  my $viewport = "<g id=\"viewport\" transform=\"translate(0,0)\">\n";
+  $svg =~ s/<g id="graph\d"/$svg_javascript$viewport$&/;
+
+  # Insert final </g> above </svg>.
+  $svg =~ s/(.*)(<\/svg>)/$1<\/g>$2/;
+  $svg =~ s/<g id="graph\d"(.*?)/<g id="viewport"$1/;
+
+  if ($main::opt_svg) {
+    # --svg: write to standard output.
+    print $svg;
+  } else {
+    # Write back to temporary file.
+    open(SVG, ">$svgfile") || die "open $svgfile: $!";
+    print SVG $svg;
+    close(SVG);
+  }
+}
+
+sub SvgJavascript {
+  return <<'EOF';
+<script type="text/ecmascript"><![CDATA[
+// SVGPan
+// http://www.cyberz.org/blog/2009/12/08/svgpan-a-javascript-svg-panzoomdrag-library/
+// Local modification: if(true || ...) below to force panning, never moving.
+
+/**
+ *  SVGPan library 1.2
+ * ====================
+ *
+ * Given an unique existing element with id "viewport", including the
+ * the library into any SVG adds the following capabilities:
+ *
+ *  - Mouse panning
+ *  - Mouse zooming (using the wheel)
+ *  - Object dargging
+ *
+ * Known issues:
+ *
+ *  - Zooming (while panning) on Safari has still some issues
+ *
+ * Releases:
+ *
+ * 1.2, Sat Mar 20 08:42:50 GMT 2010, Zeng Xiaohui
+ *	Fixed a bug with browser mouse handler interaction
+ *
+ * 1.1, Wed Feb  3 17:39:33 GMT 2010, Zeng Xiaohui
+ *	Updated the zoom code to support the mouse wheel on Safari/Chrome
+ *
+ * 1.0, Andrea Leofreddi
+ *	First release
+ *
+ * This code is licensed under the following BSD license:
+ *
+ * Copyright 2009-2010 Andrea Leofreddi <a.leofreddi@itcharm.com>. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are
+ * permitted provided that the following conditions are met:
+ *
+ *    1. Redistributions of source code must retain the above copyright notice, this list of
+ *       conditions and the following disclaimer.
+ *
+ *    2. Redistributions in binary form must reproduce the above copyright notice, this list
+ *       of conditions and the following disclaimer in the documentation and/or other materials
+ *       provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Andrea Leofreddi ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+ * FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL Andrea Leofreddi OR
+ * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * The views and conclusions contained in the software and documentation are those of the
+ * authors and should not be interpreted as representing official policies, either expressed
+ * or implied, of Andrea Leofreddi.
+ */
+
+var root = document.documentElement;
+
+var state = 'none', stateTarget, stateOrigin, stateTf;
+
+setupHandlers(root);
+
+/**
+ * Register handlers
+ */
+function setupHandlers(root){
+	setAttributes(root, {
+		"onmouseup" : "add(evt)",
+		"onmousedown" : "handleMouseDown(evt)",
+		"onmousemove" : "handleMouseMove(evt)",
+		"onmouseup" : "handleMouseUp(evt)",
+		//"onmouseout" : "handleMouseUp(evt)", // Decomment this to stop the pan functionality when dragging out of the SVG element
+	});
+
+	if(navigator.userAgent.toLowerCase().indexOf('webkit') >= 0)
+		window.addEventListener('mousewheel', handleMouseWheel, false); // Chrome/Safari
+	else
+		window.addEventListener('DOMMouseScroll', handleMouseWheel, false); // Others
+
+	var g = svgDoc.getElementById("svg");
+	g.width = "100%";
+	g.height = "100%";
+}
+
+/**
+ * Instance an SVGPoint object with given event coordinates.
+ */
+function getEventPoint(evt) {
+	var p = root.createSVGPoint();
+
+	p.x = evt.clientX;
+	p.y = evt.clientY;
+
+	return p;
+}
+
+/**
+ * Sets the current transform matrix of an element.
+ */
+function setCTM(element, matrix) {
+	var s = "matrix(" + matrix.a + "," + matrix.b + "," + matrix.c + "," + matrix.d + "," + matrix.e + "," + matrix.f + ")";
+
+	element.setAttribute("transform", s);
+}
+
+/**
+ * Dumps a matrix to a string (useful for debug).
+ */
+function dumpMatrix(matrix) {
+	var s = "[ " + matrix.a + ", " + matrix.c + ", " + matrix.e + "\n  " + matrix.b + ", " + matrix.d + ", " + matrix.f + "\n  0, 0, 1 ]";
+
+	return s;
+}
+
+/**
+ * Sets attributes of an element.
+ */
+function setAttributes(element, attributes){
+	for (i in attributes)
+		element.setAttributeNS(null, i, attributes[i]);
+}
+
+/**
+ * Handle mouse move event.
+ */
+function handleMouseWheel(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var delta;
+
+	if(evt.wheelDelta)
+		delta = evt.wheelDelta / 3600; // Chrome/Safari
+	else
+		delta = evt.detail / -90; // Mozilla
+
+	var z = 1 + delta; // Zoom factor: 0.9/1.1
+
+	var g = svgDoc.getElementById("viewport");
+
+	var p = getEventPoint(evt);
+
+	p = p.matrixTransform(g.getCTM().inverse());
+
+	// Compute new scale matrix in current mouse position
+	var k = root.createSVGMatrix().translate(p.x, p.y).scale(z).translate(-p.x, -p.y);
+
+        setCTM(g, g.getCTM().multiply(k));
+
+	stateTf = stateTf.multiply(k.inverse());
+}
+
+/**
+ * Handle mouse move event.
+ */
+function handleMouseMove(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var g = svgDoc.getElementById("viewport");
+
+	if(state == 'pan') {
+		// Pan mode
+		var p = getEventPoint(evt).matrixTransform(stateTf);
+
+		setCTM(g, stateTf.inverse().translate(p.x - stateOrigin.x, p.y - stateOrigin.y));
+	} else if(state == 'move') {
+		// Move mode
+		var p = getEventPoint(evt).matrixTransform(g.getCTM().inverse());
+
+		setCTM(stateTarget, root.createSVGMatrix().translate(p.x - stateOrigin.x, p.y - stateOrigin.y).multiply(g.getCTM().inverse()).multiply(stateTarget.getCTM()));
+
+		stateOrigin = p;
+	}
+}
+
+/**
+ * Handle click event.
+ */
+function handleMouseDown(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var g = svgDoc.getElementById("viewport");
+
+	if(true || evt.target.tagName == "svg") {
+		// Pan mode
+		state = 'pan';
+
+		stateTf = g.getCTM().inverse();
+
+		stateOrigin = getEventPoint(evt).matrixTransform(stateTf);
+	} else {
+		// Move mode
+		state = 'move';
+
+		stateTarget = evt.target;
+
+		stateTf = g.getCTM().inverse();
+
+		stateOrigin = getEventPoint(evt).matrixTransform(stateTf);
+	}
+}
+
+/**
+ * Handle mouse button release event.
+ */
+function handleMouseUp(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	if(state == 'pan' || state == 'move') {
+		// Quit pan mode
+		state = '';
+	}
+}
+
+]]></script>
+EOF
+}
+
+# Provides a map from fullname to shortname for cases where the
+# shortname is ambiguous.  The symlist has both the fullname and
+# shortname for all symbols, which is usually fine, but sometimes --
+# such as overloaded functions -- two different fullnames can map to
+# the same shortname.  In that case, we use the address of the
+# function to disambiguate the two.  This function fills in a map that
+# maps fullnames to modified shortnames in such cases.  If a fullname
+# is not present in the map, the 'normal' shortname provided by the
+# symlist is the appropriate one to use.
+sub FillFullnameToShortnameMap {
+  my $symbols = shift;
+  my $fullname_to_shortname_map = shift;
+  my $shortnames_seen_once = {};
+  my $shortnames_seen_more_than_once = {};
+
+  foreach my $symlist (values(%{$symbols})) {
+    # TODO(csilvers): deal with inlined symbols too.
+    my $shortname = $symlist->[0];
+    my $fullname = $symlist->[2];
+    if ($fullname !~ /<[0-9a-fA-F]+>$/) {  # fullname doesn't end in an address
+      next;       # the only collisions we care about are when addresses differ
+    }
+    if (defined($shortnames_seen_once->{$shortname}) &&
+        $shortnames_seen_once->{$shortname} ne $fullname) {
+      $shortnames_seen_more_than_once->{$shortname} = 1;
+    } else {
+      $shortnames_seen_once->{$shortname} = $fullname;
+    }
+  }
+
+  foreach my $symlist (values(%{$symbols})) {
+    my $shortname = $symlist->[0];
+    my $fullname = $symlist->[2];
+    # TODO(csilvers): take in a list of addresses we care about, and only
+    # store in the map if $symlist->[1] is in that list.  Saves space.
+    next if defined($fullname_to_shortname_map->{$fullname});
+    if (defined($shortnames_seen_more_than_once->{$shortname})) {
+      if ($fullname =~ /<0*([^>]*)>$/) {   # fullname has address at end of it
+        $fullname_to_shortname_map->{$fullname} = "$shortname\@$1";
+      }
+    }
+  }
+}
+
+# Return a small number that identifies the argument.
+# Multiple calls with the same argument will return the same number.
+# Calls with different arguments will return different numbers.
+sub ShortIdFor {
+  my $key = shift;
+  my $id = $main::uniqueid{$key};
+  if (!defined($id)) {
+    $id = keys(%main::uniqueid) + 1;
+    $main::uniqueid{$key} = $id;
+  }
+  return $id;
+}
+
+# Translate a stack of addresses into a stack of symbols
+sub TranslateStack {
+  my $symbols = shift;
+  my $fullname_to_shortname_map = shift;
+  my $k = shift;
+
+  my @addrs = split(/\n/, $k);
+  my @result = ();
+  for (my $i = 0; $i <= $#addrs; $i++) {
+    my $a = $addrs[$i];
+
+    # Skip large addresses since they sometimes show up as fake entries on RH9
+    if (length($a) > 8 && $a gt "7fffffffffffffff") {
+      next;
+    }
+
+    if ($main::opt_disasm || $main::opt_list) {
+      # We want just the address for the key
+      push(@result, $a);
+      next;
+    }
+
+    my $symlist = $symbols->{$a};
+    if (!defined($symlist)) {
+      $symlist = [$a, "", $a];
+    }
+
+    # We can have a sequence of symbols for a particular entry
+    # (more than one symbol in the case of inlining).  Callers
+    # come before callees in symlist, so walk backwards since
+    # the translated stack should contain callees before callers.
+    for (my $j = $#{$symlist}; $j >= 2; $j -= 3) {
+      my $func = $symlist->[$j-2];
+      my $fileline = $symlist->[$j-1];
+      my $fullfunc = $symlist->[$j];
+      if (defined($fullname_to_shortname_map->{$fullfunc})) {
+        $func = $fullname_to_shortname_map->{$fullfunc};
+      }
+      if ($j > 2) {
+        $func = "$func (inline)";
+      }
+
+      # Do not merge nodes corresponding to Callback::Run since that
+      # causes confusing cycles in dot display.  Instead, we synthesize
+      # a unique name for this frame per caller.
+      if ($func =~ m/Callback.*::Run$/) {
+        my $caller = ($i > 0) ? $addrs[$i-1] : 0;
+        $func = "Run#" . ShortIdFor($caller);
+      }
+
+      if ($main::opt_addresses) {
+        push(@result, "$a $func $fileline");
+      } elsif ($main::opt_lines) {
+        if ($func eq '??' && $fileline eq '??:0') {
+          push(@result, "$a");
+        } else {
+          push(@result, "$func $fileline");
+        }
+      } elsif ($main::opt_functions) {
+        if ($func eq '??') {
+          push(@result, "$a");
+        } else {
+          push(@result, $func);
+        }
+      } elsif ($main::opt_files) {
+        if ($fileline eq '??:0' || $fileline eq '') {
+          push(@result, "$a");
+        } else {
+          my $f = $fileline;
+          $f =~ s/:\d+$//;
+          push(@result, $f);
+        }
+      } else {
+        push(@result, $a);
+        last;  # Do not print inlined info
+      }
+    }
+  }
+
+  # print join(",", @addrs), " => ", join(",", @result), "\n";
+  return @result;
+}
+
+# Generate percent string for a number and a total
+sub Percent {
+  my $num = shift;
+  my $tot = shift;
+  if ($tot != 0) {
+    return sprintf("%.1f%%", $num * 100.0 / $tot);
+  } else {
+    return ($num == 0) ? "nan" : (($num > 0) ? "+inf" : "-inf");
+  }
+}
+
+# Generate pretty-printed form of number
+sub Unparse {
+  my $num = shift;
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    if ($main::opt_inuse_objects || $main::opt_alloc_objects) {
+      return sprintf("%d", $num);
+    } else {
+      if ($main::opt_show_bytes) {
+        return sprintf("%d", $num);
+      } else {
+        return sprintf("%.1f", $num / 1048576.0);
+      }
+    }
+  } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) {
+    return sprintf("%.3f", $num / 1e9); # Convert nanoseconds to seconds
+  } else {
+    return sprintf("%d", $num);
+  }
+}
+
+# Alternate pretty-printed form: 0 maps to "."
+sub UnparseAlt {
+  my $num = shift;
+  if ($num == 0) {
+    return ".";
+  } else {
+    return Unparse($num);
+  }
+}
+
+# Alternate pretty-printed form: 0 maps to ""
+sub HtmlPrintNumber {
+  my $num = shift;
+  if ($num == 0) {
+    return "";
+  } else {
+    return Unparse($num);
+  }
+}
+
+# Return output units
+sub Units {
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    if ($main::opt_inuse_objects || $main::opt_alloc_objects) {
+      return "objects";
+    } else {
+      if ($main::opt_show_bytes) {
+        return "B";
+      } else {
+        return "MB";
+      }
+    }
+  } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) {
+    return "seconds";
+  } else {
+    return "samples";
+  }
+}
+
+##### Profile manipulation code #####
+
+# Generate flattened profile:
+# If count is charged to stack [a,b,c,d], in generated profile,
+# it will be charged to [a]
+sub FlatProfile {
+  my $profile = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    if ($#addrs >= 0) {
+      AddEntry($result, $addrs[0], $count);
+    }
+  }
+  return $result;
+}
+
+# Generate cumulative profile:
+# If count is charged to stack [a,b,c,d], in generated profile,
+# it will be charged to [a], [b], [c], [d]
+sub CumulativeProfile {
+  my $profile = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    foreach my $a (@addrs) {
+      AddEntry($result, $a, $count);
+    }
+  }
+  return $result;
+}
+
+# If the second-youngest PC on the stack is always the same, returns
+# that pc.  Otherwise, returns undef.
+sub IsSecondPcAlwaysTheSame {
+  my $profile = shift;
+
+  my $second_pc = undef;
+  foreach my $k (keys(%{$profile})) {
+    my @addrs = split(/\n/, $k);
+    if ($#addrs < 1) {
+      return undef;
+    }
+    if (not defined $second_pc) {
+      $second_pc = $addrs[1];
+    } else {
+      if ($second_pc ne $addrs[1]) {
+        return undef;
+      }
+    }
+  }
+  return $second_pc;
+}
+
+sub ExtractSymbolNameInlineStack {
+  my $symbols = shift;
+  my $address = shift;
+
+  my @stack = ();
+
+  if (exists $symbols->{$address}) {
+    my @localinlinestack = @{$symbols->{$address}};
+    for (my $i = $#localinlinestack; $i > 0; $i-=3) {
+      my $file = $localinlinestack[$i-1];
+      my $fn = $localinlinestack[$i-0];
+
+      if ($file eq "?" || $file eq ":0") {
+        $file = "??:0";
+      }
+      if ($fn eq '??') {
+        # If we can't get the symbol name, at least use the file information.
+        $fn = $file;
+      }
+      my $suffix = "[inline]";
+      if ($i == 2) {
+        $suffix = "";
+      }
+      push (@stack, $fn.$suffix);
+    }
+  }
+  else {
+    # If we can't get a symbol name, at least fill in the address.
+    push (@stack, $address);
+  }
+
+  return @stack;
+}
+
+sub ExtractSymbolLocation {
+  my $symbols = shift;
+  my $address = shift;
+  # 'addr2line' outputs "??:0" for unknown locations; we do the
+  # same to be consistent.
+  my $location = "??:0:unknown";
+  if (exists $symbols->{$address}) {
+    my $file = $symbols->{$address}->[1];
+    if ($file eq "?") {
+      $file = "??:0"
+    }
+    $location = $file . ":" . $symbols->{$address}->[0];
+  }
+  return $location;
+}
+
+# Extracts a graph of calls.
+sub ExtractCalls {
+  my $symbols = shift;
+  my $profile = shift;
+
+  my $calls = {};
+  while( my ($stack_trace, $count) = each %$profile ) {
+    my @address = split(/\n/, $stack_trace);
+    my $destination = ExtractSymbolLocation($symbols, $address[0]);
+    AddEntry($calls, $destination, $count);
+    for (my $i = 1; $i <= $#address; $i++) {
+      my $source = ExtractSymbolLocation($symbols, $address[$i]);
+      my $call = "$source -> $destination";
+      AddEntry($calls, $call, $count);
+      $destination = $source;
+    }
+  }
+
+  return $calls;
+}
+
+sub FilterFrames {
+  my $symbols = shift;
+  my $profile = shift;
+
+  if ($main::opt_retain eq '' && $main::opt_exclude eq '') {
+    return $profile;
+  }
+
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my @path = ();
+    foreach my $a (@addrs) {
+      my $sym;
+      if (exists($symbols->{$a})) {
+        $sym = $symbols->{$a}->[0];
+      } else {
+        $sym = $a;
+      }
+      if ($main::opt_retain ne '' && $sym !~ m/$main::opt_retain/) {
+        next;
+      }
+      if ($main::opt_exclude ne '' && $sym =~ m/$main::opt_exclude/) {
+        next;
+      }
+      push(@path, $a);
+    }
+    if (scalar(@path) > 0) {
+      my $reduced_path = join("\n", @path);
+      AddEntry($result, $reduced_path, $count);
+    }
+  }
+
+  return $result;
+}
+
+sub PrintCollapsedStacks {
+  my $symbols = shift;
+  my $profile = shift;
+
+  while (my ($stack_trace, $count) = each %$profile) {
+    my @address = split(/\n/, $stack_trace);
+    my @names = reverse ( map { ExtractSymbolNameInlineStack($symbols, $_) } @address );
+    printf("%s %d\n", join(";", @names), $count);
+  }
+}
+
+sub RemoveUninterestingFrames {
+  my $symbols = shift;
+  my $profile = shift;
+
+  # List of function names to skip
+  my %skip = ();
+  my $skip_regexp = 'NOMATCH';
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    foreach my $name ('je_calloc',
+                      'cfree',
+                      'je_malloc',
+                      'newImpl',
+                      'void* newImpl',
+                      'je_free',
+                      'je_memalign',
+                      'je_posix_memalign',
+                      'je_aligned_alloc',
+                      'pvalloc',
+                      'je_valloc',
+                      'je_realloc',
+                      'je_mallocx',
+                      'je_rallocx',
+                      'je_xallocx',
+                      'je_dallocx',
+                      'je_sdallocx',
+                      'je_sdallocx_noflags',
+                      'tc_calloc',
+                      'tc_cfree',
+                      'tc_malloc',
+                      'tc_free',
+                      'tc_memalign',
+                      'tc_posix_memalign',
+                      'tc_pvalloc',
+                      'tc_valloc',
+                      'tc_realloc',
+                      'tc_new',
+                      'tc_delete',
+                      'tc_newarray',
+                      'tc_deletearray',
+                      'tc_new_nothrow',
+                      'tc_newarray_nothrow',
+                      'do_malloc',
+                      '::do_malloc',   # new name -- got moved to an unnamed ns
+                      '::do_malloc_or_cpp_alloc',
+                      'DoSampledAllocation',
+                      'simple_alloc::allocate',
+                      '__malloc_alloc_template::allocate',
+                      '__builtin_delete',
+                      '__builtin_new',
+                      '__builtin_vec_delete',
+                      '__builtin_vec_new',
+                      'operator new',
+                      'operator new[]',
+                      # The entry to our memory-allocation routines on OS X
+                      'malloc_zone_malloc',
+                      'malloc_zone_calloc',
+                      'malloc_zone_valloc',
+                      'malloc_zone_realloc',
+                      'malloc_zone_memalign',
+                      'malloc_zone_free',
+                      # These mark the beginning/end of our custom sections
+                      '__start_google_malloc',
+                      '__stop_google_malloc',
+                      '__start_malloc_hook',
+                      '__stop_malloc_hook') {
+      $skip{$name} = 1;
+      $skip{"_" . $name} = 1;   # Mach (OS X) adds a _ prefix to everything
+    }
+    # TODO: Remove TCMalloc once everything has been
+    # moved into the tcmalloc:: namespace and we have flushed
+    # old code out of the system.
+    $skip_regexp = "TCMalloc|^tcmalloc::";
+  } elsif ($main::profile_type eq 'contention') {
+    foreach my $vname ('base::RecordLockProfileData',
+                       'base::SubmitMutexProfileData',
+                       'base::SubmitSpinLockProfileData',
+                       'Mutex::Unlock',
+                       'Mutex::UnlockSlow',
+                       'Mutex::ReaderUnlock',
+                       'MutexLock::~MutexLock',
+                       'SpinLock::Unlock',
+                       'SpinLock::SlowUnlock',
+                       'SpinLockHolder::~SpinLockHolder') {
+      $skip{$vname} = 1;
+    }
+  } elsif ($main::profile_type eq 'cpu') {
+    # Drop signal handlers used for CPU profile collection
+    # TODO(dpeng): this should not be necessary; it's taken
+    # care of by the general 2nd-pc mechanism below.
+    foreach my $name ('ProfileData::Add',           # historical
+                      'ProfileData::prof_handler',  # historical
+                      'CpuProfiler::prof_handler',
+                      '__FRAME_END__',
+                      '__pthread_sighandler',
+                      '__restore') {
+      $skip{$name} = 1;
+    }
+  } else {
+    # Nothing skipped for unknown types
+  }
+
+  if ($main::profile_type eq 'cpu') {
+    # If all the second-youngest program counters are the same,
+    # this STRONGLY suggests that it is an artifact of measurement,
+    # i.e., stack frames pushed by the CPU profiler signal handler.
+    # Hence, we delete them.
+    # (The topmost PC is read from the signal structure, not from
+    # the stack, so it does not get involved.)
+    while (my $second_pc = IsSecondPcAlwaysTheSame($profile)) {
+      my $result = {};
+      my $func = '';
+      if (exists($symbols->{$second_pc})) {
+        $second_pc = $symbols->{$second_pc}->[0];
+      }
+      print STDERR "Removing $second_pc from all stack traces.\n";
+      foreach my $k (keys(%{$profile})) {
+        my $count = $profile->{$k};
+        my @addrs = split(/\n/, $k);
+        splice @addrs, 1, 1;
+        my $reduced_path = join("\n", @addrs);
+        AddEntry($result, $reduced_path, $count);
+      }
+      $profile = $result;
+    }
+  }
+
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my @path = ();
+    foreach my $a (@addrs) {
+      if (exists($symbols->{$a})) {
+        my $func = $symbols->{$a}->[0];
+        if ($skip{$func} || ($func =~ m/$skip_regexp/)) {
+          # Throw away the portion of the backtrace seen so far, under the
+          # assumption that previous frames were for functions internal to the
+          # allocator.
+          @path = ();
+          next;
+        }
+      }
+      push(@path, $a);
+    }
+    my $reduced_path = join("\n", @path);
+    AddEntry($result, $reduced_path, $count);
+  }
+
+  $result = FilterFrames($symbols, $result);
+
+  return $result;
+}
+
+# Reduce profile to granularity given by user
+sub ReduceProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $result = {};
+  my $fullname_to_shortname_map = {};
+  FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map);
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k);
+    my @path = ();
+    my %seen = ();
+    $seen{''} = 1;      # So that empty keys are skipped
+    foreach my $e (@translated) {
+      # To avoid double-counting due to recursion, skip a stack-trace
+      # entry if it has already been seen
+      if (!$seen{$e}) {
+        $seen{$e} = 1;
+        push(@path, $e);
+      }
+    }
+    my $reduced_path = join("\n", @path);
+    AddEntry($result, $reduced_path, $count);
+  }
+  return $result;
+}
+
+# Does the specified symbol array match the regexp?
+sub SymbolMatches {
+  my $sym = shift;
+  my $re = shift;
+  if (defined($sym)) {
+    for (my $i = 0; $i < $#{$sym}; $i += 3) {
+      if ($sym->[$i] =~ m/$re/ || $sym->[$i+1] =~ m/$re/) {
+        return 1;
+      }
+    }
+  }
+  return 0;
+}
+
+# Focus only on paths involving specified regexps
+sub FocusProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $focus = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    foreach my $a (@addrs) {
+      # Reply if it matches either the address/shortname/fileline
+      if (($a =~ m/$focus/) || SymbolMatches($symbols->{$a}, $focus)) {
+        AddEntry($result, $k, $count);
+        last;
+      }
+    }
+  }
+  return $result;
+}
+
+# Focus only on paths not involving specified regexps
+sub IgnoreProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $ignore = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my $matched = 0;
+    foreach my $a (@addrs) {
+      # Reply if it matches either the address/shortname/fileline
+      if (($a =~ m/$ignore/) || SymbolMatches($symbols->{$a}, $ignore)) {
+        $matched = 1;
+        last;
+      }
+    }
+    if (!$matched) {
+      AddEntry($result, $k, $count);
+    }
+  }
+  return $result;
+}
+
+# Get total count in profile
+sub TotalProfile {
+  my $profile = shift;
+  my $result = 0;
+  foreach my $k (keys(%{$profile})) {
+    $result += $profile->{$k};
+  }
+  return $result;
+}
+
+# Add A to B
+sub AddProfile {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  # add all keys in A
+  foreach my $k (keys(%{$A})) {
+    my $v = $A->{$k};
+    AddEntry($R, $k, $v);
+  }
+  # add all keys in B
+  foreach my $k (keys(%{$B})) {
+    my $v = $B->{$k};
+    AddEntry($R, $k, $v);
+  }
+  return $R;
+}
+
+# Merges symbol maps
+sub MergeSymbols {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  foreach my $k (keys(%{$A})) {
+    $R->{$k} = $A->{$k};
+  }
+  if (defined($B)) {
+    foreach my $k (keys(%{$B})) {
+      $R->{$k} = $B->{$k};
+    }
+  }
+  return $R;
+}
+
+
+# Add A to B
+sub AddPcs {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  # add all keys in A
+  foreach my $k (keys(%{$A})) {
+    $R->{$k} = 1
+  }
+  # add all keys in B
+  foreach my $k (keys(%{$B})) {
+    $R->{$k} = 1
+  }
+  return $R;
+}
+
+# Subtract B from A
+sub SubtractProfile {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  foreach my $k (keys(%{$A})) {
+    my $v = $A->{$k} - GetEntry($B, $k);
+    if ($v < 0 && $main::opt_drop_negative) {
+      $v = 0;
+    }
+    AddEntry($R, $k, $v);
+  }
+  if (!$main::opt_drop_negative) {
+    # Take care of when subtracted profile has more entries
+    foreach my $k (keys(%{$B})) {
+      if (!exists($A->{$k})) {
+        AddEntry($R, $k, 0 - $B->{$k});
+      }
+    }
+  }
+  return $R;
+}
+
+# Get entry from profile; zero if not present
+sub GetEntry {
+  my $profile = shift;
+  my $k = shift;
+  if (exists($profile->{$k})) {
+    return $profile->{$k};
+  } else {
+    return 0;
+  }
+}
+
+# Add entry to specified profile
+sub AddEntry {
+  my $profile = shift;
+  my $k = shift;
+  my $n = shift;
+  if (!exists($profile->{$k})) {
+    $profile->{$k} = 0;
+  }
+  $profile->{$k} += $n;
+}
+
+# Add a stack of entries to specified profile, and add them to the $pcs
+# list.
+sub AddEntries {
+  my $profile = shift;
+  my $pcs = shift;
+  my $stack = shift;
+  my $count = shift;
+  my @k = ();
+
+  foreach my $e (split(/\s+/, $stack)) {
+    my $pc = HexExtend($e);
+    $pcs->{$pc} = 1;
+    push @k, $pc;
+  }
+  AddEntry($profile, (join "\n", @k), $count);
+}
+
+##### Code to profile a server dynamically #####
+
+sub CheckSymbolPage {
+  my $url = SymbolPageURL();
+  my $command = ShellEscape(@URL_FETCHER, $url);
+  open(SYMBOL, "$command |") or error($command);
+  my $line = <SYMBOL>;
+  $line =~ s/\r//g;         # turn windows-looking lines into unix-looking lines
+  close(SYMBOL);
+  unless (defined($line)) {
+    error("$url doesn't exist\n");
+  }
+
+  if ($line =~ /^num_symbols:\s+(\d+)$/) {
+    if ($1 == 0) {
+      error("Stripped binary. No symbols available.\n");
+    }
+  } else {
+    error("Failed to get the number of symbols from $url\n");
+  }
+}
+
+sub IsProfileURL {
+  my $profile_name = shift;
+  if (-f $profile_name) {
+    printf STDERR "Using local file $profile_name.\n";
+    return 0;
+  }
+  return 1;
+}
+
+sub ParseProfileURL {
+  my $profile_name = shift;
+
+  if (!defined($profile_name) || $profile_name eq "") {
+    return ();
+  }
+
+  # Split profile URL - matches all non-empty strings, so no test.
+  $profile_name =~ m,^(https?://)?([^/]+)(.*?)(/|$PROFILES)?$,;
+
+  my $proto = $1 || "http://";
+  my $hostport = $2;
+  my $prefix = $3;
+  my $profile = $4 || "/";
+
+  my $host = $hostport;
+  $host =~ s/:.*//;
+
+  my $baseurl = "$proto$hostport$prefix";
+  return ($host, $baseurl, $profile);
+}
+
+# We fetch symbols from the first profile argument.
+sub SymbolPageURL {
+  my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]);
+  return "$baseURL$SYMBOL_PAGE";
+}
+
+sub FetchProgramName() {
+  my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]);
+  my $url = "$baseURL$PROGRAM_NAME_PAGE";
+  my $command_line = ShellEscape(@URL_FETCHER, $url);
+  open(CMDLINE, "$command_line |") or error($command_line);
+  my $cmdline = <CMDLINE>;
+  $cmdline =~ s/\r//g;   # turn windows-looking lines into unix-looking lines
+  close(CMDLINE);
+  error("Failed to get program name from $url\n") unless defined($cmdline);
+  $cmdline =~ s/\x00.+//;  # Remove argv[1] and latters.
+  $cmdline =~ s!\n!!g;  # Remove LFs.
+  return $cmdline;
+}
+
+# Gee, curl's -L (--location) option isn't reliable at least
+# with its 7.12.3 version.  Curl will forget to post data if
+# there is a redirection.  This function is a workaround for
+# curl.  Redirection happens on borg hosts.
+sub ResolveRedirectionForCurl {
+  my $url = shift;
+  my $command_line = ShellEscape(@URL_FETCHER, "--head", $url);
+  open(CMDLINE, "$command_line |") or error($command_line);
+  while (<CMDLINE>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    if (/^Location: (.*)/) {
+      $url = $1;
+    }
+  }
+  close(CMDLINE);
+  return $url;
+}
+
+# Add a timeout flat to URL_FETCHER.  Returns a new list.
+sub AddFetchTimeout {
+  my $timeout = shift;
+  my @fetcher = @_;
+  if (defined($timeout)) {
+    if (join(" ", @fetcher) =~ m/\bcurl -s/) {
+      push(@fetcher, "--max-time", sprintf("%d", $timeout));
+    } elsif (join(" ", @fetcher) =~ m/\brpcget\b/) {
+      push(@fetcher, sprintf("--deadline=%d", $timeout));
+    }
+  }
+  return @fetcher;
+}
+
+# Reads a symbol map from the file handle name given as $1, returning
+# the resulting symbol map.  Also processes variables relating to symbols.
+# Currently, the only variable processed is 'binary=<value>' which updates
+# $main::prog to have the correct program name.
+sub ReadSymbols {
+  my $in = shift;
+  my $map = {};
+  while (<$in>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    # Removes all the leading zeroes from the symbols, see comment below.
+    if (m/^0x0*([0-9a-f]+)\s+(.+)/) {
+      $map->{$1} = $2;
+    } elsif (m/^---/) {
+      last;
+    } elsif (m/^([a-z][^=]*)=(.*)$/ ) {
+      my ($variable, $value) = ($1, $2);
+      for ($variable, $value) {
+        s/^\s+//;
+        s/\s+$//;
+      }
+      if ($variable eq "binary") {
+        if ($main::prog ne $UNKNOWN_BINARY && $main::prog ne $value) {
+          printf STDERR ("Warning: Mismatched binary name '%s', using '%s'.\n",
+                         $main::prog, $value);
+        }
+        $main::prog = $value;
+      } else {
+        printf STDERR ("Ignoring unknown variable in symbols list: " .
+            "'%s' = '%s'\n", $variable, $value);
+      }
+    }
+  }
+  return $map;
+}
+
+sub URLEncode {
+  my $str = shift;
+  $str =~ s/([^A-Za-z0-9\-_.!~*'()])/ sprintf "%%%02x", ord $1 /eg;
+  return $str;
+}
+
+sub AppendSymbolFilterParams {
+  my $url = shift;
+  my @params = ();
+  if ($main::opt_retain ne '') {
+    push(@params, sprintf("retain=%s", URLEncode($main::opt_retain)));
+  }
+  if ($main::opt_exclude ne '') {
+    push(@params, sprintf("exclude=%s", URLEncode($main::opt_exclude)));
+  }
+  if (scalar @params > 0) {
+    $url = sprintf("%s?%s", $url, join("&", @params));
+  }
+  return $url;
+}
+
+# Fetches and processes symbols to prepare them for use in the profile output
+# code.  If the optional 'symbol_map' arg is not given, fetches symbols from
+# $SYMBOL_PAGE for all PC values found in profile.  Otherwise, the raw symbols
+# are assumed to have already been fetched into 'symbol_map' and are simply
+# extracted and processed.
+sub FetchSymbols {
+  my $pcset = shift;
+  my $symbol_map = shift;
+
+  my %seen = ();
+  my @pcs = grep { !$seen{$_}++ } keys(%$pcset);  # uniq
+
+  if (!defined($symbol_map)) {
+    my $post_data = join("+", sort((map {"0x" . "$_"} @pcs)));
+
+    open(POSTFILE, ">$main::tmpfile_sym");
+    print POSTFILE $post_data;
+    close(POSTFILE);
+
+    my $url = SymbolPageURL();
+
+    my $command_line;
+    if (join(" ", @URL_FETCHER) =~ m/\bcurl -s/) {
+      $url = ResolveRedirectionForCurl($url);
+      $url = AppendSymbolFilterParams($url);
+      $command_line = ShellEscape(@URL_FETCHER, "-d", "\@$main::tmpfile_sym",
+                                  $url);
+    } else {
+      $url = AppendSymbolFilterParams($url);
+      $command_line = (ShellEscape(@URL_FETCHER, "--post", $url)
+                       . " < " . ShellEscape($main::tmpfile_sym));
+    }
+    # We use c++filt in case $SYMBOL_PAGE gives us mangled symbols.
+    my $escaped_cppfilt = ShellEscape($obj_tool_map{"c++filt"});
+    open(SYMBOL, "$command_line | $escaped_cppfilt |") or error($command_line);
+    $symbol_map = ReadSymbols(*SYMBOL{IO});
+    close(SYMBOL);
+  }
+
+  my $symbols = {};
+  foreach my $pc (@pcs) {
+    my $fullname;
+    # For 64 bits binaries, symbols are extracted with 8 leading zeroes.
+    # Then /symbol reads the long symbols in as uint64, and outputs
+    # the result with a "0x%08llx" format which get rid of the zeroes.
+    # By removing all the leading zeroes in both $pc and the symbols from
+    # /symbol, the symbols match and are retrievable from the map.
+    my $shortpc = $pc;
+    $shortpc =~ s/^0*//;
+    # Each line may have a list of names, which includes the function
+    # and also other functions it has inlined.  They are separated (in
+    # PrintSymbolizedProfile), by --, which is illegal in function names.
+    my $fullnames;
+    if (defined($symbol_map->{$shortpc})) {
+      $fullnames = $symbol_map->{$shortpc};
+    } else {
+      $fullnames = "0x" . $pc;  # Just use addresses
+    }
+    my $sym = [];
+    $symbols->{$pc} = $sym;
+    foreach my $fullname (split("--", $fullnames)) {
+      my $name = ShortFunctionName($fullname);
+      push(@{$sym}, $name, "?", $fullname);
+    }
+  }
+  return $symbols;
+}
+
+sub BaseName {
+  my $file_name = shift;
+  $file_name =~ s!^.*/!!;  # Remove directory name
+  return $file_name;
+}
+
+sub MakeProfileBaseName {
+  my ($binary_name, $profile_name) = @_;
+  my ($host, $baseURL, $path) = ParseProfileURL($profile_name);
+  my $binary_shortname = BaseName($binary_name);
+  return sprintf("%s.%s.%s",
+                 $binary_shortname, $main::op_time, $host);
+}
+
+sub FetchDynamicProfile {
+  my $binary_name = shift;
+  my $profile_name = shift;
+  my $fetch_name_only = shift;
+  my $encourage_patience = shift;
+
+  if (!IsProfileURL($profile_name)) {
+    return $profile_name;
+  } else {
+    my ($host, $baseURL, $path) = ParseProfileURL($profile_name);
+    if ($path eq "" || $path eq "/") {
+      # Missing type specifier defaults to cpu-profile
+      $path = $PROFILE_PAGE;
+    }
+
+    my $profile_file = MakeProfileBaseName($binary_name, $profile_name);
+
+    my $url = "$baseURL$path";
+    my $fetch_timeout = undef;
+    if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE/) {
+      if ($path =~ m/[?]/) {
+        $url .= "&";
+      } else {
+        $url .= "?";
+      }
+      $url .= sprintf("seconds=%d", $main::opt_seconds);
+      $fetch_timeout = $main::opt_seconds * 1.01 + 60;
+      # Set $profile_type for consumption by PrintSymbolizedProfile.
+      $main::profile_type = 'cpu';
+    } else {
+      # For non-CPU profiles, we add a type-extension to
+      # the target profile file name.
+      my $suffix = $path;
+      $suffix =~ s,/,.,g;
+      $profile_file .= $suffix;
+      # Set $profile_type for consumption by PrintSymbolizedProfile.
+      if ($path =~ m/$HEAP_PAGE/) {
+        $main::profile_type = 'heap';
+      } elsif ($path =~ m/$GROWTH_PAGE/) {
+        $main::profile_type = 'growth';
+      } elsif ($path =~ m/$CONTENTION_PAGE/) {
+        $main::profile_type = 'contention';
+      }
+    }
+
+    my $profile_dir = $ENV{"JEPROF_TMPDIR"} || ($ENV{HOME} . "/jeprof");
+    if (! -d $profile_dir) {
+      mkdir($profile_dir)
+          || die("Unable to create profile directory $profile_dir: $!\n");
+    }
+    my $tmp_profile = "$profile_dir/.tmp.$profile_file";
+    my $real_profile = "$profile_dir/$profile_file";
+
+    if ($fetch_name_only > 0) {
+      return $real_profile;
+    }
+
+    my @fetcher = AddFetchTimeout($fetch_timeout, @URL_FETCHER);
+    my $cmd = ShellEscape(@fetcher, $url) . " > " . ShellEscape($tmp_profile);
+    if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE|$CENSUSPROFILE_PAGE/){
+      print STDERR "Gathering CPU profile from $url for $main::opt_seconds seconds to\n  ${real_profile}\n";
+      if ($encourage_patience) {
+        print STDERR "Be patient...\n";
+      }
+    } else {
+      print STDERR "Fetching $path profile from $url to\n  ${real_profile}\n";
+    }
+
+    (system($cmd) == 0) || error("Failed to get profile: $cmd: $!\n");
+    (system("mv", $tmp_profile, $real_profile) == 0) || error("Unable to rename profile\n");
+    print STDERR "Wrote profile to $real_profile\n";
+    $main::collected_profile = $real_profile;
+    return $main::collected_profile;
+  }
+}
+
+# Collect profiles in parallel
+sub FetchDynamicProfiles {
+  my $items = scalar(@main::pfile_args);
+  my $levels = log($items) / log(2);
+
+  if ($items == 1) {
+    $main::profile_files[0] = FetchDynamicProfile($main::prog, $main::pfile_args[0], 0, 1);
+  } else {
+    # math rounding issues
+    if ((2 ** $levels) < $items) {
+     $levels++;
+    }
+    my $count = scalar(@main::pfile_args);
+    for (my $i = 0; $i < $count; $i++) {
+      $main::profile_files[$i] = FetchDynamicProfile($main::prog, $main::pfile_args[$i], 1, 0);
+    }
+    print STDERR "Fetching $count profiles, Be patient...\n";
+    FetchDynamicProfilesRecurse($levels, 0, 0);
+    $main::collected_profile = join(" \\\n    ", @main::profile_files);
+  }
+}
+
+# Recursively fork a process to get enough processes
+# collecting profiles
+sub FetchDynamicProfilesRecurse {
+  my $maxlevel = shift;
+  my $level = shift;
+  my $position = shift;
+
+  if (my $pid = fork()) {
+    $position = 0 | ($position << 1);
+    TryCollectProfile($maxlevel, $level, $position);
+    wait;
+  } else {
+    $position = 1 | ($position << 1);
+    TryCollectProfile($maxlevel, $level, $position);
+    cleanup();
+    exit(0);
+  }
+}
+
+# Collect a single profile
+sub TryCollectProfile {
+  my $maxlevel = shift;
+  my $level = shift;
+  my $position = shift;
+
+  if ($level >= ($maxlevel - 1)) {
+    if ($position < scalar(@main::pfile_args)) {
+      FetchDynamicProfile($main::prog, $main::pfile_args[$position], 0, 0);
+    }
+  } else {
+    FetchDynamicProfilesRecurse($maxlevel, $level+1, $position);
+  }
+}
+
+##### Parsing code #####
+
+# Provide a small streaming-read module to handle very large
+# cpu-profile files.  Stream in chunks along a sliding window.
+# Provides an interface to get one 'slot', correctly handling
+# endian-ness differences.  A slot is one 32-bit or 64-bit word
+# (depending on the input profile).  We tell endianness and bit-size
+# for the profile by looking at the first 8 bytes: in cpu profiles,
+# the second slot is always 3 (we'll accept anything that's not 0).
+BEGIN {
+  package CpuProfileStream;
+
+  sub new {
+    my ($class, $file, $fname) = @_;
+    my $self = { file        => $file,
+                 base        => 0,
+                 stride      => 512 * 1024,   # must be a multiple of bitsize/8
+                 slots       => [],
+                 unpack_code => "",           # N for big-endian, V for little
+                 perl_is_64bit => 1,          # matters if profile is 64-bit
+    };
+    bless $self, $class;
+    # Let unittests adjust the stride
+    if ($main::opt_test_stride > 0) {
+      $self->{stride} = $main::opt_test_stride;
+    }
+    # Read the first two slots to figure out bitsize and endianness.
+    my $slots = $self->{slots};
+    my $str;
+    read($self->{file}, $str, 8);
+    # Set the global $address_length based on what we see here.
+    # 8 is 32-bit (8 hexadecimal chars); 16 is 64-bit (16 hexadecimal chars).
+    $address_length = ($str eq (chr(0)x8)) ? 16 : 8;
+    if ($address_length == 8) {
+      if (substr($str, 6, 2) eq chr(0)x2) {
+        $self->{unpack_code} = 'V';  # Little-endian.
+      } elsif (substr($str, 4, 2) eq chr(0)x2) {
+        $self->{unpack_code} = 'N';  # Big-endian
+      } else {
+        ::error("$fname: header size >= 2**16\n");
+      }
+      @$slots = unpack($self->{unpack_code} . "*", $str);
+    } else {
+      # If we're a 64-bit profile, check if we're a 64-bit-capable
+      # perl.  Otherwise, each slot will be represented as a float
+      # instead of an int64, losing precision and making all the
+      # 64-bit addresses wrong.  We won't complain yet, but will
+      # later if we ever see a value that doesn't fit in 32 bits.
+      my $has_q = 0;
+      eval { $has_q = pack("Q", "1") ? 1 : 1; };
+      if (!$has_q) {
+        $self->{perl_is_64bit} = 0;
+      }
+      read($self->{file}, $str, 8);
+      if (substr($str, 4, 4) eq chr(0)x4) {
+        # We'd love to use 'Q', but it's a) not universal, b) not endian-proof.
+        $self->{unpack_code} = 'V';  # Little-endian.
+      } elsif (substr($str, 0, 4) eq chr(0)x4) {
+        $self->{unpack_code} = 'N';  # Big-endian
+      } else {
+        ::error("$fname: header size >= 2**32\n");
+      }
+      my @pair = unpack($self->{unpack_code} . "*", $str);
+      # Since we know one of the pair is 0, it's fine to just add them.
+      @$slots = (0, $pair[0] + $pair[1]);
+    }
+    return $self;
+  }
+
+  # Load more data when we access slots->get(X) which is not yet in memory.
+  sub overflow {
+    my ($self) = @_;
+    my $slots = $self->{slots};
+    $self->{base} += $#$slots + 1;   # skip over data we're replacing
+    my $str;
+    read($self->{file}, $str, $self->{stride});
+    if ($address_length == 8) {      # the 32-bit case
+      # This is the easy case: unpack provides 32-bit unpacking primitives.
+      @$slots = unpack($self->{unpack_code} . "*", $str);
+    } else {
+      # We need to unpack 32 bits at a time and combine.
+      my @b32_values = unpack($self->{unpack_code} . "*", $str);
+      my @b64_values = ();
+      for (my $i = 0; $i < $#b32_values; $i += 2) {
+        # TODO(csilvers): if this is a 32-bit perl, the math below
+        #    could end up in a too-large int, which perl will promote
+        #    to a double, losing necessary precision.  Deal with that.
+        #    Right now, we just die.
+        my ($lo, $hi) = ($b32_values[$i], $b32_values[$i+1]);
+        if ($self->{unpack_code} eq 'N') {    # big-endian
+          ($lo, $hi) = ($hi, $lo);
+        }
+        my $value = $lo + $hi * (2**32);
+        if (!$self->{perl_is_64bit} &&   # check value is exactly represented
+            (($value % (2**32)) != $lo || int($value / (2**32)) != $hi)) {
+          ::error("Need a 64-bit perl to process this 64-bit profile.\n");
+        }
+        push(@b64_values, $value);
+      }
+      @$slots = @b64_values;
+    }
+  }
+
+  # Access the i-th long in the file (logically), or -1 at EOF.
+  sub get {
+    my ($self, $idx) = @_;
+    my $slots = $self->{slots};
+    while ($#$slots >= 0) {
+      if ($idx < $self->{base}) {
+        # The only time we expect a reference to $slots[$i - something]
+        # after referencing $slots[$i] is reading the very first header.
+        # Since $stride > |header|, that shouldn't cause any lookback
+        # errors.  And everything after the header is sequential.
+        print STDERR "Unexpected look-back reading CPU profile";
+        return -1;   # shrug, don't know what better to return
+      } elsif ($idx > $self->{base} + $#$slots) {
+        $self->overflow();
+      } else {
+        return $slots->[$idx - $self->{base}];
+      }
+    }
+    # If we get here, $slots is [], which means we've reached EOF
+    return -1;  # unique since slots is supposed to hold unsigned numbers
+  }
+}
+
+# Reads the top, 'header' section of a profile, and returns the last
+# line of the header, commonly called a 'header line'.  The header
+# section of a profile consists of zero or more 'command' lines that
+# are instructions to jeprof, which jeprof executes when reading the
+# header.  All 'command' lines start with a %.  After the command
+# lines is the 'header line', which is a profile-specific line that
+# indicates what type of profile it is, and perhaps other global
+# information about the profile.  For instance, here's a header line
+# for a heap profile:
+#   heap profile:     53:    38236 [  5525:  1284029] @ heapprofile
+# For historical reasons, the CPU profile does not contain a text-
+# readable header line.  If the profile looks like a CPU profile,
+# this function returns "".  If no header line could be found, this
+# function returns undef.
+#
+# The following commands are recognized:
+#   %warn -- emit the rest of this line to stderr, prefixed by 'WARNING:'
+#
+# The input file should be in binmode.
+sub ReadProfileHeader {
+  local *PROFILE = shift;
+  my $firstchar = "";
+  my $line = "";
+  read(PROFILE, $firstchar, 1);
+  seek(PROFILE, -1, 1);                    # unread the firstchar
+  if ($firstchar !~ /[[:print:]]/) {       # is not a text character
+    return "";
+  }
+  while (defined($line = <PROFILE>)) {
+    $line =~ s/\r//g;   # turn windows-looking lines into unix-looking lines
+    if ($line =~ /^%warn\s+(.*)/) {        # 'warn' command
+      # Note this matches both '%warn blah\n' and '%warn\n'.
+      print STDERR "WARNING: $1\n";        # print the rest of the line
+    } elsif ($line =~ /^%/) {
+      print STDERR "Ignoring unknown command from profile header: $line";
+    } else {
+      # End of commands, must be the header line.
+      return $line;
+    }
+  }
+  return undef;     # got to EOF without seeing a header line
+}
+
+sub IsSymbolizedProfileFile {
+  my $file_name = shift;
+  if (!(-e $file_name) || !(-r $file_name)) {
+    return 0;
+  }
+  # Check if the file contains a symbol-section marker.
+  open(TFILE, "<$file_name");
+  binmode TFILE;
+  my $firstline = ReadProfileHeader(*TFILE);
+  close(TFILE);
+  if (!$firstline) {
+    return 0;
+  }
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+  return $firstline =~ /^--- *$symbol_marker/;
+}
+
+# Parse profile generated by common/profiler.cc and return a reference
+# to a map:
+#      $result->{version}     Version number of profile file
+#      $result->{period}      Sampling period (in microseconds)
+#      $result->{profile}     Profile object
+#      $result->{threads}     Map of thread IDs to profile objects
+#      $result->{map}         Memory map info from profile
+#      $result->{pcs}         Hash of all PC values seen, key is hex address
+sub ReadProfile {
+  my $prog = shift;
+  my $fname = shift;
+  my $result;            # return value
+
+  $CONTENTION_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $contention_marker = $&;
+  $GROWTH_PAGE  =~ m,[^/]+$,;    # matches everything after the last slash
+  my $growth_marker = $&;
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+  $PROFILE_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $profile_marker = $&;
+  $HEAP_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $heap_marker = $&;
+
+  # Look at first line to see if it is a heap or a CPU profile.
+  # CPU profile may start with no header at all, and just binary data
+  # (starting with \0\0\0\0) -- in that case, don't try to read the
+  # whole firstline, since it may be gigabytes(!) of data.
+  open(PROFILE, "<$fname") || error("$fname: $!\n");
+  binmode PROFILE;      # New perls do UTF-8 processing
+  my $header = ReadProfileHeader(*PROFILE);
+  if (!defined($header)) {   # means "at EOF"
+    error("Profile is empty.\n");
+  }
+
+  my $symbols;
+  if ($header =~ m/^--- *$symbol_marker/o) {
+    # Verify that the user asked for a symbolized profile
+    if (!$main::use_symbolized_profile) {
+      # we have both a binary and symbolized profiles, abort
+      error("FATAL ERROR: Symbolized profile\n   $fname\ncannot be used with " .
+            "a binary arg. Try again without passing\n   $prog\n");
+    }
+    # Read the symbol section of the symbolized profile file.
+    $symbols = ReadSymbols(*PROFILE{IO});
+    # Read the next line to get the header for the remaining profile.
+    $header = ReadProfileHeader(*PROFILE) || "";
+  }
+
+  if ($header =~ m/^--- *($heap_marker|$growth_marker)/o) {
+    # Skip "--- ..." line for profile types that have their own headers.
+    $header = ReadProfileHeader(*PROFILE) || "";
+  }
+
+  $main::profile_type = '';
+
+  if ($header =~ m/^heap profile:.*$growth_marker/o) {
+    $main::profile_type = 'growth';
+    $result =  ReadHeapProfile($prog, *PROFILE, $header);
+  } elsif ($header =~ m/^heap profile:/) {
+    $main::profile_type = 'heap';
+    $result =  ReadHeapProfile($prog, *PROFILE, $header);
+  } elsif ($header =~ m/^heap/) {
+    $main::profile_type = 'heap';
+    $result = ReadThreadedHeapProfile($prog, $fname, $header);
+  } elsif ($header =~ m/^--- *$contention_marker/o) {
+    $main::profile_type = 'contention';
+    $result = ReadSynchProfile($prog, *PROFILE);
+  } elsif ($header =~ m/^--- *Stacks:/) {
+    print STDERR
+      "Old format contention profile: mistakenly reports " .
+      "condition variable signals as lock contentions.\n";
+    $main::profile_type = 'contention';
+    $result = ReadSynchProfile($prog, *PROFILE);
+  } elsif ($header =~ m/^--- *$profile_marker/) {
+    # the binary cpu profile data starts immediately after this line
+    $main::profile_type = 'cpu';
+    $result = ReadCPUProfile($prog, $fname, *PROFILE);
+  } else {
+    if (defined($symbols)) {
+      # a symbolized profile contains a format we don't recognize, bail out
+      error("$fname: Cannot recognize profile section after symbols.\n");
+    }
+    # no ascii header present -- must be a CPU profile
+    $main::profile_type

[truncated 1061424 bytes]

Technical snapshot

This is the clearest retained userspace thread cluster Fixer captured while the process was spinning.

Representative thread backtrace

  • Command: /usr/bin/redis-server 127.0.0.1:6379
  • Why Fixer classified it this way: The trace repeatedly returns to a poll-family syscall without meaningful blocking, which suggests a busy event-loop wakeup.
  • Thread summary: thread backtraces show 1 thread(s) around 0x00007fbc72ca5ffe in ?? () and 1 thread(s) around 0x00007fbc72ca5ffe in ?? ()
  • Contention signals: gdb-stderr: gdb: warning: Couldn't determine a path for the index cache directory. warning: "target:<path>": could not open as an executable file: Input<path> error. warning: `target:<path>': can't open to read symbols: Input<path> error. warning: Could not load vsyscall page because no executable was specified
  • Repeated loop: epoll_wait -> openat -> read
  • Top syscalls: epoll_wait x50, close x49, openat x49, read x49
  • Package: redis-tools 5:8.0.6-1
  • Kernel: 6.17.10+deb14-amd64
  • Distribution: debian
0x00007fbc72ca5ffe in ?? ()
0x00007fbc72c9a7a4 in ?? ()
0x00007fbc73ef75c8 in ?? ()
0x0000000000000000 in ?? ()

Possible duplicates

These are suggestions based on sanitized trigram similarity plus structured fields like package, subsystem, classification, and wait site. They are not auto-merged.

redis-check-rdb is stuck in a likely busy poll loop: 5.07% of sampled CPU passed through unresolved offset, with repeated thread backtraces show 1 thread(s) around 0x00007fbc72ca5ffe in ?? () and 1 thread(s) around 0x00007fbc72ca5ffe in ?? ().

highpackage: redis-toolssource: redisscore: 106reports: 2patch attempt readysimilarity: 99%

Why this looks related: same classification, same package, same source package, same subsystem, same target, same wait site, very similar public summary

Last seen: 2026-05-19 03:16 UTC. Public page: /issues/019dd8e9-fd09-7ce2-ad17-85a44beba62d. Public JSON: /v1/issues/019dd8e9-fd09-7ce2-ad17-85a44beba62d

redis-check-rdb is stuck in a likely busy poll loop: 4.24% of sampled CPU passed through unresolved offset, with repeated thread backtraces show 1 thread(s) around 0x00007fbc72ca5ffe in ?? () and 1 thread(s) around 0x00007fbc72ca5ffe in ?? ().

highpackage: redis-toolssource: redisscore: 106reports: 2patch attempt readysimilarity: 99%

Why this looks related: same classification, same package, same source package, same subsystem, same target, same wait site, very similar public summary

Last seen: 2026-05-19 02:02 UTC. Public page: /issues/019ddbca-7201-7d22-ba77-0c6c084821ee. Public JSON: /v1/issues/019ddbca-7201-7d22-ba77-0c6c084821ee

redis-check-rdb is stuck in a likely busy poll loop: 5.36% of sampled CPU passed through num_to_str, with repeated thread backtraces show 1 thread(s) around 0x00007fbc72ca5ffe in ?? () and 1 thread(s) around 0x00007fbc72ca5ffe in ?? ().

highpackage: redis-toolssource: redisscore: 106reports: 2patch attempt readysimilarity: 99%

Why this looks related: same classification, same package, same source package, same subsystem, same target, same wait site, very similar public summary

Last seen: 2026-05-15 02:03 UTC. Public page: /issues/019e28a3-550a-70a1-9ff0-15b0c9603ee0. Public JSON: /v1/issues/019e28a3-550a-70a1-9ff0-15b0c9603ee0

redis-check-rdb is stuck in a likely busy poll loop: 5.11% of sampled CPU passed through __snprintf_chk@plt, with repeated thread backtraces show 1 thread(s) around 0x00007fbc72ca5ffe in ?? () and 1 thread(s) around 0x00007fbc72ca5ffe in ?? ().

highpackage: redis-toolssource: redisscore: 106reports: 2patch attempt readysimilarity: 99%

Why this looks related: same classification, same package, same source package, same subsystem, same target, same wait site, very similar public summary

Last seen: 2026-05-13 02:32 UTC. Public page: /issues/019df672-9134-74c0-a830-9c987d0858ad. Public JSON: /v1/issues/019df672-9134-74c0-a830-9c987d0858ad

postgres is stuck in a likely busy poll loop: 8.29% of sampled CPU passed through set_pte_range, with repeated thread backtraces show 1 thread(s) around 0x00007f322d29a7d2 in ?? () from target:<path>.

highpackage: postgresql-18source: postgresql-18score: 106reports: 2patch attempt readysimilarity: 70%

Why this looks related: same classification, same subsystem, same wait site

Last seen: 2026-05-13 22:23 UTC. Public page: /issues/019dfe20-db75-78f0-bf46-96613e1772e0. Public JSON: /v1/issues/019dfe20-db75-78f0-bf46-96613e1772e0

postgres is stuck in a likely busy poll loop: 7.30% of sampled CPU passed through sync_regs, with repeated thread backtraces show 1 thread(s) around 0x00007f322d29a7d2 in ?? () from target:<path>.

highpackage: postgresql-18source: postgresql-18score: 106reports: 2patch attempt readysimilarity: 70%

Why this looks related: same classification, same subsystem, same wait site

Last seen: 2026-05-18 13:02 UTC. Public page: /issues/019dca93-6037-7b91-ad59-7b3ca9e9ab37. Public JSON: /v1/issues/019dca93-6037-7b91-ad59-7b3ca9e9ab37

Worker outcome summary

This issue has 5 recorded worker attempts. Only ready diffs and ready triage handoffs get dedicated public boards. Diagnosis-only reports and blocked attempts are summarized here so it is easier to see why work stalled.

5 ready patch attempts

No ready triage handoffs, diagnosis-only reports, failed patch attempts, explained impossible attempts, or other attempt states.

Published attempts

ready patch attempt

patch

redis-check-rdb likely remains stuck in a busy-poll loop. A diagnosis report and patch proposal were created locally.

state: readycreated: 2026-05-11 21:22 UTCvalidation: ready
Published session

Prompt

## Plan Pass

You are planning a fixer patch before any edits happen.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. The original pre-edit snapshot is available at `./source` if you need to inspect it. For interpreter processes, plan from the script/application entrypoint evidence first and include the runtime only as a second investigation target unless the evidence proves a runtime bug.

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround. Inspect the relevant code, nearby callers, project contribution docs, and local helper/compat APIs, but do not edit files in this pass.

Return a short markdown plan with these exact sections:

## Problem
## Evidence Confidence
## Proposed Subject
## Patch Plan
## Risks
## Validation

Classify `## Evidence Confidence` as exactly one of `reproduced`, `observed`, or `inferred`. Use `inferred` only for a no-patch diagnosis/report plan unless you can name the extra evidence you will collect before editing; inferred source patches are blocked by Fixer because they are not pull-request-ready. For `observed` source-patch plans, plan to say in the final `## Issue Connection` that the failure was observed by Fixer and not independently reproduced. The plan must explain how the proposed code change addresses the observed issue evidence, call out any prior Fixer patch that should be improved or replaced, reject awkward control flow such as avoidable `goto` if there is a cleaner bounded alternative, name any local helper APIs or maintainer conventions the patch should follow, and keep the intended maintainer-facing explanation clear enough that someone unfamiliar with the local complaint wording can still follow the fix. In `## Validation`, name the reproducible configure/build/test entrypoint you will try from the workspace root before any focused leaf compile or smoke check, and include one bounded independent reproduction attempt for the collected failure signal when it is safe and cheap. Do not plan to claim `reproduced` unless that reproduction command or test can actually show the failure.

## Patch Pass

You are working on a bounded fixer proposal.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. Produce the smallest reasonable patch for the target repository, keep the change upstreamable, prefer the clearest control flow available, and do not keep avoidable `goto` when a simpler structure would read better. Before introducing new file, process, allocation, locking, networking, or platform APIs, inspect nearby code and project contribution docs for existing helpers or compatibility wrappers and use those local patterns unless you can explain why they do not fit. Validate from a reproducible workspace-root entrypoint before falling back to focused leaf commands; if a build or test cannot run, report the exact command, the exact blocker, and any narrower check you ran instead. During validation, also try one bounded independent reproduction of the collected failure signal when it is safe and cheap, such as a failing test, smoke command, perf/strace comparison, or before/after runtime check. Only use `reproduced` if that command or test actually reproduced the failure; otherwise keep `observed` and report the reproduction blocker. The final explanation must connect the observed issue evidence to the actual code change, not just paraphrase the diff. Write like a maintainer is going to read the patch mail cold: explain the bug in plain language, define subsystem-specific jargon the first time you need it, and make the causal story obvious. Explicitly classify evidence confidence as `reproduced`, `observed`, or `inferred`: `reproduced` means you reproduced the failure locally; `observed` means Fixer has direct crash/log/trace evidence but you did not independently reproduce it; `inferred` means the source patch is not pull-request-ready, so do not leave a source diff unless you first gather stronger observed/reproduced evidence; otherwise return a no-patch diagnosis/report. For any source-changing `observed` patch, say explicitly in `## Issue Connection` that the failure was observed by Fixer and not independently reproduced. If you introduce non-obvious state translation, index remapping, or backend split logic, add a short source comment that explains the invariant being preserved.

Start by explaining the likely root cause from the collected perf, strace, and /proc evidence. If you cannot land a safe patch, leave a diagnosis that is strong enough for an upstream bug report.

A previous Fixer patch attempt already exists for this issue. It was generated by Fixer `0.148.0`. Review that patch before changing code, improve it instead of starting blind, and clean up anything awkward or underexplained. In particular, remove avoidable `goto`, tighten the explanation of what the patch is doing, and make the resulting diff feel ready for upstream git review.
- Prior patch: `./prior-best.patch`
- Prior published session: `./prior-best-session.md`

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround. 

Keep the change narrowly scoped and summarize validation clearly.

In every authoring pass, your final response must start with `Subject: <single-line git commit subject>` and then include these markdown sections exactly:

## Commit Message
A short upstream-friendly explanation of what changed and why. Write it in plain language that a maintainer can follow without local complaint context. If you use subsystem jargon, define it immediately.

## Evidence Confidence
Exactly one word: `reproduced`, `observed`, or `inferred`. Use `reproduced` only when you reproduced the failure locally with a command or test, and include that command/test in `## Validation`. Use `observed` when Fixer has direct crash/log/trace evidence but you did not independently reproduce it. If `## Git Add Paths` lists source files for an `observed` patch, `## Issue Connection` must explicitly say the failure was observed by Fixer and not independently reproduced. Use `inferred` for profiler/strace/indirect evidence; inferred responses may be no-patch diagnoses or reports, but inferred source patches are not pull-request-ready until stronger evidence is gathered.

## Issue Connection
Write this as maintainer-facing patch mail, not as local Fixer notes. Cover four things explicitly in readable sentences: the user-visible symptom or the exact collected signal, the code-level cause or the cautious inference from evidence, the specific change you made, and the expected effect. Do not invent a reproducer, command line, crash, or user-visible failure that is not present in the evidence bundle. If the evidence is direct-but-not-reproduced, say it was observed by Fixer and not independently reproduced. If the evidence is indirect and you did not gather stronger evidence, do not leave a source diff; write a no-patch diagnosis/report instead. Include an explicit effect sentence such as `The expected effect is ...`, `This should reduce ...`, or `This prevents ...` for source patches. If the logic is non-obvious in code, mention that you added a short explanatory comment.

## Git Add Paths
List the repo-relative paths that belong in the final patch, one per line. Use `None` only when you intentionally made no source changes. Include intentionally new files, and do not list generated build artifacts.

## Validation
List the checks you ran, or say clearly that you could not run them. Include the independent reproduction command/test and result when `## Evidence Confidence` is `reproduced`; if reproduction was attempted but blocked, name the exact blocker and keep confidence at `observed` or `inferred`.

Before editing, read the plan at `./plan-output.txt` and follow it unless the code proves part of it wrong. If you change course, say so explicitly in the final write-up instead of silently drifting from the plan.

## Review Pass 1

You are reviewing a freshly generated fixer patch.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. Review the first patch pass. The original pre-edit snapshot is available at `./source` for diffing.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround.

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout. The latest author response is at `./patch-output.txt`. Inspect the current code and changed paths like a strict code reviewer. Focus on correctness, regressions, maintainability, awkward control flow such as avoidable `goto`, missing validation, weak or non-gittable commit message text, and explanations that fail to connect the observed issue evidence to the code change. Also review the maintainer experience: the patch mail should be easy to accept upstream, the user-visible bug should be explained in plain language, subsystem-specific jargon should be defined when first used, and any non-obvious state translation or index remapping in code should have a short explanatory comment. Reject patches that omit `## Evidence Confidence`, use anything other than `reproduced`, `observed`, or `inferred`, or sound more certain than that evidence level allows. Reject patches that claim `reproduced` unless `## Validation` names the concrete reproduction command or test and its result. Reject patches that invent a reproducer, command line, crash, or user-visible failure not present in the evidence bundle; indirect profiler or strace evidence may justify a cautious mitigation, but the author must say it is indirect instead of presenting it as a confirmed upstream bug. Reject patches that introduce generic libc/std APIs when nearby code or project docs provide a local compat/helper API for the same job, unless the author explicitly justifies the exception. Reject validation that only reports a leaf object/syntax build when a project-level configure/build/test entrypoint exists and was not attempted; if a full build cannot run, the author must show the exact failed command and blocker.

Do not apply code changes in this pass.

Return a short markdown review report. The first non-empty line must be exactly one of:

RESULT: ok
RESULT: fix-needed

If you choose `RESULT: fix-needed`, add a `## Findings` section with concrete, actionable items.

Response

## Plan Pass

## Problem
The evidence is internally mismatched: the report labels `redis-check-rdb`, but the captured command line, status, threads, and stack show `redis-server` sleeping in `epoll_wait`. The strace cadence is `epoll_wait(..., ~100ms)` followed by `/proc/self/stat` reads at about `0.3%` CPU, which matches normal Redis idle cron memory sampling rather than a demonstrated busy-poll defect.

## Evidence Confidence
inferred

## Proposed Subject
No source patch: diagnose idle Redis cron sampling trace mismatch

## Patch Plan
Do not edit source in this pass. A Redis upstream source patch would be speculative because the evidence is profiler/strace-only, low-CPU, and points to `redis-server`, not `redis-check-rdb`.

Reject the prior Fixer patch: it adds generated `deps/jemalloc/bin/*` files and does not address the sampled Redis event-loop path.

Conventions/helpers found: `aeMain()`/`aeProcessEvents()` drive the event loop, `serverCron()` schedules periodic work, `run_with_period()` throttles cron tasks, and `zmalloc_get_rss()`/`get_proc_stat_ll()` are the local `/proc/self/stat` helpers. Docs checked: `CONTRIBUTING.md`, `README.md`, and `tests/README.md`.

If stronger evidence is collected later, prefer a small subsystem-local change using existing helpers and straightforward bounded control flow; avoid introducing unnecessary `goto` or generic replacements for Redis wrappers.

## Risks
Changing cron timing or RSS sampling without stronger evidence could regress `INFO memory`, allocator diagnostics, and normal idle maintenance. Changing `redis-check-rdb` would not address the sampled process shown in the bundle.

## Validation
Before any focused check, try from the workspace root: `make`, then `make check` if available. If either fails, record the exact command and blocker.

Bounded independent evidence attempt: after a successful build, run a short local idle `src/redis-server` trace or smoke check to compare CPU and syscall cadence for `epoll_wait`, `openat`, `read`, and `close`. Do not claim `reproduced` unless that command demonstrates the reported runaway behavior rather than normal 100 ms cron sampling.

## Patch Pass

Subject: diagnose redis idle cron sampling trace

## Commit Message
No source change is proposed. The collected perf, strace, and `/proc` evidence points at an idle `redis-server` event loop doing normal cron memory sampling, not at a demonstrated `redis-check-rdb` busy-loop bug.

## Evidence Confidence
inferred

## Issue Connection
The likely root cause is an evidence mismatch plus normal Redis idle maintenance. The report is labeled `redis-check-rdb`, but the captured command line is `/usr/bin/redis-server 127.0.0.1:6379`, the sampled process is sleeping in `do_epoll_wait`, CPU is about `0.3%`, and strace shows `epoll_wait` timing out around every 100 ms followed by reads of `/proc/self/stat`.

In Redis, `aeProcessEvents()` drives the event loop, `serverCron()` runs periodic maintenance, `run_with_period(100)` schedules 100 ms work, and `cronUpdateMemoryStats()` calls `zmalloc_get_rss()`, which uses `get_proc_stat_ll()` to read `/proc/self/stat`. That matches the collected trace, so the evidence is indirect and internally mismatched rather than a confirmed upstream defect.

I made no source change. I also rejected the prior Fixer patch because it adds generated `deps/jemalloc/bin/*` files and does not address the observed Redis event-loop/RSS-sampling path. Contribution/style docs checked: `CONTRIBUTING.md`, `README.md`, and `tests/README.md`.

The expected effect is that no speculative upstream diff is produced until stronger evidence shows abnormal CPU use or a real tight loop distinct from normal idle cron sampling.

## Git Add Paths
None

## Validation
`make` from the workspace root completed successfully.

`make check` from the workspace root failed before running tests because the Redis test harness could not find a free port: `Can't find a non busy port in the 21079-21110 range.`

Bounded independent reproduction was attempted but blocked. `timeout 5s strace -f -ttT -e trace=epoll_wait,openat,read,close ./src/redis-server --port 0 --bind 127.0.0.1 --save '' --appendonly no --daemonize no` failed with `PTRACE_TRACEME: Operation not permitted` and `ptrace(PTRACE_SEIZE, 5): Operation not permitted`.

Plain server startup checks also could not keep an idle server running in this sandbox: `--port 0` exits because Redis is configured to listen nowhere, Unix socket startup fails with `setsockopt SO_REUSEADDR: Operation not permitted`, and TCP bind fails with `unable to bind socket, errno: 1`. Confidence remains `inferred`; no source diff was left.

## Review Pass 1

RESULT: ok

No blocking findings.

The workspace has no source diff, and `## Git Add Paths` correctly lists `None`. The author kept `## Evidence Confidence` as `inferred`, which matches the profiler/strace-only evidence and the internal mismatch between the `redis-check-rdb` artifact label and the sampled `redis-server` command line.

The response rejects the prior generated jemalloc-bin patch for the right reason: it does not address the observed Redis event-loop/RSS-sampling path. The explanation is cautious, names the relevant Redis helpers/conventions (`aeProcessEvents`, `serverCron`, `run_with_period`, `cronUpdateMemoryStats`, `zmalloc_get_rss`, `get_proc_stat_ll`), and does not invent a confirmed bug or reproducer.

Validation is acceptable: project-level `make` was attempted and passed; `make check` was attempted and reports the exact port-range blocker; independent reproduction attempts and sandbox blockers are explicitly named.

Diff

--- a/deps/jemalloc/bin/jemalloc-config	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jemalloc-config	2026-05-12 01:21:04.880670582 +0400
@@ -0,0 +1,83 @@
+#!/bin/sh
+
+usage() {
+	cat <<EOF
+Usage:
+  /usr/local/bin/jemalloc-config <option>
+Options:
+  --help | -h  : Print usage.
+  --version    : Print jemalloc version.
+  --revision   : Print shared library revision number.
+  --config     : Print configure options used to build jemalloc.
+  --prefix     : Print installation directory prefix.
+  --bindir     : Print binary installation directory.
+  --datadir    : Print data installation directory.
+  --includedir : Print include installation directory.
+  --libdir     : Print library installation directory.
+  --mandir     : Print manual page installation directory.
+  --cc         : Print compiler used to build jemalloc.
+  --cflags     : Print compiler flags used to build jemalloc.
+  --cppflags   : Print preprocessor flags used to build jemalloc.
+  --cxxflags   : Print C++ compiler flags used to build jemalloc.
+  --ldflags    : Print library flags used to build jemalloc.
+  --libs       : Print libraries jemalloc was linked against.
+EOF
+}
+
+prefix="/usr/local"
+exec_prefix="/usr/local"
+
+case "$1" in
+--help | -h)
+	usage
+	exit 0
+	;;
+--version)
+	echo "5.3.0-0-g0"
+	;;
+--revision)
+	echo "2"
+	;;
+--config)
+	echo "--disable-cxx --with-version=5.3.0-0-g0 --with-lg-quantum=3 --disable-cache-oblivious --with-jemalloc-prefix=je_ CFLAGS= LDFLAGS="
+	;;
+--prefix)
+	echo "/usr/local"
+	;;
+--bindir)
+	echo "/usr/local/bin"
+	;;
+--datadir)
+	echo "/usr/local/share"
+	;;
+--includedir)
+	echo "/usr/local/include"
+	;;
+--libdir)
+	echo "/usr/local/lib"
+	;;
+--mandir)
+	echo "/usr/local/share/man"
+	;;
+--cc)
+	echo "gcc"
+	;;
+--cflags)
+	echo "-std=gnu11 -Wall -Wextra -Wsign-compare -Wundef -Wno-format-zero-length -Wpointer-arith -Wno-missing-braces -Wno-missing-field-initializers -Wno-missing-attributes -pipe -g3 -fvisibility=hidden -Wimplicit-fallthrough -O3 -funroll-loops"
+	;;
+--cppflags)
+	echo "-D_GNU_SOURCE -D_REENTRANT"
+	;;
+--cxxflags)
+	echo ""
+	;;
+--ldflags)
+	echo " "
+	;;
+--libs)
+	echo "-lm  -pthread"
+	;;
+*)
+	usage
+	exit 1
+esac
--- a/deps/jemalloc/bin/jemalloc.sh	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jemalloc.sh	2026-05-12 01:21:04.886367177 +0400
@@ -0,0 +1,9 @@
+#!/bin/sh
+
+prefix=/usr/local
+exec_prefix=/usr/local
+libdir=${exec_prefix}/lib
+
+LD_PRELOAD=${libdir}/libjemalloc.so.2
+export LD_PRELOAD
+exec "$@"
--- a/deps/jemalloc/bin/jeprof	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jeprof	2026-05-12 01:21:04.892529790 +0400
@@ -0,0 +1,5723 @@
+#! /usr/bin/env perl
+
+# Copyright (c) 1998-2007, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+# ---
+# Program for printing the profile generated by common/profiler.cc,
+# or by the heap profiler (common/debugallocation.cc)
+#
+# The profile contains a sequence of entries of the form:
+#       <count> <stack trace>
+# This program parses the profile, and generates user-readable
+# output.
+#
+# Examples:
+#
+# % tools/jeprof "program" "profile"
+#   Enters "interactive" mode
+#
+# % tools/jeprof --text "program" "profile"
+#   Generates one line per procedure
+#
+# % tools/jeprof --gv "program" "profile"
+#   Generates annotated call-graph and displays via "gv"
+#
+# % tools/jeprof --gv --focus=Mutex "program" "profile"
+#   Restrict to code paths that involve an entry that matches "Mutex"
+#
+# % tools/jeprof --gv --focus=Mutex --ignore=string "program" "profile"
+#   Restrict to code paths that involve an entry that matches "Mutex"
+#   and does not match "string"
+#
+# % tools/jeprof --list=IBF_CheckDocid "program" "profile"
+#   Generates disassembly listing of all routines with at least one
+#   sample that match the --list=<regexp> pattern.  The listing is
+#   annotated with the flat and cumulative sample counts at each line.
+#
+# % tools/jeprof --disasm=IBF_CheckDocid "program" "profile"
+#   Generates disassembly listing of all routines with at least one
+#   sample that match the --disasm=<regexp> pattern.  The listing is
+#   annotated with the flat and cumulative sample counts at each PC value.
+#
+# TODO: Use color to indicate files?
+
+use strict;
+use warnings;
+use Getopt::Long;
+use Cwd;
+
+my $JEPROF_VERSION = "5.3.0-0-g0";
+my $PPROF_VERSION = "2.0";
+
+# These are the object tools we use which can come from a
+# user-specified location using --tools, from the JEPROF_TOOLS
+# environment variable, or from the environment.
+my %obj_tool_map = (
+  "objdump" => "objdump",
+  "nm" => "nm",
+  "addr2line" => "addr2line",
+  "c++filt" => "c++filt",
+  ## ConfigureObjTools may add architecture-specific entries:
+  #"nm_pdb" => "nm-pdb",       # for reading windows (PDB-format) executables
+  #"addr2line_pdb" => "addr2line-pdb",                                # ditto
+  #"otool" => "otool",         # equivalent of objdump on OS X
+);
+# NOTE: these are lists, so you can put in commandline flags if you want.
+my @DOT = ("dot");          # leave non-absolute, since it may be in /usr/local
+my @GV = ("gv");
+my @EVINCE = ("evince");    # could also be xpdf or perhaps acroread
+my @KCACHEGRIND = ("kcachegrind");
+my @PS2PDF = ("ps2pdf");
+# These are used for dynamic profiles
+my @URL_FETCHER = ("curl", "-s", "--fail");
+
+# These are the web pages that servers need to support for dynamic profiles
+my $HEAP_PAGE = "/pprof/heap";
+my $PROFILE_PAGE = "/pprof/profile";   # must support cgi-param "?seconds=#"
+my $PMUPROFILE_PAGE = "/pprof/pmuprofile(?:\\?.*)?"; # must support cgi-param
+                                                # ?seconds=#&event=x&period=n
+my $GROWTH_PAGE = "/pprof/growth";
+my $CONTENTION_PAGE = "/pprof/contention";
+my $WALL_PAGE = "/pprof/wall(?:\\?.*)?";  # accepts options like namefilter
+my $FILTEREDPROFILE_PAGE = "/pprof/filteredprofile(?:\\?.*)?";
+my $CENSUSPROFILE_PAGE = "/pprof/censusprofile(?:\\?.*)?"; # must support cgi-param
+                                                       # "?seconds=#",
+                                                       # "?tags_regexp=#" and
+                                                       # "?type=#".
+my $SYMBOL_PAGE = "/pprof/symbol";     # must support symbol lookup via POST
+my $PROGRAM_NAME_PAGE = "/pprof/cmdline";
+
+# These are the web pages that can be named on the command line.
+# All the alternatives must begin with /.
+my $PROFILES = "($HEAP_PAGE|$PROFILE_PAGE|$PMUPROFILE_PAGE|" .
+               "$GROWTH_PAGE|$CONTENTION_PAGE|$WALL_PAGE|" .
+               "$FILTEREDPROFILE_PAGE|$CENSUSPROFILE_PAGE)";
+
+# default binary name
+my $UNKNOWN_BINARY = "(unknown)";
+
+# There is a pervasive dependency on the length (in hex characters,
+# i.e., nibbles) of an address, distinguishing between 32-bit and
+# 64-bit profiles.  To err on the safe size, default to 64-bit here:
+my $address_length = 16;
+
+my $dev_null = "/dev/null";
+if (! -e $dev_null && $^O =~ /MSWin/) {    # $^O is the OS perl was built for
+  $dev_null = "nul";
+}
+
+# A list of paths to search for shared object files
+my @prefix_list = ();
+
+# Special routine name that should not have any symbols.
+# Used as separator to parse "addr2line -i" output.
+my $sep_symbol = '_fini';
+my $sep_address = undef;
+
+##### Argument parsing #####
+
+sub usage_string {
+  return <<EOF;
+Usage:
+jeprof [options] <program> <profiles>
+   <profiles> is a space separated list of profile names.
+jeprof [options] <symbolized-profiles>
+   <symbolized-profiles> is a list of profile files where each file contains
+   the necessary symbol mappings  as well as profile data (likely generated
+   with --raw).
+jeprof [options] <profile>
+   <profile> is a remote form.  Symbols are obtained from host:port$SYMBOL_PAGE
+
+   Each name can be:
+   /path/to/profile        - a path to a profile file
+   host:port[/<service>]   - a location of a service to get profile from
+
+   The /<service> can be $HEAP_PAGE, $PROFILE_PAGE, /pprof/pmuprofile,
+                         $GROWTH_PAGE, $CONTENTION_PAGE, /pprof/wall,
+                         $CENSUSPROFILE_PAGE, or /pprof/filteredprofile.
+   For instance:
+     jeprof http://myserver.com:80$HEAP_PAGE
+   If /<service> is omitted, the service defaults to $PROFILE_PAGE (cpu profiling).
+jeprof --symbols <program>
+   Maps addresses to symbol names.  In this mode, stdin should be a
+   list of library mappings, in the same format as is found in the heap-
+   and cpu-profile files (this loosely matches that of /proc/self/maps
+   on linux), followed by a list of hex addresses to map, one per line.
+
+   For more help with querying remote servers, including how to add the
+   necessary server-side support code, see this filename (or one like it):
+
+   /usr/doc/gperftools-$PPROF_VERSION/pprof_remote_servers.html
+
+Options:
+   --cum               Sort by cumulative data
+   --base=<base>       Subtract <base> from <profile> before display
+   --interactive       Run in interactive mode (interactive "help" gives help) [default]
+   --seconds=<n>       Length of time for dynamic profiles [default=30 secs]
+   --add_lib=<file>    Read additional symbols and line info from the given library
+   --lib_prefix=<dir>  Comma separated list of library path prefixes
+
+Reporting Granularity:
+   --addresses         Report at address level
+   --lines             Report at source line level
+   --functions         Report at function level [default]
+   --files             Report at source file level
+
+Output type:
+   --text              Generate text report
+   --callgrind         Generate callgrind format to stdout
+   --gv                Generate Postscript and display
+   --evince            Generate PDF and display
+   --web               Generate SVG and display
+   --list=<regexp>     Generate source listing of matching routines
+   --disasm=<regexp>   Generate disassembly of matching routines
+   --symbols           Print demangled symbol names found at given addresses
+   --dot               Generate DOT file to stdout
+   --ps                Generate Postcript to stdout
+   --pdf               Generate PDF to stdout
+   --svg               Generate SVG to stdout
+   --gif               Generate GIF to stdout
+   --raw               Generate symbolized jeprof data (useful with remote fetch)
+   --collapsed         Generate collapsed stacks for building flame graphs
+                       (see http://www.brendangregg.com/flamegraphs.html)
+
+Heap-Profile Options:
+   --inuse_space       Display in-use (mega)bytes [default]
+   --inuse_objects     Display in-use objects
+   --alloc_space       Display allocated (mega)bytes
+   --alloc_objects     Display allocated objects
+   --show_bytes        Display space in bytes
+   --drop_negative     Ignore negative differences
+
+Contention-profile options:
+   --total_delay       Display total delay at each region [default]
+   --contentions       Display number of delays at each region
+   --mean_delay        Display mean delay at each region
+
+Call-graph Options:
+   --nodecount=<n>     Show at most so many nodes [default=80]
+   --nodefraction=<f>  Hide nodes below <f>*total [default=.005]
+   --edgefraction=<f>  Hide edges below <f>*total [default=.001]
+   --maxdegree=<n>     Max incoming/outgoing edges per node [default=8]
+   --focus=<regexp>    Focus on backtraces with nodes matching <regexp>
+   --thread=<n>        Show profile for thread <n>
+   --ignore=<regexp>   Ignore backtraces with nodes matching <regexp>
+   --scale=<n>         Set GV scaling [default=0]
+   --heapcheck         Make nodes with non-0 object counts
+                       (i.e. direct leak generators) more visible
+   --retain=<regexp>   Retain only nodes that match <regexp>
+   --exclude=<regexp>  Exclude all nodes that match <regexp>
+
+Miscellaneous:
+   --tools=<prefix or binary:fullpath>[,...]   \$PATH for object tool pathnames
+   --test              Run unit tests
+   --help              This message
+   --version           Version information
+   --debug-syms-by-id  (Linux only) Find debug symbol files by build ID as well as by name
+
+Environment Variables:
+   JEPROF_TMPDIR        Profiles directory. Defaults to \$HOME/jeprof
+   JEPROF_TOOLS         Prefix for object tools pathnames
+
+Examples:
+
+jeprof /bin/ls ls.prof
+                       Enters "interactive" mode
+jeprof --text /bin/ls ls.prof
+                       Outputs one line per procedure
+jeprof --web /bin/ls ls.prof
+                       Displays annotated call-graph in web browser
+jeprof --gv /bin/ls ls.prof
+                       Displays annotated call-graph via 'gv'
+jeprof --gv --focus=Mutex /bin/ls ls.prof
+                       Restricts to code paths including a .*Mutex.* entry
+jeprof --gv --focus=Mutex --ignore=string /bin/ls ls.prof
+                       Code paths including Mutex but not string
+jeprof --list=getdir /bin/ls ls.prof
+                       (Per-line) annotated source listing for getdir()
+jeprof --disasm=getdir /bin/ls ls.prof
+                       (Per-PC) annotated disassembly for getdir()
+
+jeprof http://localhost:1234/
+                       Enters "interactive" mode
+jeprof --text localhost:1234
+                       Outputs one line per procedure for localhost:1234
+jeprof --raw localhost:1234 > ./local.raw
+jeprof --text ./local.raw
+                       Fetches a remote profile for later analysis and then
+                       analyzes it in text mode.
+EOF
+}
+
+sub version_string {
+  return <<EOF
+jeprof (part of jemalloc $JEPROF_VERSION)
+based on pprof (part of gperftools $PPROF_VERSION)
+
+Copyright 1998-2007 Google Inc.
+
+This is BSD licensed software; see the source for copying conditions
+and license information.
+There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A
+PARTICULAR PURPOSE.
+EOF
+}
+
+sub usage {
+  my $msg = shift;
+  print STDERR "$msg\n\n";
+  print STDERR usage_string();
+  print STDERR "\nFATAL ERROR: $msg\n";    # just as a reminder
+  exit(1);
+}
+
+sub Init() {
+  # Setup tmp-file name and handler to clean it up.
+  # We do this in the very beginning so that we can use
+  # error() and cleanup() function anytime here after.
+  $main::tmpfile_sym = "/tmp/jeprof$$.sym";
+  $main::tmpfile_ps = "/tmp/jeprof$$";
+  $main::next_tmpfile = 0;
+  $SIG{'INT'} = \&sighandler;
+
+  # Cache from filename/linenumber to source code
+  $main::source_cache = ();
+
+  $main::opt_help = 0;
+  $main::opt_version = 0;
+
+  $main::opt_cum = 0;
+  $main::opt_base = '';
+  $main::opt_addresses = 0;
+  $main::opt_lines = 0;
+  $main::opt_functions = 0;
+  $main::opt_files = 0;
+  $main::opt_lib_prefix = "";
+
+  $main::opt_text = 0;
+  $main::opt_callgrind = 0;
+  $main::opt_list = "";
+  $main::opt_disasm = "";
+  $main::opt_symbols = 0;
+  $main::opt_gv = 0;
+  $main::opt_evince = 0;
+  $main::opt_web = 0;
+  $main::opt_dot = 0;
+  $main::opt_ps = 0;
+  $main::opt_pdf = 0;
+  $main::opt_gif = 0;
+  $main::opt_svg = 0;
+  $main::opt_raw = 0;
+  $main::opt_collapsed = 0;
+
+  $main::opt_nodecount = 80;
+  $main::opt_nodefraction = 0.005;
+  $main::opt_edgefraction = 0.001;
+  $main::opt_maxdegree = 8;
+  $main::opt_focus = '';
+  $main::opt_thread = undef;
+  $main::opt_ignore = '';
+  $main::opt_scale = 0;
+  $main::opt_heapcheck = 0;
+  $main::opt_retain = '';
+  $main::opt_exclude = '';
+  $main::opt_seconds = 30;
+  $main::opt_lib = "";
+
+  $main::opt_inuse_space   = 0;
+  $main::opt_inuse_objects = 0;
+  $main::opt_alloc_space   = 0;
+  $main::opt_alloc_objects = 0;
+  $main::opt_show_bytes    = 0;
+  $main::opt_drop_negative = 0;
+  $main::opt_interactive   = 0;
+
+  $main::opt_total_delay = 0;
+  $main::opt_contentions = 0;
+  $main::opt_mean_delay = 0;
+
+  $main::opt_tools   = "";
+  $main::opt_debug   = 0;
+  $main::opt_test    = 0;
+  $main::opt_debug_syms_by_id = 0;
+
+  # These are undocumented flags used only by unittests.
+  $main::opt_test_stride = 0;
+
+  # Are we using $SYMBOL_PAGE?
+  $main::use_symbol_page = 0;
+
+  # Files returned by TempName.
+  %main::tempnames = ();
+
+  # Type of profile we are dealing with
+  # Supported types:
+  #     cpu
+  #     heap
+  #     growth
+  #     contention
+  $main::profile_type = '';     # Empty type means "unknown"
+
+  GetOptions("help!"          => \$main::opt_help,
+             "version!"       => \$main::opt_version,
+             "cum!"           => \$main::opt_cum,
+             "base=s"         => \$main::opt_base,
+             "seconds=i"      => \$main::opt_seconds,
+             "add_lib=s"      => \$main::opt_lib,
+             "lib_prefix=s"   => \$main::opt_lib_prefix,
+             "functions!"     => \$main::opt_functions,
+             "lines!"         => \$main::opt_lines,
+             "addresses!"     => \$main::opt_addresses,
+             "files!"         => \$main::opt_files,
+             "text!"          => \$main::opt_text,
+             "callgrind!"     => \$main::opt_callgrind,
+             "list=s"         => \$main::opt_list,
+             "disasm=s"       => \$main::opt_disasm,
+             "symbols!"       => \$main::opt_symbols,
+             "gv!"            => \$main::opt_gv,
+             "evince!"        => \$main::opt_evince,
+             "web!"           => \$main::opt_web,
+             "dot!"           => \$main::opt_dot,
+             "ps!"            => \$main::opt_ps,
+             "pdf!"           => \$main::opt_pdf,
+             "svg!"           => \$main::opt_svg,
+             "gif!"           => \$main::opt_gif,
+             "raw!"           => \$main::opt_raw,
+             "collapsed!"     => \$main::opt_collapsed,
+             "interactive!"   => \$main::opt_interactive,
+             "nodecount=i"    => \$main::opt_nodecount,
+             "nodefraction=f" => \$main::opt_nodefraction,
+             "edgefraction=f" => \$main::opt_edgefraction,
+             "maxdegree=i"    => \$main::opt_maxdegree,
+             "focus=s"        => \$main::opt_focus,
+             "thread=s"       => \$main::opt_thread,
+             "ignore=s"       => \$main::opt_ignore,
+             "scale=i"        => \$main::opt_scale,
+             "heapcheck"      => \$main::opt_heapcheck,
+             "retain=s"       => \$main::opt_retain,
+             "exclude=s"      => \$main::opt_exclude,
+             "inuse_space!"   => \$main::opt_inuse_space,
+             "inuse_objects!" => \$main::opt_inuse_objects,
+             "alloc_space!"   => \$main::opt_alloc_space,
+             "alloc_objects!" => \$main::opt_alloc_objects,
+             "show_bytes!"    => \$main::opt_show_bytes,
+             "drop_negative!" => \$main::opt_drop_negative,
+             "total_delay!"   => \$main::opt_total_delay,
+             "contentions!"   => \$main::opt_contentions,
+             "mean_delay!"    => \$main::opt_mean_delay,
+             "tools=s"        => \$main::opt_tools,
+             "test!"          => \$main::opt_test,
+             "debug!"         => \$main::opt_debug,
+             "debug-syms-by-id!" => \$main::opt_debug_syms_by_id,
+             # Undocumented flags used only by unittests:
+             "test_stride=i"  => \$main::opt_test_stride,
+      ) || usage("Invalid option(s)");
+
+  # Deal with the standard --help and --version
+  if ($main::opt_help) {
+    print usage_string();
+    exit(0);
+  }
+
+  if ($main::opt_version) {
+    print version_string();
+    exit(0);
+  }
+
+  # Disassembly/listing/symbols mode requires address-level info
+  if ($main::opt_disasm || $main::opt_list || $main::opt_symbols) {
+    $main::opt_functions = 0;
+    $main::opt_lines = 0;
+    $main::opt_addresses = 1;
+    $main::opt_files = 0;
+  }
+
+  # Check heap-profiling flags
+  if ($main::opt_inuse_space +
+      $main::opt_inuse_objects +
+      $main::opt_alloc_space +
+      $main::opt_alloc_objects > 1) {
+    usage("Specify at most on of --inuse/--alloc options");
+  }
+
+  # Check output granularities
+  my $grains =
+      $main::opt_functions +
+      $main::opt_lines +
+      $main::opt_addresses +
+      $main::opt_files +
+      0;
+  if ($grains > 1) {
+    usage("Only specify one output granularity option");
+  }
+  if ($grains == 0) {
+    $main::opt_functions = 1;
+  }
+
+  # Check output modes
+  my $modes =
+      $main::opt_text +
+      $main::opt_callgrind +
+      ($main::opt_list eq '' ? 0 : 1) +
+      ($main::opt_disasm eq '' ? 0 : 1) +
+      ($main::opt_symbols == 0 ? 0 : 1) +
+      $main::opt_gv +
+      $main::opt_evince +
+      $main::opt_web +
+      $main::opt_dot +
+      $main::opt_ps +
+      $main::opt_pdf +
+      $main::opt_svg +
+      $main::opt_gif +
+      $main::opt_raw +
+      $main::opt_collapsed +
+      $main::opt_interactive +
+      0;
+  if ($modes > 1) {
+    usage("Only specify one output mode");
+  }
+  if ($modes == 0) {
+    if (-t STDOUT) {  # If STDOUT is a tty, activate interactive mode
+      $main::opt_interactive = 1;
+    } else {
+      $main::opt_text = 1;
+    }
+  }
+
+  if ($main::opt_test) {
+    RunUnitTests();
+    # Should not return
+    exit(1);
+  }
+
+  # Binary name and profile arguments list
+  $main::prog = "";
+  @main::pfile_args = ();
+
+  # Remote profiling without a binary (using $SYMBOL_PAGE instead)
+  if (@ARGV > 0) {
+    if (IsProfileURL($ARGV[0])) {
+      $main::use_symbol_page = 1;
+    } elsif (IsSymbolizedProfileFile($ARGV[0])) {
+      $main::use_symbolized_profile = 1;
+      $main::prog = $UNKNOWN_BINARY;  # will be set later from the profile file
+    }
+  }
+
+  if ($main::use_symbol_page || $main::use_symbolized_profile) {
+    # We don't need a binary!
+    my %disabled = ('--lines' => $main::opt_lines,
+                    '--disasm' => $main::opt_disasm);
+    for my $option (keys %disabled) {
+      usage("$option cannot be used without a binary") if $disabled{$option};
+    }
+    # Set $main::prog later...
+    scalar(@ARGV) || usage("Did not specify profile file");
+  } elsif ($main::opt_symbols) {
+    # --symbols needs a binary-name (to run nm on, etc) but not profiles
+    $main::prog = shift(@ARGV) || usage("Did not specify program");
+  } else {
+    $main::prog = shift(@ARGV) || usage("Did not specify program");
+    scalar(@ARGV) || usage("Did not specify profile file");
+  }
+
+  # Parse profile file/location arguments
+  foreach my $farg (@ARGV) {
+    if ($farg =~ m/(.*)\@([0-9]+)(|\/.*)$/ ) {
+      my $machine = $1;
+      my $num_machines = $2;
+      my $path = $3;
+      for (my $i = 0; $i < $num_machines; $i++) {
+        unshift(@main::pfile_args, "$i.$machine$path");
+      }
+    } else {
+      unshift(@main::pfile_args, $farg);
+    }
+  }
+
+  if ($main::use_symbol_page) {
+    unless (IsProfileURL($main::pfile_args[0])) {
+      error("The first profile should be a remote form to use $SYMBOL_PAGE\n");
+    }
+    CheckSymbolPage();
+    $main::prog = FetchProgramName();
+  } elsif (!$main::use_symbolized_profile) {  # may not need objtools!
+    ConfigureObjTools($main::prog)
+  }
+
+  # Break the opt_lib_prefix into the prefix_list array
+  @prefix_list = split (',', $main::opt_lib_prefix);
+
+  # Remove trailing / from the prefixes, in the list to prevent
+  # searching things like /my/path//lib/mylib.so
+  foreach (@prefix_list) {
+    s|/+$||;
+  }
+
+  # Flag to prevent us from trying over and over to use
+  #  elfutils if it's not installed (used only with
+  #  --debug-syms-by-id option).
+  $main::gave_up_on_elfutils = 0;
+}
+
+sub FilterAndPrint {
+  my ($profile, $symbols, $libs, $thread) = @_;
+
+  # Get total data in profile
+  my $total = TotalProfile($profile);
+
+  # Remove uniniteresting stack items
+  $profile = RemoveUninterestingFrames($symbols, $profile);
+
+  # Focus?
+  if ($main::opt_focus ne '') {
+    $profile = FocusProfile($symbols, $profile, $main::opt_focus);
+  }
+
+  # Ignore?
+  if ($main::opt_ignore ne '') {
+    $profile = IgnoreProfile($symbols, $profile, $main::opt_ignore);
+  }
+
+  my $calls = ExtractCalls($symbols, $profile);
+
+  # Reduce profiles to required output granularity, and also clean
+  # each stack trace so a given entry exists at most once.
+  my $reduced = ReduceProfile($symbols, $profile);
+
+  # Get derived profiles
+  my $flat = FlatProfile($reduced);
+  my $cumulative = CumulativeProfile($reduced);
+
+  # Print
+  if (!$main::opt_interactive) {
+    if ($main::opt_disasm) {
+      PrintDisassembly($libs, $flat, $cumulative, $main::opt_disasm);
+    } elsif ($main::opt_list) {
+      PrintListing($total, $libs, $flat, $cumulative, $main::opt_list, 0);
+    } elsif ($main::opt_text) {
+      # Make sure the output is empty when have nothing to report
+      # (only matters when --heapcheck is given but we must be
+      # compatible with old branches that did not pass --heapcheck always):
+      if ($total != 0) {
+        printf("Total%s: %s %s\n",
+               (defined($thread) ? " (t$thread)" : ""),
+               Unparse($total), Units());
+      }
+      PrintText($symbols, $flat, $cumulative, -1);
+    } elsif ($main::opt_raw) {
+      PrintSymbolizedProfile($symbols, $profile, $main::prog);
+    } elsif ($main::opt_collapsed) {
+      PrintCollapsedStacks($symbols, $profile);
+    } elsif ($main::opt_callgrind) {
+      PrintCallgrind($calls);
+    } else {
+      if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) {
+        if ($main::opt_gv) {
+          RunGV(TempName($main::next_tmpfile, "ps"), "");
+        } elsif ($main::opt_evince) {
+          RunEvince(TempName($main::next_tmpfile, "pdf"), "");
+        } elsif ($main::opt_web) {
+          my $tmp = TempName($main::next_tmpfile, "svg");
+          RunWeb($tmp);
+          # The command we run might hand the file name off
+          # to an already running browser instance and then exit.
+          # Normally, we'd remove $tmp on exit (right now),
+          # but fork a child to remove $tmp a little later, so that the
+          # browser has time to load it first.
+          delete $main::tempnames{$tmp};
+          if (fork() == 0) {
+            sleep 5;
+            unlink($tmp);
+            exit(0);
+          }
+        }
+      } else {
+        cleanup();
+        exit(1);
+      }
+    }
+  } else {
+    InteractiveMode($profile, $symbols, $libs, $total);
+  }
+}
+
+sub Main() {
+  Init();
+  $main::collected_profile = undef;
+  @main::profile_files = ();
+  $main::op_time = time();
+
+  # Printing symbols is special and requires a lot less info that most.
+  if ($main::opt_symbols) {
+    PrintSymbols(*STDIN);   # Get /proc/maps and symbols output from stdin
+    return;
+  }
+
+  # Fetch all profile data
+  FetchDynamicProfiles();
+
+  # this will hold symbols that we read from the profile files
+  my $symbol_map = {};
+
+  # Read one profile, pick the last item on the list
+  my $data = ReadProfile($main::prog, pop(@main::profile_files));
+  my $profile = $data->{profile};
+  my $pcs = $data->{pcs};
+  my $libs = $data->{libs};   # Info about main program and shared libraries
+  $symbol_map = MergeSymbols($symbol_map, $data->{symbols});
+
+  # Add additional profiles, if available.
+  if (scalar(@main::profile_files) > 0) {
+    foreach my $pname (@main::profile_files) {
+      my $data2 = ReadProfile($main::prog, $pname);
+      $profile = AddProfile($profile, $data2->{profile});
+      $pcs = AddPcs($pcs, $data2->{pcs});
+      $symbol_map = MergeSymbols($symbol_map, $data2->{symbols});
+    }
+  }
+
+  # Subtract base from profile, if specified
+  if ($main::opt_base ne '') {
+    my $base = ReadProfile($main::prog, $main::opt_base);
+    $profile = SubtractProfile($profile, $base->{profile});
+    $pcs = AddPcs($pcs, $base->{pcs});
+    $symbol_map = MergeSymbols($symbol_map, $base->{symbols});
+  }
+
+  # Collect symbols
+  my $symbols;
+  if ($main::use_symbolized_profile) {
+    $symbols = FetchSymbols($pcs, $symbol_map);
+  } elsif ($main::use_symbol_page) {
+    $symbols = FetchSymbols($pcs);
+  } else {
+    # TODO(csilvers): $libs uses the /proc/self/maps data from profile1,
+    # which may differ from the data from subsequent profiles, especially
+    # if they were run on different machines.  Use appropriate libs for
+    # each pc somehow.
+    $symbols = ExtractSymbols($libs, $pcs);
+  }
+
+  if (!defined($main::opt_thread)) {
+    FilterAndPrint($profile, $symbols, $libs);
+  }
+  if (defined($data->{threads})) {
+    foreach my $thread (sort { $a <=> $b } keys(%{$data->{threads}})) {
+      if (defined($main::opt_thread) &&
+          ($main::opt_thread eq '*' || $main::opt_thread == $thread)) {
+        my $thread_profile = $data->{threads}{$thread};
+        FilterAndPrint($thread_profile, $symbols, $libs, $thread);
+      }
+    }
+  }
+
+  cleanup();
+  exit(0);
+}
+
+##### Entry Point #####
+
+Main();
+
+# Temporary code to detect if we're running on a Goobuntu system.
+# These systems don't have the right stuff installed for the special
+# Readline libraries to work, so as a temporary workaround, we default
+# to using the normal stdio code, rather than the fancier readline-based
+# code
+sub ReadlineMightFail {
+  if (-e '/lib/libtermcap.so.2') {
+    return 0;  # libtermcap exists, so readline should be okay
+  } else {
+    return 1;
+  }
+}
+
+sub RunGV {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  if (!system(ShellEscape(@GV, "--version") . " >$dev_null 2>&1")) {
+    # Options using double dash are supported by this gv version.
+    # Also, turn on noantialias to better handle bug in gv for
+    # postscript files with large dimensions.
+    # TODO: Maybe we should not pass the --noantialias flag
+    # if the gv version is known to work properly without the flag.
+    system(ShellEscape(@GV, "--scale=$main::opt_scale", "--noantialias", $fname)
+           . $bg);
+  } else {
+    # Old gv version - only supports options that use single dash.
+    print STDERR ShellEscape(@GV, "-scale", $main::opt_scale) . "\n";
+    system(ShellEscape(@GV, "-scale", "$main::opt_scale", $fname) . $bg);
+  }
+}
+
+sub RunEvince {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  system(ShellEscape(@EVINCE, $fname) . $bg);
+}
+
+sub RunWeb {
+  my $fname = shift;
+  print STDERR "Loading web page file:///$fname\n";
+
+  if (`uname` =~ /Darwin/) {
+    # OS X: open will use standard preference for SVG files.
+    system("/usr/bin/open", $fname);
+    return;
+  }
+
+  # Some kind of Unix; try generic symlinks, then specific browsers.
+  # (Stop once we find one.)
+  # Works best if the browser is already running.
+  my @alt = (
+    "/etc/alternatives/gnome-www-browser",
+    "/etc/alternatives/x-www-browser",
+    "google-chrome",
+    "firefox",
+  );
+  foreach my $b (@alt) {
+    if (system($b, $fname) == 0) {
+      return;
+    }
+  }
+
+  print STDERR "Could not load web browser.\n";
+}
+
+sub RunKcachegrind {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  print STDERR "Starting '@KCACHEGRIND " . $fname . $bg . "'\n";
+  system(ShellEscape(@KCACHEGRIND, $fname) . $bg);
+}
+
+
+##### Interactive helper routines #####
+
+sub InteractiveMode {
+  $| = 1;  # Make output unbuffered for interactive mode
+  my ($orig_profile, $symbols, $libs, $total) = @_;
+
+  print STDERR "Welcome to jeprof!  For help, type 'help'.\n";
+
+  # Use ReadLine if it's installed and input comes from a console.
+  if ( -t STDIN &&
+       !ReadlineMightFail() &&
+       defined(eval {require Term::ReadLine}) ) {
+    my $term = new Term::ReadLine 'jeprof';
+    while ( defined ($_ = $term->readline('(jeprof) '))) {
+      $term->addhistory($_) if /\S/;
+      if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) {
+        last;    # exit when we get an interactive command to quit
+      }
+    }
+  } else {       # don't have readline
+    while (1) {
+      print STDERR "(jeprof) ";
+      $_ = <STDIN>;
+      last if ! defined $_ ;
+      s/\r//g;         # turn windows-looking lines into unix-looking lines
+
+      # Save some flags that might be reset by InteractiveCommand()
+      my $save_opt_lines = $main::opt_lines;
+
+      if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) {
+        last;    # exit when we get an interactive command to quit
+      }
+
+      # Restore flags
+      $main::opt_lines = $save_opt_lines;
+    }
+  }
+}
+
+# Takes two args: orig profile, and command to run.
+# Returns 1 if we should keep going, or 0 if we were asked to quit
+sub InteractiveCommand {
+  my($orig_profile, $symbols, $libs, $total, $command) = @_;
+  $_ = $command;                # just to make future m//'s easier
+  if (!defined($_)) {
+    print STDERR "\n";
+    return 0;
+  }
+  if (m/^\s*quit/) {
+    return 0;
+  }
+  if (m/^\s*help/) {
+    InteractiveHelpMessage();
+    return 1;
+  }
+  # Clear all the mode options -- mode is controlled by "$command"
+  $main::opt_text = 0;
+  $main::opt_callgrind = 0;
+  $main::opt_disasm = 0;
+  $main::opt_list = 0;
+  $main::opt_gv = 0;
+  $main::opt_evince = 0;
+  $main::opt_cum = 0;
+
+  if (m/^\s*(text|top)(\d*)\s*(.*)/) {
+    $main::opt_text = 1;
+
+    my $line_limit = ($2 ne "") ? int($2) : 10;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($3);
+
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintText($symbols, $flat, $cumulative, $line_limit);
+    return 1;
+  }
+  if (m/^\s*callgrind\s*([^ \n]*)/) {
+    $main::opt_callgrind = 1;
+
+    # Get derived profiles
+    my $calls = ExtractCalls($symbols, $orig_profile);
+    my $filename = $1;
+    if ( $1 eq '' ) {
+      $filename = TempName($main::next_tmpfile, "callgrind");
+    }
+    PrintCallgrind($calls, $filename);
+    if ( $1 eq '' ) {
+      RunKcachegrind($filename, " & ");
+      $main::next_tmpfile++;
+    }
+
+    return 1;
+  }
+  if (m/^\s*(web)?list\s*(.+)/) {
+    my $html = (defined($1) && ($1 eq "web"));
+    $main::opt_list = 1;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($2);
+
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintListing($total, $libs, $flat, $cumulative, $routine, $html);
+    return 1;
+  }
+  if (m/^\s*disasm\s*(.+)/) {
+    $main::opt_disasm = 1;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($1);
+
+    # Process current profile to account for various settings
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintDisassembly($libs, $flat, $cumulative, $routine);
+    return 1;
+  }
+  if (m/^\s*(gv|web|evince)\s*(.*)/) {
+    $main::opt_gv = 0;
+    $main::opt_evince = 0;
+    $main::opt_web = 0;
+    if ($1 eq "gv") {
+      $main::opt_gv = 1;
+    } elsif ($1 eq "evince") {
+      $main::opt_evince = 1;
+    } elsif ($1 eq "web") {
+      $main::opt_web = 1;
+    }
+
+    my $focus;
+    my $ignore;
+    ($focus, $ignore) = ParseInteractiveArgs($2);
+
+    # Process current profile to account for various settings
+    my $profile = ProcessProfile($total, $orig_profile, $symbols,
+                                 $focus, $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) {
+      if ($main::opt_gv) {
+        RunGV(TempName($main::next_tmpfile, "ps"), " &");
+      } elsif ($main::opt_evince) {
+        RunEvince(TempName($main::next_tmpfile, "pdf"), " &");
+      } elsif ($main::opt_web) {
+        RunWeb(TempName($main::next_tmpfile, "svg"));
+      }
+      $main::next_tmpfile++;
+    }
+    return 1;
+  }
+  if (m/^\s*$/) {
+    return 1;
+  }
+  print STDERR "Unknown command: try 'help'.\n";
+  return 1;
+}
+
+
+sub ProcessProfile {
+  my $total_count = shift;
+  my $orig_profile = shift;
+  my $symbols = shift;
+  my $focus = shift;
+  my $ignore = shift;
+
+  # Process current profile to account for various settings
+  my $profile = $orig_profile;
+  printf("Total: %s %s\n", Unparse($total_count), Units());
+  if ($focus ne '') {
+    $profile = FocusProfile($symbols, $profile, $focus);
+    my $focus_count = TotalProfile($profile);
+    printf("After focusing on '%s': %s %s of %s (%0.1f%%)\n",
+           $focus,
+           Unparse($focus_count), Units(),
+           Unparse($total_count), ($focus_count*100.0) / $total_count);
+  }
+  if ($ignore ne '') {
+    $profile = IgnoreProfile($symbols, $profile, $ignore);
+    my $ignore_count = TotalProfile($profile);
+    printf("After ignoring '%s': %s %s of %s (%0.1f%%)\n",
+           $ignore,
+           Unparse($ignore_count), Units(),
+           Unparse($total_count),
+           ($ignore_count*100.0) / $total_count);
+  }
+
+  return $profile;
+}
+
+sub InteractiveHelpMessage {
+  print STDERR <<ENDOFHELP;
+Interactive jeprof mode
+
+Commands:
+  gv
+  gv [focus] [-ignore1] [-ignore2]
+      Show graphical hierarchical display of current profile.  Without
+      any arguments, shows all samples in the profile.  With the optional
+      "focus" argument, restricts the samples shown to just those where
+      the "focus" regular expression matches a routine name on the stack
+      trace.
+
+  web
+  web [focus] [-ignore1] [-ignore2]
+      Like GV, but displays profile in your web browser instead of using
+      Ghostview. Works best if your web browser is already running.
+      To change the browser that gets used:
+      On Linux, set the /etc/alternatives/gnome-www-browser symlink.
+      On OS X, change the Finder association for SVG files.
+
+  list [routine_regexp] [-ignore1] [-ignore2]
+      Show source listing of routines whose names match "routine_regexp"
+
+  weblist [routine_regexp] [-ignore1] [-ignore2]
+     Displays a source listing of routines whose names match "routine_regexp"
+     in a web browser.  You can click on source lines to view the
+     corresponding disassembly.
+
+  top [--cum] [-ignore1] [-ignore2]
+  top20 [--cum] [-ignore1] [-ignore2]
+  top37 [--cum] [-ignore1] [-ignore2]
+      Show top lines ordered by flat profile count, or cumulative count
+      if --cum is specified.  If a number is present after 'top', the
+      top K routines will be shown (defaults to showing the top 10)
+
+  disasm [routine_regexp] [-ignore1] [-ignore2]
+      Show disassembly of routines whose names match "routine_regexp",
+      annotated with sample counts.
+
+  callgrind
+  callgrind [filename]
+      Generates callgrind file. If no filename is given, kcachegrind is called.
+
+  help - This listing
+  quit or ^D - End jeprof
+
+For commands that accept optional -ignore tags, samples where any routine in
+the stack trace matches the regular expression in any of the -ignore
+parameters will be ignored.
+
+Further pprof details are available at this location (or one similar):
+
+ /usr/doc/gperftools-$PPROF_VERSION/cpu_profiler.html
+ /usr/doc/gperftools-$PPROF_VERSION/heap_profiler.html
+
+ENDOFHELP
+}
+sub ParseInteractiveArgs {
+  my $args = shift;
+  my $focus = "";
+  my $ignore = "";
+  my @x = split(/ +/, $args);
+  foreach $a (@x) {
+    if ($a =~ m/^(--|-)lines$/) {
+      $main::opt_lines = 1;
+    } elsif ($a =~ m/^(--|-)cum$/) {
+      $main::opt_cum = 1;
+    } elsif ($a =~ m/^-(.*)/) {
+      $ignore .= (($ignore ne "") ? "|" : "" ) . $1;
+    } else {
+      $focus .= (($focus ne "") ? "|" : "" ) . $a;
+    }
+  }
+  if ($ignore ne "") {
+    print STDERR "Ignoring samples in call stacks that match '$ignore'\n";
+  }
+  return ($focus, $ignore);
+}
+
+##### Output code #####
+
+sub TempName {
+  my $fnum = shift;
+  my $ext = shift;
+  my $file = "$main::tmpfile_ps.$fnum.$ext";
+  $main::tempnames{$file} = 1;
+  return $file;
+}
+
+# Print profile data in packed binary format (64-bit) to standard out
+sub PrintProfileData {
+  my $profile = shift;
+
+  # print header (64-bit style)
+  # (zero) (header-size) (version) (sample-period) (zero)
+  print pack('L*', 0, 0, 3, 0, 0, 0, 1, 0, 0, 0);
+
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    if ($#addrs >= 0) {
+      my $depth = $#addrs + 1;
+      # int(foo / 2**32) is the only reliable way to get rid of bottom
+      # 32 bits on both 32- and 64-bit systems.
+      print pack('L*', $count & 0xFFFFFFFF, int($count / 2**32));
+      print pack('L*', $depth & 0xFFFFFFFF, int($depth / 2**32));
+
+      foreach my $full_addr (@addrs) {
+        my $addr = $full_addr;
+        $addr =~ s/0x0*//;  # strip off leading 0x, zeroes
+        if (length($addr) > 16) {
+          print STDERR "Invalid address in profile: $full_addr\n";
+          next;
+        }
+        my $low_addr = substr($addr, -8);       # get last 8 hex chars
+        my $high_addr = substr($addr, -16, 8);  # get up to 8 more hex chars
+        print pack('L*', hex('0x' . $low_addr), hex('0x' . $high_addr));
+      }
+    }
+  }
+}
+
+# Print symbols and profile data
+sub PrintSymbolizedProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $prog = shift;
+
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+
+  print '--- ', $symbol_marker, "\n";
+  if (defined($prog)) {
+    print 'binary=', $prog, "\n";
+  }
+  while (my ($pc, $name) = each(%{$symbols})) {
+    my $sep = ' ';
+    print '0x', $pc;
+    # We have a list of function names, which include the inlined
+    # calls.  They are separated (and terminated) by --, which is
+    # illegal in function names.
+    for (my $j = 2; $j <= $#{$name}; $j += 3) {
+      print $sep, $name->[$j];
+      $sep = '--';
+    }
+    print "\n";
+  }
+  print '---', "\n";
+
+  my $profile_marker;
+  if ($main::profile_type eq 'heap') {
+    $HEAP_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } elsif ($main::profile_type eq 'growth') {
+    $GROWTH_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } elsif ($main::profile_type eq 'contention') {
+    $CONTENTION_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } else { # elsif ($main::profile_type eq 'cpu')
+    $PROFILE_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  }
+
+  print '--- ', $profile_marker, "\n";
+  if (defined($main::collected_profile)) {
+    # if used with remote fetch, simply dump the collected profile to output.
+    open(SRC, "<$main::collected_profile");
+    while (<SRC>) {
+      print $_;
+    }
+    close(SRC);
+  } else {
+    # --raw/http: For everything to work correctly for non-remote profiles, we
+    # would need to extend PrintProfileData() to handle all possible profile
+    # types, re-enable the code that is currently disabled in ReadCPUProfile()
+    # and FixCallerAddresses(), and remove the remote profile dumping code in
+    # the block above.
+    die "--raw/http: jeprof can only dump remote profiles for --raw\n";
+    # dump a cpu-format profile to standard out
+    PrintProfileData($profile);
+  }
+}
+
+# Print text output
+sub PrintText {
+  my $symbols = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $line_limit = shift;
+
+  my $total = TotalProfile($flat);
+
+  # Which profile to sort by?
+  my $s = $main::opt_cum ? $cumulative : $flat;
+
+  my $running_sum = 0;
+  my $lines = 0;
+  foreach my $k (sort { GetEntry($s, $b) <=> GetEntry($s, $a) || $a cmp $b }
+                 keys(%{$cumulative})) {
+    my $f = GetEntry($flat, $k);
+    my $c = GetEntry($cumulative, $k);
+    $running_sum += $f;
+
+    my $sym = $k;
+    if (exists($symbols->{$k})) {
+      $sym = $symbols->{$k}->[0] . " " . $symbols->{$k}->[1];
+      if ($main::opt_addresses) {
+        $sym = $k . " " . $sym;
+      }
+    }
+
+    if ($f != 0 || $c != 0) {
+      printf("%8s %6s %6s %8s %6s %s\n",
+             Unparse($f),
+             Percent($f, $total),
+             Percent($running_sum, $total),
+             Unparse($c),
+             Percent($c, $total),
+             $sym);
+    }
+    $lines++;
+    last if ($line_limit >= 0 && $lines >= $line_limit);
+  }
+}
+
+# Callgrind format has a compression for repeated function and file
+# names.  You show the name the first time, and just use its number
+# subsequently.  This can cut down the file to about a third or a
+# quarter of its uncompressed size.  $key and $val are the key/value
+# pair that would normally be printed by callgrind; $map is a map from
+# value to number.
+sub CompressedCGName {
+  my($key, $val, $map) = @_;
+  my $idx = $map->{$val};
+  # For very short keys, providing an index hurts rather than helps.
+  if (length($val) <= 3) {
+    return "$key=$val\n";
+  } elsif (defined($idx)) {
+    return "$key=($idx)\n";
+  } else {
+    # scalar(keys $map) gives the number of items in the map.
+    $idx = scalar(keys(%{$map})) + 1;
+    $map->{$val} = $idx;
+    return "$key=($idx) $val\n";
+  }
+}
+
+# Print the call graph in a way that's suiteable for callgrind.
+sub PrintCallgrind {
+  my $calls = shift;
+  my $filename;
+  my %filename_to_index_map;
+  my %fnname_to_index_map;
+
+  if ($main::opt_interactive) {
+    $filename = shift;
+    print STDERR "Writing callgrind file to '$filename'.\n"
+  } else {
+    $filename = "&STDOUT";
+  }
+  open(CG, ">$filename");
+  printf CG ("events: Hits\n\n");
+  foreach my $call ( map { $_->[0] }
+                     sort { $a->[1] cmp $b ->[1] ||
+                            $a->[2] <=> $b->[2] }
+                     map { /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/;
+                           [$_, $1, $2] }
+                     keys %$calls ) {
+    my $count = int($calls->{$call});
+    $call =~ /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/;
+    my ( $caller_file, $caller_line, $caller_function,
+         $callee_file, $callee_line, $callee_function ) =
+       ( $1, $2, $3, $5, $6, $7 );
+
+    # TODO(csilvers): for better compression, collect all the
+    # caller/callee_files and functions first, before printing
+    # anything, and only compress those referenced more than once.
+    printf CG CompressedCGName("fl", $caller_file, \%filename_to_index_map);
+    printf CG CompressedCGName("fn", $caller_function, \%fnname_to_index_map);
+    if (defined $6) {
+      printf CG CompressedCGName("cfl", $callee_file, \%filename_to_index_map);
+      printf CG CompressedCGName("cfn", $callee_function, \%fnname_to_index_map);
+      printf CG ("calls=$count $callee_line\n");
+    }
+    printf CG ("$caller_line $count\n\n");
+  }
+}
+
+# Print disassembly for all all routines that match $main::opt_disasm
+sub PrintDisassembly {
+  my $libs = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $disasm_opts = shift;
+
+  my $total = TotalProfile($flat);
+
+  foreach my $lib (@{$libs}) {
+    my $symbol_table = GetProcedureBoundaries($lib->[0], $disasm_opts);
+    my $offset = AddressSub($lib->[1], $lib->[3]);
+    foreach my $routine (sort ByName keys(%{$symbol_table})) {
+      my $start_addr = $symbol_table->{$routine}->[0];
+      my $end_addr = $symbol_table->{$routine}->[1];
+      # See if there are any samples in this routine
+      my $length = hex(AddressSub($end_addr, $start_addr));
+      my $addr = AddressAdd($start_addr, $offset);
+      for (my $i = 0; $i < $length; $i++) {
+        if (defined($cumulative->{$addr})) {
+          PrintDisassembledFunction($lib->[0], $offset,
+                                    $routine, $flat, $cumulative,
+                                    $start_addr, $end_addr, $total);
+          last;
+        }
+        $addr = AddressInc($addr);
+      }
+    }
+  }
+}
+
+# Return reference to array of tuples of the form:
+#       [start_address, filename, linenumber, instruction, limit_address]
+# E.g.,
+#       ["0x806c43d", "/foo/bar.cc", 131, "ret", "0x806c440"]
+sub Disassemble {
+  my $prog = shift;
+  my $offset = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+
+  my $objdump = $obj_tool_map{"objdump"};
+  my $cmd = ShellEscape($objdump, "-C", "-d", "-l", "--no-show-raw-insn",
+                        "--start-address=0x$start_addr",
+                        "--stop-address=0x$end_addr", $prog);
+  open(OBJDUMP, "$cmd |") || error("$cmd: $!\n");
+  my @result = ();
+  my $filename = "";
+  my $linenumber = -1;
+  my $last = ["", "", "", ""];
+  while (<OBJDUMP>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    chop;
+    if (m|\s*([^:\s]+):(\d+)\s*$|) {
+      # Location line of the form:
+      #   <filename>:<linenumber>
+      $filename = $1;
+      $linenumber = $2;
+    } elsif (m/^ +([0-9a-f]+):\s*(.*)/) {
+      # Disassembly line -- zero-extend address to full length
+      my $addr = HexExtend($1);
+      my $k = AddressAdd($addr, $offset);
+      $last->[4] = $k;   # Store ending address for previous instruction
+      $last = [$k, $filename, $linenumber, $2, $end_addr];
+      push(@result, $last);
+    }
+  }
+  close(OBJDUMP);
+  return @result;
+}
+
+# The input file should contain lines of the form /proc/maps-like
+# output (same format as expected from the profiles) or that looks
+# like hex addresses (like "0xDEADBEEF").  We will parse all
+# /proc/maps output, and for all the hex addresses, we will output
+# "short" symbol names, one per line, in the same order as the input.
+sub PrintSymbols {
+  my $maps_and_symbols_file = shift;
+
+  # ParseLibraries expects pcs to be in a set.  Fine by us...
+  my @pclist = ();   # pcs in sorted order
+  my $pcs = {};
+  my $map = "";
+  foreach my $line (<$maps_and_symbols_file>) {
+    $line =~ s/\r//g;    # turn windows-looking lines into unix-looking lines
+    if ($line =~ /\b(0x[0-9a-f]+)\b/i) {
+      push(@pclist, HexExtend($1));
+      $pcs->{$pclist[-1]} = 1;
+    } else {
+      $map .= $line;
+    }
+  }
+
+  my $libs = ParseLibraries($main::prog, $map, $pcs);
+  my $symbols = ExtractSymbols($libs, $pcs);
+
+  foreach my $pc (@pclist) {
+    # ->[0] is the shortname, ->[2] is the full name
+    print(($symbols->{$pc}->[0] || "??") . "\n");
+  }
+}
+
+
+# For sorting functions by name
+sub ByName {
+  return ShortFunctionName($a) cmp ShortFunctionName($b);
+}
+
+# Print source-listing for all all routines that match $list_opts
+sub PrintListing {
+  my $total = shift;
+  my $libs = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $list_opts = shift;
+  my $html = shift;
+
+  my $output = \*STDOUT;
+  my $fname = "";
+
+  if ($html) {
+    # Arrange to write the output to a temporary file
+    $fname = TempName($main::next_tmpfile, "html");
+    $main::next_tmpfile++;
+    if (!open(TEMP, ">$fname")) {
+      print STDERR "$fname: $!\n";
+      return;
+    }
+    $output = \*TEMP;
+    print $output HtmlListingHeader();
+    printf $output ("<div class=\"legend\">%s<br>Total: %s %s</div>\n",
+                    $main::prog, Unparse($total), Units());
+  }
+
+  my $listed = 0;
+  foreach my $lib (@{$libs}) {
+    my $symbol_table = GetProcedureBoundaries($lib->[0], $list_opts);
+    my $offset = AddressSub($lib->[1], $lib->[3]);
+    foreach my $routine (sort ByName keys(%{$symbol_table})) {
+      # Print if there are any samples in this routine
+      my $start_addr = $symbol_table->{$routine}->[0];
+      my $end_addr = $symbol_table->{$routine}->[1];
+      my $length = hex(AddressSub($end_addr, $start_addr));
+      my $addr = AddressAdd($start_addr, $offset);
+      for (my $i = 0; $i < $length; $i++) {
+        if (defined($cumulative->{$addr})) {
+          $listed += PrintSource(
+            $lib->[0], $offset,
+            $routine, $flat, $cumulative,
+            $start_addr, $end_addr,
+            $html,
+            $output);
+          last;
+        }
+        $addr = AddressInc($addr);
+      }
+    }
+  }
+
+  if ($html) {
+    if ($listed > 0) {
+      print $output HtmlListingFooter();
+      close($output);
+      RunWeb($fname);
+    } else {
+      close($output);
+      unlink($fname);
+    }
+  }
+}
+
+sub HtmlListingHeader {
+  return <<'EOF';
+<DOCTYPE html>
+<html>
+<head>
+<title>Pprof listing</title>
+<style type="text/css">
+body {
+  font-family: sans-serif;
+}
+h1 {
+  font-size: 1.5em;
+  margin-bottom: 4px;
+}
+.legend {
+  font-size: 1.25em;
+}
+.line {
+  color: #aaaaaa;
+}
+.nop {
+  color: #aaaaaa;
+}
+.unimportant {
+  color: #cccccc;
+}
+.disasmloc {
+  color: #000000;
+}
+.deadsrc {
+  cursor: pointer;
+}
+.deadsrc:hover {
+  background-color: #eeeeee;
+}
+.livesrc {
+  color: #0000ff;
+  cursor: pointer;
+}
+.livesrc:hover {
+  background-color: #eeeeee;
+}
+.asm {
+  color: #008800;
+  display: none;
+}
+</style>
+<script type="text/javascript">
+function jeprof_toggle_asm(e) {
+  var target;
+  if (!e) e = window.event;
+  if (e.target) target = e.target;
+  else if (e.srcElement) target = e.srcElement;
+
+  if (target) {
+    var asm = target.nextSibling;
+    if (asm && asm.className == "asm") {
+      asm.style.display = (asm.style.display == "block" ? "" : "block");
+      e.preventDefault();
+      return false;
+    }
+  }
+}
+</script>
+</head>
+<body>
+EOF
+}
+
+sub HtmlListingFooter {
+  return <<'EOF';
+</body>
+</html>
+EOF
+}
+
+sub HtmlEscape {
+  my $text = shift;
+  $text =~ s/&/&amp;/g;
+  $text =~ s/</&lt;/g;
+  $text =~ s/>/&gt;/g;
+  return $text;
+}
+
+# Returns the indentation of the line, if it has any non-whitespace
+# characters.  Otherwise, returns -1.
+sub Indentation {
+  my $line = shift;
+  if (m/^(\s*)\S/) {
+    return length($1);
+  } else {
+    return -1;
+  }
+}
+
+# If the symbol table contains inlining info, Disassemble() may tag an
+# instruction with a location inside an inlined function.  But for
+# source listings, we prefer to use the location in the function we
+# are listing.  So use MapToSymbols() to fetch full location
+# information for each instruction and then pick out the first
+# location from a location list (location list contains callers before
+# callees in case of inlining).
+#
+# After this routine has run, each entry in $instructions contains:
+#   [0] start address
+#   [1] filename for function we are listing
+#   [2] line number for function we are listing
+#   [3] disassembly
+#   [4] limit address
+#   [5] most specific filename (may be different from [1] due to inlining)
+#   [6] most specific line number (may be different from [2] due to inlining)
+sub GetTopLevelLineNumbers {
+  my ($lib, $offset, $instructions) = @_;
+  my $pcs = [];
+  for (my $i = 0; $i <= $#{$instructions}; $i++) {
+    push(@{$pcs}, $instructions->[$i]->[0]);
+  }
+  my $symbols = {};
+  MapToSymbols($lib, $offset, $pcs, $symbols);
+  for (my $i = 0; $i <= $#{$instructions}; $i++) {
+    my $e = $instructions->[$i];
+    push(@{$e}, $e->[1]);
+    push(@{$e}, $e->[2]);
+    my $addr = $e->[0];
+    my $sym = $symbols->{$addr};
+    if (defined($sym)) {
+      if ($#{$sym} >= 2 && $sym->[1] =~ m/^(.*):(\d+)$/) {
+        $e->[1] = $1;  # File name
+        $e->[2] = $2;  # Line number
+      }
+    }
+  }
+}
+
+# Print source-listing for one routine
+sub PrintSource {
+  my $prog = shift;
+  my $offset = shift;
+  my $routine = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+  my $html = shift;
+  my $output = shift;
+
+  # Disassemble all instructions (just to get line numbers)
+  my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr);
+  GetTopLevelLineNumbers($prog, $offset, \@instructions);
+
+  # Hack 1: assume that the first source file encountered in the
+  # disassembly contains the routine
+  my $filename = undef;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    if ($instructions[$i]->[2] >= 0) {
+      $filename = $instructions[$i]->[1];
+      last;
+    }
+  }
+  if (!defined($filename)) {
+    print STDERR "no filename found in $routine\n";
+    return 0;
+  }
+
+  # Hack 2: assume that the largest line number from $filename is the
+  # end of the procedure.  This is typically safe since if P1 contains
+  # an inlined call to P2, then P2 usually occurs earlier in the
+  # source file.  If this does not work, we might have to compute a
+  # density profile or just print all regions we find.
+  my $lastline = 0;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    my $f = $instructions[$i]->[1];
+    my $l = $instructions[$i]->[2];
+    if (($f eq $filename) && ($l > $lastline)) {
+      $lastline = $l;
+    }
+  }
+
+  # Hack 3: assume the first source location from "filename" is the start of
+  # the source code.
+  my $firstline = 1;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    if ($instructions[$i]->[1] eq $filename) {
+      $firstline = $instructions[$i]->[2];
+      last;
+    }
+  }
+
+  # Hack 4: Extend last line forward until its indentation is less than
+  # the indentation we saw on $firstline
+  my $oldlastline = $lastline;
+  {
+    if (!open(FILE, "<$filename")) {
+      print STDERR "$filename: $!\n";
+      return 0;
+    }
+    my $l = 0;
+    my $first_indentation = -1;
+    while (<FILE>) {
+      s/\r//g;         # turn windows-looking lines into unix-looking lines
+      $l++;
+      my $indent = Indentation($_);
+      if ($l >= $firstline) {
+        if ($first_indentation < 0 && $indent >= 0) {
+          $first_indentation = $indent;
+          last if ($first_indentation == 0);
+        }
+      }
+      if ($l >= $lastline && $indent >= 0) {
+        if ($indent >= $first_indentation) {
+          $lastline = $l+1;
+        } else {
+          last;
+        }
+      }
+    }
+    close(FILE);
+  }
+
+  # Assign all samples to the range $firstline,$lastline,
+  # Hack 4: If an instruction does not occur in the range, its samples
+  # are moved to the next instruction that occurs in the range.
+  my $samples1 = {};        # Map from line number to flat count
+  my $samples2 = {};        # Map from line number to cumulative count
+  my $running1 = 0;         # Unassigned flat counts
+  my $running2 = 0;         # Unassigned cumulative counts
+  my $total1 = 0;           # Total flat counts
+  my $total2 = 0;           # Total cumulative counts
+  my %disasm = ();          # Map from line number to disassembly
+  my $running_disasm = "";  # Unassigned disassembly
+  my $skip_marker = "---\n";
+  if ($html) {
+    $skip_marker = "";
+    for (my $l = $firstline; $l <= $lastline; $l++) {
+      $disasm{$l} = "";
+    }
+  }
+  my $last_dis_filename = '';
+  my $last_dis_linenum = -1;
+  my $last_touched_line = -1;  # To detect gaps in disassembly for a line
+  foreach my $e (@instructions) {
+    # Add up counts for all address that fall inside this instruction
+    my $c1 = 0;
+    my $c2 = 0;
+    for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) {
+      $c1 += GetEntry($flat, $a);
+      $c2 += GetEntry($cumulative, $a);
+    }
+
+    if ($html) {
+      my $dis = sprintf("      %6s %6s \t\t%8s: %s ",
+                        HtmlPrintNumber($c1),
+                        HtmlPrintNumber($c2),
+                        UnparseAddress($offset, $e->[0]),
+                        CleanDisassembly($e->[3]));
+
+      # Append the most specific source line associated with this instruction
+      if (length($dis) < 80) { $dis .= (' ' x (80 - length($dis))) };
+      $dis = HtmlEscape($dis);
+      my $f = $e->[5];
+      my $l = $e->[6];
+      if ($f ne $last_dis_filename) {
+        $dis .= sprintf("<span class=disasmloc>%s:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      } elsif ($l ne $last_dis_linenum) {
+        # De-emphasize the unchanged file name portion
+        $dis .= sprintf("<span class=unimportant>%s</span>" .
+                        "<span class=disasmloc>:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      } else {
+        # De-emphasize the entire location
+        $dis .= sprintf("<span class=unimportant>%s:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      }
+      $last_dis_filename = $f;
+      $last_dis_linenum = $l;
+      $running_disasm .= $dis;
+      $running_disasm .= "\n";
+    }
+
+    $running1 += $c1;
+    $running2 += $c2;
+    $total1 += $c1;
+    $total2 += $c2;
+    my $file = $e->[1];
+    my $line = $e->[2];
+    if (($file eq $filename) &&
+        ($line >= $firstline) &&
+        ($line <= $lastline)) {
+      # Assign all accumulated samples to this line
+      AddEntry($samples1, $line, $running1);
+      AddEntry($samples2, $line, $running2);
+      $running1 = 0;
+      $running2 = 0;
+      if ($html) {
+        if ($line != $last_touched_line && $disasm{$line} ne '') {
+          $disasm{$line} .= "\n";
+        }
+        $disasm{$line} .= $running_disasm;
+        $running_disasm = '';
+        $last_touched_line = $line;
+      }
+    }
+  }
+
+  # Assign any leftover samples to $lastline
+  AddEntry($samples1, $lastline, $running1);
+  AddEntry($samples2, $lastline, $running2);
+  if ($html) {
+    if ($lastline != $last_touched_line && $disasm{$lastline} ne '') {
+      $disasm{$lastline} .= "\n";
+    }
+    $disasm{$lastline} .= $running_disasm;
+  }
+
+  if ($html) {
+    printf $output (
+      "<h1>%s</h1>%s\n<pre onClick=\"jeprof_toggle_asm()\">\n" .
+      "Total:%6s %6s (flat / cumulative %s)\n",
+      HtmlEscape(ShortFunctionName($routine)),
+      HtmlEscape(CleanFileName($filename)),
+      Unparse($total1),
+      Unparse($total2),
+      Units());
+  } else {
+    printf $output (
+      "ROUTINE ====================== %s in %s\n" .
+      "%6s %6s Total %s (flat / cumulative)\n",
+      ShortFunctionName($routine),
+      CleanFileName($filename),
+      Unparse($total1),
+      Unparse($total2),
+      Units());
+  }
+  if (!open(FILE, "<$filename")) {
+    print STDERR "$filename: $!\n";
+    return 0;
+  }
+  my $l = 0;
+  while (<FILE>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    $l++;
+    if ($l >= $firstline - 5 &&
+        (($l <= $oldlastline + 5) || ($l <= $lastline))) {
+      chop;
+      my $text = $_;
+      if ($l == $firstline) { print $output $skip_marker; }
+      my $n1 = GetEntry($samples1, $l);
+      my $n2 = GetEntry($samples2, $l);
+      if ($html) {
+        # Emit a span that has one of the following classes:
+        #    livesrc -- has samples
+        #    deadsrc -- has disassembly, but with no samples
+        #    nop     -- has no matching disasembly
+        # Also emit an optional span containing disassembly.
+        my $dis = $disasm{$l};
+        my $asm = "";
+        if (defined($dis) && $dis ne '') {
+          $asm = "<span class=\"asm\">" . $dis . "</span>";
+        }
+        my $source_class = (($n1 + $n2 > 0)
+                            ? "livesrc"
+                            : (($asm ne "") ? "deadsrc" : "nop"));
+        printf $output (
+          "<span class=\"line\">%5d</span> " .
+          "<span class=\"%s\">%6s %6s %s</span>%s\n",
+          $l, $source_class,
+          HtmlPrintNumber($n1),
+          HtmlPrintNumber($n2),
+          HtmlEscape($text),
+          $asm);
+      } else {
+        printf $output(
+          "%6s %6s %4d: %s\n",
+          UnparseAlt($n1),
+          UnparseAlt($n2),
+          $l,
+          $text);
+      }
+      if ($l == $lastline)  { print $output $skip_marker; }
+    };
+  }
+  close(FILE);
+  if ($html) {
+    print $output "</pre>\n";
+  }
+  return 1;
+}
+
+# Return the source line for the specified file/linenumber.
+# Returns undef if not found.
+sub SourceLine {
+  my $file = shift;
+  my $line = shift;
+
+  # Look in cache
+  if (!defined($main::source_cache{$file})) {
+    if (100 < scalar keys(%main::source_cache)) {
+      # Clear the cache when it gets too big
+      $main::source_cache = ();
+    }
+
+    # Read all lines from the file
+    if (!open(FILE, "<$file")) {
+      print STDERR "$file: $!\n";
+      $main::source_cache{$file} = [];  # Cache the negative result
+      return undef;
+    }
+    my $lines = [];
+    push(@{$lines}, "");        # So we can use 1-based line numbers as indices
+    while (<FILE>) {
+      push(@{$lines}, $_);
+    }
+    close(FILE);
+
+    # Save the lines in the cache
+    $main::source_cache{$file} = $lines;
+  }
+
+  my $lines = $main::source_cache{$file};
+  if (($line < 0) || ($line > $#{$lines})) {
+    return undef;
+  } else {
+    return $lines->[$line];
+  }
+}
+
+# Print disassembly for one routine with interspersed source if available
+sub PrintDisassembledFunction {
+  my $prog = shift;
+  my $offset = shift;
+  my $routine = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+  my $total = shift;
+
+  # Disassemble all instructions
+  my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr);
+
+  # Make array of counts per instruction
+  my @flat_count = ();
+  my @cum_count = ();
+  my $flat_total = 0;
+  my $cum_total = 0;
+  foreach my $e (@instructions) {
+    # Add up counts for all address that fall inside this instruction
+    my $c1 = 0;
+    my $c2 = 0;
+    for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) {
+      $c1 += GetEntry($flat, $a);
+      $c2 += GetEntry($cumulative, $a);
+    }
+    push(@flat_count, $c1);
+    push(@cum_count, $c2);
+    $flat_total += $c1;
+    $cum_total += $c2;
+  }
+
+  # Print header with total counts
+  printf("ROUTINE ====================== %s\n" .
+         "%6s %6s %s (flat, cumulative) %.1f%% of total\n",
+         ShortFunctionName($routine),
+         Unparse($flat_total),
+         Unparse($cum_total),
+         Units(),
+         ($cum_total * 100.0) / $total);
+
+  # Process instructions in order
+  my $current_file = "";
+  for (my $i = 0; $i <= $#instructions; ) {
+    my $e = $instructions[$i];
+
+    # Print the new file name whenever we switch files
+    if ($e->[1] ne $current_file) {
+      $current_file = $e->[1];
+      my $fname = $current_file;
+      $fname =~ s|^\./||;   # Trim leading "./"
+
+      # Shorten long file names
+      if (length($fname) >= 58) {
+        $fname = "..." . substr($fname, -55);
+      }
+      printf("-------------------- %s\n", $fname);
+    }
+
+    # TODO: Compute range of lines to print together to deal with
+    # small reorderings.
+    my $first_line = $e->[2];
+    my $last_line = $first_line;
+    my %flat_sum = ();
+    my %cum_sum = ();
+    for (my $l = $first_line; $l <= $last_line; $l++) {
+      $flat_sum{$l} = 0;
+      $cum_sum{$l} = 0;
+    }
+
+    # Find run of instructions for this range of source lines
+    my $first_inst = $i;
+    while (($i <= $#instructions) &&
+           ($instructions[$i]->[2] >= $first_line) &&
+           ($instructions[$i]->[2] <= $last_line)) {
+      $e = $instructions[$i];
+      $flat_sum{$e->[2]} += $flat_count[$i];
+      $cum_sum{$e->[2]} += $cum_count[$i];
+      $i++;
+    }
+    my $last_inst = $i - 1;
+
+    # Print source lines
+    for (my $l = $first_line; $l <= $last_line; $l++) {
+      my $line = SourceLine($current_file, $l);
+      if (!defined($line)) {
+        $line = "?\n";
+        next;
+      } else {
+        $line =~ s/^\s+//;
+      }
+      printf("%6s %6s %5d: %s",
+             UnparseAlt($flat_sum{$l}),
+             UnparseAlt($cum_sum{$l}),
+             $l,
+             $line);
+    }
+
+    # Print disassembly
+    for (my $x = $first_inst; $x <= $last_inst; $x++) {
+      my $e = $instructions[$x];
+      printf("%6s %6s    %8s: %6s\n",
+             UnparseAlt($flat_count[$x]),
+             UnparseAlt($cum_count[$x]),
+             UnparseAddress($offset, $e->[0]),
+             CleanDisassembly($e->[3]));
+    }
+  }
+}
+
+# Print DOT graph
+sub PrintDot {
+  my $prog = shift;
+  my $symbols = shift;
+  my $raw = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $overall_total = shift;
+
+  # Get total
+  my $local_total = TotalProfile($flat);
+  my $nodelimit = int($main::opt_nodefraction * $local_total);
+  my $edgelimit = int($main::opt_edgefraction * $local_total);
+  my $nodecount = $main::opt_nodecount;
+
+  # Find nodes to include
+  my @list = (sort { abs(GetEntry($cumulative, $b)) <=>
+                     abs(GetEntry($cumulative, $a))
+                     || $a cmp $b }
+              keys(%{$cumulative}));
+  my $last = $nodecount - 1;
+  if ($last > $#list) {
+    $last = $#list;
+  }
+  while (($last >= 0) &&
+         (abs(GetEntry($cumulative, $list[$last])) <= $nodelimit)) {
+    $last--;
+  }
+  if ($last < 0) {
+    print STDERR "No nodes to print\n";
+    return 0;
+  }
+
+  if ($nodelimit > 0 || $edgelimit > 0) {
+    printf STDERR ("Dropping nodes with <= %s %s; edges with <= %s abs(%s)\n",
+                   Unparse($nodelimit), Units(),
+                   Unparse($edgelimit), Units());
+  }
+
+  # Open DOT output file
+  my $output;
+  my $escaped_dot = ShellEscape(@DOT);
+  my $escaped_ps2pdf = ShellEscape(@PS2PDF);
+  if ($main::opt_gv) {
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "ps"));
+    $output = "| $escaped_dot -Tps2 >$escaped_outfile";
+  } elsif ($main::opt_evince) {
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "pdf"));
+    $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - $escaped_outfile";
+  } elsif ($main::opt_ps) {
+    $output = "| $escaped_dot -Tps2";
+  } elsif ($main::opt_pdf) {
+    $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - -";
+  } elsif ($main::opt_web || $main::opt_svg) {
+    # We need to post-process the SVG, so write to a temporary file always.
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "svg"));
+    $output = "| $escaped_dot -Tsvg >$escaped_outfile";
+  } elsif ($main::opt_gif) {
+    $output = "| $escaped_dot -Tgif";
+  } else {
+    $output = ">&STDOUT";
+  }
+  open(DOT, $output) || error("$output: $!\n");
+
+  # Title
+  printf DOT ("digraph \"%s; %s %s\" {\n",
+              $prog,
+              Unparse($overall_total),
+              Units());
+  if ($main::opt_pdf) {
+    # The output is more printable if we set the page size for dot.
+    printf DOT ("size=\"8,11\"\n");
+  }
+  printf DOT ("node [width=0.375,height=0.25];\n");
+
+  # Print legend
+  printf DOT ("Legend [shape=box,fontsize=24,shape=plaintext," .
+              "label=\"%s\\l%s\\l%s\\l%s\\l%s\\l\"];\n",
+              $prog,
+              sprintf("Total %s: %s", Units(), Unparse($overall_total)),
+              sprintf("Focusing on: %s", Unparse($local_total)),
+              sprintf("Dropped nodes with <= %s abs(%s)",
+                      Unparse($nodelimit), Units()),
+              sprintf("Dropped edges with <= %s %s",
+                      Unparse($edgelimit), Units())
+              );
+
+  # Print nodes
+  my %node = ();
+  my $nextnode = 1;
+  foreach my $a (@list[0..$last]) {
+    # Pick font size
+    my $f = GetEntry($flat, $a);
+    my $c = GetEntry($cumulative, $a);
+
+    my $fs = 8;
+    if ($local_total > 0) {
+      $fs = 8 + (50.0 * sqrt(abs($f * 1.0 / $local_total)));
+    }
+
+    $node{$a} = $nextnode++;
+    my $sym = $a;
+    $sym =~ s/\s+/\\n/g;
+    $sym =~ s/::/\\n/g;
+
+    # Extra cumulative info to print for non-leaves
+    my $extra = "";
+    if ($f != $c) {
+      $extra = sprintf("\\rof %s (%s)",
+                       Unparse($c),
+                       Percent($c, $local_total));
+    }
+    my $style = "";
+    if ($main::opt_heapcheck) {
+      if ($f > 0) {
+        # make leak-causing nodes more visible (add a background)
+        $style = ",style=filled,fillcolor=gray"
+      } elsif ($f < 0) {
+        # make anti-leak-causing nodes (which almost never occur)
+        # stand out as well (triple border)
+        $style = ",peripheries=3"
+      }
+    }
+
+    printf DOT ("N%d [label=\"%s\\n%s (%s)%s\\r" .
+                "\",shape=box,fontsize=%.1f%s];\n",
+                $node{$a},
+                $sym,
+                Unparse($f),
+                Percent($f, $local_total),
+                $extra,
+                $fs,
+                $style,
+               );
+  }
+
+  # Get edges and counts per edge
+  my %edge = ();
+  my $n;
+  my $fullname_to_shortname_map = {};
+  FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map);
+  foreach my $k (keys(%{$raw})) {
+    # TODO: omit low %age edges
+    $n = $raw->{$k};
+    my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k);
+    for (my $i = 1; $i <= $#translated; $i++) {
+      my $src = $translated[$i];
+      my $dst = $translated[$i-1];
+      #next if ($src eq $dst);  # Avoid self-edges?
+      if (exists($node{$src}) && exists($node{$dst})) {
+        my $edge_label = "$src\001$dst";
+        if (!exists($edge{$edge_label})) {
+          $edge{$edge_label} = 0;
+        }
+        $edge{$edge_label} += $n;
+      }
+    }
+  }
+
+  # Print edges (process in order of decreasing counts)
+  my %indegree = ();   # Number of incoming edges added per node so far
+  my %outdegree = ();  # Number of outgoing edges added per node so far
+  foreach my $e (sort { $edge{$b} <=> $edge{$a} } keys(%edge)) {
+    my @x = split(/\001/, $e);
+    $n = $edge{$e};
+
+    # Initialize degree of kept incoming and outgoing edges if necessary
+    my $src = $x[0];
+    my $dst = $x[1];
+    if (!exists($outdegree{$src})) { $outdegree{$src} = 0; }
+    if (!exists($indegree{$dst})) { $indegree{$dst} = 0; }
+
+    my $keep;
+    if ($indegree{$dst} == 0) {
+      # Keep edge if needed for reachability
+      $keep = 1;
+    } elsif (abs($n) <= $edgelimit) {
+      # Drop if we are below --edgefraction
+      $keep = 0;
+    } elsif ($outdegree{$src} >= $main::opt_maxdegree ||
+             $indegree{$dst} >= $main::opt_maxdegree) {
+      # Keep limited number of in/out edges per node
+      $keep = 0;
+    } else {
+      $keep = 1;
+    }
+
+    if ($keep) {
+      $outdegree{$src}++;
+      $indegree{$dst}++;
+
+      # Compute line width based on edge count
+      my $fraction = abs($local_total ? (3 * ($n / $local_total)) : 0);
+      if ($fraction > 1) { $fraction = 1; }
+      my $w = $fraction * 2;
+      if ($w < 1 && ($main::opt_web || $main::opt_svg)) {
+        # SVG output treats line widths < 1 poorly.
+        $w = 1;
+      }
+
+      # Dot sometimes segfaults if given edge weights that are too large, so
+      # we cap the weights at a large value
+      my $edgeweight = abs($n) ** 0.7;
+      if ($edgeweight > 100000) { $edgeweight = 100000; }
+      $edgeweight = int($edgeweight);
+
+      my $style = sprintf("setlinewidth(%f)", $w);
+      if ($x[1] =~ m/\(inline\)/) {
+        $style .= ",dashed";
+      }
+
+      # Use a slightly squashed function of the edge count as the weight
+      printf DOT ("N%s -> N%s [label=%s, weight=%d, style=\"%s\"];\n",
+                  $node{$x[0]},
+                  $node{$x[1]},
+                  Unparse($n),
+                  $edgeweight,
+                  $style);
+    }
+  }
+
+  print DOT ("}\n");
+  close(DOT);
+
+  if ($main::opt_web || $main::opt_svg) {
+    # Rewrite SVG to be more usable inside web browser.
+    RewriteSvg(TempName($main::next_tmpfile, "svg"));
+  }
+
+  return 1;
+}
+
+sub RewriteSvg {
+  my $svgfile = shift;
+
+  open(SVG, $svgfile) || die "open temp svg: $!";
+  my @svg = <SVG>;
+  close(SVG);
+  unlink $svgfile;
+  my $svg = join('', @svg);
+
+  # Dot's SVG output is
+  #
+  #    <svg width="___" height="___"
+  #     viewBox="___" xmlns=...>
+  #    <g id="graph0" transform="...">
+  #    ...
+  #    </g>
+  #    </svg>
+  #
+  # Change it to
+  #
+  #    <svg width="100%" height="100%"
+  #     xmlns=...>
+  #    $svg_javascript
+  #    <g id="viewport" transform="translate(0,0)">
+  #    <g id="graph0" transform="...">
+  #    ...
+  #    </g>
+  #    </g>
+  #    </svg>
+
+  # Fix width, height; drop viewBox.
+  $svg =~ s/(?s)<svg width="[^"]+" height="[^"]+"(.*?)viewBox="[^"]+"/<svg width="100%" height="100%"$1/;
+
+  # Insert script, viewport <g> above first <g>
+  my $svg_javascript = SvgJavascript();
+  my $viewport = "<g id=\"viewport\" transform=\"translate(0,0)\">\n";
+  $svg =~ s/<g id="graph\d"/$svg_javascript$viewport$&/;
+
+  # Insert final </g> above </svg>.
+  $svg =~ s/(.*)(<\/svg>)/$1<\/g>$2/;
+  $svg =~ s/<g id="graph\d"(.*?)/<g id="viewport"$1/;
+
+  if ($main::opt_svg) {
+    # --svg: write to standard output.
+    print $svg;
+  } else {
+    # Write back to temporary file.
+    open(SVG, ">$svgfile") || die "open $svgfile: $!";
+    print SVG $svg;
+    close(SVG);
+  }
+}
+
+sub SvgJavascript {
+  return <<'EOF';
+<script type="text/ecmascript"><![CDATA[
+// SVGPan
+// http://www.cyberz.org/blog/2009/12/08/svgpan-a-javascript-svg-panzoomdrag-library/
+// Local modification: if(true || ...) below to force panning, never moving.
+
+/**
+ *  SVGPan library 1.2
+ * ====================
+ *
+ * Given an unique existing element with id "viewport", including the
+ * the library into any SVG adds the following capabilities:
+ *
+ *  - Mouse panning
+ *  - Mouse zooming (using the wheel)
+ *  - Object dargging
+ *
+ * Known issues:
+ *
+ *  - Zooming (while panning) on Safari has still some issues
+ *
+ * Releases:
+ *
+ * 1.2, Sat Mar 20 08:42:50 GMT 2010, Zeng Xiaohui
+ *	Fixed a bug with browser mouse handler interaction
+ *
+ * 1.1, Wed Feb  3 17:39:33 GMT 2010, Zeng Xiaohui
+ *	Updated the zoom code to support the mouse wheel on Safari/Chrome
+ *
+ * 1.0, Andrea Leofreddi
+ *	First release
+ *
+ * This code is licensed under the following BSD license:
+ *
+ * Copyright 2009-2010 Andrea Leofreddi <a.leofreddi@itcharm.com>. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are
+ * permitted provided that the following conditions are met:
+ *
+ *    1. Redistributions of source code must retain the above copyright notice, this list of
+ *       conditions and the following disclaimer.
+ *
+ *    2. Redistributions in binary form must reproduce the above copyright notice, this list
+ *       of conditions and the following disclaimer in the documentation and/or other materials
+ *       provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Andrea Leofreddi ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+ * FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL Andrea Leofreddi OR
+ * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * The views and conclusions contained in the software and documentation are those of the
+ * authors and should not be interpreted as representing official policies, either expressed
+ * or implied, of Andrea Leofreddi.
+ */
+
+var root = document.documentElement;
+
+var state = 'none', stateTarget, stateOrigin, stateTf;
+
+setupHandlers(root);
+
+/**
+ * Register handlers
+ */
+function setupHandlers(root){
+	setAttributes(root, {
+		"onmouseup" : "add(evt)",
+		"onmousedown" : "handleMouseDown(evt)",
+		"onmousemove" : "handleMouseMove(evt)",
+		"onmouseup" : "handleMouseUp(evt)",
+		//"onmouseout" : "handleMouseUp(evt)", // Decomment this to stop the pan functionality when dragging out of the SVG element
+	});
+
+	if(navigator.userAgent.toLowerCase().indexOf('webkit') >= 0)
+		window.addEventListener('mousewheel', handleMouseWheel, false); // Chrome/Safari
+	else
+		window.addEventListener('DOMMouseScroll', handleMouseWheel, false); // Others
+
+	var g = svgDoc.getElementById("svg");
+	g.width = "100%";
+	g.height = "100%";
+}
+
+/**
+ * Instance an SVGPoint object with given event coordinates.
+ */
+function getEventPoint(evt) {
+	var p = root.createSVGPoint();
+
+	p.x = evt.clientX;
+	p.y = evt.clientY;
+
+	return p;
+}
+
+/**
+ * Sets the current transform matrix of an element.
+ */
+function setCTM(element, matrix) {
+	var s = "matrix(" + matrix.a + "," + matrix.b + "," + matrix.c + "," + matrix.d + "," + matrix.e + "," + matrix.f + ")";
+
+	element.setAttribute("transform", s);
+}
+
+/**
+ * Dumps a matrix to a string (useful for debug).
+ */
+function dumpMatrix(matrix) {
+	var s = "[ " + matrix.a + ", " + matrix.c + ", " + matrix.e + "\n  " + matrix.b + ", " + matrix.d + ", " + matrix.f + "\n  0, 0, 1 ]";
+
+	return s;
+}
+
+/**
+ * Sets attributes of an element.
+ */
+function setAttributes(element, attributes){
+	for (i in attributes)
+		element.setAttributeNS(null, i, attributes[i]);
+}
+
+/**
+ * Handle mouse move event.
+ */
+function handleMouseWheel(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var delta;
+
+	if(evt.wheelDelta)
+		delta = evt.wheelDelta / 3600; // Chrome/Safari
+	else
+		delta = evt.detail / -90; // Mozilla
+
+	var z = 1 + delta; // Zoom factor: 0.9/1.1
+
+	var g = svgDoc.getElementById("viewport");
+
+	var p = getEventPoint(evt);
+
+	p = p.matrixTransform(g.getCTM().inverse());
+
+	// Compute new scale matrix in current mouse position
+	var k = root.createSVGMatrix().translate(p.x, p.y).scale(z).translate(-p.x, -p.y);
+
+        setCTM(g, g.getCTM().multiply(k));
+
+	stateTf = stateTf.multiply(k.inverse());
+}
+
+/**
+ * Handle mouse move event.
+ */
+function handleMouseMove(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var g = svgDoc.getElementById("viewport");
+
+	if(state == 'pan') {
+		// Pan mode
+		var p = getEventPoint(evt).matrixTransform(stateTf);
+
+		setCTM(g, stateTf.inverse().translate(p.x - stateOrigin.x, p.y - stateOrigin.y));
+	} else if(state == 'move') {
+		// Move mode
+		var p = getEventPoint(evt).matrixTransform(g.getCTM().inverse());
+
+		setCTM(stateTarget, root.createSVGMatrix().translate(p.x - stateOrigin.x, p.y - stateOrigin.y).multiply(g.getCTM().inverse()).multiply(stateTarget.getCTM()));
+
+		stateOrigin = p;
+	}
+}
+
+/**
+ * Handle click event.
+ */
+function handleMouseDown(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var g = svgDoc.getElementById("viewport");
+
+	if(true || evt.target.tagName == "svg") {
+		// Pan mode
+		state = 'pan';
+
+		stateTf = g.getCTM().inverse();
+
+		stateOrigin = getEventPoint(evt).matrixTransform(stateTf);
+	} else {
+		// Move mode
+		state = 'move';
+
+		stateTarget = evt.target;
+
+		stateTf = g.getCTM().inverse();
+
+		stateOrigin = getEventPoint(evt).matrixTransform(stateTf);
+	}
+}
+
+/**
+ * Handle mouse button release event.
+ */
+function handleMouseUp(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	if(state == 'pan' || state == 'move') {
+		// Quit pan mode
+		state = '';
+	}
+}
+
+]]></script>
+EOF
+}
+
+# Provides a map from fullname to shortname for cases where the
+# shortname is ambiguous.  The symlist has both the fullname and
+# shortname for all symbols, which is usually fine, but sometimes --
+# such as overloaded functions -- two different fullnames can map to
+# the same shortname.  In that case, we use the address of the
+# function to disambiguate the two.  This function fills in a map that
+# maps fullnames to modified shortnames in such cases.  If a fullname
+# is not present in the map, the 'normal' shortname provided by the
+# symlist is the appropriate one to use.
+sub FillFullnameToShortnameMap {
+  my $symbols = shift;
+  my $fullname_to_shortname_map = shift;
+  my $shortnames_seen_once = {};
+  my $shortnames_seen_more_than_once = {};
+
+  foreach my $symlist (values(%{$symbols})) {
+    # TODO(csilvers): deal with inlined symbols too.
+    my $shortname = $symlist->[0];
+    my $fullname = $symlist->[2];
+    if ($fullname !~ /<[0-9a-fA-F]+>$/) {  # fullname doesn't end in an address
+      next;       # the only collisions we care about are when addresses differ
+    }
+    if (defined($shortnames_seen_once->{$shortname}) &&
+        $shortnames_seen_once->{$shortname} ne $fullname) {
+      $shortnames_seen_more_than_once->{$shortname} = 1;
+    } else {
+      $shortnames_seen_once->{$shortname} = $fullname;
+    }
+  }
+
+  foreach my $symlist (values(%{$symbols})) {
+    my $shortname = $symlist->[0];
+    my $fullname = $symlist->[2];
+    # TODO(csilvers): take in a list of addresses we care about, and only
+    # store in the map if $symlist->[1] is in that list.  Saves space.
+    next if defined($fullname_to_shortname_map->{$fullname});
+    if (defined($shortnames_seen_more_than_once->{$shortname})) {
+      if ($fullname =~ /<0*([^>]*)>$/) {   # fullname has address at end of it
+        $fullname_to_shortname_map->{$fullname} = "$shortname\@$1";
+      }
+    }
+  }
+}
+
+# Return a small number that identifies the argument.
+# Multiple calls with the same argument will return the same number.
+# Calls with different arguments will return different numbers.
+sub ShortIdFor {
+  my $key = shift;
+  my $id = $main::uniqueid{$key};
+  if (!defined($id)) {
+    $id = keys(%main::uniqueid) + 1;
+    $main::uniqueid{$key} = $id;
+  }
+  return $id;
+}
+
+# Translate a stack of addresses into a stack of symbols
+sub TranslateStack {
+  my $symbols = shift;
+  my $fullname_to_shortname_map = shift;
+  my $k = shift;
+
+  my @addrs = split(/\n/, $k);
+  my @result = ();
+  for (my $i = 0; $i <= $#addrs; $i++) {
+    my $a = $addrs[$i];
+
+    # Skip large addresses since they sometimes show up as fake entries on RH9
+    if (length($a) > 8 && $a gt "7fffffffffffffff") {
+      next;
+    }
+
+    if ($main::opt_disasm || $main::opt_list) {
+      # We want just the address for the key
+      push(@result, $a);
+      next;
+    }
+
+    my $symlist = $symbols->{$a};
+    if (!defined($symlist)) {
+      $symlist = [$a, "", $a];
+    }
+
+    # We can have a sequence of symbols for a particular entry
+    # (more than one symbol in the case of inlining).  Callers
+    # come before callees in symlist, so walk backwards since
+    # the translated stack should contain callees before callers.
+    for (my $j = $#{$symlist}; $j >= 2; $j -= 3) {
+      my $func = $symlist->[$j-2];
+      my $fileline = $symlist->[$j-1];
+      my $fullfunc = $symlist->[$j];
+      if (defined($fullname_to_shortname_map->{$fullfunc})) {
+        $func = $fullname_to_shortname_map->{$fullfunc};
+      }
+      if ($j > 2) {
+        $func = "$func (inline)";
+      }
+
+      # Do not merge nodes corresponding to Callback::Run since that
+      # causes confusing cycles in dot display.  Instead, we synthesize
+      # a unique name for this frame per caller.
+      if ($func =~ m/Callback.*::Run$/) {
+        my $caller = ($i > 0) ? $addrs[$i-1] : 0;
+        $func = "Run#" . ShortIdFor($caller);
+      }
+
+      if ($main::opt_addresses) {
+        push(@result, "$a $func $fileline");
+      } elsif ($main::opt_lines) {
+        if ($func eq '??' && $fileline eq '??:0') {
+          push(@result, "$a");
+        } else {
+          push(@result, "$func $fileline");
+        }
+      } elsif ($main::opt_functions) {
+        if ($func eq '??') {
+          push(@result, "$a");
+        } else {
+          push(@result, $func);
+        }
+      } elsif ($main::opt_files) {
+        if ($fileline eq '??:0' || $fileline eq '') {
+          push(@result, "$a");
+        } else {
+          my $f = $fileline;
+          $f =~ s/:\d+$//;
+          push(@result, $f);
+        }
+      } else {
+        push(@result, $a);
+        last;  # Do not print inlined info
+      }
+    }
+  }
+
+  # print join(",", @addrs), " => ", join(",", @result), "\n";
+  return @result;
+}
+
+# Generate percent string for a number and a total
+sub Percent {
+  my $num = shift;
+  my $tot = shift;
+  if ($tot != 0) {
+    return sprintf("%.1f%%", $num * 100.0 / $tot);
+  } else {
+    return ($num == 0) ? "nan" : (($num > 0) ? "+inf" : "-inf");
+  }
+}
+
+# Generate pretty-printed form of number
+sub Unparse {
+  my $num = shift;
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    if ($main::opt_inuse_objects || $main::opt_alloc_objects) {
+      return sprintf("%d", $num);
+    } else {
+      if ($main::opt_show_bytes) {
+        return sprintf("%d", $num);
+      } else {
+        return sprintf("%.1f", $num / 1048576.0);
+      }
+    }
+  } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) {
+    return sprintf("%.3f", $num / 1e9); # Convert nanoseconds to seconds
+  } else {
+    return sprintf("%d", $num);
+  }
+}
+
+# Alternate pretty-printed form: 0 maps to "."
+sub UnparseAlt {
+  my $num = shift;
+  if ($num == 0) {
+    return ".";
+  } else {
+    return Unparse($num);
+  }
+}
+
+# Alternate pretty-printed form: 0 maps to ""
+sub HtmlPrintNumber {
+  my $num = shift;
+  if ($num == 0) {
+    return "";
+  } else {
+    return Unparse($num);
+  }
+}
+
+# Return output units
+sub Units {
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    if ($main::opt_inuse_objects || $main::opt_alloc_objects) {
+      return "objects";
+    } else {
+      if ($main::opt_show_bytes) {
+        return "B";
+      } else {
+        return "MB";
+      }
+    }
+  } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) {
+    return "seconds";
+  } else {
+    return "samples";
+  }
+}
+
+##### Profile manipulation code #####
+
+# Generate flattened profile:
+# If count is charged to stack [a,b,c,d], in generated profile,
+# it will be charged to [a]
+sub FlatProfile {
+  my $profile = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    if ($#addrs >= 0) {
+      AddEntry($result, $addrs[0], $count);
+    }
+  }
+  return $result;
+}
+
+# Generate cumulative profile:
+# If count is charged to stack [a,b,c,d], in generated profile,
+# it will be charged to [a], [b], [c], [d]
+sub CumulativeProfile {
+  my $profile = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    foreach my $a (@addrs) {
+      AddEntry($result, $a, $count);
+    }
+  }
+  return $result;
+}
+
+# If the second-youngest PC on the stack is always the same, returns
+# that pc.  Otherwise, returns undef.
+sub IsSecondPcAlwaysTheSame {
+  my $profile = shift;
+
+  my $second_pc = undef;
+  foreach my $k (keys(%{$profile})) {
+    my @addrs = split(/\n/, $k);
+    if ($#addrs < 1) {
+      return undef;
+    }
+    if (not defined $second_pc) {
+      $second_pc = $addrs[1];
+    } else {
+      if ($second_pc ne $addrs[1]) {
+        return undef;
+      }
+    }
+  }
+  return $second_pc;
+}
+
+sub ExtractSymbolNameInlineStack {
+  my $symbols = shift;
+  my $address = shift;
+
+  my @stack = ();
+
+  if (exists $symbols->{$address}) {
+    my @localinlinestack = @{$symbols->{$address}};
+    for (my $i = $#localinlinestack; $i > 0; $i-=3) {
+      my $file = $localinlinestack[$i-1];
+      my $fn = $localinlinestack[$i-0];
+
+      if ($file eq "?" || $file eq ":0") {
+        $file = "??:0";
+      }
+      if ($fn eq '??') {
+        # If we can't get the symbol name, at least use the file information.
+        $fn = $file;
+      }
+      my $suffix = "[inline]";
+      if ($i == 2) {
+        $suffix = "";
+      }
+      push (@stack, $fn.$suffix);
+    }
+  }
+  else {
+    # If we can't get a symbol name, at least fill in the address.
+    push (@stack, $address);
+  }
+
+  return @stack;
+}
+
+sub ExtractSymbolLocation {
+  my $symbols = shift;
+  my $address = shift;
+  # 'addr2line' outputs "??:0" for unknown locations; we do the
+  # same to be consistent.
+  my $location = "??:0:unknown";
+  if (exists $symbols->{$address}) {
+    my $file = $symbols->{$address}->[1];
+    if ($file eq "?") {
+      $file = "??:0"
+    }
+    $location = $file . ":" . $symbols->{$address}->[0];
+  }
+  return $location;
+}
+
+# Extracts a graph of calls.
+sub ExtractCalls {
+  my $symbols = shift;
+  my $profile = shift;
+
+  my $calls = {};
+  while( my ($stack_trace, $count) = each %$profile ) {
+    my @address = split(/\n/, $stack_trace);
+    my $destination = ExtractSymbolLocation($symbols, $address[0]);
+    AddEntry($calls, $destination, $count);
+    for (my $i = 1; $i <= $#address; $i++) {
+      my $source = ExtractSymbolLocation($symbols, $address[$i]);
+      my $call = "$source -> $destination";
+      AddEntry($calls, $call, $count);
+      $destination = $source;
+    }
+  }
+
+  return $calls;
+}
+
+sub FilterFrames {
+  my $symbols = shift;
+  my $profile = shift;
+
+  if ($main::opt_retain eq '' && $main::opt_exclude eq '') {
+    return $profile;
+  }
+
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my @path = ();
+    foreach my $a (@addrs) {
+      my $sym;
+      if (exists($symbols->{$a})) {
+        $sym = $symbols->{$a}->[0];
+      } else {
+        $sym = $a;
+      }
+      if ($main::opt_retain ne '' && $sym !~ m/$main::opt_retain/) {
+        next;
+      }
+      if ($main::opt_exclude ne '' && $sym =~ m/$main::opt_exclude/) {
+        next;
+      }
+      push(@path, $a);
+    }
+    if (scalar(@path) > 0) {
+      my $reduced_path = join("\n", @path);
+      AddEntry($result, $reduced_path, $count);
+    }
+  }
+
+  return $result;
+}
+
+sub PrintCollapsedStacks {
+  my $symbols = shift;
+  my $profile = shift;
+
+  while (my ($stack_trace, $count) = each %$profile) {
+    my @address = split(/\n/, $stack_trace);
+    my @names = reverse ( map { ExtractSymbolNameInlineStack($symbols, $_) } @address );
+    printf("%s %d\n", join(";", @names), $count);
+  }
+}
+
+sub RemoveUninterestingFrames {
+  my $symbols = shift;
+  my $profile = shift;
+
+  # List of function names to skip
+  my %skip = ();
+  my $skip_regexp = 'NOMATCH';
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    foreach my $name ('je_calloc',
+                      'cfree',
+                      'je_malloc',
+                      'newImpl',
+                      'void* newImpl',
+                      'je_free',
+                      'je_memalign',
+                      'je_posix_memalign',
+                      'je_aligned_alloc',
+                      'pvalloc',
+                      'je_valloc',
+                      'je_realloc',
+                      'je_mallocx',
+                      'je_rallocx',
+                      'je_xallocx',
+                      'je_dallocx',
+                      'je_sdallocx',
+                      'je_sdallocx_noflags',
+                      'tc_calloc',
+                      'tc_cfree',
+                      'tc_malloc',
+                      'tc_free',
+                      'tc_memalign',
+                      'tc_posix_memalign',
+                      'tc_pvalloc',
+                      'tc_valloc',
+                      'tc_realloc',
+                      'tc_new',
+                      'tc_delete',
+                      'tc_newarray',
+                      'tc_deletearray',
+                      'tc_new_nothrow',
+                      'tc_newarray_nothrow',
+                      'do_malloc',
+                      '::do_malloc',   # new name -- got moved to an unnamed ns
+                      '::do_malloc_or_cpp_alloc',
+                      'DoSampledAllocation',
+                      'simple_alloc::allocate',
+                      '__malloc_alloc_template::allocate',
+                      '__builtin_delete',
+                      '__builtin_new',
+                      '__builtin_vec_delete',
+                      '__builtin_vec_new',
+                      'operator new',
+                      'operator new[]',
+                      # The entry to our memory-allocation routines on OS X
+                      'malloc_zone_malloc',
+                      'malloc_zone_calloc',
+                      'malloc_zone_valloc',
+                      'malloc_zone_realloc',
+                      'malloc_zone_memalign',
+                      'malloc_zone_free',
+                      # These mark the beginning/end of our custom sections
+                      '__start_google_malloc',
+                      '__stop_google_malloc',
+                      '__start_malloc_hook',
+                      '__stop_malloc_hook') {
+      $skip{$name} = 1;
+      $skip{"_" . $name} = 1;   # Mach (OS X) adds a _ prefix to everything
+    }
+    # TODO: Remove TCMalloc once everything has been
+    # moved into the tcmalloc:: namespace and we have flushed
+    # old code out of the system.
+    $skip_regexp = "TCMalloc|^tcmalloc::";
+  } elsif ($main::profile_type eq 'contention') {
+    foreach my $vname ('base::RecordLockProfileData',
+                       'base::SubmitMutexProfileData',
+                       'base::SubmitSpinLockProfileData',
+                       'Mutex::Unlock',
+                       'Mutex::UnlockSlow',
+                       'Mutex::ReaderUnlock',
+                       'MutexLock::~MutexLock',
+                       'SpinLock::Unlock',
+                       'SpinLock::SlowUnlock',
+                       'SpinLockHolder::~SpinLockHolder') {
+      $skip{$vname} = 1;
+    }
+  } elsif ($main::profile_type eq 'cpu') {
+    # Drop signal handlers used for CPU profile collection
+    # TODO(dpeng): this should not be necessary; it's taken
+    # care of by the general 2nd-pc mechanism below.
+    foreach my $name ('ProfileData::Add',           # historical
+                      'ProfileData::prof_handler',  # historical
+                      'CpuProfiler::prof_handler',
+                      '__FRAME_END__',
+                      '__pthread_sighandler',
+                      '__restore') {
+      $skip{$name} = 1;
+    }
+  } else {
+    # Nothing skipped for unknown types
+  }
+
+  if ($main::profile_type eq 'cpu') {
+    # If all the second-youngest program counters are the same,
+    # this STRONGLY suggests that it is an artifact of measurement,
+    # i.e., stack frames pushed by the CPU profiler signal handler.
+    # Hence, we delete them.
+    # (The topmost PC is read from the signal structure, not from
+    # the stack, so it does not get involved.)
+    while (my $second_pc = IsSecondPcAlwaysTheSame($profile)) {
+      my $result = {};
+      my $func = '';
+      if (exists($symbols->{$second_pc})) {
+        $second_pc = $symbols->{$second_pc}->[0];
+      }
+      print STDERR "Removing $second_pc from all stack traces.\n";
+      foreach my $k (keys(%{$profile})) {
+        my $count = $profile->{$k};
+        my @addrs = split(/\n/, $k);
+        splice @addrs, 1, 1;
+        my $reduced_path = join("\n", @addrs);
+        AddEntry($result, $reduced_path, $count);
+      }
+      $profile = $result;
+    }
+  }
+
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my @path = ();
+    foreach my $a (@addrs) {
+      if (exists($symbols->{$a})) {
+        my $func = $symbols->{$a}->[0];
+        if ($skip{$func} || ($func =~ m/$skip_regexp/)) {
+          # Throw away the portion of the backtrace seen so far, under the
+          # assumption that previous frames were for functions internal to the
+          # allocator.
+          @path = ();
+          next;
+        }
+      }
+      push(@path, $a);
+    }
+    my $reduced_path = join("\n", @path);
+    AddEntry($result, $reduced_path, $count);
+  }
+
+  $result = FilterFrames($symbols, $result);
+
+  return $result;
+}
+
+# Reduce profile to granularity given by user
+sub ReduceProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $result = {};
+  my $fullname_to_shortname_map = {};
+  FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map);
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k);
+    my @path = ();
+    my %seen = ();
+    $seen{''} = 1;      # So that empty keys are skipped
+    foreach my $e (@translated) {
+      # To avoid double-counting due to recursion, skip a stack-trace
+      # entry if it has already been seen
+      if (!$seen{$e}) {
+        $seen{$e} = 1;
+        push(@path, $e);
+      }
+    }
+    my $reduced_path = join("\n", @path);
+    AddEntry($result, $reduced_path, $count);
+  }
+  return $result;
+}
+
+# Does the specified symbol array match the regexp?
+sub SymbolMatches {
+  my $sym = shift;
+  my $re = shift;
+  if (defined($sym)) {
+    for (my $i = 0; $i < $#{$sym}; $i += 3) {
+      if ($sym->[$i] =~ m/$re/ || $sym->[$i+1] =~ m/$re/) {
+        return 1;
+      }
+    }
+  }
+  return 0;
+}
+
+# Focus only on paths involving specified regexps
+sub FocusProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $focus = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    foreach my $a (@addrs) {
+      # Reply if it matches either the address/shortname/fileline
+      if (($a =~ m/$focus/) || SymbolMatches($symbols->{$a}, $focus)) {
+        AddEntry($result, $k, $count);
+        last;
+      }
+    }
+  }
+  return $result;
+}
+
+# Focus only on paths not involving specified regexps
+sub IgnoreProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $ignore = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my $matched = 0;
+    foreach my $a (@addrs) {
+      # Reply if it matches either the address/shortname/fileline
+      if (($a =~ m/$ignore/) || SymbolMatches($symbols->{$a}, $ignore)) {
+        $matched = 1;
+        last;
+      }
+    }
+    if (!$matched) {
+      AddEntry($result, $k, $count);
+    }
+  }
+  return $result;
+}
+
+# Get total count in profile
+sub TotalProfile {
+  my $profile = shift;
+  my $result = 0;
+  foreach my $k (keys(%{$profile})) {
+    $result += $profile->{$k};
+  }
+  return $result;
+}
+
+# Add A to B
+sub AddProfile {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  # add all keys in A
+  foreach my $k (keys(%{$A})) {
+    my $v = $A->{$k};
+    AddEntry($R, $k, $v);
+  }
+  # add all keys in B
+  foreach my $k (keys(%{$B})) {
+    my $v = $B->{$k};
+    AddEntry($R, $k, $v);
+  }
+  return $R;
+}
+
+# Merges symbol maps
+sub MergeSymbols {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  foreach my $k (keys(%{$A})) {
+    $R->{$k} = $A->{$k};
+  }
+  if (defined($B)) {
+    foreach my $k (keys(%{$B})) {
+      $R->{$k} = $B->{$k};
+    }
+  }
+  return $R;
+}
+
+
+# Add A to B
+sub AddPcs {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  # add all keys in A
+  foreach my $k (keys(%{$A})) {
+    $R->{$k} = 1
+  }
+  # add all keys in B
+  foreach my $k (keys(%{$B})) {
+    $R->{$k} = 1
+  }
+  return $R;
+}
+
+# Subtract B from A
+sub SubtractProfile {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  foreach my $k (keys(%{$A})) {
+    my $v = $A->{$k} - GetEntry($B, $k);
+    if ($v < 0 && $main::opt_drop_negative) {
+      $v = 0;
+    }
+    AddEntry($R, $k, $v);
+  }
+  if (!$main::opt_drop_negative) {
+    # Take care of when subtracted profile has more entries
+    foreach my $k (keys(%{$B})) {
+      if (!exists($A->{$k})) {
+        AddEntry($R, $k, 0 - $B->{$k});
+      }
+    }
+  }
+  return $R;
+}
+
+# Get entry from profile; zero if not present
+sub GetEntry {
+  my $profile = shift;
+  my $k = shift;
+  if (exists($profile->{$k})) {
+    return $profile->{$k};
+  } else {
+    return 0;
+  }
+}
+
+# Add entry to specified profile
+sub AddEntry {
+  my $profile = shift;
+  my $k = shift;
+  my $n = shift;
+  if (!exists($profile->{$k})) {
+    $profile->{$k} = 0;
+  }
+  $profile->{$k} += $n;
+}
+
+# Add a stack of entries to specified profile, and add them to the $pcs
+# list.
+sub AddEntries {
+  my $profile = shift;
+  my $pcs = shift;
+  my $stack = shift;
+  my $count = shift;
+  my @k = ();
+
+  foreach my $e (split(/\s+/, $stack)) {
+    my $pc = HexExtend($e);
+    $pcs->{$pc} = 1;
+    push @k, $pc;
+  }
+  AddEntry($profile, (join "\n", @k), $count);
+}
+
+##### Code to profile a server dynamically #####
+
+sub CheckSymbolPage {
+  my $url = SymbolPageURL();
+  my $command = ShellEscape(@URL_FETCHER, $url);
+  open(SYMBOL, "$command |") or error($command);
+  my $line = <SYMBOL>;
+  $line =~ s/\r//g;         # turn windows-looking lines into unix-looking lines
+  close(SYMBOL);
+  unless (defined($line)) {
+    error("$url doesn't exist\n");
+  }
+
+  if ($line =~ /^num_symbols:\s+(\d+)$/) {
+    if ($1 == 0) {
+      error("Stripped binary. No symbols available.\n");
+    }
+  } else {
+    error("Failed to get the number of symbols from $url\n");
+  }
+}
+
+sub IsProfileURL {
+  my $profile_name = shift;
+  if (-f $profile_name) {
+    printf STDERR "Using local file $profile_name.\n";
+    return 0;
+  }
+  return 1;
+}
+
+sub ParseProfileURL {
+  my $profile_name = shift;
+
+  if (!defined($profile_name) || $profile_name eq "") {
+    return ();
+  }
+
+  # Split profile URL - matches all non-empty strings, so no test.
+  $profile_name =~ m,^(https?://)?([^/]+)(.*?)(/|$PROFILES)?$,;
+
+  my $proto = $1 || "http://";
+  my $hostport = $2;
+  my $prefix = $3;
+  my $profile = $4 || "/";
+
+  my $host = $hostport;
+  $host =~ s/:.*//;
+
+  my $baseurl = "$proto$hostport$prefix";
+  return ($host, $baseurl, $profile);
+}
+
+# We fetch symbols from the first profile argument.
+sub SymbolPageURL {
+  my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]);
+  return "$baseURL$SYMBOL_PAGE";
+}
+
+sub FetchProgramName() {
+  my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]);
+  my $url = "$baseURL$PROGRAM_NAME_PAGE";
+  my $command_line = ShellEscape(@URL_FETCHER, $url);
+  open(CMDLINE, "$command_line |") or error($command_line);
+  my $cmdline = <CMDLINE>;
+  $cmdline =~ s/\r//g;   # turn windows-looking lines into unix-looking lines
+  close(CMDLINE);
+  error("Failed to get program name from $url\n") unless defined($cmdline);
+  $cmdline =~ s/\x00.+//;  # Remove argv[1] and latters.
+  $cmdline =~ s!\n!!g;  # Remove LFs.
+  return $cmdline;
+}
+
+# Gee, curl's -L (--location) option isn't reliable at least
+# with its 7.12.3 version.  Curl will forget to post data if
+# there is a redirection.  This function is a workaround for
+# curl.  Redirection happens on borg hosts.
+sub ResolveRedirectionForCurl {
+  my $url = shift;
+  my $command_line = ShellEscape(@URL_FETCHER, "--head", $url);
+  open(CMDLINE, "$command_line |") or error($command_line);
+  while (<CMDLINE>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    if (/^Location: (.*)/) {
+      $url = $1;
+    }
+  }
+  close(CMDLINE);
+  return $url;
+}
+
+# Add a timeout flat to URL_FETCHER.  Returns a new list.
+sub AddFetchTimeout {
+  my $timeout = shift;
+  my @fetcher = @_;
+  if (defined($timeout)) {
+    if (join(" ", @fetcher) =~ m/\bcurl -s/) {
+      push(@fetcher, "--max-time", sprintf("%d", $timeout));
+    } elsif (join(" ", @fetcher) =~ m/\brpcget\b/) {
+      push(@fetcher, sprintf("--deadline=%d", $timeout));
+    }
+  }
+  return @fetcher;
+}
+
+# Reads a symbol map from the file handle name given as $1, returning
+# the resulting symbol map.  Also processes variables relating to symbols.
+# Currently, the only variable processed is 'binary=<value>' which updates
+# $main::prog to have the correct program name.
+sub ReadSymbols {
+  my $in = shift;
+  my $map = {};
+  while (<$in>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    # Removes all the leading zeroes from the symbols, see comment below.
+    if (m/^0x0*([0-9a-f]+)\s+(.+)/) {
+      $map->{$1} = $2;
+    } elsif (m/^---/) {
+      last;
+    } elsif (m/^([a-z][^=]*)=(.*)$/ ) {
+      my ($variable, $value) = ($1, $2);
+      for ($variable, $value) {
+        s/^\s+//;
+        s/\s+$//;
+      }
+      if ($variable eq "binary") {
+        if ($main::prog ne $UNKNOWN_BINARY && $main::prog ne $value) {
+          printf STDERR ("Warning: Mismatched binary name '%s', using '%s'.\n",
+                         $main::prog, $value);
+        }
+        $main::prog = $value;
+      } else {
+        printf STDERR ("Ignoring unknown variable in symbols list: " .
+            "'%s' = '%s'\n", $variable, $value);
+      }
+    }
+  }
+  return $map;
+}
+
+sub URLEncode {
+  my $str = shift;
+  $str =~ s/([^A-Za-z0-9\-_.!~*'()])/ sprintf "%%%02x", ord $1 /eg;
+  return $str;
+}
+
+sub AppendSymbolFilterParams {
+  my $url = shift;
+  my @params = ();
+  if ($main::opt_retain ne '') {
+    push(@params, sprintf("retain=%s", URLEncode($main::opt_retain)));
+  }
+  if ($main::opt_exclude ne '') {
+    push(@params, sprintf("exclude=%s", URLEncode($main::opt_exclude)));
+  }
+  if (scalar @params > 0) {
+    $url = sprintf("%s?%s", $url, join("&", @params));
+  }
+  return $url;
+}
+
+# Fetches and processes symbols to prepare them for use in the profile output
+# code.  If the optional 'symbol_map' arg is not given, fetches symbols from
+# $SYMBOL_PAGE for all PC values found in profile.  Otherwise, the raw symbols
+# are assumed to have already been fetched into 'symbol_map' and are simply
+# extracted and processed.
+sub FetchSymbols {
+  my $pcset = shift;
+  my $symbol_map = shift;
+
+  my %seen = ();
+  my @pcs = grep { !$seen{$_}++ } keys(%$pcset);  # uniq
+
+  if (!defined($symbol_map)) {
+    my $post_data = join("+", sort((map {"0x" . "$_"} @pcs)));
+
+    open(POSTFILE, ">$main::tmpfile_sym");
+    print POSTFILE $post_data;
+    close(POSTFILE);
+
+    my $url = SymbolPageURL();
+
+    my $command_line;
+    if (join(" ", @URL_FETCHER) =~ m/\bcurl -s/) {
+      $url = ResolveRedirectionForCurl($url);
+      $url = AppendSymbolFilterParams($url);
+      $command_line = ShellEscape(@URL_FETCHER, "-d", "\@$main::tmpfile_sym",
+                                  $url);
+    } else {
+      $url = AppendSymbolFilterParams($url);
+      $command_line = (ShellEscape(@URL_FETCHER, "--post", $url)
+                       . " < " . ShellEscape($main::tmpfile_sym));
+    }
+    # We use c++filt in case $SYMBOL_PAGE gives us mangled symbols.
+    my $escaped_cppfilt = ShellEscape($obj_tool_map{"c++filt"});
+    open(SYMBOL, "$command_line | $escaped_cppfilt |") or error($command_line);
+    $symbol_map = ReadSymbols(*SYMBOL{IO});
+    close(SYMBOL);
+  }
+
+  my $symbols = {};
+  foreach my $pc (@pcs) {
+    my $fullname;
+    # For 64 bits binaries, symbols are extracted with 8 leading zeroes.
+    # Then /symbol reads the long symbols in as uint64, and outputs
+    # the result with a "0x%08llx" format which get rid of the zeroes.
+    # By removing all the leading zeroes in both $pc and the symbols from
+    # /symbol, the symbols match and are retrievable from the map.
+    my $shortpc = $pc;
+    $shortpc =~ s/^0*//;
+    # Each line may have a list of names, which includes the function
+    # and also other functions it has inlined.  They are separated (in
+    # PrintSymbolizedProfile), by --, which is illegal in function names.
+    my $fullnames;
+    if (defined($symbol_map->{$shortpc})) {
+      $fullnames = $symbol_map->{$shortpc};
+    } else {
+      $fullnames = "0x" . $pc;  # Just use addresses
+    }
+    my $sym = [];
+    $symbols->{$pc} = $sym;
+    foreach my $fullname (split("--", $fullnames)) {
+      my $name = ShortFunctionName($fullname);
+      push(@{$sym}, $name, "?", $fullname);
+    }
+  }
+  return $symbols;
+}
+
+sub BaseName {
+  my $file_name = shift;
+  $file_name =~ s!^.*/!!;  # Remove directory name
+  return $file_name;
+}
+
+sub MakeProfileBaseName {
+  my ($binary_name, $profile_name) = @_;
+  my ($host, $baseURL, $path) = ParseProfileURL($profile_name);
+  my $binary_shortname = BaseName($binary_name);
+  return sprintf("%s.%s.%s",
+                 $binary_shortname, $main::op_time, $host);
+}
+
+sub FetchDynamicProfile {
+  my $binary_name = shift;
+  my $profile_name = shift;
+  my $fetch_name_only = shift;
+  my $encourage_patience = shift;
+
+  if (!IsProfileURL($profile_name)) {
+    return $profile_name;
+  } else {
+    my ($host, $baseURL, $path) = ParseProfileURL($profile_name);
+    if ($path eq "" || $path eq "/") {
+      # Missing type specifier defaults to cpu-profile
+      $path = $PROFILE_PAGE;
+    }
+
+    my $profile_file = MakeProfileBaseName($binary_name, $profile_name);
+
+    my $url = "$baseURL$path";
+    my $fetch_timeout = undef;
+    if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE/) {
+      if ($path =~ m/[?]/) {
+        $url .= "&";
+      } else {
+        $url .= "?";
+      }
+      $url .= sprintf("seconds=%d", $main::opt_seconds);
+      $fetch_timeout = $main::opt_seconds * 1.01 + 60;
+      # Set $profile_type for consumption by PrintSymbolizedProfile.
+      $main::profile_type = 'cpu';
+    } else {
+      # For non-CPU profiles, we add a type-extension to
+      # the target profile file name.
+      my $suffix = $path;
+      $suffix =~ s,/,.,g;
+      $profile_file .= $suffix;
+      # Set $profile_type for consumption by PrintSymbolizedProfile.
+      if ($path =~ m/$HEAP_PAGE/) {
+        $main::profile_type = 'heap';
+      } elsif ($path =~ m/$GROWTH_PAGE/) {
+        $main::profile_type = 'growth';
+      } elsif ($path =~ m/$CONTENTION_PAGE/) {
+        $main::profile_type = 'contention';
+      }
+    }
+
+    my $profile_dir = $ENV{"JEPROF_TMPDIR"} || ($ENV{HOME} . "/jeprof");
+    if (! -d $profile_dir) {
+      mkdir($profile_dir)
+          || die("Unable to create profile directory $profile_dir: $!\n");
+    }
+    my $tmp_profile = "$profile_dir/.tmp.$profile_file";
+    my $real_profile = "$profile_dir/$profile_file";
+
+    if ($fetch_name_only > 0) {
+      return $real_profile;
+    }
+
+    my @fetcher = AddFetchTimeout($fetch_timeout, @URL_FETCHER);
+    my $cmd = ShellEscape(@fetcher, $url) . " > " . ShellEscape($tmp_profile);
+    if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE|$CENSUSPROFILE_PAGE/){
+      print STDERR "Gathering CPU profile from $url for $main::opt_seconds seconds to\n  ${real_profile}\n";
+      if ($encourage_patience) {
+        print STDERR "Be patient...\n";
+      }
+    } else {
+      print STDERR "Fetching $path profile from $url to\n  ${real_profile}\n";
+    }
+
+    (system($cmd) == 0) || error("Failed to get profile: $cmd: $!\n");
+    (system("mv", $tmp_profile, $real_profile) == 0) || error("Unable to rename profile\n");
+    print STDERR "Wrote profile to $real_profile\n";
+    $main::collected_profile = $real_profile;
+    return $main::collected_profile;
+  }
+}
+
+# Collect profiles in parallel
+sub FetchDynamicProfiles {
+  my $items = scalar(@main::pfile_args);
+  my $levels = log($items) / log(2);
+
+  if ($items == 1) {
+    $main::profile_files[0] = FetchDynamicProfile($main::prog, $main::pfile_args[0], 0, 1);
+  } else {
+    # math rounding issues
+    if ((2 ** $levels) < $items) {
+     $levels++;
+    }
+    my $count = scalar(@main::pfile_args);
+    for (my $i = 0; $i < $count; $i++) {
+      $main::profile_files[$i] = FetchDynamicProfile($main::prog, $main::pfile_args[$i], 1, 0);
+    }
+    print STDERR "Fetching $count profiles, Be patient...\n";
+    FetchDynamicProfilesRecurse($levels, 0, 0);
+    $main::collected_profile = join(" \\\n    ", @main::profile_files);
+  }
+}
+
+# Recursively fork a process to get enough processes
+# collecting profiles
+sub FetchDynamicProfilesRecurse {
+  my $maxlevel = shift;
+  my $level = shift;
+  my $position = shift;
+
+  if (my $pid = fork()) {
+    $position = 0 | ($position << 1);
+    TryCollectProfile($maxlevel, $level, $position);
+    wait;
+  } else {
+    $position = 1 | ($position << 1);
+    TryCollectProfile($maxlevel, $level, $position);
+    cleanup();
+    exit(0);
+  }
+}
+
+# Collect a single profile
+sub TryCollectProfile {
+  my $maxlevel = shift;
+  my $level = shift;
+  my $position = shift;
+
+  if ($level >= ($maxlevel - 1)) {
+    if ($position < scalar(@main::pfile_args)) {
+      FetchDynamicProfile($main::prog, $main::pfile_args[$position], 0, 0);
+    }
+  } else {
+    FetchDynamicProfilesRecurse($maxlevel, $level+1, $position);
+  }
+}
+
+##### Parsing code #####
+
+# Provide a small streaming-read module to handle very large
+# cpu-profile files.  Stream in chunks along a sliding window.
+# Provides an interface to get one 'slot', correctly handling
+# endian-ness differences.  A slot is one 32-bit or 64-bit word
+# (depending on the input profile).  We tell endianness and bit-size
+# for the profile by looking at the first 8 bytes: in cpu profiles,
+# the second slot is always 3 (we'll accept anything that's not 0).
+BEGIN {
+  package CpuProfileStream;
+
+  sub new {
+    my ($class, $file, $fname) = @_;
+    my $self = { file        => $file,
+                 base        => 0,
+                 stride      => 512 * 1024,   # must be a multiple of bitsize/8
+                 slots       => [],
+                 unpack_code => "",           # N for big-endian, V for little
+                 perl_is_64bit => 1,          # matters if profile is 64-bit
+    };
+    bless $self, $class;
+    # Let unittests adjust the stride
+    if ($main::opt_test_stride > 0) {
+      $self->{stride} = $main::opt_test_stride;
+    }
+    # Read the first two slots to figure out bitsize and endianness.
+    my $slots = $self->{slots};
+    my $str;
+    read($self->{file}, $str, 8);
+    # Set the global $address_length based on what we see here.
+    # 8 is 32-bit (8 hexadecimal chars); 16 is 64-bit (16 hexadecimal chars).
+    $address_length = ($str eq (chr(0)x8)) ? 16 : 8;
+    if ($address_length == 8) {
+      if (substr($str, 6, 2) eq chr(0)x2) {
+        $self->{unpack_code} = 'V';  # Little-endian.
+      } elsif (substr($str, 4, 2) eq chr(0)x2) {
+        $self->{unpack_code} = 'N';  # Big-endian
+      } else {
+        ::error("$fname: header size >= 2**16\n");
+      }
+      @$slots = unpack($self->{unpack_code} . "*", $str);
+    } else {
+      # If we're a 64-bit profile, check if we're a 64-bit-capable
+      # perl.  Otherwise, each slot will be represented as a float
+      # instead of an int64, losing precision and making all the
+      # 64-bit addresses wrong.  We won't complain yet, but will
+      # later if we ever see a value that doesn't fit in 32 bits.
+      my $has_q = 0;
+      eval { $has_q = pack("Q", "1") ? 1 : 1; };
+      if (!$has_q) {
+        $self->{perl_is_64bit} = 0;
+      }
+      read($self->{file}, $str, 8);
+      if (substr($str, 4, 4) eq chr(0)x4) {
+        # We'd love to use 'Q', but it's a) not universal, b) not endian-proof.
+        $self->{unpack_code} = 'V';  # Little-endian.
+      } elsif (substr($str, 0, 4) eq chr(0)x4) {
+        $self->{unpack_code} = 'N';  # Big-endian
+      } else {
+        ::error("$fname: header size >= 2**32\n");
+      }
+      my @pair = unpack($self->{unpack_code} . "*", $str);
+      # Since we know one of the pair is 0, it's fine to just add them.
+      @$slots = (0, $pair[0] + $pair[1]);
+    }
+    return $self;
+  }
+
+  # Load more data when we access slots->get(X) which is not yet in memory.
+  sub overflow {
+    my ($self) = @_;
+    my $slots = $self->{slots};
+    $self->{base} += $#$slots + 1;   # skip over data we're replacing
+    my $str;
+    read($self->{file}, $str, $self->{stride});
+    if ($address_length == 8) {      # the 32-bit case
+      # This is the easy case: unpack provides 32-bit unpacking primitives.
+      @$slots = unpack($self->{unpack_code} . "*", $str);
+    } else {
+      # We need to unpack 32 bits at a time and combine.
+      my @b32_values = unpack($self->{unpack_code} . "*", $str);
+      my @b64_values = ();
+      for (my $i = 0; $i < $#b32_values; $i += 2) {
+        # TODO(csilvers): if this is a 32-bit perl, the math below
+        #    could end up in a too-large int, which perl will promote
+        #    to a double, losing necessary precision.  Deal with that.
+        #    Right now, we just die.
+        my ($lo, $hi) = ($b32_values[$i], $b32_values[$i+1]);
+        if ($self->{unpack_code} eq 'N') {    # big-endian
+          ($lo, $hi) = ($hi, $lo);
+        }
+        my $value = $lo + $hi * (2**32);
+        if (!$self->{perl_is_64bit} &&   # check value is exactly represented
+            (($value % (2**32)) != $lo || int($value / (2**32)) != $hi)) {
+          ::error("Need a 64-bit perl to process this 64-bit profile.\n");
+        }
+        push(@b64_values, $value);
+      }
+      @$slots = @b64_values;
+    }
+  }
+
+  # Access the i-th long in the file (logically), or -1 at EOF.
+  sub get {
+    my ($self, $idx) = @_;
+    my $slots = $self->{slots};
+    while ($#$slots >= 0) {
+      if ($idx < $self->{base}) {
+        # The only time we expect a reference to $slots[$i - something]
+        # after referencing $slots[$i] is reading the very first header.
+        # Since $stride > |header|, that shouldn't cause any lookback
+        # errors.  And everything after the header is sequential.
+        print STDERR "Unexpected look-back reading CPU profile";
+        return -1;   # shrug, don't know what better to return
+      } elsif ($idx > $self->{base} + $#$slots) {
+        $self->overflow();
+      } else {
+        return $slots->[$idx - $self->{base}];
+      }
+    }
+    # If we get here, $slots is [], which means we've reached EOF
+    return -1;  # unique since slots is supposed to hold unsigned numbers
+  }
+}
+
+# Reads the top, 'header' section of a profile, and returns the last
+# line of the header, commonly called a 'header line'.  The header
+# section of a profile consists of zero or more 'command' lines that
+# are instructions to jeprof, which jeprof executes when reading the
+# header.  All 'command' lines start with a %.  After the command
+# lines is the 'header line', which is a profile-specific line that
+# indicates what type of profile it is, and perhaps other global
+# information about the profile.  For instance, here's a header line
+# for a heap profile:
+#   heap profile:     53:    38236 [  5525:  1284029] @ heapprofile
+# For historical reasons, the CPU profile does not contain a text-
+# readable header line.  If the profile looks like a CPU profile,
+# this function returns "".  If no header line could be found, this
+# function returns undef.
+#
+# The following commands are recognized:
+#   %warn -- emit the rest of this line to stderr, prefixed by 'WARNING:'
+#
+# The input file should be in binmode.
+sub ReadProfileHeader {
+  local *PROFILE = shift;
+  my $firstchar = "";
+  my $line = "";
+  read(PROFILE, $firstchar, 1);
+  seek(PROFILE, -1, 1);                    # unread the firstchar
+  if ($firstchar !~ /[[:print:]]/) {       # is not a text character
+    return "";
+  }
+  while (defined($line = <PROFILE>)) {
+    $line =~ s/\r//g;   # turn windows-looking lines into unix-looking lines
+    if ($line =~ /^%warn\s+(.*)/) {        # 'warn' command
+      # Note this matches both '%warn blah\n' and '%warn\n'.
+      print STDERR "WARNING: $1\n";        # print the rest of the line
+    } elsif ($line =~ /^%/) {
+      print STDERR "Ignoring unknown command from profile header: $line";
+    } else {
+      # End of commands, must be the header line.
+      return $line;
+    }
+  }
+  return undef;     # got to EOF without seeing a header line
+}
+
+sub IsSymbolizedProfileFile {
+  my $file_name = shift;
+  if (!(-e $file_name) || !(-r $file_name)) {
+    return 0;
+  }
+  # Check if the file contains a symbol-section marker.
+  open(TFILE, "<$file_name");
+  binmode TFILE;
+  my $firstline = ReadProfileHeader(*TFILE);
+  close(TFILE);
+  if (!$firstline) {
+    return 0;
+  }
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+  return $firstline =~ /^--- *$symbol_marker/;
+}
+
+# Parse profile generated by common/profiler.cc and return a reference
+# to a map:
+#      $result->{version}     Version number of profile file
+#      $result->{period}      Sampling period (in microseconds)
+#      $result->{profile}     Profile object
+#      $result->{threads}     Map of thread IDs to profile objects
+#      $result->{map}         Memory map info from profile
+#      $result->{pcs}         Hash of all PC values seen, key is hex address
+sub ReadProfile {
+  my $prog = shift;
+  my $fname = shift;
+  my $result;            # return value
+
+  $CONTENTION_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $contention_marker = $&;
+  $GROWTH_PAGE  =~ m,[^/]+$,;    # matches everything after the last slash
+  my $growth_marker = $&;
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+  $PROFILE_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $profile_marker = $&;
+  $HEAP_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $heap_marker = $&;
+
+  # Look at first line to see if it is a heap or a CPU profile.
+  # CPU profile may start with no header at all, and just binary data
+  # (starting with \0\0\0\0) -- in that case, don't try to read the
+  # whole firstline, since it may be gigabytes(!) of data.
+  open(PROFILE, "<$fname") || error("$fname: $!\n");
+  binmode PROFILE;      # New perls do UTF-8 processing
+  my $header = ReadProfileHeader(*PROFILE);
+  if (!defined($header)) {   # means "at EOF"
+    error("Profile is empty.\n");
+  }
+
+  my $symbols;
+  if ($header =~ m/^--- *$symbol_marker/o) {
+    # Verify that the user asked for a symbolized profile
+    if (!$main::use_symbolized_profile) {
+      # we have both a binary and symbolized profiles, abort
+      error("FATAL ERROR: Symbolized profile\n   $fname\ncannot be used with " .
+            "a binary arg. Try again without passing\n   $prog\n");
+    }
+    # Read the symbol section of the symbolized profile file.
+    $symbols = ReadSymbols(*PROFILE{IO});
+    # Read the next line to get the header for the remaining profile.
+    $header = ReadProfileHeader(*PROFILE) || "";
+  }
+
+  if ($header =~ m/^--- *($heap_marker|$growth_marker)/o) {
+    # Skip "--- ..." line for profile types that have their own headers.
+    $header = ReadProfileHeader(*PROFILE) || "";
+  }
+
+  $main::profile_type = '';
+
+  if ($header =~ m/^heap profile:.*$growth_marker/o) {
+    $main::profile_type = 'growth';
+    $result =  ReadHeapProfile($prog, *PROFILE, $header);
+  } elsif ($header =~ m/^heap profile:/) {
+    $main::profile_type = 'heap';
+    $result =  ReadHeapProfile($prog, *PROFILE, $header);
+  } elsif ($header =~ m/^heap/) {
+    $main::profile_type = 'heap';
+    $result = ReadThreadedHeapProfile($prog, $fname, $header);
+  } elsif ($header =~ m/^--- *$contention_marker/o) {
+    $main::profile_type = 'contention';
+    $result = ReadSynchProfile($prog, *PROFILE);
+  } elsif ($header =~ m/^--- *Stacks:/) {
+    print STDERR
+      "Old format contention profile: mistakenly reports " .
+      "condition variable signals as lock contentions.\n";
+    $main::profile_type = 'contention';
+    $result = ReadSynchProfile($prog, *PROFILE);
+  } elsif ($header =~ m/^--- *$profile_marker/) {
+    # the binary cpu profile data starts immediately after this line
+    $main::profile_type = 'cpu';
+    $result = ReadCPUProfile($prog, $fname, *PROFILE);
+  } else {
+    if (defined($symbols)) {
+      # a symbolized profile contains a format we don't recognize, bail out
+      error("$fname: Cannot recognize profile section after symbols.\n");
+    }
+    # no ascii header present -- must be a CPU profile
+    $main::profile_type

[truncated 1061424 bytes]

ready patch attempt

patch

redis-check-rdb likely remains stuck in a busy-poll loop. A diagnosis report and patch proposal were created locally.

state: readycreated: 2026-05-11 21:22 UTCvalidation: ready
Published session

Prompt

## Plan Pass

You are planning a fixer patch before any edits happen.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. The original pre-edit snapshot is available at `./source` if you need to inspect it. For interpreter processes, plan from the script/application entrypoint evidence first and include the runtime only as a second investigation target unless the evidence proves a runtime bug.

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround. Inspect the relevant code, nearby callers, project contribution docs, and local helper/compat APIs, but do not edit files in this pass.

Return a short markdown plan with these exact sections:

## Problem
## Evidence Confidence
## Proposed Subject
## Patch Plan
## Risks
## Validation

Classify `## Evidence Confidence` as exactly one of `reproduced`, `observed`, or `inferred`. Use `inferred` only for a no-patch diagnosis/report plan unless you can name the extra evidence you will collect before editing; inferred source patches are blocked by Fixer because they are not pull-request-ready. For `observed` source-patch plans, plan to say in the final `## Issue Connection` that the failure was observed by Fixer and not independently reproduced. The plan must explain how the proposed code change addresses the observed issue evidence, call out any prior Fixer patch that should be improved or replaced, reject awkward control flow such as avoidable `goto` if there is a cleaner bounded alternative, name any local helper APIs or maintainer conventions the patch should follow, and keep the intended maintainer-facing explanation clear enough that someone unfamiliar with the local complaint wording can still follow the fix. In `## Validation`, name the reproducible configure/build/test entrypoint you will try from the workspace root before any focused leaf compile or smoke check, and include one bounded independent reproduction attempt for the collected failure signal when it is safe and cheap. Do not plan to claim `reproduced` unless that reproduction command or test can actually show the failure.

## Patch Pass

You are working on a bounded fixer proposal.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. Produce the smallest reasonable patch for the target repository, keep the change upstreamable, prefer the clearest control flow available, and do not keep avoidable `goto` when a simpler structure would read better. Before introducing new file, process, allocation, locking, networking, or platform APIs, inspect nearby code and project contribution docs for existing helpers or compatibility wrappers and use those local patterns unless you can explain why they do not fit. Validate from a reproducible workspace-root entrypoint before falling back to focused leaf commands; if a build or test cannot run, report the exact command, the exact blocker, and any narrower check you ran instead. During validation, also try one bounded independent reproduction of the collected failure signal when it is safe and cheap, such as a failing test, smoke command, perf/strace comparison, or before/after runtime check. Only use `reproduced` if that command or test actually reproduced the failure; otherwise keep `observed` and report the reproduction blocker. The final explanation must connect the observed issue evidence to the actual code change, not just paraphrase the diff. Write like a maintainer is going to read the patch mail cold: explain the bug in plain language, define subsystem-specific jargon the first time you need it, and make the causal story obvious. Explicitly classify evidence confidence as `reproduced`, `observed`, or `inferred`: `reproduced` means you reproduced the failure locally; `observed` means Fixer has direct crash/log/trace evidence but you did not independently reproduce it; `inferred` means the source patch is not pull-request-ready, so do not leave a source diff unless you first gather stronger observed/reproduced evidence; otherwise return a no-patch diagnosis/report. For any source-changing `observed` patch, say explicitly in `## Issue Connection` that the failure was observed by Fixer and not independently reproduced. If you introduce non-obvious state translation, index remapping, or backend split logic, add a short source comment that explains the invariant being preserved.

Start by explaining the likely root cause from the collected perf, strace, and /proc evidence. If you cannot land a safe patch, leave a diagnosis that is strong enough for an upstream bug report.

A previous Fixer patch attempt already exists for this issue. It was generated by Fixer `0.148.0`. Review that patch before changing code, improve it instead of starting blind, and clean up anything awkward or underexplained. In particular, remove avoidable `goto`, tighten the explanation of what the patch is doing, and make the resulting diff feel ready for upstream git review.
- Prior patch: `./prior-best.patch`
- Prior published session: `./prior-best-session.md`

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround. 

Keep the change narrowly scoped and summarize validation clearly.

In every authoring pass, your final response must start with `Subject: <single-line git commit subject>` and then include these markdown sections exactly:

## Commit Message
A short upstream-friendly explanation of what changed and why. Write it in plain language that a maintainer can follow without local complaint context. If you use subsystem jargon, define it immediately.

## Evidence Confidence
Exactly one word: `reproduced`, `observed`, or `inferred`. Use `reproduced` only when you reproduced the failure locally with a command or test, and include that command/test in `## Validation`. Use `observed` when Fixer has direct crash/log/trace evidence but you did not independently reproduce it. If `## Git Add Paths` lists source files for an `observed` patch, `## Issue Connection` must explicitly say the failure was observed by Fixer and not independently reproduced. Use `inferred` for profiler/strace/indirect evidence; inferred responses may be no-patch diagnoses or reports, but inferred source patches are not pull-request-ready until stronger evidence is gathered.

## Issue Connection
Write this as maintainer-facing patch mail, not as local Fixer notes. Cover four things explicitly in readable sentences: the user-visible symptom or the exact collected signal, the code-level cause or the cautious inference from evidence, the specific change you made, and the expected effect. Do not invent a reproducer, command line, crash, or user-visible failure that is not present in the evidence bundle. If the evidence is direct-but-not-reproduced, say it was observed by Fixer and not independently reproduced. If the evidence is indirect and you did not gather stronger evidence, do not leave a source diff; write a no-patch diagnosis/report instead. Include an explicit effect sentence such as `The expected effect is ...`, `This should reduce ...`, or `This prevents ...` for source patches. If the logic is non-obvious in code, mention that you added a short explanatory comment.

## Git Add Paths
List the repo-relative paths that belong in the final patch, one per line. Use `None` only when you intentionally made no source changes. Include intentionally new files, and do not list generated build artifacts.

## Validation
List the checks you ran, or say clearly that you could not run them. Include the independent reproduction command/test and result when `## Evidence Confidence` is `reproduced`; if reproduction was attempted but blocked, name the exact blocker and keep confidence at `observed` or `inferred`.

Before editing, read the plan at `./plan-output.txt` and follow it unless the code proves part of it wrong. If you change course, say so explicitly in the final write-up instead of silently drifting from the plan.

## Review Pass 1

You are reviewing a freshly generated fixer patch.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. Review the first patch pass. The original pre-edit snapshot is available at `./source` for diffing.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround.

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout. The latest author response is at `./patch-output.txt`. Inspect the current code and changed paths like a strict code reviewer. Focus on correctness, regressions, maintainability, awkward control flow such as avoidable `goto`, missing validation, weak or non-gittable commit message text, and explanations that fail to connect the observed issue evidence to the code change. Also review the maintainer experience: the patch mail should be easy to accept upstream, the user-visible bug should be explained in plain language, subsystem-specific jargon should be defined when first used, and any non-obvious state translation or index remapping in code should have a short explanatory comment. Reject patches that omit `## Evidence Confidence`, use anything other than `reproduced`, `observed`, or `inferred`, or sound more certain than that evidence level allows. Reject patches that claim `reproduced` unless `## Validation` names the concrete reproduction command or test and its result. Reject patches that invent a reproducer, command line, crash, or user-visible failure not present in the evidence bundle; indirect profiler or strace evidence may justify a cautious mitigation, but the author must say it is indirect instead of presenting it as a confirmed upstream bug. Reject patches that introduce generic libc/std APIs when nearby code or project docs provide a local compat/helper API for the same job, unless the author explicitly justifies the exception. Reject validation that only reports a leaf object/syntax build when a project-level configure/build/test entrypoint exists and was not attempted; if a full build cannot run, the author must show the exact failed command and blocker.

Do not apply code changes in this pass.

Return a short markdown review report. The first non-empty line must be exactly one of:

RESULT: ok
RESULT: fix-needed

If you choose `RESULT: fix-needed`, add a `## Findings` section with concrete, actionable items.

Response

## Plan Pass

## Problem
The evidence is internally mismatched: the report labels `redis-check-rdb`, but the captured command line, status, threads, and stack show `redis-server` sleeping in `epoll_wait`. The strace cadence is `epoll_wait(..., ~100ms)` followed by `/proc/self/stat` reads at about `0.3%` CPU, which matches normal Redis idle cron memory sampling rather than a demonstrated busy-poll defect.

## Evidence Confidence
inferred

## Proposed Subject
No source patch: diagnose idle Redis cron sampling trace mismatch

## Patch Plan
Do not edit source in this pass. A Redis upstream source patch would be speculative because the evidence is profiler/strace-only, low-CPU, and points to `redis-server`, not `redis-check-rdb`.

Reject the prior Fixer patch: it adds generated `deps/jemalloc/bin/*` files and does not address the sampled Redis event-loop path.

Conventions/helpers found: `aeMain()`/`aeProcessEvents()` drive the event loop, `serverCron()` schedules periodic work, `run_with_period()` throttles cron tasks, and `zmalloc_get_rss()`/`get_proc_stat_ll()` are the local `/proc/self/stat` helpers. Docs checked: `CONTRIBUTING.md`, `README.md`, and `tests/README.md`.

If stronger evidence is collected later, prefer a small subsystem-local change using existing helpers and straightforward bounded control flow; avoid introducing unnecessary `goto` or generic replacements for Redis wrappers.

## Risks
Changing cron timing or RSS sampling without stronger evidence could regress `INFO memory`, allocator diagnostics, and normal idle maintenance. Changing `redis-check-rdb` would not address the sampled process shown in the bundle.

## Validation
Before any focused check, try from the workspace root: `make`, then `make check` if available. If either fails, record the exact command and blocker.

Bounded independent evidence attempt: after a successful build, run a short local idle `src/redis-server` trace or smoke check to compare CPU and syscall cadence for `epoll_wait`, `openat`, `read`, and `close`. Do not claim `reproduced` unless that command demonstrates the reported runaway behavior rather than normal 100 ms cron sampling.

## Patch Pass

Subject: diagnose redis idle cron sampling trace

## Commit Message
No source change is proposed. The collected perf, strace, and `/proc` evidence points at an idle `redis-server` event loop doing normal cron memory sampling, not at a demonstrated `redis-check-rdb` busy-loop bug.

## Evidence Confidence
inferred

## Issue Connection
The likely root cause is an evidence mismatch plus normal Redis idle maintenance. The report is labeled `redis-check-rdb`, but the captured command line is `/usr/bin/redis-server 127.0.0.1:6379`, the sampled process is sleeping in `do_epoll_wait`, CPU is about `0.3%`, and strace shows `epoll_wait` timing out around every 100 ms followed by reads of `/proc/self/stat`.

In Redis, `aeProcessEvents()` drives the event loop, `serverCron()` runs periodic maintenance, `run_with_period(100)` schedules 100 ms work, and `cronUpdateMemoryStats()` calls `zmalloc_get_rss()`, which uses `get_proc_stat_ll()` to read `/proc/self/stat`. That matches the collected trace, so the evidence is indirect and internally mismatched rather than a confirmed upstream defect.

I made no source change. I also rejected the prior Fixer patch because it adds generated `deps/jemalloc/bin/*` files and does not address the observed Redis event-loop/RSS-sampling path. Contribution/style docs checked: `CONTRIBUTING.md`, `README.md`, and `tests/README.md`.

The expected effect is that no speculative upstream diff is produced until stronger evidence shows abnormal CPU use or a real tight loop distinct from normal idle cron sampling.

## Git Add Paths
None

## Validation
`make` from the workspace root completed successfully.

`make check` from the workspace root failed before running tests because the Redis test harness could not find a free port: `Can't find a non busy port in the 21079-21110 range.`

Bounded independent reproduction was attempted but blocked. `timeout 5s strace -f -ttT -e trace=epoll_wait,openat,read,close ./src/redis-server --port 0 --bind 127.0.0.1 --save '' --appendonly no --daemonize no` failed with `PTRACE_TRACEME: Operation not permitted` and `ptrace(PTRACE_SEIZE, 5): Operation not permitted`.

Plain server startup checks also could not keep an idle server running in this sandbox: `--port 0` exits because Redis is configured to listen nowhere, Unix socket startup fails with `setsockopt SO_REUSEADDR: Operation not permitted`, and TCP bind fails with `unable to bind socket, errno: 1`. Confidence remains `inferred`; no source diff was left.

## Review Pass 1

RESULT: ok

No blocking findings.

The workspace has no source diff, and `## Git Add Paths` correctly lists `None`. The author kept `## Evidence Confidence` as `inferred`, which matches the profiler/strace-only evidence and the internal mismatch between the `redis-check-rdb` artifact label and the sampled `redis-server` command line.

The response rejects the prior generated jemalloc-bin patch for the right reason: it does not address the observed Redis event-loop/RSS-sampling path. The explanation is cautious, names the relevant Redis helpers/conventions (`aeProcessEvents`, `serverCron`, `run_with_period`, `cronUpdateMemoryStats`, `zmalloc_get_rss`, `get_proc_stat_ll`), and does not invent a confirmed bug or reproducer.

Validation is acceptable: project-level `make` was attempted and passed; `make check` was attempted and reports the exact port-range blocker; independent reproduction attempts and sandbox blockers are explicitly named.

Diff

--- a/deps/jemalloc/bin/jemalloc-config	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jemalloc-config	2026-05-12 01:21:04.880670582 +0400
@@ -0,0 +1,83 @@
+#!/bin/sh
+
+usage() {
+	cat <<EOF
+Usage:
+  /usr/local/bin/jemalloc-config <option>
+Options:
+  --help | -h  : Print usage.
+  --version    : Print jemalloc version.
+  --revision   : Print shared library revision number.
+  --config     : Print configure options used to build jemalloc.
+  --prefix     : Print installation directory prefix.
+  --bindir     : Print binary installation directory.
+  --datadir    : Print data installation directory.
+  --includedir : Print include installation directory.
+  --libdir     : Print library installation directory.
+  --mandir     : Print manual page installation directory.
+  --cc         : Print compiler used to build jemalloc.
+  --cflags     : Print compiler flags used to build jemalloc.
+  --cppflags   : Print preprocessor flags used to build jemalloc.
+  --cxxflags   : Print C++ compiler flags used to build jemalloc.
+  --ldflags    : Print library flags used to build jemalloc.
+  --libs       : Print libraries jemalloc was linked against.
+EOF
+}
+
+prefix="/usr/local"
+exec_prefix="/usr/local"
+
+case "$1" in
+--help | -h)
+	usage
+	exit 0
+	;;
+--version)
+	echo "5.3.0-0-g0"
+	;;
+--revision)
+	echo "2"
+	;;
+--config)
+	echo "--disable-cxx --with-version=5.3.0-0-g0 --with-lg-quantum=3 --disable-cache-oblivious --with-jemalloc-prefix=je_ CFLAGS= LDFLAGS="
+	;;
+--prefix)
+	echo "/usr/local"
+	;;
+--bindir)
+	echo "/usr/local/bin"
+	;;
+--datadir)
+	echo "/usr/local/share"
+	;;
+--includedir)
+	echo "/usr/local/include"
+	;;
+--libdir)
+	echo "/usr/local/lib"
+	;;
+--mandir)
+	echo "/usr/local/share/man"
+	;;
+--cc)
+	echo "gcc"
+	;;
+--cflags)
+	echo "-std=gnu11 -Wall -Wextra -Wsign-compare -Wundef -Wno-format-zero-length -Wpointer-arith -Wno-missing-braces -Wno-missing-field-initializers -Wno-missing-attributes -pipe -g3 -fvisibility=hidden -Wimplicit-fallthrough -O3 -funroll-loops"
+	;;
+--cppflags)
+	echo "-D_GNU_SOURCE -D_REENTRANT"
+	;;
+--cxxflags)
+	echo ""
+	;;
+--ldflags)
+	echo " "
+	;;
+--libs)
+	echo "-lm  -pthread"
+	;;
+*)
+	usage
+	exit 1
+esac
--- a/deps/jemalloc/bin/jemalloc.sh	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jemalloc.sh	2026-05-12 01:21:04.886367177 +0400
@@ -0,0 +1,9 @@
+#!/bin/sh
+
+prefix=/usr/local
+exec_prefix=/usr/local
+libdir=${exec_prefix}/lib
+
+LD_PRELOAD=${libdir}/libjemalloc.so.2
+export LD_PRELOAD
+exec "$@"
--- a/deps/jemalloc/bin/jeprof	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jeprof	2026-05-12 01:21:04.892529790 +0400
@@ -0,0 +1,5723 @@
+#! /usr/bin/env perl
+
+# Copyright (c) 1998-2007, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+# ---
+# Program for printing the profile generated by common/profiler.cc,
+# or by the heap profiler (common/debugallocation.cc)
+#
+# The profile contains a sequence of entries of the form:
+#       <count> <stack trace>
+# This program parses the profile, and generates user-readable
+# output.
+#
+# Examples:
+#
+# % tools/jeprof "program" "profile"
+#   Enters "interactive" mode
+#
+# % tools/jeprof --text "program" "profile"
+#   Generates one line per procedure
+#
+# % tools/jeprof --gv "program" "profile"
+#   Generates annotated call-graph and displays via "gv"
+#
+# % tools/jeprof --gv --focus=Mutex "program" "profile"
+#   Restrict to code paths that involve an entry that matches "Mutex"
+#
+# % tools/jeprof --gv --focus=Mutex --ignore=string "program" "profile"
+#   Restrict to code paths that involve an entry that matches "Mutex"
+#   and does not match "string"
+#
+# % tools/jeprof --list=IBF_CheckDocid "program" "profile"
+#   Generates disassembly listing of all routines with at least one
+#   sample that match the --list=<regexp> pattern.  The listing is
+#   annotated with the flat and cumulative sample counts at each line.
+#
+# % tools/jeprof --disasm=IBF_CheckDocid "program" "profile"
+#   Generates disassembly listing of all routines with at least one
+#   sample that match the --disasm=<regexp> pattern.  The listing is
+#   annotated with the flat and cumulative sample counts at each PC value.
+#
+# TODO: Use color to indicate files?
+
+use strict;
+use warnings;
+use Getopt::Long;
+use Cwd;
+
+my $JEPROF_VERSION = "5.3.0-0-g0";
+my $PPROF_VERSION = "2.0";
+
+# These are the object tools we use which can come from a
+# user-specified location using --tools, from the JEPROF_TOOLS
+# environment variable, or from the environment.
+my %obj_tool_map = (
+  "objdump" => "objdump",
+  "nm" => "nm",
+  "addr2line" => "addr2line",
+  "c++filt" => "c++filt",
+  ## ConfigureObjTools may add architecture-specific entries:
+  #"nm_pdb" => "nm-pdb",       # for reading windows (PDB-format) executables
+  #"addr2line_pdb" => "addr2line-pdb",                                # ditto
+  #"otool" => "otool",         # equivalent of objdump on OS X
+);
+# NOTE: these are lists, so you can put in commandline flags if you want.
+my @DOT = ("dot");          # leave non-absolute, since it may be in /usr/local
+my @GV = ("gv");
+my @EVINCE = ("evince");    # could also be xpdf or perhaps acroread
+my @KCACHEGRIND = ("kcachegrind");
+my @PS2PDF = ("ps2pdf");
+# These are used for dynamic profiles
+my @URL_FETCHER = ("curl", "-s", "--fail");
+
+# These are the web pages that servers need to support for dynamic profiles
+my $HEAP_PAGE = "/pprof/heap";
+my $PROFILE_PAGE = "/pprof/profile";   # must support cgi-param "?seconds=#"
+my $PMUPROFILE_PAGE = "/pprof/pmuprofile(?:\\?.*)?"; # must support cgi-param
+                                                # ?seconds=#&event=x&period=n
+my $GROWTH_PAGE = "/pprof/growth";
+my $CONTENTION_PAGE = "/pprof/contention";
+my $WALL_PAGE = "/pprof/wall(?:\\?.*)?";  # accepts options like namefilter
+my $FILTEREDPROFILE_PAGE = "/pprof/filteredprofile(?:\\?.*)?";
+my $CENSUSPROFILE_PAGE = "/pprof/censusprofile(?:\\?.*)?"; # must support cgi-param
+                                                       # "?seconds=#",
+                                                       # "?tags_regexp=#" and
+                                                       # "?type=#".
+my $SYMBOL_PAGE = "/pprof/symbol";     # must support symbol lookup via POST
+my $PROGRAM_NAME_PAGE = "/pprof/cmdline";
+
+# These are the web pages that can be named on the command line.
+# All the alternatives must begin with /.
+my $PROFILES = "($HEAP_PAGE|$PROFILE_PAGE|$PMUPROFILE_PAGE|" .
+               "$GROWTH_PAGE|$CONTENTION_PAGE|$WALL_PAGE|" .
+               "$FILTEREDPROFILE_PAGE|$CENSUSPROFILE_PAGE)";
+
+# default binary name
+my $UNKNOWN_BINARY = "(unknown)";
+
+# There is a pervasive dependency on the length (in hex characters,
+# i.e., nibbles) of an address, distinguishing between 32-bit and
+# 64-bit profiles.  To err on the safe size, default to 64-bit here:
+my $address_length = 16;
+
+my $dev_null = "/dev/null";
+if (! -e $dev_null && $^O =~ /MSWin/) {    # $^O is the OS perl was built for
+  $dev_null = "nul";
+}
+
+# A list of paths to search for shared object files
+my @prefix_list = ();
+
+# Special routine name that should not have any symbols.
+# Used as separator to parse "addr2line -i" output.
+my $sep_symbol = '_fini';
+my $sep_address = undef;
+
+##### Argument parsing #####
+
+sub usage_string {
+  return <<EOF;
+Usage:
+jeprof [options] <program> <profiles>
+   <profiles> is a space separated list of profile names.
+jeprof [options] <symbolized-profiles>
+   <symbolized-profiles> is a list of profile files where each file contains
+   the necessary symbol mappings  as well as profile data (likely generated
+   with --raw).
+jeprof [options] <profile>
+   <profile> is a remote form.  Symbols are obtained from host:port$SYMBOL_PAGE
+
+   Each name can be:
+   /path/to/profile        - a path to a profile file
+   host:port[/<service>]   - a location of a service to get profile from
+
+   The /<service> can be $HEAP_PAGE, $PROFILE_PAGE, /pprof/pmuprofile,
+                         $GROWTH_PAGE, $CONTENTION_PAGE, /pprof/wall,
+                         $CENSUSPROFILE_PAGE, or /pprof/filteredprofile.
+   For instance:
+     jeprof http://myserver.com:80$HEAP_PAGE
+   If /<service> is omitted, the service defaults to $PROFILE_PAGE (cpu profiling).
+jeprof --symbols <program>
+   Maps addresses to symbol names.  In this mode, stdin should be a
+   list of library mappings, in the same format as is found in the heap-
+   and cpu-profile files (this loosely matches that of /proc/self/maps
+   on linux), followed by a list of hex addresses to map, one per line.
+
+   For more help with querying remote servers, including how to add the
+   necessary server-side support code, see this filename (or one like it):
+
+   /usr/doc/gperftools-$PPROF_VERSION/pprof_remote_servers.html
+
+Options:
+   --cum               Sort by cumulative data
+   --base=<base>       Subtract <base> from <profile> before display
+   --interactive       Run in interactive mode (interactive "help" gives help) [default]
+   --seconds=<n>       Length of time for dynamic profiles [default=30 secs]
+   --add_lib=<file>    Read additional symbols and line info from the given library
+   --lib_prefix=<dir>  Comma separated list of library path prefixes
+
+Reporting Granularity:
+   --addresses         Report at address level
+   --lines             Report at source line level
+   --functions         Report at function level [default]
+   --files             Report at source file level
+
+Output type:
+   --text              Generate text report
+   --callgrind         Generate callgrind format to stdout
+   --gv                Generate Postscript and display
+   --evince            Generate PDF and display
+   --web               Generate SVG and display
+   --list=<regexp>     Generate source listing of matching routines
+   --disasm=<regexp>   Generate disassembly of matching routines
+   --symbols           Print demangled symbol names found at given addresses
+   --dot               Generate DOT file to stdout
+   --ps                Generate Postcript to stdout
+   --pdf               Generate PDF to stdout
+   --svg               Generate SVG to stdout
+   --gif               Generate GIF to stdout
+   --raw               Generate symbolized jeprof data (useful with remote fetch)
+   --collapsed         Generate collapsed stacks for building flame graphs
+                       (see http://www.brendangregg.com/flamegraphs.html)
+
+Heap-Profile Options:
+   --inuse_space       Display in-use (mega)bytes [default]
+   --inuse_objects     Display in-use objects
+   --alloc_space       Display allocated (mega)bytes
+   --alloc_objects     Display allocated objects
+   --show_bytes        Display space in bytes
+   --drop_negative     Ignore negative differences
+
+Contention-profile options:
+   --total_delay       Display total delay at each region [default]
+   --contentions       Display number of delays at each region
+   --mean_delay        Display mean delay at each region
+
+Call-graph Options:
+   --nodecount=<n>     Show at most so many nodes [default=80]
+   --nodefraction=<f>  Hide nodes below <f>*total [default=.005]
+   --edgefraction=<f>  Hide edges below <f>*total [default=.001]
+   --maxdegree=<n>     Max incoming/outgoing edges per node [default=8]
+   --focus=<regexp>    Focus on backtraces with nodes matching <regexp>
+   --thread=<n>        Show profile for thread <n>
+   --ignore=<regexp>   Ignore backtraces with nodes matching <regexp>
+   --scale=<n>         Set GV scaling [default=0]
+   --heapcheck         Make nodes with non-0 object counts
+                       (i.e. direct leak generators) more visible
+   --retain=<regexp>   Retain only nodes that match <regexp>
+   --exclude=<regexp>  Exclude all nodes that match <regexp>
+
+Miscellaneous:
+   --tools=<prefix or binary:fullpath>[,...]   \$PATH for object tool pathnames
+   --test              Run unit tests
+   --help              This message
+   --version           Version information
+   --debug-syms-by-id  (Linux only) Find debug symbol files by build ID as well as by name
+
+Environment Variables:
+   JEPROF_TMPDIR        Profiles directory. Defaults to \$HOME/jeprof
+   JEPROF_TOOLS         Prefix for object tools pathnames
+
+Examples:
+
+jeprof /bin/ls ls.prof
+                       Enters "interactive" mode
+jeprof --text /bin/ls ls.prof
+                       Outputs one line per procedure
+jeprof --web /bin/ls ls.prof
+                       Displays annotated call-graph in web browser
+jeprof --gv /bin/ls ls.prof
+                       Displays annotated call-graph via 'gv'
+jeprof --gv --focus=Mutex /bin/ls ls.prof
+                       Restricts to code paths including a .*Mutex.* entry
+jeprof --gv --focus=Mutex --ignore=string /bin/ls ls.prof
+                       Code paths including Mutex but not string
+jeprof --list=getdir /bin/ls ls.prof
+                       (Per-line) annotated source listing for getdir()
+jeprof --disasm=getdir /bin/ls ls.prof
+                       (Per-PC) annotated disassembly for getdir()
+
+jeprof http://localhost:1234/
+                       Enters "interactive" mode
+jeprof --text localhost:1234
+                       Outputs one line per procedure for localhost:1234
+jeprof --raw localhost:1234 > ./local.raw
+jeprof --text ./local.raw
+                       Fetches a remote profile for later analysis and then
+                       analyzes it in text mode.
+EOF
+}
+
+sub version_string {
+  return <<EOF
+jeprof (part of jemalloc $JEPROF_VERSION)
+based on pprof (part of gperftools $PPROF_VERSION)
+
+Copyright 1998-2007 Google Inc.
+
+This is BSD licensed software; see the source for copying conditions
+and license information.
+There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A
+PARTICULAR PURPOSE.
+EOF
+}
+
+sub usage {
+  my $msg = shift;
+  print STDERR "$msg\n\n";
+  print STDERR usage_string();
+  print STDERR "\nFATAL ERROR: $msg\n";    # just as a reminder
+  exit(1);
+}
+
+sub Init() {
+  # Setup tmp-file name and handler to clean it up.
+  # We do this in the very beginning so that we can use
+  # error() and cleanup() function anytime here after.
+  $main::tmpfile_sym = "/tmp/jeprof$$.sym";
+  $main::tmpfile_ps = "/tmp/jeprof$$";
+  $main::next_tmpfile = 0;
+  $SIG{'INT'} = \&sighandler;
+
+  # Cache from filename/linenumber to source code
+  $main::source_cache = ();
+
+  $main::opt_help = 0;
+  $main::opt_version = 0;
+
+  $main::opt_cum = 0;
+  $main::opt_base = '';
+  $main::opt_addresses = 0;
+  $main::opt_lines = 0;
+  $main::opt_functions = 0;
+  $main::opt_files = 0;
+  $main::opt_lib_prefix = "";
+
+  $main::opt_text = 0;
+  $main::opt_callgrind = 0;
+  $main::opt_list = "";
+  $main::opt_disasm = "";
+  $main::opt_symbols = 0;
+  $main::opt_gv = 0;
+  $main::opt_evince = 0;
+  $main::opt_web = 0;
+  $main::opt_dot = 0;
+  $main::opt_ps = 0;
+  $main::opt_pdf = 0;
+  $main::opt_gif = 0;
+  $main::opt_svg = 0;
+  $main::opt_raw = 0;
+  $main::opt_collapsed = 0;
+
+  $main::opt_nodecount = 80;
+  $main::opt_nodefraction = 0.005;
+  $main::opt_edgefraction = 0.001;
+  $main::opt_maxdegree = 8;
+  $main::opt_focus = '';
+  $main::opt_thread = undef;
+  $main::opt_ignore = '';
+  $main::opt_scale = 0;
+  $main::opt_heapcheck = 0;
+  $main::opt_retain = '';
+  $main::opt_exclude = '';
+  $main::opt_seconds = 30;
+  $main::opt_lib = "";
+
+  $main::opt_inuse_space   = 0;
+  $main::opt_inuse_objects = 0;
+  $main::opt_alloc_space   = 0;
+  $main::opt_alloc_objects = 0;
+  $main::opt_show_bytes    = 0;
+  $main::opt_drop_negative = 0;
+  $main::opt_interactive   = 0;
+
+  $main::opt_total_delay = 0;
+  $main::opt_contentions = 0;
+  $main::opt_mean_delay = 0;
+
+  $main::opt_tools   = "";
+  $main::opt_debug   = 0;
+  $main::opt_test    = 0;
+  $main::opt_debug_syms_by_id = 0;
+
+  # These are undocumented flags used only by unittests.
+  $main::opt_test_stride = 0;
+
+  # Are we using $SYMBOL_PAGE?
+  $main::use_symbol_page = 0;
+
+  # Files returned by TempName.
+  %main::tempnames = ();
+
+  # Type of profile we are dealing with
+  # Supported types:
+  #     cpu
+  #     heap
+  #     growth
+  #     contention
+  $main::profile_type = '';     # Empty type means "unknown"
+
+  GetOptions("help!"          => \$main::opt_help,
+             "version!"       => \$main::opt_version,
+             "cum!"           => \$main::opt_cum,
+             "base=s"         => \$main::opt_base,
+             "seconds=i"      => \$main::opt_seconds,
+             "add_lib=s"      => \$main::opt_lib,
+             "lib_prefix=s"   => \$main::opt_lib_prefix,
+             "functions!"     => \$main::opt_functions,
+             "lines!"         => \$main::opt_lines,
+             "addresses!"     => \$main::opt_addresses,
+             "files!"         => \$main::opt_files,
+             "text!"          => \$main::opt_text,
+             "callgrind!"     => \$main::opt_callgrind,
+             "list=s"         => \$main::opt_list,
+             "disasm=s"       => \$main::opt_disasm,
+             "symbols!"       => \$main::opt_symbols,
+             "gv!"            => \$main::opt_gv,
+             "evince!"        => \$main::opt_evince,
+             "web!"           => \$main::opt_web,
+             "dot!"           => \$main::opt_dot,
+             "ps!"            => \$main::opt_ps,
+             "pdf!"           => \$main::opt_pdf,
+             "svg!"           => \$main::opt_svg,
+             "gif!"           => \$main::opt_gif,
+             "raw!"           => \$main::opt_raw,
+             "collapsed!"     => \$main::opt_collapsed,
+             "interactive!"   => \$main::opt_interactive,
+             "nodecount=i"    => \$main::opt_nodecount,
+             "nodefraction=f" => \$main::opt_nodefraction,
+             "edgefraction=f" => \$main::opt_edgefraction,
+             "maxdegree=i"    => \$main::opt_maxdegree,
+             "focus=s"        => \$main::opt_focus,
+             "thread=s"       => \$main::opt_thread,
+             "ignore=s"       => \$main::opt_ignore,
+             "scale=i"        => \$main::opt_scale,
+             "heapcheck"      => \$main::opt_heapcheck,
+             "retain=s"       => \$main::opt_retain,
+             "exclude=s"      => \$main::opt_exclude,
+             "inuse_space!"   => \$main::opt_inuse_space,
+             "inuse_objects!" => \$main::opt_inuse_objects,
+             "alloc_space!"   => \$main::opt_alloc_space,
+             "alloc_objects!" => \$main::opt_alloc_objects,
+             "show_bytes!"    => \$main::opt_show_bytes,
+             "drop_negative!" => \$main::opt_drop_negative,
+             "total_delay!"   => \$main::opt_total_delay,
+             "contentions!"   => \$main::opt_contentions,
+             "mean_delay!"    => \$main::opt_mean_delay,
+             "tools=s"        => \$main::opt_tools,
+             "test!"          => \$main::opt_test,
+             "debug!"         => \$main::opt_debug,
+             "debug-syms-by-id!" => \$main::opt_debug_syms_by_id,
+             # Undocumented flags used only by unittests:
+             "test_stride=i"  => \$main::opt_test_stride,
+      ) || usage("Invalid option(s)");
+
+  # Deal with the standard --help and --version
+  if ($main::opt_help) {
+    print usage_string();
+    exit(0);
+  }
+
+  if ($main::opt_version) {
+    print version_string();
+    exit(0);
+  }
+
+  # Disassembly/listing/symbols mode requires address-level info
+  if ($main::opt_disasm || $main::opt_list || $main::opt_symbols) {
+    $main::opt_functions = 0;
+    $main::opt_lines = 0;
+    $main::opt_addresses = 1;
+    $main::opt_files = 0;
+  }
+
+  # Check heap-profiling flags
+  if ($main::opt_inuse_space +
+      $main::opt_inuse_objects +
+      $main::opt_alloc_space +
+      $main::opt_alloc_objects > 1) {
+    usage("Specify at most on of --inuse/--alloc options");
+  }
+
+  # Check output granularities
+  my $grains =
+      $main::opt_functions +
+      $main::opt_lines +
+      $main::opt_addresses +
+      $main::opt_files +
+      0;
+  if ($grains > 1) {
+    usage("Only specify one output granularity option");
+  }
+  if ($grains == 0) {
+    $main::opt_functions = 1;
+  }
+
+  # Check output modes
+  my $modes =
+      $main::opt_text +
+      $main::opt_callgrind +
+      ($main::opt_list eq '' ? 0 : 1) +
+      ($main::opt_disasm eq '' ? 0 : 1) +
+      ($main::opt_symbols == 0 ? 0 : 1) +
+      $main::opt_gv +
+      $main::opt_evince +
+      $main::opt_web +
+      $main::opt_dot +
+      $main::opt_ps +
+      $main::opt_pdf +
+      $main::opt_svg +
+      $main::opt_gif +
+      $main::opt_raw +
+      $main::opt_collapsed +
+      $main::opt_interactive +
+      0;
+  if ($modes > 1) {
+    usage("Only specify one output mode");
+  }
+  if ($modes == 0) {
+    if (-t STDOUT) {  # If STDOUT is a tty, activate interactive mode
+      $main::opt_interactive = 1;
+    } else {
+      $main::opt_text = 1;
+    }
+  }
+
+  if ($main::opt_test) {
+    RunUnitTests();
+    # Should not return
+    exit(1);
+  }
+
+  # Binary name and profile arguments list
+  $main::prog = "";
+  @main::pfile_args = ();
+
+  # Remote profiling without a binary (using $SYMBOL_PAGE instead)
+  if (@ARGV > 0) {
+    if (IsProfileURL($ARGV[0])) {
+      $main::use_symbol_page = 1;
+    } elsif (IsSymbolizedProfileFile($ARGV[0])) {
+      $main::use_symbolized_profile = 1;
+      $main::prog = $UNKNOWN_BINARY;  # will be set later from the profile file
+    }
+  }
+
+  if ($main::use_symbol_page || $main::use_symbolized_profile) {
+    # We don't need a binary!
+    my %disabled = ('--lines' => $main::opt_lines,
+                    '--disasm' => $main::opt_disasm);
+    for my $option (keys %disabled) {
+      usage("$option cannot be used without a binary") if $disabled{$option};
+    }
+    # Set $main::prog later...
+    scalar(@ARGV) || usage("Did not specify profile file");
+  } elsif ($main::opt_symbols) {
+    # --symbols needs a binary-name (to run nm on, etc) but not profiles
+    $main::prog = shift(@ARGV) || usage("Did not specify program");
+  } else {
+    $main::prog = shift(@ARGV) || usage("Did not specify program");
+    scalar(@ARGV) || usage("Did not specify profile file");
+  }
+
+  # Parse profile file/location arguments
+  foreach my $farg (@ARGV) {
+    if ($farg =~ m/(.*)\@([0-9]+)(|\/.*)$/ ) {
+      my $machine = $1;
+      my $num_machines = $2;
+      my $path = $3;
+      for (my $i = 0; $i < $num_machines; $i++) {
+        unshift(@main::pfile_args, "$i.$machine$path");
+      }
+    } else {
+      unshift(@main::pfile_args, $farg);
+    }
+  }
+
+  if ($main::use_symbol_page) {
+    unless (IsProfileURL($main::pfile_args[0])) {
+      error("The first profile should be a remote form to use $SYMBOL_PAGE\n");
+    }
+    CheckSymbolPage();
+    $main::prog = FetchProgramName();
+  } elsif (!$main::use_symbolized_profile) {  # may not need objtools!
+    ConfigureObjTools($main::prog)
+  }
+
+  # Break the opt_lib_prefix into the prefix_list array
+  @prefix_list = split (',', $main::opt_lib_prefix);
+
+  # Remove trailing / from the prefixes, in the list to prevent
+  # searching things like /my/path//lib/mylib.so
+  foreach (@prefix_list) {
+    s|/+$||;
+  }
+
+  # Flag to prevent us from trying over and over to use
+  #  elfutils if it's not installed (used only with
+  #  --debug-syms-by-id option).
+  $main::gave_up_on_elfutils = 0;
+}
+
+sub FilterAndPrint {
+  my ($profile, $symbols, $libs, $thread) = @_;
+
+  # Get total data in profile
+  my $total = TotalProfile($profile);
+
+  # Remove uniniteresting stack items
+  $profile = RemoveUninterestingFrames($symbols, $profile);
+
+  # Focus?
+  if ($main::opt_focus ne '') {
+    $profile = FocusProfile($symbols, $profile, $main::opt_focus);
+  }
+
+  # Ignore?
+  if ($main::opt_ignore ne '') {
+    $profile = IgnoreProfile($symbols, $profile, $main::opt_ignore);
+  }
+
+  my $calls = ExtractCalls($symbols, $profile);
+
+  # Reduce profiles to required output granularity, and also clean
+  # each stack trace so a given entry exists at most once.
+  my $reduced = ReduceProfile($symbols, $profile);
+
+  # Get derived profiles
+  my $flat = FlatProfile($reduced);
+  my $cumulative = CumulativeProfile($reduced);
+
+  # Print
+  if (!$main::opt_interactive) {
+    if ($main::opt_disasm) {
+      PrintDisassembly($libs, $flat, $cumulative, $main::opt_disasm);
+    } elsif ($main::opt_list) {
+      PrintListing($total, $libs, $flat, $cumulative, $main::opt_list, 0);
+    } elsif ($main::opt_text) {
+      # Make sure the output is empty when have nothing to report
+      # (only matters when --heapcheck is given but we must be
+      # compatible with old branches that did not pass --heapcheck always):
+      if ($total != 0) {
+        printf("Total%s: %s %s\n",
+               (defined($thread) ? " (t$thread)" : ""),
+               Unparse($total), Units());
+      }
+      PrintText($symbols, $flat, $cumulative, -1);
+    } elsif ($main::opt_raw) {
+      PrintSymbolizedProfile($symbols, $profile, $main::prog);
+    } elsif ($main::opt_collapsed) {
+      PrintCollapsedStacks($symbols, $profile);
+    } elsif ($main::opt_callgrind) {
+      PrintCallgrind($calls);
+    } else {
+      if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) {
+        if ($main::opt_gv) {
+          RunGV(TempName($main::next_tmpfile, "ps"), "");
+        } elsif ($main::opt_evince) {
+          RunEvince(TempName($main::next_tmpfile, "pdf"), "");
+        } elsif ($main::opt_web) {
+          my $tmp = TempName($main::next_tmpfile, "svg");
+          RunWeb($tmp);
+          # The command we run might hand the file name off
+          # to an already running browser instance and then exit.
+          # Normally, we'd remove $tmp on exit (right now),
+          # but fork a child to remove $tmp a little later, so that the
+          # browser has time to load it first.
+          delete $main::tempnames{$tmp};
+          if (fork() == 0) {
+            sleep 5;
+            unlink($tmp);
+            exit(0);
+          }
+        }
+      } else {
+        cleanup();
+        exit(1);
+      }
+    }
+  } else {
+    InteractiveMode($profile, $symbols, $libs, $total);
+  }
+}
+
+sub Main() {
+  Init();
+  $main::collected_profile = undef;
+  @main::profile_files = ();
+  $main::op_time = time();
+
+  # Printing symbols is special and requires a lot less info that most.
+  if ($main::opt_symbols) {
+    PrintSymbols(*STDIN);   # Get /proc/maps and symbols output from stdin
+    return;
+  }
+
+  # Fetch all profile data
+  FetchDynamicProfiles();
+
+  # this will hold symbols that we read from the profile files
+  my $symbol_map = {};
+
+  # Read one profile, pick the last item on the list
+  my $data = ReadProfile($main::prog, pop(@main::profile_files));
+  my $profile = $data->{profile};
+  my $pcs = $data->{pcs};
+  my $libs = $data->{libs};   # Info about main program and shared libraries
+  $symbol_map = MergeSymbols($symbol_map, $data->{symbols});
+
+  # Add additional profiles, if available.
+  if (scalar(@main::profile_files) > 0) {
+    foreach my $pname (@main::profile_files) {
+      my $data2 = ReadProfile($main::prog, $pname);
+      $profile = AddProfile($profile, $data2->{profile});
+      $pcs = AddPcs($pcs, $data2->{pcs});
+      $symbol_map = MergeSymbols($symbol_map, $data2->{symbols});
+    }
+  }
+
+  # Subtract base from profile, if specified
+  if ($main::opt_base ne '') {
+    my $base = ReadProfile($main::prog, $main::opt_base);
+    $profile = SubtractProfile($profile, $base->{profile});
+    $pcs = AddPcs($pcs, $base->{pcs});
+    $symbol_map = MergeSymbols($symbol_map, $base->{symbols});
+  }
+
+  # Collect symbols
+  my $symbols;
+  if ($main::use_symbolized_profile) {
+    $symbols = FetchSymbols($pcs, $symbol_map);
+  } elsif ($main::use_symbol_page) {
+    $symbols = FetchSymbols($pcs);
+  } else {
+    # TODO(csilvers): $libs uses the /proc/self/maps data from profile1,
+    # which may differ from the data from subsequent profiles, especially
+    # if they were run on different machines.  Use appropriate libs for
+    # each pc somehow.
+    $symbols = ExtractSymbols($libs, $pcs);
+  }
+
+  if (!defined($main::opt_thread)) {
+    FilterAndPrint($profile, $symbols, $libs);
+  }
+  if (defined($data->{threads})) {
+    foreach my $thread (sort { $a <=> $b } keys(%{$data->{threads}})) {
+      if (defined($main::opt_thread) &&
+          ($main::opt_thread eq '*' || $main::opt_thread == $thread)) {
+        my $thread_profile = $data->{threads}{$thread};
+        FilterAndPrint($thread_profile, $symbols, $libs, $thread);
+      }
+    }
+  }
+
+  cleanup();
+  exit(0);
+}
+
+##### Entry Point #####
+
+Main();
+
+# Temporary code to detect if we're running on a Goobuntu system.
+# These systems don't have the right stuff installed for the special
+# Readline libraries to work, so as a temporary workaround, we default
+# to using the normal stdio code, rather than the fancier readline-based
+# code
+sub ReadlineMightFail {
+  if (-e '/lib/libtermcap.so.2') {
+    return 0;  # libtermcap exists, so readline should be okay
+  } else {
+    return 1;
+  }
+}
+
+sub RunGV {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  if (!system(ShellEscape(@GV, "--version") . " >$dev_null 2>&1")) {
+    # Options using double dash are supported by this gv version.
+    # Also, turn on noantialias to better handle bug in gv for
+    # postscript files with large dimensions.
+    # TODO: Maybe we should not pass the --noantialias flag
+    # if the gv version is known to work properly without the flag.
+    system(ShellEscape(@GV, "--scale=$main::opt_scale", "--noantialias", $fname)
+           . $bg);
+  } else {
+    # Old gv version - only supports options that use single dash.
+    print STDERR ShellEscape(@GV, "-scale", $main::opt_scale) . "\n";
+    system(ShellEscape(@GV, "-scale", "$main::opt_scale", $fname) . $bg);
+  }
+}
+
+sub RunEvince {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  system(ShellEscape(@EVINCE, $fname) . $bg);
+}
+
+sub RunWeb {
+  my $fname = shift;
+  print STDERR "Loading web page file:///$fname\n";
+
+  if (`uname` =~ /Darwin/) {
+    # OS X: open will use standard preference for SVG files.
+    system("/usr/bin/open", $fname);
+    return;
+  }
+
+  # Some kind of Unix; try generic symlinks, then specific browsers.
+  # (Stop once we find one.)
+  # Works best if the browser is already running.
+  my @alt = (
+    "/etc/alternatives/gnome-www-browser",
+    "/etc/alternatives/x-www-browser",
+    "google-chrome",
+    "firefox",
+  );
+  foreach my $b (@alt) {
+    if (system($b, $fname) == 0) {
+      return;
+    }
+  }
+
+  print STDERR "Could not load web browser.\n";
+}
+
+sub RunKcachegrind {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  print STDERR "Starting '@KCACHEGRIND " . $fname . $bg . "'\n";
+  system(ShellEscape(@KCACHEGRIND, $fname) . $bg);
+}
+
+
+##### Interactive helper routines #####
+
+sub InteractiveMode {
+  $| = 1;  # Make output unbuffered for interactive mode
+  my ($orig_profile, $symbols, $libs, $total) = @_;
+
+  print STDERR "Welcome to jeprof!  For help, type 'help'.\n";
+
+  # Use ReadLine if it's installed and input comes from a console.
+  if ( -t STDIN &&
+       !ReadlineMightFail() &&
+       defined(eval {require Term::ReadLine}) ) {
+    my $term = new Term::ReadLine 'jeprof';
+    while ( defined ($_ = $term->readline('(jeprof) '))) {
+      $term->addhistory($_) if /\S/;
+      if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) {
+        last;    # exit when we get an interactive command to quit
+      }
+    }
+  } else {       # don't have readline
+    while (1) {
+      print STDERR "(jeprof) ";
+      $_ = <STDIN>;
+      last if ! defined $_ ;
+      s/\r//g;         # turn windows-looking lines into unix-looking lines
+
+      # Save some flags that might be reset by InteractiveCommand()
+      my $save_opt_lines = $main::opt_lines;
+
+      if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) {
+        last;    # exit when we get an interactive command to quit
+      }
+
+      # Restore flags
+      $main::opt_lines = $save_opt_lines;
+    }
+  }
+}
+
+# Takes two args: orig profile, and command to run.
+# Returns 1 if we should keep going, or 0 if we were asked to quit
+sub InteractiveCommand {
+  my($orig_profile, $symbols, $libs, $total, $command) = @_;
+  $_ = $command;                # just to make future m//'s easier
+  if (!defined($_)) {
+    print STDERR "\n";
+    return 0;
+  }
+  if (m/^\s*quit/) {
+    return 0;
+  }
+  if (m/^\s*help/) {
+    InteractiveHelpMessage();
+    return 1;
+  }
+  # Clear all the mode options -- mode is controlled by "$command"
+  $main::opt_text = 0;
+  $main::opt_callgrind = 0;
+  $main::opt_disasm = 0;
+  $main::opt_list = 0;
+  $main::opt_gv = 0;
+  $main::opt_evince = 0;
+  $main::opt_cum = 0;
+
+  if (m/^\s*(text|top)(\d*)\s*(.*)/) {
+    $main::opt_text = 1;
+
+    my $line_limit = ($2 ne "") ? int($2) : 10;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($3);
+
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintText($symbols, $flat, $cumulative, $line_limit);
+    return 1;
+  }
+  if (m/^\s*callgrind\s*([^ \n]*)/) {
+    $main::opt_callgrind = 1;
+
+    # Get derived profiles
+    my $calls = ExtractCalls($symbols, $orig_profile);
+    my $filename = $1;
+    if ( $1 eq '' ) {
+      $filename = TempName($main::next_tmpfile, "callgrind");
+    }
+    PrintCallgrind($calls, $filename);
+    if ( $1 eq '' ) {
+      RunKcachegrind($filename, " & ");
+      $main::next_tmpfile++;
+    }
+
+    return 1;
+  }
+  if (m/^\s*(web)?list\s*(.+)/) {
+    my $html = (defined($1) && ($1 eq "web"));
+    $main::opt_list = 1;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($2);
+
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintListing($total, $libs, $flat, $cumulative, $routine, $html);
+    return 1;
+  }
+  if (m/^\s*disasm\s*(.+)/) {
+    $main::opt_disasm = 1;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($1);
+
+    # Process current profile to account for various settings
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintDisassembly($libs, $flat, $cumulative, $routine);
+    return 1;
+  }
+  if (m/^\s*(gv|web|evince)\s*(.*)/) {
+    $main::opt_gv = 0;
+    $main::opt_evince = 0;
+    $main::opt_web = 0;
+    if ($1 eq "gv") {
+      $main::opt_gv = 1;
+    } elsif ($1 eq "evince") {
+      $main::opt_evince = 1;
+    } elsif ($1 eq "web") {
+      $main::opt_web = 1;
+    }
+
+    my $focus;
+    my $ignore;
+    ($focus, $ignore) = ParseInteractiveArgs($2);
+
+    # Process current profile to account for various settings
+    my $profile = ProcessProfile($total, $orig_profile, $symbols,
+                                 $focus, $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) {
+      if ($main::opt_gv) {
+        RunGV(TempName($main::next_tmpfile, "ps"), " &");
+      } elsif ($main::opt_evince) {
+        RunEvince(TempName($main::next_tmpfile, "pdf"), " &");
+      } elsif ($main::opt_web) {
+        RunWeb(TempName($main::next_tmpfile, "svg"));
+      }
+      $main::next_tmpfile++;
+    }
+    return 1;
+  }
+  if (m/^\s*$/) {
+    return 1;
+  }
+  print STDERR "Unknown command: try 'help'.\n";
+  return 1;
+}
+
+
+sub ProcessProfile {
+  my $total_count = shift;
+  my $orig_profile = shift;
+  my $symbols = shift;
+  my $focus = shift;
+  my $ignore = shift;
+
+  # Process current profile to account for various settings
+  my $profile = $orig_profile;
+  printf("Total: %s %s\n", Unparse($total_count), Units());
+  if ($focus ne '') {
+    $profile = FocusProfile($symbols, $profile, $focus);
+    my $focus_count = TotalProfile($profile);
+    printf("After focusing on '%s': %s %s of %s (%0.1f%%)\n",
+           $focus,
+           Unparse($focus_count), Units(),
+           Unparse($total_count), ($focus_count*100.0) / $total_count);
+  }
+  if ($ignore ne '') {
+    $profile = IgnoreProfile($symbols, $profile, $ignore);
+    my $ignore_count = TotalProfile($profile);
+    printf("After ignoring '%s': %s %s of %s (%0.1f%%)\n",
+           $ignore,
+           Unparse($ignore_count), Units(),
+           Unparse($total_count),
+           ($ignore_count*100.0) / $total_count);
+  }
+
+  return $profile;
+}
+
+sub InteractiveHelpMessage {
+  print STDERR <<ENDOFHELP;
+Interactive jeprof mode
+
+Commands:
+  gv
+  gv [focus] [-ignore1] [-ignore2]
+      Show graphical hierarchical display of current profile.  Without
+      any arguments, shows all samples in the profile.  With the optional
+      "focus" argument, restricts the samples shown to just those where
+      the "focus" regular expression matches a routine name on the stack
+      trace.
+
+  web
+  web [focus] [-ignore1] [-ignore2]
+      Like GV, but displays profile in your web browser instead of using
+      Ghostview. Works best if your web browser is already running.
+      To change the browser that gets used:
+      On Linux, set the /etc/alternatives/gnome-www-browser symlink.
+      On OS X, change the Finder association for SVG files.
+
+  list [routine_regexp] [-ignore1] [-ignore2]
+      Show source listing of routines whose names match "routine_regexp"
+
+  weblist [routine_regexp] [-ignore1] [-ignore2]
+     Displays a source listing of routines whose names match "routine_regexp"
+     in a web browser.  You can click on source lines to view the
+     corresponding disassembly.
+
+  top [--cum] [-ignore1] [-ignore2]
+  top20 [--cum] [-ignore1] [-ignore2]
+  top37 [--cum] [-ignore1] [-ignore2]
+      Show top lines ordered by flat profile count, or cumulative count
+      if --cum is specified.  If a number is present after 'top', the
+      top K routines will be shown (defaults to showing the top 10)
+
+  disasm [routine_regexp] [-ignore1] [-ignore2]
+      Show disassembly of routines whose names match "routine_regexp",
+      annotated with sample counts.
+
+  callgrind
+  callgrind [filename]
+      Generates callgrind file. If no filename is given, kcachegrind is called.
+
+  help - This listing
+  quit or ^D - End jeprof
+
+For commands that accept optional -ignore tags, samples where any routine in
+the stack trace matches the regular expression in any of the -ignore
+parameters will be ignored.
+
+Further pprof details are available at this location (or one similar):
+
+ /usr/doc/gperftools-$PPROF_VERSION/cpu_profiler.html
+ /usr/doc/gperftools-$PPROF_VERSION/heap_profiler.html
+
+ENDOFHELP
+}
+sub ParseInteractiveArgs {
+  my $args = shift;
+  my $focus = "";
+  my $ignore = "";
+  my @x = split(/ +/, $args);
+  foreach $a (@x) {
+    if ($a =~ m/^(--|-)lines$/) {
+      $main::opt_lines = 1;
+    } elsif ($a =~ m/^(--|-)cum$/) {
+      $main::opt_cum = 1;
+    } elsif ($a =~ m/^-(.*)/) {
+      $ignore .= (($ignore ne "") ? "|" : "" ) . $1;
+    } else {
+      $focus .= (($focus ne "") ? "|" : "" ) . $a;
+    }
+  }
+  if ($ignore ne "") {
+    print STDERR "Ignoring samples in call stacks that match '$ignore'\n";
+  }
+  return ($focus, $ignore);
+}
+
+##### Output code #####
+
+sub TempName {
+  my $fnum = shift;
+  my $ext = shift;
+  my $file = "$main::tmpfile_ps.$fnum.$ext";
+  $main::tempnames{$file} = 1;
+  return $file;
+}
+
+# Print profile data in packed binary format (64-bit) to standard out
+sub PrintProfileData {
+  my $profile = shift;
+
+  # print header (64-bit style)
+  # (zero) (header-size) (version) (sample-period) (zero)
+  print pack('L*', 0, 0, 3, 0, 0, 0, 1, 0, 0, 0);
+
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    if ($#addrs >= 0) {
+      my $depth = $#addrs + 1;
+      # int(foo / 2**32) is the only reliable way to get rid of bottom
+      # 32 bits on both 32- and 64-bit systems.
+      print pack('L*', $count & 0xFFFFFFFF, int($count / 2**32));
+      print pack('L*', $depth & 0xFFFFFFFF, int($depth / 2**32));
+
+      foreach my $full_addr (@addrs) {
+        my $addr = $full_addr;
+        $addr =~ s/0x0*//;  # strip off leading 0x, zeroes
+        if (length($addr) > 16) {
+          print STDERR "Invalid address in profile: $full_addr\n";
+          next;
+        }
+        my $low_addr = substr($addr, -8);       # get last 8 hex chars
+        my $high_addr = substr($addr, -16, 8);  # get up to 8 more hex chars
+        print pack('L*', hex('0x' . $low_addr), hex('0x' . $high_addr));
+      }
+    }
+  }
+}
+
+# Print symbols and profile data
+sub PrintSymbolizedProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $prog = shift;
+
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+
+  print '--- ', $symbol_marker, "\n";
+  if (defined($prog)) {
+    print 'binary=', $prog, "\n";
+  }
+  while (my ($pc, $name) = each(%{$symbols})) {
+    my $sep = ' ';
+    print '0x', $pc;
+    # We have a list of function names, which include the inlined
+    # calls.  They are separated (and terminated) by --, which is
+    # illegal in function names.
+    for (my $j = 2; $j <= $#{$name}; $j += 3) {
+      print $sep, $name->[$j];
+      $sep = '--';
+    }
+    print "\n";
+  }
+  print '---', "\n";
+
+  my $profile_marker;
+  if ($main::profile_type eq 'heap') {
+    $HEAP_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } elsif ($main::profile_type eq 'growth') {
+    $GROWTH_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } elsif ($main::profile_type eq 'contention') {
+    $CONTENTION_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } else { # elsif ($main::profile_type eq 'cpu')
+    $PROFILE_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  }
+
+  print '--- ', $profile_marker, "\n";
+  if (defined($main::collected_profile)) {
+    # if used with remote fetch, simply dump the collected profile to output.
+    open(SRC, "<$main::collected_profile");
+    while (<SRC>) {
+      print $_;
+    }
+    close(SRC);
+  } else {
+    # --raw/http: For everything to work correctly for non-remote profiles, we
+    # would need to extend PrintProfileData() to handle all possible profile
+    # types, re-enable the code that is currently disabled in ReadCPUProfile()
+    # and FixCallerAddresses(), and remove the remote profile dumping code in
+    # the block above.
+    die "--raw/http: jeprof can only dump remote profiles for --raw\n";
+    # dump a cpu-format profile to standard out
+    PrintProfileData($profile);
+  }
+}
+
+# Print text output
+sub PrintText {
+  my $symbols = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $line_limit = shift;
+
+  my $total = TotalProfile($flat);
+
+  # Which profile to sort by?
+  my $s = $main::opt_cum ? $cumulative : $flat;
+
+  my $running_sum = 0;
+  my $lines = 0;
+  foreach my $k (sort { GetEntry($s, $b) <=> GetEntry($s, $a) || $a cmp $b }
+                 keys(%{$cumulative})) {
+    my $f = GetEntry($flat, $k);
+    my $c = GetEntry($cumulative, $k);
+    $running_sum += $f;
+
+    my $sym = $k;
+    if (exists($symbols->{$k})) {
+      $sym = $symbols->{$k}->[0] . " " . $symbols->{$k}->[1];
+      if ($main::opt_addresses) {
+        $sym = $k . " " . $sym;
+      }
+    }
+
+    if ($f != 0 || $c != 0) {
+      printf("%8s %6s %6s %8s %6s %s\n",
+             Unparse($f),
+             Percent($f, $total),
+             Percent($running_sum, $total),
+             Unparse($c),
+             Percent($c, $total),
+             $sym);
+    }
+    $lines++;
+    last if ($line_limit >= 0 && $lines >= $line_limit);
+  }
+}
+
+# Callgrind format has a compression for repeated function and file
+# names.  You show the name the first time, and just use its number
+# subsequently.  This can cut down the file to about a third or a
+# quarter of its uncompressed size.  $key and $val are the key/value
+# pair that would normally be printed by callgrind; $map is a map from
+# value to number.
+sub CompressedCGName {
+  my($key, $val, $map) = @_;
+  my $idx = $map->{$val};
+  # For very short keys, providing an index hurts rather than helps.
+  if (length($val) <= 3) {
+    return "$key=$val\n";
+  } elsif (defined($idx)) {
+    return "$key=($idx)\n";
+  } else {
+    # scalar(keys $map) gives the number of items in the map.
+    $idx = scalar(keys(%{$map})) + 1;
+    $map->{$val} = $idx;
+    return "$key=($idx) $val\n";
+  }
+}
+
+# Print the call graph in a way that's suiteable for callgrind.
+sub PrintCallgrind {
+  my $calls = shift;
+  my $filename;
+  my %filename_to_index_map;
+  my %fnname_to_index_map;
+
+  if ($main::opt_interactive) {
+    $filename = shift;
+    print STDERR "Writing callgrind file to '$filename'.\n"
+  } else {
+    $filename = "&STDOUT";
+  }
+  open(CG, ">$filename");
+  printf CG ("events: Hits\n\n");
+  foreach my $call ( map { $_->[0] }
+                     sort { $a->[1] cmp $b ->[1] ||
+                            $a->[2] <=> $b->[2] }
+                     map { /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/;
+                           [$_, $1, $2] }
+                     keys %$calls ) {
+    my $count = int($calls->{$call});
+    $call =~ /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/;
+    my ( $caller_file, $caller_line, $caller_function,
+         $callee_file, $callee_line, $callee_function ) =
+       ( $1, $2, $3, $5, $6, $7 );
+
+    # TODO(csilvers): for better compression, collect all the
+    # caller/callee_files and functions first, before printing
+    # anything, and only compress those referenced more than once.
+    printf CG CompressedCGName("fl", $caller_file, \%filename_to_index_map);
+    printf CG CompressedCGName("fn", $caller_function, \%fnname_to_index_map);
+    if (defined $6) {
+      printf CG CompressedCGName("cfl", $callee_file, \%filename_to_index_map);
+      printf CG CompressedCGName("cfn", $callee_function, \%fnname_to_index_map);
+      printf CG ("calls=$count $callee_line\n");
+    }
+    printf CG ("$caller_line $count\n\n");
+  }
+}
+
+# Print disassembly for all all routines that match $main::opt_disasm
+sub PrintDisassembly {
+  my $libs = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $disasm_opts = shift;
+
+  my $total = TotalProfile($flat);
+
+  foreach my $lib (@{$libs}) {
+    my $symbol_table = GetProcedureBoundaries($lib->[0], $disasm_opts);
+    my $offset = AddressSub($lib->[1], $lib->[3]);
+    foreach my $routine (sort ByName keys(%{$symbol_table})) {
+      my $start_addr = $symbol_table->{$routine}->[0];
+      my $end_addr = $symbol_table->{$routine}->[1];
+      # See if there are any samples in this routine
+      my $length = hex(AddressSub($end_addr, $start_addr));
+      my $addr = AddressAdd($start_addr, $offset);
+      for (my $i = 0; $i < $length; $i++) {
+        if (defined($cumulative->{$addr})) {
+          PrintDisassembledFunction($lib->[0], $offset,
+                                    $routine, $flat, $cumulative,
+                                    $start_addr, $end_addr, $total);
+          last;
+        }
+        $addr = AddressInc($addr);
+      }
+    }
+  }
+}
+
+# Return reference to array of tuples of the form:
+#       [start_address, filename, linenumber, instruction, limit_address]
+# E.g.,
+#       ["0x806c43d", "/foo/bar.cc", 131, "ret", "0x806c440"]
+sub Disassemble {
+  my $prog = shift;
+  my $offset = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+
+  my $objdump = $obj_tool_map{"objdump"};
+  my $cmd = ShellEscape($objdump, "-C", "-d", "-l", "--no-show-raw-insn",
+                        "--start-address=0x$start_addr",
+                        "--stop-address=0x$end_addr", $prog);
+  open(OBJDUMP, "$cmd |") || error("$cmd: $!\n");
+  my @result = ();
+  my $filename = "";
+  my $linenumber = -1;
+  my $last = ["", "", "", ""];
+  while (<OBJDUMP>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    chop;
+    if (m|\s*([^:\s]+):(\d+)\s*$|) {
+      # Location line of the form:
+      #   <filename>:<linenumber>
+      $filename = $1;
+      $linenumber = $2;
+    } elsif (m/^ +([0-9a-f]+):\s*(.*)/) {
+      # Disassembly line -- zero-extend address to full length
+      my $addr = HexExtend($1);
+      my $k = AddressAdd($addr, $offset);
+      $last->[4] = $k;   # Store ending address for previous instruction
+      $last = [$k, $filename, $linenumber, $2, $end_addr];
+      push(@result, $last);
+    }
+  }
+  close(OBJDUMP);
+  return @result;
+}
+
+# The input file should contain lines of the form /proc/maps-like
+# output (same format as expected from the profiles) or that looks
+# like hex addresses (like "0xDEADBEEF").  We will parse all
+# /proc/maps output, and for all the hex addresses, we will output
+# "short" symbol names, one per line, in the same order as the input.
+sub PrintSymbols {
+  my $maps_and_symbols_file = shift;
+
+  # ParseLibraries expects pcs to be in a set.  Fine by us...
+  my @pclist = ();   # pcs in sorted order
+  my $pcs = {};
+  my $map = "";
+  foreach my $line (<$maps_and_symbols_file>) {
+    $line =~ s/\r//g;    # turn windows-looking lines into unix-looking lines
+    if ($line =~ /\b(0x[0-9a-f]+)\b/i) {
+      push(@pclist, HexExtend($1));
+      $pcs->{$pclist[-1]} = 1;
+    } else {
+      $map .= $line;
+    }
+  }
+
+  my $libs = ParseLibraries($main::prog, $map, $pcs);
+  my $symbols = ExtractSymbols($libs, $pcs);
+
+  foreach my $pc (@pclist) {
+    # ->[0] is the shortname, ->[2] is the full name
+    print(($symbols->{$pc}->[0] || "??") . "\n");
+  }
+}
+
+
+# For sorting functions by name
+sub ByName {
+  return ShortFunctionName($a) cmp ShortFunctionName($b);
+}
+
+# Print source-listing for all all routines that match $list_opts
+sub PrintListing {
+  my $total = shift;
+  my $libs = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $list_opts = shift;
+  my $html = shift;
+
+  my $output = \*STDOUT;
+  my $fname = "";
+
+  if ($html) {
+    # Arrange to write the output to a temporary file
+    $fname = TempName($main::next_tmpfile, "html");
+    $main::next_tmpfile++;
+    if (!open(TEMP, ">$fname")) {
+      print STDERR "$fname: $!\n";
+      return;
+    }
+    $output = \*TEMP;
+    print $output HtmlListingHeader();
+    printf $output ("<div class=\"legend\">%s<br>Total: %s %s</div>\n",
+                    $main::prog, Unparse($total), Units());
+  }
+
+  my $listed = 0;
+  foreach my $lib (@{$libs}) {
+    my $symbol_table = GetProcedureBoundaries($lib->[0], $list_opts);
+    my $offset = AddressSub($lib->[1], $lib->[3]);
+    foreach my $routine (sort ByName keys(%{$symbol_table})) {
+      # Print if there are any samples in this routine
+      my $start_addr = $symbol_table->{$routine}->[0];
+      my $end_addr = $symbol_table->{$routine}->[1];
+      my $length = hex(AddressSub($end_addr, $start_addr));
+      my $addr = AddressAdd($start_addr, $offset);
+      for (my $i = 0; $i < $length; $i++) {
+        if (defined($cumulative->{$addr})) {
+          $listed += PrintSource(
+            $lib->[0], $offset,
+            $routine, $flat, $cumulative,
+            $start_addr, $end_addr,
+            $html,
+            $output);
+          last;
+        }
+        $addr = AddressInc($addr);
+      }
+    }
+  }
+
+  if ($html) {
+    if ($listed > 0) {
+      print $output HtmlListingFooter();
+      close($output);
+      RunWeb($fname);
+    } else {
+      close($output);
+      unlink($fname);
+    }
+  }
+}
+
+sub HtmlListingHeader {
+  return <<'EOF';
+<DOCTYPE html>
+<html>
+<head>
+<title>Pprof listing</title>
+<style type="text/css">
+body {
+  font-family: sans-serif;
+}
+h1 {
+  font-size: 1.5em;
+  margin-bottom: 4px;
+}
+.legend {
+  font-size: 1.25em;
+}
+.line {
+  color: #aaaaaa;
+}
+.nop {
+  color: #aaaaaa;
+}
+.unimportant {
+  color: #cccccc;
+}
+.disasmloc {
+  color: #000000;
+}
+.deadsrc {
+  cursor: pointer;
+}
+.deadsrc:hover {
+  background-color: #eeeeee;
+}
+.livesrc {
+  color: #0000ff;
+  cursor: pointer;
+}
+.livesrc:hover {
+  background-color: #eeeeee;
+}
+.asm {
+  color: #008800;
+  display: none;
+}
+</style>
+<script type="text/javascript">
+function jeprof_toggle_asm(e) {
+  var target;
+  if (!e) e = window.event;
+  if (e.target) target = e.target;
+  else if (e.srcElement) target = e.srcElement;
+
+  if (target) {
+    var asm = target.nextSibling;
+    if (asm && asm.className == "asm") {
+      asm.style.display = (asm.style.display == "block" ? "" : "block");
+      e.preventDefault();
+      return false;
+    }
+  }
+}
+</script>
+</head>
+<body>
+EOF
+}
+
+sub HtmlListingFooter {
+  return <<'EOF';
+</body>
+</html>
+EOF
+}
+
+sub HtmlEscape {
+  my $text = shift;
+  $text =~ s/&/&amp;/g;
+  $text =~ s/</&lt;/g;
+  $text =~ s/>/&gt;/g;
+  return $text;
+}
+
+# Returns the indentation of the line, if it has any non-whitespace
+# characters.  Otherwise, returns -1.
+sub Indentation {
+  my $line = shift;
+  if (m/^(\s*)\S/) {
+    return length($1);
+  } else {
+    return -1;
+  }
+}
+
+# If the symbol table contains inlining info, Disassemble() may tag an
+# instruction with a location inside an inlined function.  But for
+# source listings, we prefer to use the location in the function we
+# are listing.  So use MapToSymbols() to fetch full location
+# information for each instruction and then pick out the first
+# location from a location list (location list contains callers before
+# callees in case of inlining).
+#
+# After this routine has run, each entry in $instructions contains:
+#   [0] start address
+#   [1] filename for function we are listing
+#   [2] line number for function we are listing
+#   [3] disassembly
+#   [4] limit address
+#   [5] most specific filename (may be different from [1] due to inlining)
+#   [6] most specific line number (may be different from [2] due to inlining)
+sub GetTopLevelLineNumbers {
+  my ($lib, $offset, $instructions) = @_;
+  my $pcs = [];
+  for (my $i = 0; $i <= $#{$instructions}; $i++) {
+    push(@{$pcs}, $instructions->[$i]->[0]);
+  }
+  my $symbols = {};
+  MapToSymbols($lib, $offset, $pcs, $symbols);
+  for (my $i = 0; $i <= $#{$instructions}; $i++) {
+    my $e = $instructions->[$i];
+    push(@{$e}, $e->[1]);
+    push(@{$e}, $e->[2]);
+    my $addr = $e->[0];
+    my $sym = $symbols->{$addr};
+    if (defined($sym)) {
+      if ($#{$sym} >= 2 && $sym->[1] =~ m/^(.*):(\d+)$/) {
+        $e->[1] = $1;  # File name
+        $e->[2] = $2;  # Line number
+      }
+    }
+  }
+}
+
+# Print source-listing for one routine
+sub PrintSource {
+  my $prog = shift;
+  my $offset = shift;
+  my $routine = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+  my $html = shift;
+  my $output = shift;
+
+  # Disassemble all instructions (just to get line numbers)
+  my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr);
+  GetTopLevelLineNumbers($prog, $offset, \@instructions);
+
+  # Hack 1: assume that the first source file encountered in the
+  # disassembly contains the routine
+  my $filename = undef;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    if ($instructions[$i]->[2] >= 0) {
+      $filename = $instructions[$i]->[1];
+      last;
+    }
+  }
+  if (!defined($filename)) {
+    print STDERR "no filename found in $routine\n";
+    return 0;
+  }
+
+  # Hack 2: assume that the largest line number from $filename is the
+  # end of the procedure.  This is typically safe since if P1 contains
+  # an inlined call to P2, then P2 usually occurs earlier in the
+  # source file.  If this does not work, we might have to compute a
+  # density profile or just print all regions we find.
+  my $lastline = 0;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    my $f = $instructions[$i]->[1];
+    my $l = $instructions[$i]->[2];
+    if (($f eq $filename) && ($l > $lastline)) {
+      $lastline = $l;
+    }
+  }
+
+  # Hack 3: assume the first source location from "filename" is the start of
+  # the source code.
+  my $firstline = 1;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    if ($instructions[$i]->[1] eq $filename) {
+      $firstline = $instructions[$i]->[2];
+      last;
+    }
+  }
+
+  # Hack 4: Extend last line forward until its indentation is less than
+  # the indentation we saw on $firstline
+  my $oldlastline = $lastline;
+  {
+    if (!open(FILE, "<$filename")) {
+      print STDERR "$filename: $!\n";
+      return 0;
+    }
+    my $l = 0;
+    my $first_indentation = -1;
+    while (<FILE>) {
+      s/\r//g;         # turn windows-looking lines into unix-looking lines
+      $l++;
+      my $indent = Indentation($_);
+      if ($l >= $firstline) {
+        if ($first_indentation < 0 && $indent >= 0) {
+          $first_indentation = $indent;
+          last if ($first_indentation == 0);
+        }
+      }
+      if ($l >= $lastline && $indent >= 0) {
+        if ($indent >= $first_indentation) {
+          $lastline = $l+1;
+        } else {
+          last;
+        }
+      }
+    }
+    close(FILE);
+  }
+
+  # Assign all samples to the range $firstline,$lastline,
+  # Hack 4: If an instruction does not occur in the range, its samples
+  # are moved to the next instruction that occurs in the range.
+  my $samples1 = {};        # Map from line number to flat count
+  my $samples2 = {};        # Map from line number to cumulative count
+  my $running1 = 0;         # Unassigned flat counts
+  my $running2 = 0;         # Unassigned cumulative counts
+  my $total1 = 0;           # Total flat counts
+  my $total2 = 0;           # Total cumulative counts
+  my %disasm = ();          # Map from line number to disassembly
+  my $running_disasm = "";  # Unassigned disassembly
+  my $skip_marker = "---\n";
+  if ($html) {
+    $skip_marker = "";
+    for (my $l = $firstline; $l <= $lastline; $l++) {
+      $disasm{$l} = "";
+    }
+  }
+  my $last_dis_filename = '';
+  my $last_dis_linenum = -1;
+  my $last_touched_line = -1;  # To detect gaps in disassembly for a line
+  foreach my $e (@instructions) {
+    # Add up counts for all address that fall inside this instruction
+    my $c1 = 0;
+    my $c2 = 0;
+    for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) {
+      $c1 += GetEntry($flat, $a);
+      $c2 += GetEntry($cumulative, $a);
+    }
+
+    if ($html) {
+      my $dis = sprintf("      %6s %6s \t\t%8s: %s ",
+                        HtmlPrintNumber($c1),
+                        HtmlPrintNumber($c2),
+                        UnparseAddress($offset, $e->[0]),
+                        CleanDisassembly($e->[3]));
+
+      # Append the most specific source line associated with this instruction
+      if (length($dis) < 80) { $dis .= (' ' x (80 - length($dis))) };
+      $dis = HtmlEscape($dis);
+      my $f = $e->[5];
+      my $l = $e->[6];
+      if ($f ne $last_dis_filename) {
+        $dis .= sprintf("<span class=disasmloc>%s:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      } elsif ($l ne $last_dis_linenum) {
+        # De-emphasize the unchanged file name portion
+        $dis .= sprintf("<span class=unimportant>%s</span>" .
+                        "<span class=disasmloc>:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      } else {
+        # De-emphasize the entire location
+        $dis .= sprintf("<span class=unimportant>%s:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      }
+      $last_dis_filename = $f;
+      $last_dis_linenum = $l;
+      $running_disasm .= $dis;
+      $running_disasm .= "\n";
+    }
+
+    $running1 += $c1;
+    $running2 += $c2;
+    $total1 += $c1;
+    $total2 += $c2;
+    my $file = $e->[1];
+    my $line = $e->[2];
+    if (($file eq $filename) &&
+        ($line >= $firstline) &&
+        ($line <= $lastline)) {
+      # Assign all accumulated samples to this line
+      AddEntry($samples1, $line, $running1);
+      AddEntry($samples2, $line, $running2);
+      $running1 = 0;
+      $running2 = 0;
+      if ($html) {
+        if ($line != $last_touched_line && $disasm{$line} ne '') {
+          $disasm{$line} .= "\n";
+        }
+        $disasm{$line} .= $running_disasm;
+        $running_disasm = '';
+        $last_touched_line = $line;
+      }
+    }
+  }
+
+  # Assign any leftover samples to $lastline
+  AddEntry($samples1, $lastline, $running1);
+  AddEntry($samples2, $lastline, $running2);
+  if ($html) {
+    if ($lastline != $last_touched_line && $disasm{$lastline} ne '') {
+      $disasm{$lastline} .= "\n";
+    }
+    $disasm{$lastline} .= $running_disasm;
+  }
+
+  if ($html) {
+    printf $output (
+      "<h1>%s</h1>%s\n<pre onClick=\"jeprof_toggle_asm()\">\n" .
+      "Total:%6s %6s (flat / cumulative %s)\n",
+      HtmlEscape(ShortFunctionName($routine)),
+      HtmlEscape(CleanFileName($filename)),
+      Unparse($total1),
+      Unparse($total2),
+      Units());
+  } else {
+    printf $output (
+      "ROUTINE ====================== %s in %s\n" .
+      "%6s %6s Total %s (flat / cumulative)\n",
+      ShortFunctionName($routine),
+      CleanFileName($filename),
+      Unparse($total1),
+      Unparse($total2),
+      Units());
+  }
+  if (!open(FILE, "<$filename")) {
+    print STDERR "$filename: $!\n";
+    return 0;
+  }
+  my $l = 0;
+  while (<FILE>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    $l++;
+    if ($l >= $firstline - 5 &&
+        (($l <= $oldlastline + 5) || ($l <= $lastline))) {
+      chop;
+      my $text = $_;
+      if ($l == $firstline) { print $output $skip_marker; }
+      my $n1 = GetEntry($samples1, $l);
+      my $n2 = GetEntry($samples2, $l);
+      if ($html) {
+        # Emit a span that has one of the following classes:
+        #    livesrc -- has samples
+        #    deadsrc -- has disassembly, but with no samples
+        #    nop     -- has no matching disasembly
+        # Also emit an optional span containing disassembly.
+        my $dis = $disasm{$l};
+        my $asm = "";
+        if (defined($dis) && $dis ne '') {
+          $asm = "<span class=\"asm\">" . $dis . "</span>";
+        }
+        my $source_class = (($n1 + $n2 > 0)
+                            ? "livesrc"
+                            : (($asm ne "") ? "deadsrc" : "nop"));
+        printf $output (
+          "<span class=\"line\">%5d</span> " .
+          "<span class=\"%s\">%6s %6s %s</span>%s\n",
+          $l, $source_class,
+          HtmlPrintNumber($n1),
+          HtmlPrintNumber($n2),
+          HtmlEscape($text),
+          $asm);
+      } else {
+        printf $output(
+          "%6s %6s %4d: %s\n",
+          UnparseAlt($n1),
+          UnparseAlt($n2),
+          $l,
+          $text);
+      }
+      if ($l == $lastline)  { print $output $skip_marker; }
+    };
+  }
+  close(FILE);
+  if ($html) {
+    print $output "</pre>\n";
+  }
+  return 1;
+}
+
+# Return the source line for the specified file/linenumber.
+# Returns undef if not found.
+sub SourceLine {
+  my $file = shift;
+  my $line = shift;
+
+  # Look in cache
+  if (!defined($main::source_cache{$file})) {
+    if (100 < scalar keys(%main::source_cache)) {
+      # Clear the cache when it gets too big
+      $main::source_cache = ();
+    }
+
+    # Read all lines from the file
+    if (!open(FILE, "<$file")) {
+      print STDERR "$file: $!\n";
+      $main::source_cache{$file} = [];  # Cache the negative result
+      return undef;
+    }
+    my $lines = [];
+    push(@{$lines}, "");        # So we can use 1-based line numbers as indices
+    while (<FILE>) {
+      push(@{$lines}, $_);
+    }
+    close(FILE);
+
+    # Save the lines in the cache
+    $main::source_cache{$file} = $lines;
+  }
+
+  my $lines = $main::source_cache{$file};
+  if (($line < 0) || ($line > $#{$lines})) {
+    return undef;
+  } else {
+    return $lines->[$line];
+  }
+}
+
+# Print disassembly for one routine with interspersed source if available
+sub PrintDisassembledFunction {
+  my $prog = shift;
+  my $offset = shift;
+  my $routine = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+  my $total = shift;
+
+  # Disassemble all instructions
+  my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr);
+
+  # Make array of counts per instruction
+  my @flat_count = ();
+  my @cum_count = ();
+  my $flat_total = 0;
+  my $cum_total = 0;
+  foreach my $e (@instructions) {
+    # Add up counts for all address that fall inside this instruction
+    my $c1 = 0;
+    my $c2 = 0;
+    for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) {
+      $c1 += GetEntry($flat, $a);
+      $c2 += GetEntry($cumulative, $a);
+    }
+    push(@flat_count, $c1);
+    push(@cum_count, $c2);
+    $flat_total += $c1;
+    $cum_total += $c2;
+  }
+
+  # Print header with total counts
+  printf("ROUTINE ====================== %s\n" .
+         "%6s %6s %s (flat, cumulative) %.1f%% of total\n",
+         ShortFunctionName($routine),
+         Unparse($flat_total),
+         Unparse($cum_total),
+         Units(),
+         ($cum_total * 100.0) / $total);
+
+  # Process instructions in order
+  my $current_file = "";
+  for (my $i = 0; $i <= $#instructions; ) {
+    my $e = $instructions[$i];
+
+    # Print the new file name whenever we switch files
+    if ($e->[1] ne $current_file) {
+      $current_file = $e->[1];
+      my $fname = $current_file;
+      $fname =~ s|^\./||;   # Trim leading "./"
+
+      # Shorten long file names
+      if (length($fname) >= 58) {
+        $fname = "..." . substr($fname, -55);
+      }
+      printf("-------------------- %s\n", $fname);
+    }
+
+    # TODO: Compute range of lines to print together to deal with
+    # small reorderings.
+    my $first_line = $e->[2];
+    my $last_line = $first_line;
+    my %flat_sum = ();
+    my %cum_sum = ();
+    for (my $l = $first_line; $l <= $last_line; $l++) {
+      $flat_sum{$l} = 0;
+      $cum_sum{$l} = 0;
+    }
+
+    # Find run of instructions for this range of source lines
+    my $first_inst = $i;
+    while (($i <= $#instructions) &&
+           ($instructions[$i]->[2] >= $first_line) &&
+           ($instructions[$i]->[2] <= $last_line)) {
+      $e = $instructions[$i];
+      $flat_sum{$e->[2]} += $flat_count[$i];
+      $cum_sum{$e->[2]} += $cum_count[$i];
+      $i++;
+    }
+    my $last_inst = $i - 1;
+
+    # Print source lines
+    for (my $l = $first_line; $l <= $last_line; $l++) {
+      my $line = SourceLine($current_file, $l);
+      if (!defined($line)) {
+        $line = "?\n";
+        next;
+      } else {
+        $line =~ s/^\s+//;
+      }
+      printf("%6s %6s %5d: %s",
+             UnparseAlt($flat_sum{$l}),
+             UnparseAlt($cum_sum{$l}),
+             $l,
+             $line);
+    }
+
+    # Print disassembly
+    for (my $x = $first_inst; $x <= $last_inst; $x++) {
+      my $e = $instructions[$x];
+      printf("%6s %6s    %8s: %6s\n",
+             UnparseAlt($flat_count[$x]),
+             UnparseAlt($cum_count[$x]),
+             UnparseAddress($offset, $e->[0]),
+             CleanDisassembly($e->[3]));
+    }
+  }
+}
+
+# Print DOT graph
+sub PrintDot {
+  my $prog = shift;
+  my $symbols = shift;
+  my $raw = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $overall_total = shift;
+
+  # Get total
+  my $local_total = TotalProfile($flat);
+  my $nodelimit = int($main::opt_nodefraction * $local_total);
+  my $edgelimit = int($main::opt_edgefraction * $local_total);
+  my $nodecount = $main::opt_nodecount;
+
+  # Find nodes to include
+  my @list = (sort { abs(GetEntry($cumulative, $b)) <=>
+                     abs(GetEntry($cumulative, $a))
+                     || $a cmp $b }
+              keys(%{$cumulative}));
+  my $last = $nodecount - 1;
+  if ($last > $#list) {
+    $last = $#list;
+  }
+  while (($last >= 0) &&
+         (abs(GetEntry($cumulative, $list[$last])) <= $nodelimit)) {
+    $last--;
+  }
+  if ($last < 0) {
+    print STDERR "No nodes to print\n";
+    return 0;
+  }
+
+  if ($nodelimit > 0 || $edgelimit > 0) {
+    printf STDERR ("Dropping nodes with <= %s %s; edges with <= %s abs(%s)\n",
+                   Unparse($nodelimit), Units(),
+                   Unparse($edgelimit), Units());
+  }
+
+  # Open DOT output file
+  my $output;
+  my $escaped_dot = ShellEscape(@DOT);
+  my $escaped_ps2pdf = ShellEscape(@PS2PDF);
+  if ($main::opt_gv) {
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "ps"));
+    $output = "| $escaped_dot -Tps2 >$escaped_outfile";
+  } elsif ($main::opt_evince) {
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "pdf"));
+    $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - $escaped_outfile";
+  } elsif ($main::opt_ps) {
+    $output = "| $escaped_dot -Tps2";
+  } elsif ($main::opt_pdf) {
+    $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - -";
+  } elsif ($main::opt_web || $main::opt_svg) {
+    # We need to post-process the SVG, so write to a temporary file always.
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "svg"));
+    $output = "| $escaped_dot -Tsvg >$escaped_outfile";
+  } elsif ($main::opt_gif) {
+    $output = "| $escaped_dot -Tgif";
+  } else {
+    $output = ">&STDOUT";
+  }
+  open(DOT, $output) || error("$output: $!\n");
+
+  # Title
+  printf DOT ("digraph \"%s; %s %s\" {\n",
+              $prog,
+              Unparse($overall_total),
+              Units());
+  if ($main::opt_pdf) {
+    # The output is more printable if we set the page size for dot.
+    printf DOT ("size=\"8,11\"\n");
+  }
+  printf DOT ("node [width=0.375,height=0.25];\n");
+
+  # Print legend
+  printf DOT ("Legend [shape=box,fontsize=24,shape=plaintext," .
+              "label=\"%s\\l%s\\l%s\\l%s\\l%s\\l\"];\n",
+              $prog,
+              sprintf("Total %s: %s", Units(), Unparse($overall_total)),
+              sprintf("Focusing on: %s", Unparse($local_total)),
+              sprintf("Dropped nodes with <= %s abs(%s)",
+                      Unparse($nodelimit), Units()),
+              sprintf("Dropped edges with <= %s %s",
+                      Unparse($edgelimit), Units())
+              );
+
+  # Print nodes
+  my %node = ();
+  my $nextnode = 1;
+  foreach my $a (@list[0..$last]) {
+    # Pick font size
+    my $f = GetEntry($flat, $a);
+    my $c = GetEntry($cumulative, $a);
+
+    my $fs = 8;
+    if ($local_total > 0) {
+      $fs = 8 + (50.0 * sqrt(abs($f * 1.0 / $local_total)));
+    }
+
+    $node{$a} = $nextnode++;
+    my $sym = $a;
+    $sym =~ s/\s+/\\n/g;
+    $sym =~ s/::/\\n/g;
+
+    # Extra cumulative info to print for non-leaves
+    my $extra = "";
+    if ($f != $c) {
+      $extra = sprintf("\\rof %s (%s)",
+                       Unparse($c),
+                       Percent($c, $local_total));
+    }
+    my $style = "";
+    if ($main::opt_heapcheck) {
+      if ($f > 0) {
+        # make leak-causing nodes more visible (add a background)
+        $style = ",style=filled,fillcolor=gray"
+      } elsif ($f < 0) {
+        # make anti-leak-causing nodes (which almost never occur)
+        # stand out as well (triple border)
+        $style = ",peripheries=3"
+      }
+    }
+
+    printf DOT ("N%d [label=\"%s\\n%s (%s)%s\\r" .
+                "\",shape=box,fontsize=%.1f%s];\n",
+                $node{$a},
+                $sym,
+                Unparse($f),
+                Percent($f, $local_total),
+                $extra,
+                $fs,
+                $style,
+               );
+  }
+
+  # Get edges and counts per edge
+  my %edge = ();
+  my $n;
+  my $fullname_to_shortname_map = {};
+  FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map);
+  foreach my $k (keys(%{$raw})) {
+    # TODO: omit low %age edges
+    $n = $raw->{$k};
+    my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k);
+    for (my $i = 1; $i <= $#translated; $i++) {
+      my $src = $translated[$i];
+      my $dst = $translated[$i-1];
+      #next if ($src eq $dst);  # Avoid self-edges?
+      if (exists($node{$src}) && exists($node{$dst})) {
+        my $edge_label = "$src\001$dst";
+        if (!exists($edge{$edge_label})) {
+          $edge{$edge_label} = 0;
+        }
+        $edge{$edge_label} += $n;
+      }
+    }
+  }
+
+  # Print edges (process in order of decreasing counts)
+  my %indegree = ();   # Number of incoming edges added per node so far
+  my %outdegree = ();  # Number of outgoing edges added per node so far
+  foreach my $e (sort { $edge{$b} <=> $edge{$a} } keys(%edge)) {
+    my @x = split(/\001/, $e);
+    $n = $edge{$e};
+
+    # Initialize degree of kept incoming and outgoing edges if necessary
+    my $src = $x[0];
+    my $dst = $x[1];
+    if (!exists($outdegree{$src})) { $outdegree{$src} = 0; }
+    if (!exists($indegree{$dst})) { $indegree{$dst} = 0; }
+
+    my $keep;
+    if ($indegree{$dst} == 0) {
+      # Keep edge if needed for reachability
+      $keep = 1;
+    } elsif (abs($n) <= $edgelimit) {
+      # Drop if we are below --edgefraction
+      $keep = 0;
+    } elsif ($outdegree{$src} >= $main::opt_maxdegree ||
+             $indegree{$dst} >= $main::opt_maxdegree) {
+      # Keep limited number of in/out edges per node
+      $keep = 0;
+    } else {
+      $keep = 1;
+    }
+
+    if ($keep) {
+      $outdegree{$src}++;
+      $indegree{$dst}++;
+
+      # Compute line width based on edge count
+      my $fraction = abs($local_total ? (3 * ($n / $local_total)) : 0);
+      if ($fraction > 1) { $fraction = 1; }
+      my $w = $fraction * 2;
+      if ($w < 1 && ($main::opt_web || $main::opt_svg)) {
+        # SVG output treats line widths < 1 poorly.
+        $w = 1;
+      }
+
+      # Dot sometimes segfaults if given edge weights that are too large, so
+      # we cap the weights at a large value
+      my $edgeweight = abs($n) ** 0.7;
+      if ($edgeweight > 100000) { $edgeweight = 100000; }
+      $edgeweight = int($edgeweight);
+
+      my $style = sprintf("setlinewidth(%f)", $w);
+      if ($x[1] =~ m/\(inline\)/) {
+        $style .= ",dashed";
+      }
+
+      # Use a slightly squashed function of the edge count as the weight
+      printf DOT ("N%s -> N%s [label=%s, weight=%d, style=\"%s\"];\n",
+                  $node{$x[0]},
+                  $node{$x[1]},
+                  Unparse($n),
+                  $edgeweight,
+                  $style);
+    }
+  }
+
+  print DOT ("}\n");
+  close(DOT);
+
+  if ($main::opt_web || $main::opt_svg) {
+    # Rewrite SVG to be more usable inside web browser.
+    RewriteSvg(TempName($main::next_tmpfile, "svg"));
+  }
+
+  return 1;
+}
+
+sub RewriteSvg {
+  my $svgfile = shift;
+
+  open(SVG, $svgfile) || die "open temp svg: $!";
+  my @svg = <SVG>;
+  close(SVG);
+  unlink $svgfile;
+  my $svg = join('', @svg);
+
+  # Dot's SVG output is
+  #
+  #    <svg width="___" height="___"
+  #     viewBox="___" xmlns=...>
+  #    <g id="graph0" transform="...">
+  #    ...
+  #    </g>
+  #    </svg>
+  #
+  # Change it to
+  #
+  #    <svg width="100%" height="100%"
+  #     xmlns=...>
+  #    $svg_javascript
+  #    <g id="viewport" transform="translate(0,0)">
+  #    <g id="graph0" transform="...">
+  #    ...
+  #    </g>
+  #    </g>
+  #    </svg>
+
+  # Fix width, height; drop viewBox.
+  $svg =~ s/(?s)<svg width="[^"]+" height="[^"]+"(.*?)viewBox="[^"]+"/<svg width="100%" height="100%"$1/;
+
+  # Insert script, viewport <g> above first <g>
+  my $svg_javascript = SvgJavascript();
+  my $viewport = "<g id=\"viewport\" transform=\"translate(0,0)\">\n";
+  $svg =~ s/<g id="graph\d"/$svg_javascript$viewport$&/;
+
+  # Insert final </g> above </svg>.
+  $svg =~ s/(.*)(<\/svg>)/$1<\/g>$2/;
+  $svg =~ s/<g id="graph\d"(.*?)/<g id="viewport"$1/;
+
+  if ($main::opt_svg) {
+    # --svg: write to standard output.
+    print $svg;
+  } else {
+    # Write back to temporary file.
+    open(SVG, ">$svgfile") || die "open $svgfile: $!";
+    print SVG $svg;
+    close(SVG);
+  }
+}
+
+sub SvgJavascript {
+  return <<'EOF';
+<script type="text/ecmascript"><![CDATA[
+// SVGPan
+// http://www.cyberz.org/blog/2009/12/08/svgpan-a-javascript-svg-panzoomdrag-library/
+// Local modification: if(true || ...) below to force panning, never moving.
+
+/**
+ *  SVGPan library 1.2
+ * ====================
+ *
+ * Given an unique existing element with id "viewport", including the
+ * the library into any SVG adds the following capabilities:
+ *
+ *  - Mouse panning
+ *  - Mouse zooming (using the wheel)
+ *  - Object dargging
+ *
+ * Known issues:
+ *
+ *  - Zooming (while panning) on Safari has still some issues
+ *
+ * Releases:
+ *
+ * 1.2, Sat Mar 20 08:42:50 GMT 2010, Zeng Xiaohui
+ *	Fixed a bug with browser mouse handler interaction
+ *
+ * 1.1, Wed Feb  3 17:39:33 GMT 2010, Zeng Xiaohui
+ *	Updated the zoom code to support the mouse wheel on Safari/Chrome
+ *
+ * 1.0, Andrea Leofreddi
+ *	First release
+ *
+ * This code is licensed under the following BSD license:
+ *
+ * Copyright 2009-2010 Andrea Leofreddi <a.leofreddi@itcharm.com>. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are
+ * permitted provided that the following conditions are met:
+ *
+ *    1. Redistributions of source code must retain the above copyright notice, this list of
+ *       conditions and the following disclaimer.
+ *
+ *    2. Redistributions in binary form must reproduce the above copyright notice, this list
+ *       of conditions and the following disclaimer in the documentation and/or other materials
+ *       provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Andrea Leofreddi ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+ * FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL Andrea Leofreddi OR
+ * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * The views and conclusions contained in the software and documentation are those of the
+ * authors and should not be interpreted as representing official policies, either expressed
+ * or implied, of Andrea Leofreddi.
+ */
+
+var root = document.documentElement;
+
+var state = 'none', stateTarget, stateOrigin, stateTf;
+
+setupHandlers(root);
+
+/**
+ * Register handlers
+ */
+function setupHandlers(root){
+	setAttributes(root, {
+		"onmouseup" : "add(evt)",
+		"onmousedown" : "handleMouseDown(evt)",
+		"onmousemove" : "handleMouseMove(evt)",
+		"onmouseup" : "handleMouseUp(evt)",
+		//"onmouseout" : "handleMouseUp(evt)", // Decomment this to stop the pan functionality when dragging out of the SVG element
+	});
+
+	if(navigator.userAgent.toLowerCase().indexOf('webkit') >= 0)
+		window.addEventListener('mousewheel', handleMouseWheel, false); // Chrome/Safari
+	else
+		window.addEventListener('DOMMouseScroll', handleMouseWheel, false); // Others
+
+	var g = svgDoc.getElementById("svg");
+	g.width = "100%";
+	g.height = "100%";
+}
+
+/**
+ * Instance an SVGPoint object with given event coordinates.
+ */
+function getEventPoint(evt) {
+	var p = root.createSVGPoint();
+
+	p.x = evt.clientX;
+	p.y = evt.clientY;
+
+	return p;
+}
+
+/**
+ * Sets the current transform matrix of an element.
+ */
+function setCTM(element, matrix) {
+	var s = "matrix(" + matrix.a + "," + matrix.b + "," + matrix.c + "," + matrix.d + "," + matrix.e + "," + matrix.f + ")";
+
+	element.setAttribute("transform", s);
+}
+
+/**
+ * Dumps a matrix to a string (useful for debug).
+ */
+function dumpMatrix(matrix) {
+	var s = "[ " + matrix.a + ", " + matrix.c + ", " + matrix.e + "\n  " + matrix.b + ", " + matrix.d + ", " + matrix.f + "\n  0, 0, 1 ]";
+
+	return s;
+}
+
+/**
+ * Sets attributes of an element.
+ */
+function setAttributes(element, attributes){
+	for (i in attributes)
+		element.setAttributeNS(null, i, attributes[i]);
+}
+
+/**
+ * Handle mouse move event.
+ */
+function handleMouseWheel(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var delta;
+
+	if(evt.wheelDelta)
+		delta = evt.wheelDelta / 3600; // Chrome/Safari
+	else
+		delta = evt.detail / -90; // Mozilla
+
+	var z = 1 + delta; // Zoom factor: 0.9/1.1
+
+	var g = svgDoc.getElementById("viewport");
+
+	var p = getEventPoint(evt);
+
+	p = p.matrixTransform(g.getCTM().inverse());
+
+	// Compute new scale matrix in current mouse position
+	var k = root.createSVGMatrix().translate(p.x, p.y).scale(z).translate(-p.x, -p.y);
+
+        setCTM(g, g.getCTM().multiply(k));
+
+	stateTf = stateTf.multiply(k.inverse());
+}
+
+/**
+ * Handle mouse move event.
+ */
+function handleMouseMove(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var g = svgDoc.getElementById("viewport");
+
+	if(state == 'pan') {
+		// Pan mode
+		var p = getEventPoint(evt).matrixTransform(stateTf);
+
+		setCTM(g, stateTf.inverse().translate(p.x - stateOrigin.x, p.y - stateOrigin.y));
+	} else if(state == 'move') {
+		// Move mode
+		var p = getEventPoint(evt).matrixTransform(g.getCTM().inverse());
+
+		setCTM(stateTarget, root.createSVGMatrix().translate(p.x - stateOrigin.x, p.y - stateOrigin.y).multiply(g.getCTM().inverse()).multiply(stateTarget.getCTM()));
+
+		stateOrigin = p;
+	}
+}
+
+/**
+ * Handle click event.
+ */
+function handleMouseDown(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var g = svgDoc.getElementById("viewport");
+
+	if(true || evt.target.tagName == "svg") {
+		// Pan mode
+		state = 'pan';
+
+		stateTf = g.getCTM().inverse();
+
+		stateOrigin = getEventPoint(evt).matrixTransform(stateTf);
+	} else {
+		// Move mode
+		state = 'move';
+
+		stateTarget = evt.target;
+
+		stateTf = g.getCTM().inverse();
+
+		stateOrigin = getEventPoint(evt).matrixTransform(stateTf);
+	}
+}
+
+/**
+ * Handle mouse button release event.
+ */
+function handleMouseUp(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	if(state == 'pan' || state == 'move') {
+		// Quit pan mode
+		state = '';
+	}
+}
+
+]]></script>
+EOF
+}
+
+# Provides a map from fullname to shortname for cases where the
+# shortname is ambiguous.  The symlist has both the fullname and
+# shortname for all symbols, which is usually fine, but sometimes --
+# such as overloaded functions -- two different fullnames can map to
+# the same shortname.  In that case, we use the address of the
+# function to disambiguate the two.  This function fills in a map that
+# maps fullnames to modified shortnames in such cases.  If a fullname
+# is not present in the map, the 'normal' shortname provided by the
+# symlist is the appropriate one to use.
+sub FillFullnameToShortnameMap {
+  my $symbols = shift;
+  my $fullname_to_shortname_map = shift;
+  my $shortnames_seen_once = {};
+  my $shortnames_seen_more_than_once = {};
+
+  foreach my $symlist (values(%{$symbols})) {
+    # TODO(csilvers): deal with inlined symbols too.
+    my $shortname = $symlist->[0];
+    my $fullname = $symlist->[2];
+    if ($fullname !~ /<[0-9a-fA-F]+>$/) {  # fullname doesn't end in an address
+      next;       # the only collisions we care about are when addresses differ
+    }
+    if (defined($shortnames_seen_once->{$shortname}) &&
+        $shortnames_seen_once->{$shortname} ne $fullname) {
+      $shortnames_seen_more_than_once->{$shortname} = 1;
+    } else {
+      $shortnames_seen_once->{$shortname} = $fullname;
+    }
+  }
+
+  foreach my $symlist (values(%{$symbols})) {
+    my $shortname = $symlist->[0];
+    my $fullname = $symlist->[2];
+    # TODO(csilvers): take in a list of addresses we care about, and only
+    # store in the map if $symlist->[1] is in that list.  Saves space.
+    next if defined($fullname_to_shortname_map->{$fullname});
+    if (defined($shortnames_seen_more_than_once->{$shortname})) {
+      if ($fullname =~ /<0*([^>]*)>$/) {   # fullname has address at end of it
+        $fullname_to_shortname_map->{$fullname} = "$shortname\@$1";
+      }
+    }
+  }
+}
+
+# Return a small number that identifies the argument.
+# Multiple calls with the same argument will return the same number.
+# Calls with different arguments will return different numbers.
+sub ShortIdFor {
+  my $key = shift;
+  my $id = $main::uniqueid{$key};
+  if (!defined($id)) {
+    $id = keys(%main::uniqueid) + 1;
+    $main::uniqueid{$key} = $id;
+  }
+  return $id;
+}
+
+# Translate a stack of addresses into a stack of symbols
+sub TranslateStack {
+  my $symbols = shift;
+  my $fullname_to_shortname_map = shift;
+  my $k = shift;
+
+  my @addrs = split(/\n/, $k);
+  my @result = ();
+  for (my $i = 0; $i <= $#addrs; $i++) {
+    my $a = $addrs[$i];
+
+    # Skip large addresses since they sometimes show up as fake entries on RH9
+    if (length($a) > 8 && $a gt "7fffffffffffffff") {
+      next;
+    }
+
+    if ($main::opt_disasm || $main::opt_list) {
+      # We want just the address for the key
+      push(@result, $a);
+      next;
+    }
+
+    my $symlist = $symbols->{$a};
+    if (!defined($symlist)) {
+      $symlist = [$a, "", $a];
+    }
+
+    # We can have a sequence of symbols for a particular entry
+    # (more than one symbol in the case of inlining).  Callers
+    # come before callees in symlist, so walk backwards since
+    # the translated stack should contain callees before callers.
+    for (my $j = $#{$symlist}; $j >= 2; $j -= 3) {
+      my $func = $symlist->[$j-2];
+      my $fileline = $symlist->[$j-1];
+      my $fullfunc = $symlist->[$j];
+      if (defined($fullname_to_shortname_map->{$fullfunc})) {
+        $func = $fullname_to_shortname_map->{$fullfunc};
+      }
+      if ($j > 2) {
+        $func = "$func (inline)";
+      }
+
+      # Do not merge nodes corresponding to Callback::Run since that
+      # causes confusing cycles in dot display.  Instead, we synthesize
+      # a unique name for this frame per caller.
+      if ($func =~ m/Callback.*::Run$/) {
+        my $caller = ($i > 0) ? $addrs[$i-1] : 0;
+        $func = "Run#" . ShortIdFor($caller);
+      }
+
+      if ($main::opt_addresses) {
+        push(@result, "$a $func $fileline");
+      } elsif ($main::opt_lines) {
+        if ($func eq '??' && $fileline eq '??:0') {
+          push(@result, "$a");
+        } else {
+          push(@result, "$func $fileline");
+        }
+      } elsif ($main::opt_functions) {
+        if ($func eq '??') {
+          push(@result, "$a");
+        } else {
+          push(@result, $func);
+        }
+      } elsif ($main::opt_files) {
+        if ($fileline eq '??:0' || $fileline eq '') {
+          push(@result, "$a");
+        } else {
+          my $f = $fileline;
+          $f =~ s/:\d+$//;
+          push(@result, $f);
+        }
+      } else {
+        push(@result, $a);
+        last;  # Do not print inlined info
+      }
+    }
+  }
+
+  # print join(",", @addrs), " => ", join(",", @result), "\n";
+  return @result;
+}
+
+# Generate percent string for a number and a total
+sub Percent {
+  my $num = shift;
+  my $tot = shift;
+  if ($tot != 0) {
+    return sprintf("%.1f%%", $num * 100.0 / $tot);
+  } else {
+    return ($num == 0) ? "nan" : (($num > 0) ? "+inf" : "-inf");
+  }
+}
+
+# Generate pretty-printed form of number
+sub Unparse {
+  my $num = shift;
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    if ($main::opt_inuse_objects || $main::opt_alloc_objects) {
+      return sprintf("%d", $num);
+    } else {
+      if ($main::opt_show_bytes) {
+        return sprintf("%d", $num);
+      } else {
+        return sprintf("%.1f", $num / 1048576.0);
+      }
+    }
+  } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) {
+    return sprintf("%.3f", $num / 1e9); # Convert nanoseconds to seconds
+  } else {
+    return sprintf("%d", $num);
+  }
+}
+
+# Alternate pretty-printed form: 0 maps to "."
+sub UnparseAlt {
+  my $num = shift;
+  if ($num == 0) {
+    return ".";
+  } else {
+    return Unparse($num);
+  }
+}
+
+# Alternate pretty-printed form: 0 maps to ""
+sub HtmlPrintNumber {
+  my $num = shift;
+  if ($num == 0) {
+    return "";
+  } else {
+    return Unparse($num);
+  }
+}
+
+# Return output units
+sub Units {
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    if ($main::opt_inuse_objects || $main::opt_alloc_objects) {
+      return "objects";
+    } else {
+      if ($main::opt_show_bytes) {
+        return "B";
+      } else {
+        return "MB";
+      }
+    }
+  } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) {
+    return "seconds";
+  } else {
+    return "samples";
+  }
+}
+
+##### Profile manipulation code #####
+
+# Generate flattened profile:
+# If count is charged to stack [a,b,c,d], in generated profile,
+# it will be charged to [a]
+sub FlatProfile {
+  my $profile = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    if ($#addrs >= 0) {
+      AddEntry($result, $addrs[0], $count);
+    }
+  }
+  return $result;
+}
+
+# Generate cumulative profile:
+# If count is charged to stack [a,b,c,d], in generated profile,
+# it will be charged to [a], [b], [c], [d]
+sub CumulativeProfile {
+  my $profile = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    foreach my $a (@addrs) {
+      AddEntry($result, $a, $count);
+    }
+  }
+  return $result;
+}
+
+# If the second-youngest PC on the stack is always the same, returns
+# that pc.  Otherwise, returns undef.
+sub IsSecondPcAlwaysTheSame {
+  my $profile = shift;
+
+  my $second_pc = undef;
+  foreach my $k (keys(%{$profile})) {
+    my @addrs = split(/\n/, $k);
+    if ($#addrs < 1) {
+      return undef;
+    }
+    if (not defined $second_pc) {
+      $second_pc = $addrs[1];
+    } else {
+      if ($second_pc ne $addrs[1]) {
+        return undef;
+      }
+    }
+  }
+  return $second_pc;
+}
+
+sub ExtractSymbolNameInlineStack {
+  my $symbols = shift;
+  my $address = shift;
+
+  my @stack = ();
+
+  if (exists $symbols->{$address}) {
+    my @localinlinestack = @{$symbols->{$address}};
+    for (my $i = $#localinlinestack; $i > 0; $i-=3) {
+      my $file = $localinlinestack[$i-1];
+      my $fn = $localinlinestack[$i-0];
+
+      if ($file eq "?" || $file eq ":0") {
+        $file = "??:0";
+      }
+      if ($fn eq '??') {
+        # If we can't get the symbol name, at least use the file information.
+        $fn = $file;
+      }
+      my $suffix = "[inline]";
+      if ($i == 2) {
+        $suffix = "";
+      }
+      push (@stack, $fn.$suffix);
+    }
+  }
+  else {
+    # If we can't get a symbol name, at least fill in the address.
+    push (@stack, $address);
+  }
+
+  return @stack;
+}
+
+sub ExtractSymbolLocation {
+  my $symbols = shift;
+  my $address = shift;
+  # 'addr2line' outputs "??:0" for unknown locations; we do the
+  # same to be consistent.
+  my $location = "??:0:unknown";
+  if (exists $symbols->{$address}) {
+    my $file = $symbols->{$address}->[1];
+    if ($file eq "?") {
+      $file = "??:0"
+    }
+    $location = $file . ":" . $symbols->{$address}->[0];
+  }
+  return $location;
+}
+
+# Extracts a graph of calls.
+sub ExtractCalls {
+  my $symbols = shift;
+  my $profile = shift;
+
+  my $calls = {};
+  while( my ($stack_trace, $count) = each %$profile ) {
+    my @address = split(/\n/, $stack_trace);
+    my $destination = ExtractSymbolLocation($symbols, $address[0]);
+    AddEntry($calls, $destination, $count);
+    for (my $i = 1; $i <= $#address; $i++) {
+      my $source = ExtractSymbolLocation($symbols, $address[$i]);
+      my $call = "$source -> $destination";
+      AddEntry($calls, $call, $count);
+      $destination = $source;
+    }
+  }
+
+  return $calls;
+}
+
+sub FilterFrames {
+  my $symbols = shift;
+  my $profile = shift;
+
+  if ($main::opt_retain eq '' && $main::opt_exclude eq '') {
+    return $profile;
+  }
+
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my @path = ();
+    foreach my $a (@addrs) {
+      my $sym;
+      if (exists($symbols->{$a})) {
+        $sym = $symbols->{$a}->[0];
+      } else {
+        $sym = $a;
+      }
+      if ($main::opt_retain ne '' && $sym !~ m/$main::opt_retain/) {
+        next;
+      }
+      if ($main::opt_exclude ne '' && $sym =~ m/$main::opt_exclude/) {
+        next;
+      }
+      push(@path, $a);
+    }
+    if (scalar(@path) > 0) {
+      my $reduced_path = join("\n", @path);
+      AddEntry($result, $reduced_path, $count);
+    }
+  }
+
+  return $result;
+}
+
+sub PrintCollapsedStacks {
+  my $symbols = shift;
+  my $profile = shift;
+
+  while (my ($stack_trace, $count) = each %$profile) {
+    my @address = split(/\n/, $stack_trace);
+    my @names = reverse ( map { ExtractSymbolNameInlineStack($symbols, $_) } @address );
+    printf("%s %d\n", join(";", @names), $count);
+  }
+}
+
+sub RemoveUninterestingFrames {
+  my $symbols = shift;
+  my $profile = shift;
+
+  # List of function names to skip
+  my %skip = ();
+  my $skip_regexp = 'NOMATCH';
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    foreach my $name ('je_calloc',
+                      'cfree',
+                      'je_malloc',
+                      'newImpl',
+                      'void* newImpl',
+                      'je_free',
+                      'je_memalign',
+                      'je_posix_memalign',
+                      'je_aligned_alloc',
+                      'pvalloc',
+                      'je_valloc',
+                      'je_realloc',
+                      'je_mallocx',
+                      'je_rallocx',
+                      'je_xallocx',
+                      'je_dallocx',
+                      'je_sdallocx',
+                      'je_sdallocx_noflags',
+                      'tc_calloc',
+                      'tc_cfree',
+                      'tc_malloc',
+                      'tc_free',
+                      'tc_memalign',
+                      'tc_posix_memalign',
+                      'tc_pvalloc',
+                      'tc_valloc',
+                      'tc_realloc',
+                      'tc_new',
+                      'tc_delete',
+                      'tc_newarray',
+                      'tc_deletearray',
+                      'tc_new_nothrow',
+                      'tc_newarray_nothrow',
+                      'do_malloc',
+                      '::do_malloc',   # new name -- got moved to an unnamed ns
+                      '::do_malloc_or_cpp_alloc',
+                      'DoSampledAllocation',
+                      'simple_alloc::allocate',
+                      '__malloc_alloc_template::allocate',
+                      '__builtin_delete',
+                      '__builtin_new',
+                      '__builtin_vec_delete',
+                      '__builtin_vec_new',
+                      'operator new',
+                      'operator new[]',
+                      # The entry to our memory-allocation routines on OS X
+                      'malloc_zone_malloc',
+                      'malloc_zone_calloc',
+                      'malloc_zone_valloc',
+                      'malloc_zone_realloc',
+                      'malloc_zone_memalign',
+                      'malloc_zone_free',
+                      # These mark the beginning/end of our custom sections
+                      '__start_google_malloc',
+                      '__stop_google_malloc',
+                      '__start_malloc_hook',
+                      '__stop_malloc_hook') {
+      $skip{$name} = 1;
+      $skip{"_" . $name} = 1;   # Mach (OS X) adds a _ prefix to everything
+    }
+    # TODO: Remove TCMalloc once everything has been
+    # moved into the tcmalloc:: namespace and we have flushed
+    # old code out of the system.
+    $skip_regexp = "TCMalloc|^tcmalloc::";
+  } elsif ($main::profile_type eq 'contention') {
+    foreach my $vname ('base::RecordLockProfileData',
+                       'base::SubmitMutexProfileData',
+                       'base::SubmitSpinLockProfileData',
+                       'Mutex::Unlock',
+                       'Mutex::UnlockSlow',
+                       'Mutex::ReaderUnlock',
+                       'MutexLock::~MutexLock',
+                       'SpinLock::Unlock',
+                       'SpinLock::SlowUnlock',
+                       'SpinLockHolder::~SpinLockHolder') {
+      $skip{$vname} = 1;
+    }
+  } elsif ($main::profile_type eq 'cpu') {
+    # Drop signal handlers used for CPU profile collection
+    # TODO(dpeng): this should not be necessary; it's taken
+    # care of by the general 2nd-pc mechanism below.
+    foreach my $name ('ProfileData::Add',           # historical
+                      'ProfileData::prof_handler',  # historical
+                      'CpuProfiler::prof_handler',
+                      '__FRAME_END__',
+                      '__pthread_sighandler',
+                      '__restore') {
+      $skip{$name} = 1;
+    }
+  } else {
+    # Nothing skipped for unknown types
+  }
+
+  if ($main::profile_type eq 'cpu') {
+    # If all the second-youngest program counters are the same,
+    # this STRONGLY suggests that it is an artifact of measurement,
+    # i.e., stack frames pushed by the CPU profiler signal handler.
+    # Hence, we delete them.
+    # (The topmost PC is read from the signal structure, not from
+    # the stack, so it does not get involved.)
+    while (my $second_pc = IsSecondPcAlwaysTheSame($profile)) {
+      my $result = {};
+      my $func = '';
+      if (exists($symbols->{$second_pc})) {
+        $second_pc = $symbols->{$second_pc}->[0];
+      }
+      print STDERR "Removing $second_pc from all stack traces.\n";
+      foreach my $k (keys(%{$profile})) {
+        my $count = $profile->{$k};
+        my @addrs = split(/\n/, $k);
+        splice @addrs, 1, 1;
+        my $reduced_path = join("\n", @addrs);
+        AddEntry($result, $reduced_path, $count);
+      }
+      $profile = $result;
+    }
+  }
+
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my @path = ();
+    foreach my $a (@addrs) {
+      if (exists($symbols->{$a})) {
+        my $func = $symbols->{$a}->[0];
+        if ($skip{$func} || ($func =~ m/$skip_regexp/)) {
+          # Throw away the portion of the backtrace seen so far, under the
+          # assumption that previous frames were for functions internal to the
+          # allocator.
+          @path = ();
+          next;
+        }
+      }
+      push(@path, $a);
+    }
+    my $reduced_path = join("\n", @path);
+    AddEntry($result, $reduced_path, $count);
+  }
+
+  $result = FilterFrames($symbols, $result);
+
+  return $result;
+}
+
+# Reduce profile to granularity given by user
+sub ReduceProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $result = {};
+  my $fullname_to_shortname_map = {};
+  FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map);
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k);
+    my @path = ();
+    my %seen = ();
+    $seen{''} = 1;      # So that empty keys are skipped
+    foreach my $e (@translated) {
+      # To avoid double-counting due to recursion, skip a stack-trace
+      # entry if it has already been seen
+      if (!$seen{$e}) {
+        $seen{$e} = 1;
+        push(@path, $e);
+      }
+    }
+    my $reduced_path = join("\n", @path);
+    AddEntry($result, $reduced_path, $count);
+  }
+  return $result;
+}
+
+# Does the specified symbol array match the regexp?
+sub SymbolMatches {
+  my $sym = shift;
+  my $re = shift;
+  if (defined($sym)) {
+    for (my $i = 0; $i < $#{$sym}; $i += 3) {
+      if ($sym->[$i] =~ m/$re/ || $sym->[$i+1] =~ m/$re/) {
+        return 1;
+      }
+    }
+  }
+  return 0;
+}
+
+# Focus only on paths involving specified regexps
+sub FocusProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $focus = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    foreach my $a (@addrs) {
+      # Reply if it matches either the address/shortname/fileline
+      if (($a =~ m/$focus/) || SymbolMatches($symbols->{$a}, $focus)) {
+        AddEntry($result, $k, $count);
+        last;
+      }
+    }
+  }
+  return $result;
+}
+
+# Focus only on paths not involving specified regexps
+sub IgnoreProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $ignore = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my $matched = 0;
+    foreach my $a (@addrs) {
+      # Reply if it matches either the address/shortname/fileline
+      if (($a =~ m/$ignore/) || SymbolMatches($symbols->{$a}, $ignore)) {
+        $matched = 1;
+        last;
+      }
+    }
+    if (!$matched) {
+      AddEntry($result, $k, $count);
+    }
+  }
+  return $result;
+}
+
+# Get total count in profile
+sub TotalProfile {
+  my $profile = shift;
+  my $result = 0;
+  foreach my $k (keys(%{$profile})) {
+    $result += $profile->{$k};
+  }
+  return $result;
+}
+
+# Add A to B
+sub AddProfile {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  # add all keys in A
+  foreach my $k (keys(%{$A})) {
+    my $v = $A->{$k};
+    AddEntry($R, $k, $v);
+  }
+  # add all keys in B
+  foreach my $k (keys(%{$B})) {
+    my $v = $B->{$k};
+    AddEntry($R, $k, $v);
+  }
+  return $R;
+}
+
+# Merges symbol maps
+sub MergeSymbols {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  foreach my $k (keys(%{$A})) {
+    $R->{$k} = $A->{$k};
+  }
+  if (defined($B)) {
+    foreach my $k (keys(%{$B})) {
+      $R->{$k} = $B->{$k};
+    }
+  }
+  return $R;
+}
+
+
+# Add A to B
+sub AddPcs {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  # add all keys in A
+  foreach my $k (keys(%{$A})) {
+    $R->{$k} = 1
+  }
+  # add all keys in B
+  foreach my $k (keys(%{$B})) {
+    $R->{$k} = 1
+  }
+  return $R;
+}
+
+# Subtract B from A
+sub SubtractProfile {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  foreach my $k (keys(%{$A})) {
+    my $v = $A->{$k} - GetEntry($B, $k);
+    if ($v < 0 && $main::opt_drop_negative) {
+      $v = 0;
+    }
+    AddEntry($R, $k, $v);
+  }
+  if (!$main::opt_drop_negative) {
+    # Take care of when subtracted profile has more entries
+    foreach my $k (keys(%{$B})) {
+      if (!exists($A->{$k})) {
+        AddEntry($R, $k, 0 - $B->{$k});
+      }
+    }
+  }
+  return $R;
+}
+
+# Get entry from profile; zero if not present
+sub GetEntry {
+  my $profile = shift;
+  my $k = shift;
+  if (exists($profile->{$k})) {
+    return $profile->{$k};
+  } else {
+    return 0;
+  }
+}
+
+# Add entry to specified profile
+sub AddEntry {
+  my $profile = shift;
+  my $k = shift;
+  my $n = shift;
+  if (!exists($profile->{$k})) {
+    $profile->{$k} = 0;
+  }
+  $profile->{$k} += $n;
+}
+
+# Add a stack of entries to specified profile, and add them to the $pcs
+# list.
+sub AddEntries {
+  my $profile = shift;
+  my $pcs = shift;
+  my $stack = shift;
+  my $count = shift;
+  my @k = ();
+
+  foreach my $e (split(/\s+/, $stack)) {
+    my $pc = HexExtend($e);
+    $pcs->{$pc} = 1;
+    push @k, $pc;
+  }
+  AddEntry($profile, (join "\n", @k), $count);
+}
+
+##### Code to profile a server dynamically #####
+
+sub CheckSymbolPage {
+  my $url = SymbolPageURL();
+  my $command = ShellEscape(@URL_FETCHER, $url);
+  open(SYMBOL, "$command |") or error($command);
+  my $line = <SYMBOL>;
+  $line =~ s/\r//g;         # turn windows-looking lines into unix-looking lines
+  close(SYMBOL);
+  unless (defined($line)) {
+    error("$url doesn't exist\n");
+  }
+
+  if ($line =~ /^num_symbols:\s+(\d+)$/) {
+    if ($1 == 0) {
+      error("Stripped binary. No symbols available.\n");
+    }
+  } else {
+    error("Failed to get the number of symbols from $url\n");
+  }
+}
+
+sub IsProfileURL {
+  my $profile_name = shift;
+  if (-f $profile_name) {
+    printf STDERR "Using local file $profile_name.\n";
+    return 0;
+  }
+  return 1;
+}
+
+sub ParseProfileURL {
+  my $profile_name = shift;
+
+  if (!defined($profile_name) || $profile_name eq "") {
+    return ();
+  }
+
+  # Split profile URL - matches all non-empty strings, so no test.
+  $profile_name =~ m,^(https?://)?([^/]+)(.*?)(/|$PROFILES)?$,;
+
+  my $proto = $1 || "http://";
+  my $hostport = $2;
+  my $prefix = $3;
+  my $profile = $4 || "/";
+
+  my $host = $hostport;
+  $host =~ s/:.*//;
+
+  my $baseurl = "$proto$hostport$prefix";
+  return ($host, $baseurl, $profile);
+}
+
+# We fetch symbols from the first profile argument.
+sub SymbolPageURL {
+  my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]);
+  return "$baseURL$SYMBOL_PAGE";
+}
+
+sub FetchProgramName() {
+  my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]);
+  my $url = "$baseURL$PROGRAM_NAME_PAGE";
+  my $command_line = ShellEscape(@URL_FETCHER, $url);
+  open(CMDLINE, "$command_line |") or error($command_line);
+  my $cmdline = <CMDLINE>;
+  $cmdline =~ s/\r//g;   # turn windows-looking lines into unix-looking lines
+  close(CMDLINE);
+  error("Failed to get program name from $url\n") unless defined($cmdline);
+  $cmdline =~ s/\x00.+//;  # Remove argv[1] and latters.
+  $cmdline =~ s!\n!!g;  # Remove LFs.
+  return $cmdline;
+}
+
+# Gee, curl's -L (--location) option isn't reliable at least
+# with its 7.12.3 version.  Curl will forget to post data if
+# there is a redirection.  This function is a workaround for
+# curl.  Redirection happens on borg hosts.
+sub ResolveRedirectionForCurl {
+  my $url = shift;
+  my $command_line = ShellEscape(@URL_FETCHER, "--head", $url);
+  open(CMDLINE, "$command_line |") or error($command_line);
+  while (<CMDLINE>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    if (/^Location: (.*)/) {
+      $url = $1;
+    }
+  }
+  close(CMDLINE);
+  return $url;
+}
+
+# Add a timeout flat to URL_FETCHER.  Returns a new list.
+sub AddFetchTimeout {
+  my $timeout = shift;
+  my @fetcher = @_;
+  if (defined($timeout)) {
+    if (join(" ", @fetcher) =~ m/\bcurl -s/) {
+      push(@fetcher, "--max-time", sprintf("%d", $timeout));
+    } elsif (join(" ", @fetcher) =~ m/\brpcget\b/) {
+      push(@fetcher, sprintf("--deadline=%d", $timeout));
+    }
+  }
+  return @fetcher;
+}
+
+# Reads a symbol map from the file handle name given as $1, returning
+# the resulting symbol map.  Also processes variables relating to symbols.
+# Currently, the only variable processed is 'binary=<value>' which updates
+# $main::prog to have the correct program name.
+sub ReadSymbols {
+  my $in = shift;
+  my $map = {};
+  while (<$in>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    # Removes all the leading zeroes from the symbols, see comment below.
+    if (m/^0x0*([0-9a-f]+)\s+(.+)/) {
+      $map->{$1} = $2;
+    } elsif (m/^---/) {
+      last;
+    } elsif (m/^([a-z][^=]*)=(.*)$/ ) {
+      my ($variable, $value) = ($1, $2);
+      for ($variable, $value) {
+        s/^\s+//;
+        s/\s+$//;
+      }
+      if ($variable eq "binary") {
+        if ($main::prog ne $UNKNOWN_BINARY && $main::prog ne $value) {
+          printf STDERR ("Warning: Mismatched binary name '%s', using '%s'.\n",
+                         $main::prog, $value);
+        }
+        $main::prog = $value;
+      } else {
+        printf STDERR ("Ignoring unknown variable in symbols list: " .
+            "'%s' = '%s'\n", $variable, $value);
+      }
+    }
+  }
+  return $map;
+}
+
+sub URLEncode {
+  my $str = shift;
+  $str =~ s/([^A-Za-z0-9\-_.!~*'()])/ sprintf "%%%02x", ord $1 /eg;
+  return $str;
+}
+
+sub AppendSymbolFilterParams {
+  my $url = shift;
+  my @params = ();
+  if ($main::opt_retain ne '') {
+    push(@params, sprintf("retain=%s", URLEncode($main::opt_retain)));
+  }
+  if ($main::opt_exclude ne '') {
+    push(@params, sprintf("exclude=%s", URLEncode($main::opt_exclude)));
+  }
+  if (scalar @params > 0) {
+    $url = sprintf("%s?%s", $url, join("&", @params));
+  }
+  return $url;
+}
+
+# Fetches and processes symbols to prepare them for use in the profile output
+# code.  If the optional 'symbol_map' arg is not given, fetches symbols from
+# $SYMBOL_PAGE for all PC values found in profile.  Otherwise, the raw symbols
+# are assumed to have already been fetched into 'symbol_map' and are simply
+# extracted and processed.
+sub FetchSymbols {
+  my $pcset = shift;
+  my $symbol_map = shift;
+
+  my %seen = ();
+  my @pcs = grep { !$seen{$_}++ } keys(%$pcset);  # uniq
+
+  if (!defined($symbol_map)) {
+    my $post_data = join("+", sort((map {"0x" . "$_"} @pcs)));
+
+    open(POSTFILE, ">$main::tmpfile_sym");
+    print POSTFILE $post_data;
+    close(POSTFILE);
+
+    my $url = SymbolPageURL();
+
+    my $command_line;
+    if (join(" ", @URL_FETCHER) =~ m/\bcurl -s/) {
+      $url = ResolveRedirectionForCurl($url);
+      $url = AppendSymbolFilterParams($url);
+      $command_line = ShellEscape(@URL_FETCHER, "-d", "\@$main::tmpfile_sym",
+                                  $url);
+    } else {
+      $url = AppendSymbolFilterParams($url);
+      $command_line = (ShellEscape(@URL_FETCHER, "--post", $url)
+                       . " < " . ShellEscape($main::tmpfile_sym));
+    }
+    # We use c++filt in case $SYMBOL_PAGE gives us mangled symbols.
+    my $escaped_cppfilt = ShellEscape($obj_tool_map{"c++filt"});
+    open(SYMBOL, "$command_line | $escaped_cppfilt |") or error($command_line);
+    $symbol_map = ReadSymbols(*SYMBOL{IO});
+    close(SYMBOL);
+  }
+
+  my $symbols = {};
+  foreach my $pc (@pcs) {
+    my $fullname;
+    # For 64 bits binaries, symbols are extracted with 8 leading zeroes.
+    # Then /symbol reads the long symbols in as uint64, and outputs
+    # the result with a "0x%08llx" format which get rid of the zeroes.
+    # By removing all the leading zeroes in both $pc and the symbols from
+    # /symbol, the symbols match and are retrievable from the map.
+    my $shortpc = $pc;
+    $shortpc =~ s/^0*//;
+    # Each line may have a list of names, which includes the function
+    # and also other functions it has inlined.  They are separated (in
+    # PrintSymbolizedProfile), by --, which is illegal in function names.
+    my $fullnames;
+    if (defined($symbol_map->{$shortpc})) {
+      $fullnames = $symbol_map->{$shortpc};
+    } else {
+      $fullnames = "0x" . $pc;  # Just use addresses
+    }
+    my $sym = [];
+    $symbols->{$pc} = $sym;
+    foreach my $fullname (split("--", $fullnames)) {
+      my $name = ShortFunctionName($fullname);
+      push(@{$sym}, $name, "?", $fullname);
+    }
+  }
+  return $symbols;
+}
+
+sub BaseName {
+  my $file_name = shift;
+  $file_name =~ s!^.*/!!;  # Remove directory name
+  return $file_name;
+}
+
+sub MakeProfileBaseName {
+  my ($binary_name, $profile_name) = @_;
+  my ($host, $baseURL, $path) = ParseProfileURL($profile_name);
+  my $binary_shortname = BaseName($binary_name);
+  return sprintf("%s.%s.%s",
+                 $binary_shortname, $main::op_time, $host);
+}
+
+sub FetchDynamicProfile {
+  my $binary_name = shift;
+  my $profile_name = shift;
+  my $fetch_name_only = shift;
+  my $encourage_patience = shift;
+
+  if (!IsProfileURL($profile_name)) {
+    return $profile_name;
+  } else {
+    my ($host, $baseURL, $path) = ParseProfileURL($profile_name);
+    if ($path eq "" || $path eq "/") {
+      # Missing type specifier defaults to cpu-profile
+      $path = $PROFILE_PAGE;
+    }
+
+    my $profile_file = MakeProfileBaseName($binary_name, $profile_name);
+
+    my $url = "$baseURL$path";
+    my $fetch_timeout = undef;
+    if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE/) {
+      if ($path =~ m/[?]/) {
+        $url .= "&";
+      } else {
+        $url .= "?";
+      }
+      $url .= sprintf("seconds=%d", $main::opt_seconds);
+      $fetch_timeout = $main::opt_seconds * 1.01 + 60;
+      # Set $profile_type for consumption by PrintSymbolizedProfile.
+      $main::profile_type = 'cpu';
+    } else {
+      # For non-CPU profiles, we add a type-extension to
+      # the target profile file name.
+      my $suffix = $path;
+      $suffix =~ s,/,.,g;
+      $profile_file .= $suffix;
+      # Set $profile_type for consumption by PrintSymbolizedProfile.
+      if ($path =~ m/$HEAP_PAGE/) {
+        $main::profile_type = 'heap';
+      } elsif ($path =~ m/$GROWTH_PAGE/) {
+        $main::profile_type = 'growth';
+      } elsif ($path =~ m/$CONTENTION_PAGE/) {
+        $main::profile_type = 'contention';
+      }
+    }
+
+    my $profile_dir = $ENV{"JEPROF_TMPDIR"} || ($ENV{HOME} . "/jeprof");
+    if (! -d $profile_dir) {
+      mkdir($profile_dir)
+          || die("Unable to create profile directory $profile_dir: $!\n");
+    }
+    my $tmp_profile = "$profile_dir/.tmp.$profile_file";
+    my $real_profile = "$profile_dir/$profile_file";
+
+    if ($fetch_name_only > 0) {
+      return $real_profile;
+    }
+
+    my @fetcher = AddFetchTimeout($fetch_timeout, @URL_FETCHER);
+    my $cmd = ShellEscape(@fetcher, $url) . " > " . ShellEscape($tmp_profile);
+    if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE|$CENSUSPROFILE_PAGE/){
+      print STDERR "Gathering CPU profile from $url for $main::opt_seconds seconds to\n  ${real_profile}\n";
+      if ($encourage_patience) {
+        print STDERR "Be patient...\n";
+      }
+    } else {
+      print STDERR "Fetching $path profile from $url to\n  ${real_profile}\n";
+    }
+
+    (system($cmd) == 0) || error("Failed to get profile: $cmd: $!\n");
+    (system("mv", $tmp_profile, $real_profile) == 0) || error("Unable to rename profile\n");
+    print STDERR "Wrote profile to $real_profile\n";
+    $main::collected_profile = $real_profile;
+    return $main::collected_profile;
+  }
+}
+
+# Collect profiles in parallel
+sub FetchDynamicProfiles {
+  my $items = scalar(@main::pfile_args);
+  my $levels = log($items) / log(2);
+
+  if ($items == 1) {
+    $main::profile_files[0] = FetchDynamicProfile($main::prog, $main::pfile_args[0], 0, 1);
+  } else {
+    # math rounding issues
+    if ((2 ** $levels) < $items) {
+     $levels++;
+    }
+    my $count = scalar(@main::pfile_args);
+    for (my $i = 0; $i < $count; $i++) {
+      $main::profile_files[$i] = FetchDynamicProfile($main::prog, $main::pfile_args[$i], 1, 0);
+    }
+    print STDERR "Fetching $count profiles, Be patient...\n";
+    FetchDynamicProfilesRecurse($levels, 0, 0);
+    $main::collected_profile = join(" \\\n    ", @main::profile_files);
+  }
+}
+
+# Recursively fork a process to get enough processes
+# collecting profiles
+sub FetchDynamicProfilesRecurse {
+  my $maxlevel = shift;
+  my $level = shift;
+  my $position = shift;
+
+  if (my $pid = fork()) {
+    $position = 0 | ($position << 1);
+    TryCollectProfile($maxlevel, $level, $position);
+    wait;
+  } else {
+    $position = 1 | ($position << 1);
+    TryCollectProfile($maxlevel, $level, $position);
+    cleanup();
+    exit(0);
+  }
+}
+
+# Collect a single profile
+sub TryCollectProfile {
+  my $maxlevel = shift;
+  my $level = shift;
+  my $position = shift;
+
+  if ($level >= ($maxlevel - 1)) {
+    if ($position < scalar(@main::pfile_args)) {
+      FetchDynamicProfile($main::prog, $main::pfile_args[$position], 0, 0);
+    }
+  } else {
+    FetchDynamicProfilesRecurse($maxlevel, $level+1, $position);
+  }
+}
+
+##### Parsing code #####
+
+# Provide a small streaming-read module to handle very large
+# cpu-profile files.  Stream in chunks along a sliding window.
+# Provides an interface to get one 'slot', correctly handling
+# endian-ness differences.  A slot is one 32-bit or 64-bit word
+# (depending on the input profile).  We tell endianness and bit-size
+# for the profile by looking at the first 8 bytes: in cpu profiles,
+# the second slot is always 3 (we'll accept anything that's not 0).
+BEGIN {
+  package CpuProfileStream;
+
+  sub new {
+    my ($class, $file, $fname) = @_;
+    my $self = { file        => $file,
+                 base        => 0,
+                 stride      => 512 * 1024,   # must be a multiple of bitsize/8
+                 slots       => [],
+                 unpack_code => "",           # N for big-endian, V for little
+                 perl_is_64bit => 1,          # matters if profile is 64-bit
+    };
+    bless $self, $class;
+    # Let unittests adjust the stride
+    if ($main::opt_test_stride > 0) {
+      $self->{stride} = $main::opt_test_stride;
+    }
+    # Read the first two slots to figure out bitsize and endianness.
+    my $slots = $self->{slots};
+    my $str;
+    read($self->{file}, $str, 8);
+    # Set the global $address_length based on what we see here.
+    # 8 is 32-bit (8 hexadecimal chars); 16 is 64-bit (16 hexadecimal chars).
+    $address_length = ($str eq (chr(0)x8)) ? 16 : 8;
+    if ($address_length == 8) {
+      if (substr($str, 6, 2) eq chr(0)x2) {
+        $self->{unpack_code} = 'V';  # Little-endian.
+      } elsif (substr($str, 4, 2) eq chr(0)x2) {
+        $self->{unpack_code} = 'N';  # Big-endian
+      } else {
+        ::error("$fname: header size >= 2**16\n");
+      }
+      @$slots = unpack($self->{unpack_code} . "*", $str);
+    } else {
+      # If we're a 64-bit profile, check if we're a 64-bit-capable
+      # perl.  Otherwise, each slot will be represented as a float
+      # instead of an int64, losing precision and making all the
+      # 64-bit addresses wrong.  We won't complain yet, but will
+      # later if we ever see a value that doesn't fit in 32 bits.
+      my $has_q = 0;
+      eval { $has_q = pack("Q", "1") ? 1 : 1; };
+      if (!$has_q) {
+        $self->{perl_is_64bit} = 0;
+      }
+      read($self->{file}, $str, 8);
+      if (substr($str, 4, 4) eq chr(0)x4) {
+        # We'd love to use 'Q', but it's a) not universal, b) not endian-proof.
+        $self->{unpack_code} = 'V';  # Little-endian.
+      } elsif (substr($str, 0, 4) eq chr(0)x4) {
+        $self->{unpack_code} = 'N';  # Big-endian
+      } else {
+        ::error("$fname: header size >= 2**32\n");
+      }
+      my @pair = unpack($self->{unpack_code} . "*", $str);
+      # Since we know one of the pair is 0, it's fine to just add them.
+      @$slots = (0, $pair[0] + $pair[1]);
+    }
+    return $self;
+  }
+
+  # Load more data when we access slots->get(X) which is not yet in memory.
+  sub overflow {
+    my ($self) = @_;
+    my $slots = $self->{slots};
+    $self->{base} += $#$slots + 1;   # skip over data we're replacing
+    my $str;
+    read($self->{file}, $str, $self->{stride});
+    if ($address_length == 8) {      # the 32-bit case
+      # This is the easy case: unpack provides 32-bit unpacking primitives.
+      @$slots = unpack($self->{unpack_code} . "*", $str);
+    } else {
+      # We need to unpack 32 bits at a time and combine.
+      my @b32_values = unpack($self->{unpack_code} . "*", $str);
+      my @b64_values = ();
+      for (my $i = 0; $i < $#b32_values; $i += 2) {
+        # TODO(csilvers): if this is a 32-bit perl, the math below
+        #    could end up in a too-large int, which perl will promote
+        #    to a double, losing necessary precision.  Deal with that.
+        #    Right now, we just die.
+        my ($lo, $hi) = ($b32_values[$i], $b32_values[$i+1]);
+        if ($self->{unpack_code} eq 'N') {    # big-endian
+          ($lo, $hi) = ($hi, $lo);
+        }
+        my $value = $lo + $hi * (2**32);
+        if (!$self->{perl_is_64bit} &&   # check value is exactly represented
+            (($value % (2**32)) != $lo || int($value / (2**32)) != $hi)) {
+          ::error("Need a 64-bit perl to process this 64-bit profile.\n");
+        }
+        push(@b64_values, $value);
+      }
+      @$slots = @b64_values;
+    }
+  }
+
+  # Access the i-th long in the file (logically), or -1 at EOF.
+  sub get {
+    my ($self, $idx) = @_;
+    my $slots = $self->{slots};
+    while ($#$slots >= 0) {
+      if ($idx < $self->{base}) {
+        # The only time we expect a reference to $slots[$i - something]
+        # after referencing $slots[$i] is reading the very first header.
+        # Since $stride > |header|, that shouldn't cause any lookback
+        # errors.  And everything after the header is sequential.
+        print STDERR "Unexpected look-back reading CPU profile";
+        return -1;   # shrug, don't know what better to return
+      } elsif ($idx > $self->{base} + $#$slots) {
+        $self->overflow();
+      } else {
+        return $slots->[$idx - $self->{base}];
+      }
+    }
+    # If we get here, $slots is [], which means we've reached EOF
+    return -1;  # unique since slots is supposed to hold unsigned numbers
+  }
+}
+
+# Reads the top, 'header' section of a profile, and returns the last
+# line of the header, commonly called a 'header line'.  The header
+# section of a profile consists of zero or more 'command' lines that
+# are instructions to jeprof, which jeprof executes when reading the
+# header.  All 'command' lines start with a %.  After the command
+# lines is the 'header line', which is a profile-specific line that
+# indicates what type of profile it is, and perhaps other global
+# information about the profile.  For instance, here's a header line
+# for a heap profile:
+#   heap profile:     53:    38236 [  5525:  1284029] @ heapprofile
+# For historical reasons, the CPU profile does not contain a text-
+# readable header line.  If the profile looks like a CPU profile,
+# this function returns "".  If no header line could be found, this
+# function returns undef.
+#
+# The following commands are recognized:
+#   %warn -- emit the rest of this line to stderr, prefixed by 'WARNING:'
+#
+# The input file should be in binmode.
+sub ReadProfileHeader {
+  local *PROFILE = shift;
+  my $firstchar = "";
+  my $line = "";
+  read(PROFILE, $firstchar, 1);
+  seek(PROFILE, -1, 1);                    # unread the firstchar
+  if ($firstchar !~ /[[:print:]]/) {       # is not a text character
+    return "";
+  }
+  while (defined($line = <PROFILE>)) {
+    $line =~ s/\r//g;   # turn windows-looking lines into unix-looking lines
+    if ($line =~ /^%warn\s+(.*)/) {        # 'warn' command
+      # Note this matches both '%warn blah\n' and '%warn\n'.
+      print STDERR "WARNING: $1\n";        # print the rest of the line
+    } elsif ($line =~ /^%/) {
+      print STDERR "Ignoring unknown command from profile header: $line";
+    } else {
+      # End of commands, must be the header line.
+      return $line;
+    }
+  }
+  return undef;     # got to EOF without seeing a header line
+}
+
+sub IsSymbolizedProfileFile {
+  my $file_name = shift;
+  if (!(-e $file_name) || !(-r $file_name)) {
+    return 0;
+  }
+  # Check if the file contains a symbol-section marker.
+  open(TFILE, "<$file_name");
+  binmode TFILE;
+  my $firstline = ReadProfileHeader(*TFILE);
+  close(TFILE);
+  if (!$firstline) {
+    return 0;
+  }
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+  return $firstline =~ /^--- *$symbol_marker/;
+}
+
+# Parse profile generated by common/profiler.cc and return a reference
+# to a map:
+#      $result->{version}     Version number of profile file
+#      $result->{period}      Sampling period (in microseconds)
+#      $result->{profile}     Profile object
+#      $result->{threads}     Map of thread IDs to profile objects
+#      $result->{map}         Memory map info from profile
+#      $result->{pcs}         Hash of all PC values seen, key is hex address
+sub ReadProfile {
+  my $prog = shift;
+  my $fname = shift;
+  my $result;            # return value
+
+  $CONTENTION_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $contention_marker = $&;
+  $GROWTH_PAGE  =~ m,[^/]+$,;    # matches everything after the last slash
+  my $growth_marker = $&;
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+  $PROFILE_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $profile_marker = $&;
+  $HEAP_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $heap_marker = $&;
+
+  # Look at first line to see if it is a heap or a CPU profile.
+  # CPU profile may start with no header at all, and just binary data
+  # (starting with \0\0\0\0) -- in that case, don't try to read the
+  # whole firstline, since it may be gigabytes(!) of data.
+  open(PROFILE, "<$fname") || error("$fname: $!\n");
+  binmode PROFILE;      # New perls do UTF-8 processing
+  my $header = ReadProfileHeader(*PROFILE);
+  if (!defined($header)) {   # means "at EOF"
+    error("Profile is empty.\n");
+  }
+
+  my $symbols;
+  if ($header =~ m/^--- *$symbol_marker/o) {
+    # Verify that the user asked for a symbolized profile
+    if (!$main::use_symbolized_profile) {
+      # we have both a binary and symbolized profiles, abort
+      error("FATAL ERROR: Symbolized profile\n   $fname\ncannot be used with " .
+            "a binary arg. Try again without passing\n   $prog\n");
+    }
+    # Read the symbol section of the symbolized profile file.
+    $symbols = ReadSymbols(*PROFILE{IO});
+    # Read the next line to get the header for the remaining profile.
+    $header = ReadProfileHeader(*PROFILE) || "";
+  }
+
+  if ($header =~ m/^--- *($heap_marker|$growth_marker)/o) {
+    # Skip "--- ..." line for profile types that have their own headers.
+    $header = ReadProfileHeader(*PROFILE) || "";
+  }
+
+  $main::profile_type = '';
+
+  if ($header =~ m/^heap profile:.*$growth_marker/o) {
+    $main::profile_type = 'growth';
+    $result =  ReadHeapProfile($prog, *PROFILE, $header);
+  } elsif ($header =~ m/^heap profile:/) {
+    $main::profile_type = 'heap';
+    $result =  ReadHeapProfile($prog, *PROFILE, $header);
+  } elsif ($header =~ m/^heap/) {
+    $main::profile_type = 'heap';
+    $result = ReadThreadedHeapProfile($prog, $fname, $header);
+  } elsif ($header =~ m/^--- *$contention_marker/o) {
+    $main::profile_type = 'contention';
+    $result = ReadSynchProfile($prog, *PROFILE);
+  } elsif ($header =~ m/^--- *Stacks:/) {
+    print STDERR
+      "Old format contention profile: mistakenly reports " .
+      "condition variable signals as lock contentions.\n";
+    $main::profile_type = 'contention';
+    $result = ReadSynchProfile($prog, *PROFILE);
+  } elsif ($header =~ m/^--- *$profile_marker/) {
+    # the binary cpu profile data starts immediately after this line
+    $main::profile_type = 'cpu';
+    $result = ReadCPUProfile($prog, $fname, *PROFILE);
+  } else {
+    if (defined($symbols)) {
+      # a symbolized profile contains a format we don't recognize, bail out
+      error("$fname: Cannot recognize profile section after symbols.\n");
+    }
+    # no ascii header present -- must be a CPU profile
+    $main::profile_type

[truncated 1061424 bytes]

ready patch attempt

patch

redis-check-rdb likely remains stuck in a busy-poll loop. A diagnosis report and patch proposal were created locally.

state: readycreated: 2026-05-10 13:20 UTCvalidation: ready
Published session

Prompt

## Plan Pass

You are planning a fixer patch before any edits happen.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. The original pre-edit snapshot is available at `./source` if you need to inspect it. For interpreter processes, plan from the script/application entrypoint evidence first and include the runtime only as a second investigation target unless the evidence proves a runtime bug.

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround. Inspect the relevant code, nearby callers, project contribution docs, and local helper/compat APIs, but do not edit files in this pass.

Return a short markdown plan with these exact sections:

## Problem
## Evidence Confidence
## Proposed Subject
## Patch Plan
## Risks
## Validation

Classify `## Evidence Confidence` as exactly one of `reproduced`, `observed`, or `inferred`. Use `inferred` only for a no-patch diagnosis/report plan unless you can name the extra evidence you will collect before editing; inferred source patches are blocked by Fixer because they are not pull-request-ready. For `observed` source-patch plans, plan to say in the final `## Issue Connection` that the failure was observed by Fixer and not independently reproduced. The plan must explain how the proposed code change addresses the observed issue evidence, call out any prior Fixer patch that should be improved or replaced, reject awkward control flow such as avoidable `goto` if there is a cleaner bounded alternative, name any local helper APIs or maintainer conventions the patch should follow, and keep the intended maintainer-facing explanation clear enough that someone unfamiliar with the local complaint wording can still follow the fix. In `## Validation`, name the reproducible configure/build/test entrypoint you will try from the workspace root before any focused leaf compile or smoke check, and include one bounded independent reproduction attempt for the collected failure signal when it is safe and cheap. Do not plan to claim `reproduced` unless that reproduction command or test can actually show the failure.

## Patch Pass

You are working on a bounded fixer proposal.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. Produce the smallest reasonable patch for the target repository, keep the change upstreamable, prefer the clearest control flow available, and do not keep avoidable `goto` when a simpler structure would read better. Before introducing new file, process, allocation, locking, networking, or platform APIs, inspect nearby code and project contribution docs for existing helpers or compatibility wrappers and use those local patterns unless you can explain why they do not fit. Validate from a reproducible workspace-root entrypoint before falling back to focused leaf commands; if a build or test cannot run, report the exact command, the exact blocker, and any narrower check you ran instead. During validation, also try one bounded independent reproduction of the collected failure signal when it is safe and cheap, such as a failing test, smoke command, perf/strace comparison, or before/after runtime check. Only use `reproduced` if that command or test actually reproduced the failure; otherwise keep `observed` and report the reproduction blocker. The final explanation must connect the observed issue evidence to the actual code change, not just paraphrase the diff. Write like a maintainer is going to read the patch mail cold: explain the bug in plain language, define subsystem-specific jargon the first time you need it, and make the causal story obvious. Explicitly classify evidence confidence as `reproduced`, `observed`, or `inferred`: `reproduced` means you reproduced the failure locally; `observed` means Fixer has direct crash/log/trace evidence but you did not independently reproduce it; `inferred` means the source patch is not pull-request-ready, so do not leave a source diff unless you first gather stronger observed/reproduced evidence; otherwise return a no-patch diagnosis/report. For any source-changing `observed` patch, say explicitly in `## Issue Connection` that the failure was observed by Fixer and not independently reproduced. If you introduce non-obvious state translation, index remapping, or backend split logic, add a short source comment that explains the invariant being preserved.

Start by explaining the likely root cause from the collected perf, strace, and /proc evidence. If you cannot land a safe patch, leave a diagnosis that is strong enough for an upstream bug report.

A previous Fixer patch attempt already exists for this issue. It was generated by Fixer `0.145.0`. Review that patch before changing code, improve it instead of starting blind, and clean up anything awkward or underexplained. In particular, remove avoidable `goto`, tighten the explanation of what the patch is doing, and make the resulting diff feel ready for upstream git review.
- Prior patch: `./prior-best.patch`
- Prior published session: `./prior-best-session.md`

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround. 

Keep the change narrowly scoped and summarize validation clearly.

In every authoring pass, your final response must start with `Subject: <single-line git commit subject>` and then include these markdown sections exactly:

## Commit Message
A short upstream-friendly explanation of what changed and why. Write it in plain language that a maintainer can follow without local complaint context. If you use subsystem jargon, define it immediately.

## Evidence Confidence
Exactly one word: `reproduced`, `observed`, or `inferred`. Use `reproduced` only when you reproduced the failure locally with a command or test, and include that command/test in `## Validation`. Use `observed` when Fixer has direct crash/log/trace evidence but you did not independently reproduce it. If `## Git Add Paths` lists source files for an `observed` patch, `## Issue Connection` must explicitly say the failure was observed by Fixer and not independently reproduced. Use `inferred` for profiler/strace/indirect evidence; inferred responses may be no-patch diagnoses or reports, but inferred source patches are not pull-request-ready until stronger evidence is gathered.

## Issue Connection
Write this as maintainer-facing patch mail, not as local Fixer notes. Cover four things explicitly in readable sentences: the user-visible symptom or the exact collected signal, the code-level cause or the cautious inference from evidence, the specific change you made, and the expected effect. Do not invent a reproducer, command line, crash, or user-visible failure that is not present in the evidence bundle. If the evidence is direct-but-not-reproduced, say it was observed by Fixer and not independently reproduced. If the evidence is indirect and you did not gather stronger evidence, do not leave a source diff; write a no-patch diagnosis/report instead. Include an explicit effect sentence such as `The expected effect is ...`, `This should reduce ...`, or `This prevents ...` for source patches. If the logic is non-obvious in code, mention that you added a short explanatory comment.

## Git Add Paths
List the repo-relative paths that belong in the final patch, one per line. Use `None` only when you intentionally made no source changes. Include intentionally new files, and do not list generated build artifacts.

## Validation
List the checks you ran, or say clearly that you could not run them. Include the independent reproduction command/test and result when `## Evidence Confidence` is `reproduced`; if reproduction was attempted but blocked, name the exact blocker and keep confidence at `observed` or `inferred`.

Before editing, read the plan at `./plan-output.txt` and follow it unless the code proves part of it wrong. If you change course, say so explicitly in the final write-up instead of silently drifting from the plan.

## Review Pass 1

You are reviewing a freshly generated fixer patch.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. Review the first patch pass. The original pre-edit snapshot is available at `./source` for diffing.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround.

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout. The latest author response is at `./patch-output.txt`. Inspect the current code and changed paths like a strict code reviewer. Focus on correctness, regressions, maintainability, awkward control flow such as avoidable `goto`, missing validation, weak or non-gittable commit message text, and explanations that fail to connect the observed issue evidence to the code change. Also review the maintainer experience: the patch mail should be easy to accept upstream, the user-visible bug should be explained in plain language, subsystem-specific jargon should be defined when first used, and any non-obvious state translation or index remapping in code should have a short explanatory comment. Reject patches that omit `## Evidence Confidence`, use anything other than `reproduced`, `observed`, or `inferred`, or sound more certain than that evidence level allows. Reject patches that claim `reproduced` unless `## Validation` names the concrete reproduction command or test and its result. Reject patches that invent a reproducer, command line, crash, or user-visible failure not present in the evidence bundle; indirect profiler or strace evidence may justify a cautious mitigation, but the author must say it is indirect instead of presenting it as a confirmed upstream bug. Reject patches that introduce generic libc/std APIs when nearby code or project docs provide a local compat/helper API for the same job, unless the author explicitly justifies the exception. Reject validation that only reports a leaf object/syntax build when a project-level configure/build/test entrypoint exists and was not attempted; if a full build cannot run, the author must show the exact failed command and blocker.

Do not apply code changes in this pass.

Return a short markdown review report. The first non-empty line must be exactly one of:

RESULT: ok
RESULT: fix-needed

If you choose `RESULT: fix-needed`, add a `## Findings` section with concrete, actionable items.

Response

## Plan Pass

## Problem
The bundle labels the process as `redis-check-rdb`, but the embedded command line, status, and thread names show `redis-server` sleeping in `epoll_wait`. The strace pattern is `epoll_wait(..., ~100ms)` followed by `/proc/self/stat` reads, with reported CPU around `0.3%`, which matches normal Redis idle cron/RSS sampling rather than a demonstrated busy loop.

## Evidence Confidence
inferred

## Proposed Subject
No source patch: diagnose idle Redis cron sampling evidence mismatch

## Patch Plan
Do not edit source in this pass. The evidence is profiler/trace-only and internally mismatched, so a Redis upstream source patch would be speculative.

Reject the prior Fixer patch: it adds generated `deps/jemalloc/bin/*` files and does not address the observed Redis event-loop/RSS-sampling trace.

If stronger evidence is needed before editing, collect a bounded local trace of an idle built `src/redis-server` and compare CPU plus syscall cadence. Only consider a patch if that shows abnormal CPU or a tight loop beyond Redis’ normal `serverCron()` timing.

Conventions/helpers found: `aeMain()`/`aeProcessEvents()` drive the event loop, `serverCron()` and `run_with_period()` schedule periodic work, and `zmalloc_get_rss()`/`get_proc_stat_ll()` are the local `/proc/self/stat` RSS helpers. Contribution docs checked: `CONTRIBUTING.md`, `README.md`, and `tests/README.md`.

## Risks
Changing cron or memory-stat sampling without stronger evidence could degrade `INFO memory`, allocator diagnostics, or normal maintenance behavior. Changing `redis-check-rdb` would not address the sampled process shown in the evidence.

## Validation
Before any focused check, try from the workspace root: `make`, then `make check` if available. If either fails due dependencies, generated files, or environment limits, record the exact command and failure.

Bounded independent evidence attempt: after `make`, run a short local idle `src/redis-server` smoke/trace, such as a `timeout`-bounded `strace` of `epoll_wait`, `openat`, `read`, and `close`, and check whether it merely reproduces the normal 100 ms cron `/proc/self/stat` cadence at low CPU. Do not claim `reproduced` unless the command demonstrates the reported runaway behavior.

## Patch Pass

Subject: diagnose redis idle cron sampling trace

## Commit Message
No source change is proposed. The collected perf, strace, and `/proc` evidence points at an idle `redis-server` event loop doing normal cron memory sampling, not at a demonstrated `redis-check-rdb` busy-loop bug.

## Evidence Confidence
inferred

## Issue Connection
The likely root cause from the collected evidence is normal Redis server cron activity. The report is labeled `redis-check-rdb`, but the command line is `/usr/bin/redis-server 127.0.0.1:6379`, the sampled process is sleeping in `do_epoll_wait`, CPU is about `0.3%`, and strace shows `epoll_wait` timing out around every 100 ms followed by reads of `/proc/self/stat`.

In Redis, `aeMain()` and `aeProcessEvents()` drive the event loop, `serverCron()` is the periodic maintenance callback, `run_with_period(100)` schedules 100 ms work, and `zmalloc_get_rss()` uses `get_proc_stat_ll()` to read `/proc/self/stat`. That matches the trace, so the evidence is indirect and internally mismatched rather than a confirmed upstream defect.

I made no source change and rejected the prior Fixer patch because it adds generated `deps/jemalloc/bin/*` files and does not address the observed Redis event-loop/RSS-sampling path. Contribution/style docs checked: `CONTRIBUTING.md`, `README.md`, and `tests/README.md`.

The expected effect is that no speculative upstream diff is produced until stronger evidence shows abnormal CPU use or a real tight loop distinct from normal idle cron sampling.

## Git Add Paths
None

## Validation
`make` from the workspace root completed successfully.

`make check` from the workspace root failed before running tests because the Redis test harness could not find a free port: `Can't find a non busy port in the 21079-21110 range.`

Bounded independent reproduction was attempted but blocked. `timeout 5s strace -f -ttT -e trace=epoll_wait,openat,read,close ./src/redis-server --port 0 --save '' --appendonly no --daemonize no` failed with `PTRACE_TRACEME: Operation not permitted` and `ptrace(PTRACE_SEIZE, 5): Operation not permitted`.

A local server smoke run could not stay running for comparison: `--port 0` exits because Redis is configured to listen nowhere, Unix socket startup fails with `setsockopt SO_REUSEADDR: Operation not permitted`, and TCP bind fails with `unable to bind socket, errno: 1`. Confidence remains `inferred`; no source diff was left.

## Review Pass 1

RESULT: ok

No blocking findings.

The workspace has no source diff, and `## Git Add Paths` correctly lists `None`. The author kept `## Evidence Confidence` as `inferred`, which matches the profiler/strace-only evidence and the internal mismatch between the `redis-check-rdb` artifact label and the sampled `redis-server` command line.

The response rejects the prior generated jemalloc-bin patch for the right reason: it does not address the observed Redis event-loop/RSS-sampling path. The maintainer-facing explanation is cautious, names the relevant Redis helpers/conventions (`aeMain`, `aeProcessEvents`, `serverCron`, `run_with_period`, `zmalloc_get_rss`, `get_proc_stat_ll`), and does not invent a confirmed bug or reproducer.

Validation is acceptable: project-level `make` was attempted and passed; `make check` was attempted and reports the exact port-range blocker; independent reproduction attempts and sandbox blockers are explicitly named.

Diff

--- a/deps/jemalloc/bin/jemalloc-config	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jemalloc-config	2026-05-10 17:18:05.797913375 +0400
@@ -0,0 +1,83 @@
+#!/bin/sh
+
+usage() {
+	cat <<EOF
+Usage:
+  /usr/local/bin/jemalloc-config <option>
+Options:
+  --help | -h  : Print usage.
+  --version    : Print jemalloc version.
+  --revision   : Print shared library revision number.
+  --config     : Print configure options used to build jemalloc.
+  --prefix     : Print installation directory prefix.
+  --bindir     : Print binary installation directory.
+  --datadir    : Print data installation directory.
+  --includedir : Print include installation directory.
+  --libdir     : Print library installation directory.
+  --mandir     : Print manual page installation directory.
+  --cc         : Print compiler used to build jemalloc.
+  --cflags     : Print compiler flags used to build jemalloc.
+  --cppflags   : Print preprocessor flags used to build jemalloc.
+  --cxxflags   : Print C++ compiler flags used to build jemalloc.
+  --ldflags    : Print library flags used to build jemalloc.
+  --libs       : Print libraries jemalloc was linked against.
+EOF
+}
+
+prefix="/usr/local"
+exec_prefix="/usr/local"
+
+case "$1" in
+--help | -h)
+	usage
+	exit 0
+	;;
+--version)
+	echo "5.3.0-0-g0"
+	;;
+--revision)
+	echo "2"
+	;;
+--config)
+	echo "--disable-cxx --with-version=5.3.0-0-g0 --with-lg-quantum=3 --disable-cache-oblivious --with-jemalloc-prefix=je_ CFLAGS= LDFLAGS="
+	;;
+--prefix)
+	echo "/usr/local"
+	;;
+--bindir)
+	echo "/usr/local/bin"
+	;;
+--datadir)
+	echo "/usr/local/share"
+	;;
+--includedir)
+	echo "/usr/local/include"
+	;;
+--libdir)
+	echo "/usr/local/lib"
+	;;
+--mandir)
+	echo "/usr/local/share/man"
+	;;
+--cc)
+	echo "gcc"
+	;;
+--cflags)
+	echo "-std=gnu11 -Wall -Wextra -Wsign-compare -Wundef -Wno-format-zero-length -Wpointer-arith -Wno-missing-braces -Wno-missing-field-initializers -Wno-missing-attributes -pipe -g3 -fvisibility=hidden -Wimplicit-fallthrough -O3 -funroll-loops"
+	;;
+--cppflags)
+	echo "-D_GNU_SOURCE -D_REENTRANT"
+	;;
+--cxxflags)
+	echo ""
+	;;
+--ldflags)
+	echo " "
+	;;
+--libs)
+	echo "-lm  -pthread"
+	;;
+*)
+	usage
+	exit 1
+esac
--- a/deps/jemalloc/bin/jemalloc.sh	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jemalloc.sh	2026-05-10 17:18:05.803227947 +0400
@@ -0,0 +1,9 @@
+#!/bin/sh
+
+prefix=/usr/local
+exec_prefix=/usr/local
+libdir=${exec_prefix}/lib
+
+LD_PRELOAD=${libdir}/libjemalloc.so.2
+export LD_PRELOAD
+exec "$@"
--- a/deps/jemalloc/bin/jeprof	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jeprof	2026-05-10 17:18:05.811227841 +0400
@@ -0,0 +1,5723 @@
+#! /usr/bin/env perl
+
+# Copyright (c) 1998-2007, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+# ---
+# Program for printing the profile generated by common/profiler.cc,
+# or by the heap profiler (common/debugallocation.cc)
+#
+# The profile contains a sequence of entries of the form:
+#       <count> <stack trace>
+# This program parses the profile, and generates user-readable
+# output.
+#
+# Examples:
+#
+# % tools/jeprof "program" "profile"
+#   Enters "interactive" mode
+#
+# % tools/jeprof --text "program" "profile"
+#   Generates one line per procedure
+#
+# % tools/jeprof --gv "program" "profile"
+#   Generates annotated call-graph and displays via "gv"
+#
+# % tools/jeprof --gv --focus=Mutex "program" "profile"
+#   Restrict to code paths that involve an entry that matches "Mutex"
+#
+# % tools/jeprof --gv --focus=Mutex --ignore=string "program" "profile"
+#   Restrict to code paths that involve an entry that matches "Mutex"
+#   and does not match "string"
+#
+# % tools/jeprof --list=IBF_CheckDocid "program" "profile"
+#   Generates disassembly listing of all routines with at least one
+#   sample that match the --list=<regexp> pattern.  The listing is
+#   annotated with the flat and cumulative sample counts at each line.
+#
+# % tools/jeprof --disasm=IBF_CheckDocid "program" "profile"
+#   Generates disassembly listing of all routines with at least one
+#   sample that match the --disasm=<regexp> pattern.  The listing is
+#   annotated with the flat and cumulative sample counts at each PC value.
+#
+# TODO: Use color to indicate files?
+
+use strict;
+use warnings;
+use Getopt::Long;
+use Cwd;
+
+my $JEPROF_VERSION = "5.3.0-0-g0";
+my $PPROF_VERSION = "2.0";
+
+# These are the object tools we use which can come from a
+# user-specified location using --tools, from the JEPROF_TOOLS
+# environment variable, or from the environment.
+my %obj_tool_map = (
+  "objdump" => "objdump",
+  "nm" => "nm",
+  "addr2line" => "addr2line",
+  "c++filt" => "c++filt",
+  ## ConfigureObjTools may add architecture-specific entries:
+  #"nm_pdb" => "nm-pdb",       # for reading windows (PDB-format) executables
+  #"addr2line_pdb" => "addr2line-pdb",                                # ditto
+  #"otool" => "otool",         # equivalent of objdump on OS X
+);
+# NOTE: these are lists, so you can put in commandline flags if you want.
+my @DOT = ("dot");          # leave non-absolute, since it may be in /usr/local
+my @GV = ("gv");
+my @EVINCE = ("evince");    # could also be xpdf or perhaps acroread
+my @KCACHEGRIND = ("kcachegrind");
+my @PS2PDF = ("ps2pdf");
+# These are used for dynamic profiles
+my @URL_FETCHER = ("curl", "-s", "--fail");
+
+# These are the web pages that servers need to support for dynamic profiles
+my $HEAP_PAGE = "/pprof/heap";
+my $PROFILE_PAGE = "/pprof/profile";   # must support cgi-param "?seconds=#"
+my $PMUPROFILE_PAGE = "/pprof/pmuprofile(?:\\?.*)?"; # must support cgi-param
+                                                # ?seconds=#&event=x&period=n
+my $GROWTH_PAGE = "/pprof/growth";
+my $CONTENTION_PAGE = "/pprof/contention";
+my $WALL_PAGE = "/pprof/wall(?:\\?.*)?";  # accepts options like namefilter
+my $FILTEREDPROFILE_PAGE = "/pprof/filteredprofile(?:\\?.*)?";
+my $CENSUSPROFILE_PAGE = "/pprof/censusprofile(?:\\?.*)?"; # must support cgi-param
+                                                       # "?seconds=#",
+                                                       # "?tags_regexp=#" and
+                                                       # "?type=#".
+my $SYMBOL_PAGE = "/pprof/symbol";     # must support symbol lookup via POST
+my $PROGRAM_NAME_PAGE = "/pprof/cmdline";
+
+# These are the web pages that can be named on the command line.
+# All the alternatives must begin with /.
+my $PROFILES = "($HEAP_PAGE|$PROFILE_PAGE|$PMUPROFILE_PAGE|" .
+               "$GROWTH_PAGE|$CONTENTION_PAGE|$WALL_PAGE|" .
+               "$FILTEREDPROFILE_PAGE|$CENSUSPROFILE_PAGE)";
+
+# default binary name
+my $UNKNOWN_BINARY = "(unknown)";
+
+# There is a pervasive dependency on the length (in hex characters,
+# i.e., nibbles) of an address, distinguishing between 32-bit and
+# 64-bit profiles.  To err on the safe size, default to 64-bit here:
+my $address_length = 16;
+
+my $dev_null = "/dev/null";
+if (! -e $dev_null && $^O =~ /MSWin/) {    # $^O is the OS perl was built for
+  $dev_null = "nul";
+}
+
+# A list of paths to search for shared object files
+my @prefix_list = ();
+
+# Special routine name that should not have any symbols.
+# Used as separator to parse "addr2line -i" output.
+my $sep_symbol = '_fini';
+my $sep_address = undef;
+
+##### Argument parsing #####
+
+sub usage_string {
+  return <<EOF;
+Usage:
+jeprof [options] <program> <profiles>
+   <profiles> is a space separated list of profile names.
+jeprof [options] <symbolized-profiles>
+   <symbolized-profiles> is a list of profile files where each file contains
+   the necessary symbol mappings  as well as profile data (likely generated
+   with --raw).
+jeprof [options] <profile>
+   <profile> is a remote form.  Symbols are obtained from host:port$SYMBOL_PAGE
+
+   Each name can be:
+   /path/to/profile        - a path to a profile file
+   host:port[/<service>]   - a location of a service to get profile from
+
+   The /<service> can be $HEAP_PAGE, $PROFILE_PAGE, /pprof/pmuprofile,
+                         $GROWTH_PAGE, $CONTENTION_PAGE, /pprof/wall,
+                         $CENSUSPROFILE_PAGE, or /pprof/filteredprofile.
+   For instance:
+     jeprof http://myserver.com:80$HEAP_PAGE
+   If /<service> is omitted, the service defaults to $PROFILE_PAGE (cpu profiling).
+jeprof --symbols <program>
+   Maps addresses to symbol names.  In this mode, stdin should be a
+   list of library mappings, in the same format as is found in the heap-
+   and cpu-profile files (this loosely matches that of /proc/self/maps
+   on linux), followed by a list of hex addresses to map, one per line.
+
+   For more help with querying remote servers, including how to add the
+   necessary server-side support code, see this filename (or one like it):
+
+   /usr/doc/gperftools-$PPROF_VERSION/pprof_remote_servers.html
+
+Options:
+   --cum               Sort by cumulative data
+   --base=<base>       Subtract <base> from <profile> before display
+   --interactive       Run in interactive mode (interactive "help" gives help) [default]
+   --seconds=<n>       Length of time for dynamic profiles [default=30 secs]
+   --add_lib=<file>    Read additional symbols and line info from the given library
+   --lib_prefix=<dir>  Comma separated list of library path prefixes
+
+Reporting Granularity:
+   --addresses         Report at address level
+   --lines             Report at source line level
+   --functions         Report at function level [default]
+   --files             Report at source file level
+
+Output type:
+   --text              Generate text report
+   --callgrind         Generate callgrind format to stdout
+   --gv                Generate Postscript and display
+   --evince            Generate PDF and display
+   --web               Generate SVG and display
+   --list=<regexp>     Generate source listing of matching routines
+   --disasm=<regexp>   Generate disassembly of matching routines
+   --symbols           Print demangled symbol names found at given addresses
+   --dot               Generate DOT file to stdout
+   --ps                Generate Postcript to stdout
+   --pdf               Generate PDF to stdout
+   --svg               Generate SVG to stdout
+   --gif               Generate GIF to stdout
+   --raw               Generate symbolized jeprof data (useful with remote fetch)
+   --collapsed         Generate collapsed stacks for building flame graphs
+                       (see http://www.brendangregg.com/flamegraphs.html)
+
+Heap-Profile Options:
+   --inuse_space       Display in-use (mega)bytes [default]
+   --inuse_objects     Display in-use objects
+   --alloc_space       Display allocated (mega)bytes
+   --alloc_objects     Display allocated objects
+   --show_bytes        Display space in bytes
+   --drop_negative     Ignore negative differences
+
+Contention-profile options:
+   --total_delay       Display total delay at each region [default]
+   --contentions       Display number of delays at each region
+   --mean_delay        Display mean delay at each region
+
+Call-graph Options:
+   --nodecount=<n>     Show at most so many nodes [default=80]
+   --nodefraction=<f>  Hide nodes below <f>*total [default=.005]
+   --edgefraction=<f>  Hide edges below <f>*total [default=.001]
+   --maxdegree=<n>     Max incoming/outgoing edges per node [default=8]
+   --focus=<regexp>    Focus on backtraces with nodes matching <regexp>
+   --thread=<n>        Show profile for thread <n>
+   --ignore=<regexp>   Ignore backtraces with nodes matching <regexp>
+   --scale=<n>         Set GV scaling [default=0]
+   --heapcheck         Make nodes with non-0 object counts
+                       (i.e. direct leak generators) more visible
+   --retain=<regexp>   Retain only nodes that match <regexp>
+   --exclude=<regexp>  Exclude all nodes that match <regexp>
+
+Miscellaneous:
+   --tools=<prefix or binary:fullpath>[,...]   \$PATH for object tool pathnames
+   --test              Run unit tests
+   --help              This message
+   --version           Version information
+   --debug-syms-by-id  (Linux only) Find debug symbol files by build ID as well as by name
+
+Environment Variables:
+   JEPROF_TMPDIR        Profiles directory. Defaults to \$HOME/jeprof
+   JEPROF_TOOLS         Prefix for object tools pathnames
+
+Examples:
+
+jeprof /bin/ls ls.prof
+                       Enters "interactive" mode
+jeprof --text /bin/ls ls.prof
+                       Outputs one line per procedure
+jeprof --web /bin/ls ls.prof
+                       Displays annotated call-graph in web browser
+jeprof --gv /bin/ls ls.prof
+                       Displays annotated call-graph via 'gv'
+jeprof --gv --focus=Mutex /bin/ls ls.prof
+                       Restricts to code paths including a .*Mutex.* entry
+jeprof --gv --focus=Mutex --ignore=string /bin/ls ls.prof
+                       Code paths including Mutex but not string
+jeprof --list=getdir /bin/ls ls.prof
+                       (Per-line) annotated source listing for getdir()
+jeprof --disasm=getdir /bin/ls ls.prof
+                       (Per-PC) annotated disassembly for getdir()
+
+jeprof http://localhost:1234/
+                       Enters "interactive" mode
+jeprof --text localhost:1234
+                       Outputs one line per procedure for localhost:1234
+jeprof --raw localhost:1234 > ./local.raw
+jeprof --text ./local.raw
+                       Fetches a remote profile for later analysis and then
+                       analyzes it in text mode.
+EOF
+}
+
+sub version_string {
+  return <<EOF
+jeprof (part of jemalloc $JEPROF_VERSION)
+based on pprof (part of gperftools $PPROF_VERSION)
+
+Copyright 1998-2007 Google Inc.
+
+This is BSD licensed software; see the source for copying conditions
+and license information.
+There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A
+PARTICULAR PURPOSE.
+EOF
+}
+
+sub usage {
+  my $msg = shift;
+  print STDERR "$msg\n\n";
+  print STDERR usage_string();
+  print STDERR "\nFATAL ERROR: $msg\n";    # just as a reminder
+  exit(1);
+}
+
+sub Init() {
+  # Setup tmp-file name and handler to clean it up.
+  # We do this in the very beginning so that we can use
+  # error() and cleanup() function anytime here after.
+  $main::tmpfile_sym = "/tmp/jeprof$$.sym";
+  $main::tmpfile_ps = "/tmp/jeprof$$";
+  $main::next_tmpfile = 0;
+  $SIG{'INT'} = \&sighandler;
+
+  # Cache from filename/linenumber to source code
+  $main::source_cache = ();
+
+  $main::opt_help = 0;
+  $main::opt_version = 0;
+
+  $main::opt_cum = 0;
+  $main::opt_base = '';
+  $main::opt_addresses = 0;
+  $main::opt_lines = 0;
+  $main::opt_functions = 0;
+  $main::opt_files = 0;
+  $main::opt_lib_prefix = "";
+
+  $main::opt_text = 0;
+  $main::opt_callgrind = 0;
+  $main::opt_list = "";
+  $main::opt_disasm = "";
+  $main::opt_symbols = 0;
+  $main::opt_gv = 0;
+  $main::opt_evince = 0;
+  $main::opt_web = 0;
+  $main::opt_dot = 0;
+  $main::opt_ps = 0;
+  $main::opt_pdf = 0;
+  $main::opt_gif = 0;
+  $main::opt_svg = 0;
+  $main::opt_raw = 0;
+  $main::opt_collapsed = 0;
+
+  $main::opt_nodecount = 80;
+  $main::opt_nodefraction = 0.005;
+  $main::opt_edgefraction = 0.001;
+  $main::opt_maxdegree = 8;
+  $main::opt_focus = '';
+  $main::opt_thread = undef;
+  $main::opt_ignore = '';
+  $main::opt_scale = 0;
+  $main::opt_heapcheck = 0;
+  $main::opt_retain = '';
+  $main::opt_exclude = '';
+  $main::opt_seconds = 30;
+  $main::opt_lib = "";
+
+  $main::opt_inuse_space   = 0;
+  $main::opt_inuse_objects = 0;
+  $main::opt_alloc_space   = 0;
+  $main::opt_alloc_objects = 0;
+  $main::opt_show_bytes    = 0;
+  $main::opt_drop_negative = 0;
+  $main::opt_interactive   = 0;
+
+  $main::opt_total_delay = 0;
+  $main::opt_contentions = 0;
+  $main::opt_mean_delay = 0;
+
+  $main::opt_tools   = "";
+  $main::opt_debug   = 0;
+  $main::opt_test    = 0;
+  $main::opt_debug_syms_by_id = 0;
+
+  # These are undocumented flags used only by unittests.
+  $main::opt_test_stride = 0;
+
+  # Are we using $SYMBOL_PAGE?
+  $main::use_symbol_page = 0;
+
+  # Files returned by TempName.
+  %main::tempnames = ();
+
+  # Type of profile we are dealing with
+  # Supported types:
+  #     cpu
+  #     heap
+  #     growth
+  #     contention
+  $main::profile_type = '';     # Empty type means "unknown"
+
+  GetOptions("help!"          => \$main::opt_help,
+             "version!"       => \$main::opt_version,
+             "cum!"           => \$main::opt_cum,
+             "base=s"         => \$main::opt_base,
+             "seconds=i"      => \$main::opt_seconds,
+             "add_lib=s"      => \$main::opt_lib,
+             "lib_prefix=s"   => \$main::opt_lib_prefix,
+             "functions!"     => \$main::opt_functions,
+             "lines!"         => \$main::opt_lines,
+             "addresses!"     => \$main::opt_addresses,
+             "files!"         => \$main::opt_files,
+             "text!"          => \$main::opt_text,
+             "callgrind!"     => \$main::opt_callgrind,
+             "list=s"         => \$main::opt_list,
+             "disasm=s"       => \$main::opt_disasm,
+             "symbols!"       => \$main::opt_symbols,
+             "gv!"            => \$main::opt_gv,
+             "evince!"        => \$main::opt_evince,
+             "web!"           => \$main::opt_web,
+             "dot!"           => \$main::opt_dot,
+             "ps!"            => \$main::opt_ps,
+             "pdf!"           => \$main::opt_pdf,
+             "svg!"           => \$main::opt_svg,
+             "gif!"           => \$main::opt_gif,
+             "raw!"           => \$main::opt_raw,
+             "collapsed!"     => \$main::opt_collapsed,
+             "interactive!"   => \$main::opt_interactive,
+             "nodecount=i"    => \$main::opt_nodecount,
+             "nodefraction=f" => \$main::opt_nodefraction,
+             "edgefraction=f" => \$main::opt_edgefraction,
+             "maxdegree=i"    => \$main::opt_maxdegree,
+             "focus=s"        => \$main::opt_focus,
+             "thread=s"       => \$main::opt_thread,
+             "ignore=s"       => \$main::opt_ignore,
+             "scale=i"        => \$main::opt_scale,
+             "heapcheck"      => \$main::opt_heapcheck,
+             "retain=s"       => \$main::opt_retain,
+             "exclude=s"      => \$main::opt_exclude,
+             "inuse_space!"   => \$main::opt_inuse_space,
+             "inuse_objects!" => \$main::opt_inuse_objects,
+             "alloc_space!"   => \$main::opt_alloc_space,
+             "alloc_objects!" => \$main::opt_alloc_objects,
+             "show_bytes!"    => \$main::opt_show_bytes,
+             "drop_negative!" => \$main::opt_drop_negative,
+             "total_delay!"   => \$main::opt_total_delay,
+             "contentions!"   => \$main::opt_contentions,
+             "mean_delay!"    => \$main::opt_mean_delay,
+             "tools=s"        => \$main::opt_tools,
+             "test!"          => \$main::opt_test,
+             "debug!"         => \$main::opt_debug,
+             "debug-syms-by-id!" => \$main::opt_debug_syms_by_id,
+             # Undocumented flags used only by unittests:
+             "test_stride=i"  => \$main::opt_test_stride,
+      ) || usage("Invalid option(s)");
+
+  # Deal with the standard --help and --version
+  if ($main::opt_help) {
+    print usage_string();
+    exit(0);
+  }
+
+  if ($main::opt_version) {
+    print version_string();
+    exit(0);
+  }
+
+  # Disassembly/listing/symbols mode requires address-level info
+  if ($main::opt_disasm || $main::opt_list || $main::opt_symbols) {
+    $main::opt_functions = 0;
+    $main::opt_lines = 0;
+    $main::opt_addresses = 1;
+    $main::opt_files = 0;
+  }
+
+  # Check heap-profiling flags
+  if ($main::opt_inuse_space +
+      $main::opt_inuse_objects +
+      $main::opt_alloc_space +
+      $main::opt_alloc_objects > 1) {
+    usage("Specify at most on of --inuse/--alloc options");
+  }
+
+  # Check output granularities
+  my $grains =
+      $main::opt_functions +
+      $main::opt_lines +
+      $main::opt_addresses +
+      $main::opt_files +
+      0;
+  if ($grains > 1) {
+    usage("Only specify one output granularity option");
+  }
+  if ($grains == 0) {
+    $main::opt_functions = 1;
+  }
+
+  # Check output modes
+  my $modes =
+      $main::opt_text +
+      $main::opt_callgrind +
+      ($main::opt_list eq '' ? 0 : 1) +
+      ($main::opt_disasm eq '' ? 0 : 1) +
+      ($main::opt_symbols == 0 ? 0 : 1) +
+      $main::opt_gv +
+      $main::opt_evince +
+      $main::opt_web +
+      $main::opt_dot +
+      $main::opt_ps +
+      $main::opt_pdf +
+      $main::opt_svg +
+      $main::opt_gif +
+      $main::opt_raw +
+      $main::opt_collapsed +
+      $main::opt_interactive +
+      0;
+  if ($modes > 1) {
+    usage("Only specify one output mode");
+  }
+  if ($modes == 0) {
+    if (-t STDOUT) {  # If STDOUT is a tty, activate interactive mode
+      $main::opt_interactive = 1;
+    } else {
+      $main::opt_text = 1;
+    }
+  }
+
+  if ($main::opt_test) {
+    RunUnitTests();
+    # Should not return
+    exit(1);
+  }
+
+  # Binary name and profile arguments list
+  $main::prog = "";
+  @main::pfile_args = ();
+
+  # Remote profiling without a binary (using $SYMBOL_PAGE instead)
+  if (@ARGV > 0) {
+    if (IsProfileURL($ARGV[0])) {
+      $main::use_symbol_page = 1;
+    } elsif (IsSymbolizedProfileFile($ARGV[0])) {
+      $main::use_symbolized_profile = 1;
+      $main::prog = $UNKNOWN_BINARY;  # will be set later from the profile file
+    }
+  }
+
+  if ($main::use_symbol_page || $main::use_symbolized_profile) {
+    # We don't need a binary!
+    my %disabled = ('--lines' => $main::opt_lines,
+                    '--disasm' => $main::opt_disasm);
+    for my $option (keys %disabled) {
+      usage("$option cannot be used without a binary") if $disabled{$option};
+    }
+    # Set $main::prog later...
+    scalar(@ARGV) || usage("Did not specify profile file");
+  } elsif ($main::opt_symbols) {
+    # --symbols needs a binary-name (to run nm on, etc) but not profiles
+    $main::prog = shift(@ARGV) || usage("Did not specify program");
+  } else {
+    $main::prog = shift(@ARGV) || usage("Did not specify program");
+    scalar(@ARGV) || usage("Did not specify profile file");
+  }
+
+  # Parse profile file/location arguments
+  foreach my $farg (@ARGV) {
+    if ($farg =~ m/(.*)\@([0-9]+)(|\/.*)$/ ) {
+      my $machine = $1;
+      my $num_machines = $2;
+      my $path = $3;
+      for (my $i = 0; $i < $num_machines; $i++) {
+        unshift(@main::pfile_args, "$i.$machine$path");
+      }
+    } else {
+      unshift(@main::pfile_args, $farg);
+    }
+  }
+
+  if ($main::use_symbol_page) {
+    unless (IsProfileURL($main::pfile_args[0])) {
+      error("The first profile should be a remote form to use $SYMBOL_PAGE\n");
+    }
+    CheckSymbolPage();
+    $main::prog = FetchProgramName();
+  } elsif (!$main::use_symbolized_profile) {  # may not need objtools!
+    ConfigureObjTools($main::prog)
+  }
+
+  # Break the opt_lib_prefix into the prefix_list array
+  @prefix_list = split (',', $main::opt_lib_prefix);
+
+  # Remove trailing / from the prefixes, in the list to prevent
+  # searching things like /my/path//lib/mylib.so
+  foreach (@prefix_list) {
+    s|/+$||;
+  }
+
+  # Flag to prevent us from trying over and over to use
+  #  elfutils if it's not installed (used only with
+  #  --debug-syms-by-id option).
+  $main::gave_up_on_elfutils = 0;
+}
+
+sub FilterAndPrint {
+  my ($profile, $symbols, $libs, $thread) = @_;
+
+  # Get total data in profile
+  my $total = TotalProfile($profile);
+
+  # Remove uniniteresting stack items
+  $profile = RemoveUninterestingFrames($symbols, $profile);
+
+  # Focus?
+  if ($main::opt_focus ne '') {
+    $profile = FocusProfile($symbols, $profile, $main::opt_focus);
+  }
+
+  # Ignore?
+  if ($main::opt_ignore ne '') {
+    $profile = IgnoreProfile($symbols, $profile, $main::opt_ignore);
+  }
+
+  my $calls = ExtractCalls($symbols, $profile);
+
+  # Reduce profiles to required output granularity, and also clean
+  # each stack trace so a given entry exists at most once.
+  my $reduced = ReduceProfile($symbols, $profile);
+
+  # Get derived profiles
+  my $flat = FlatProfile($reduced);
+  my $cumulative = CumulativeProfile($reduced);
+
+  # Print
+  if (!$main::opt_interactive) {
+    if ($main::opt_disasm) {
+      PrintDisassembly($libs, $flat, $cumulative, $main::opt_disasm);
+    } elsif ($main::opt_list) {
+      PrintListing($total, $libs, $flat, $cumulative, $main::opt_list, 0);
+    } elsif ($main::opt_text) {
+      # Make sure the output is empty when have nothing to report
+      # (only matters when --heapcheck is given but we must be
+      # compatible with old branches that did not pass --heapcheck always):
+      if ($total != 0) {
+        printf("Total%s: %s %s\n",
+               (defined($thread) ? " (t$thread)" : ""),
+               Unparse($total), Units());
+      }
+      PrintText($symbols, $flat, $cumulative, -1);
+    } elsif ($main::opt_raw) {
+      PrintSymbolizedProfile($symbols, $profile, $main::prog);
+    } elsif ($main::opt_collapsed) {
+      PrintCollapsedStacks($symbols, $profile);
+    } elsif ($main::opt_callgrind) {
+      PrintCallgrind($calls);
+    } else {
+      if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) {
+        if ($main::opt_gv) {
+          RunGV(TempName($main::next_tmpfile, "ps"), "");
+        } elsif ($main::opt_evince) {
+          RunEvince(TempName($main::next_tmpfile, "pdf"), "");
+        } elsif ($main::opt_web) {
+          my $tmp = TempName($main::next_tmpfile, "svg");
+          RunWeb($tmp);
+          # The command we run might hand the file name off
+          # to an already running browser instance and then exit.
+          # Normally, we'd remove $tmp on exit (right now),
+          # but fork a child to remove $tmp a little later, so that the
+          # browser has time to load it first.
+          delete $main::tempnames{$tmp};
+          if (fork() == 0) {
+            sleep 5;
+            unlink($tmp);
+            exit(0);
+          }
+        }
+      } else {
+        cleanup();
+        exit(1);
+      }
+    }
+  } else {
+    InteractiveMode($profile, $symbols, $libs, $total);
+  }
+}
+
+sub Main() {
+  Init();
+  $main::collected_profile = undef;
+  @main::profile_files = ();
+  $main::op_time = time();
+
+  # Printing symbols is special and requires a lot less info that most.
+  if ($main::opt_symbols) {
+    PrintSymbols(*STDIN);   # Get /proc/maps and symbols output from stdin
+    return;
+  }
+
+  # Fetch all profile data
+  FetchDynamicProfiles();
+
+  # this will hold symbols that we read from the profile files
+  my $symbol_map = {};
+
+  # Read one profile, pick the last item on the list
+  my $data = ReadProfile($main::prog, pop(@main::profile_files));
+  my $profile = $data->{profile};
+  my $pcs = $data->{pcs};
+  my $libs = $data->{libs};   # Info about main program and shared libraries
+  $symbol_map = MergeSymbols($symbol_map, $data->{symbols});
+
+  # Add additional profiles, if available.
+  if (scalar(@main::profile_files) > 0) {
+    foreach my $pname (@main::profile_files) {
+      my $data2 = ReadProfile($main::prog, $pname);
+      $profile = AddProfile($profile, $data2->{profile});
+      $pcs = AddPcs($pcs, $data2->{pcs});
+      $symbol_map = MergeSymbols($symbol_map, $data2->{symbols});
+    }
+  }
+
+  # Subtract base from profile, if specified
+  if ($main::opt_base ne '') {
+    my $base = ReadProfile($main::prog, $main::opt_base);
+    $profile = SubtractProfile($profile, $base->{profile});
+    $pcs = AddPcs($pcs, $base->{pcs});
+    $symbol_map = MergeSymbols($symbol_map, $base->{symbols});
+  }
+
+  # Collect symbols
+  my $symbols;
+  if ($main::use_symbolized_profile) {
+    $symbols = FetchSymbols($pcs, $symbol_map);
+  } elsif ($main::use_symbol_page) {
+    $symbols = FetchSymbols($pcs);
+  } else {
+    # TODO(csilvers): $libs uses the /proc/self/maps data from profile1,
+    # which may differ from the data from subsequent profiles, especially
+    # if they were run on different machines.  Use appropriate libs for
+    # each pc somehow.
+    $symbols = ExtractSymbols($libs, $pcs);
+  }
+
+  if (!defined($main::opt_thread)) {
+    FilterAndPrint($profile, $symbols, $libs);
+  }
+  if (defined($data->{threads})) {
+    foreach my $thread (sort { $a <=> $b } keys(%{$data->{threads}})) {
+      if (defined($main::opt_thread) &&
+          ($main::opt_thread eq '*' || $main::opt_thread == $thread)) {
+        my $thread_profile = $data->{threads}{$thread};
+        FilterAndPrint($thread_profile, $symbols, $libs, $thread);
+      }
+    }
+  }
+
+  cleanup();
+  exit(0);
+}
+
+##### Entry Point #####
+
+Main();
+
+# Temporary code to detect if we're running on a Goobuntu system.
+# These systems don't have the right stuff installed for the special
+# Readline libraries to work, so as a temporary workaround, we default
+# to using the normal stdio code, rather than the fancier readline-based
+# code
+sub ReadlineMightFail {
+  if (-e '/lib/libtermcap.so.2') {
+    return 0;  # libtermcap exists, so readline should be okay
+  } else {
+    return 1;
+  }
+}
+
+sub RunGV {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  if (!system(ShellEscape(@GV, "--version") . " >$dev_null 2>&1")) {
+    # Options using double dash are supported by this gv version.
+    # Also, turn on noantialias to better handle bug in gv for
+    # postscript files with large dimensions.
+    # TODO: Maybe we should not pass the --noantialias flag
+    # if the gv version is known to work properly without the flag.
+    system(ShellEscape(@GV, "--scale=$main::opt_scale", "--noantialias", $fname)
+           . $bg);
+  } else {
+    # Old gv version - only supports options that use single dash.
+    print STDERR ShellEscape(@GV, "-scale", $main::opt_scale) . "\n";
+    system(ShellEscape(@GV, "-scale", "$main::opt_scale", $fname) . $bg);
+  }
+}
+
+sub RunEvince {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  system(ShellEscape(@EVINCE, $fname) . $bg);
+}
+
+sub RunWeb {
+  my $fname = shift;
+  print STDERR "Loading web page file:///$fname\n";
+
+  if (`uname` =~ /Darwin/) {
+    # OS X: open will use standard preference for SVG files.
+    system("/usr/bin/open", $fname);
+    return;
+  }
+
+  # Some kind of Unix; try generic symlinks, then specific browsers.
+  # (Stop once we find one.)
+  # Works best if the browser is already running.
+  my @alt = (
+    "/etc/alternatives/gnome-www-browser",
+    "/etc/alternatives/x-www-browser",
+    "google-chrome",
+    "firefox",
+  );
+  foreach my $b (@alt) {
+    if (system($b, $fname) == 0) {
+      return;
+    }
+  }
+
+  print STDERR "Could not load web browser.\n";
+}
+
+sub RunKcachegrind {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  print STDERR "Starting '@KCACHEGRIND " . $fname . $bg . "'\n";
+  system(ShellEscape(@KCACHEGRIND, $fname) . $bg);
+}
+
+
+##### Interactive helper routines #####
+
+sub InteractiveMode {
+  $| = 1;  # Make output unbuffered for interactive mode
+  my ($orig_profile, $symbols, $libs, $total) = @_;
+
+  print STDERR "Welcome to jeprof!  For help, type 'help'.\n";
+
+  # Use ReadLine if it's installed and input comes from a console.
+  if ( -t STDIN &&
+       !ReadlineMightFail() &&
+       defined(eval {require Term::ReadLine}) ) {
+    my $term = new Term::ReadLine 'jeprof';
+    while ( defined ($_ = $term->readline('(jeprof) '))) {
+      $term->addhistory($_) if /\S/;
+      if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) {
+        last;    # exit when we get an interactive command to quit
+      }
+    }
+  } else {       # don't have readline
+    while (1) {
+      print STDERR "(jeprof) ";
+      $_ = <STDIN>;
+      last if ! defined $_ ;
+      s/\r//g;         # turn windows-looking lines into unix-looking lines
+
+      # Save some flags that might be reset by InteractiveCommand()
+      my $save_opt_lines = $main::opt_lines;
+
+      if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) {
+        last;    # exit when we get an interactive command to quit
+      }
+
+      # Restore flags
+      $main::opt_lines = $save_opt_lines;
+    }
+  }
+}
+
+# Takes two args: orig profile, and command to run.
+# Returns 1 if we should keep going, or 0 if we were asked to quit
+sub InteractiveCommand {
+  my($orig_profile, $symbols, $libs, $total, $command) = @_;
+  $_ = $command;                # just to make future m//'s easier
+  if (!defined($_)) {
+    print STDERR "\n";
+    return 0;
+  }
+  if (m/^\s*quit/) {
+    return 0;
+  }
+  if (m/^\s*help/) {
+    InteractiveHelpMessage();
+    return 1;
+  }
+  # Clear all the mode options -- mode is controlled by "$command"
+  $main::opt_text = 0;
+  $main::opt_callgrind = 0;
+  $main::opt_disasm = 0;
+  $main::opt_list = 0;
+  $main::opt_gv = 0;
+  $main::opt_evince = 0;
+  $main::opt_cum = 0;
+
+  if (m/^\s*(text|top)(\d*)\s*(.*)/) {
+    $main::opt_text = 1;
+
+    my $line_limit = ($2 ne "") ? int($2) : 10;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($3);
+
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintText($symbols, $flat, $cumulative, $line_limit);
+    return 1;
+  }
+  if (m/^\s*callgrind\s*([^ \n]*)/) {
+    $main::opt_callgrind = 1;
+
+    # Get derived profiles
+    my $calls = ExtractCalls($symbols, $orig_profile);
+    my $filename = $1;
+    if ( $1 eq '' ) {
+      $filename = TempName($main::next_tmpfile, "callgrind");
+    }
+    PrintCallgrind($calls, $filename);
+    if ( $1 eq '' ) {
+      RunKcachegrind($filename, " & ");
+      $main::next_tmpfile++;
+    }
+
+    return 1;
+  }
+  if (m/^\s*(web)?list\s*(.+)/) {
+    my $html = (defined($1) && ($1 eq "web"));
+    $main::opt_list = 1;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($2);
+
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintListing($total, $libs, $flat, $cumulative, $routine, $html);
+    return 1;
+  }
+  if (m/^\s*disasm\s*(.+)/) {
+    $main::opt_disasm = 1;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($1);
+
+    # Process current profile to account for various settings
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintDisassembly($libs, $flat, $cumulative, $routine);
+    return 1;
+  }
+  if (m/^\s*(gv|web|evince)\s*(.*)/) {
+    $main::opt_gv = 0;
+    $main::opt_evince = 0;
+    $main::opt_web = 0;
+    if ($1 eq "gv") {
+      $main::opt_gv = 1;
+    } elsif ($1 eq "evince") {
+      $main::opt_evince = 1;
+    } elsif ($1 eq "web") {
+      $main::opt_web = 1;
+    }
+
+    my $focus;
+    my $ignore;
+    ($focus, $ignore) = ParseInteractiveArgs($2);
+
+    # Process current profile to account for various settings
+    my $profile = ProcessProfile($total, $orig_profile, $symbols,
+                                 $focus, $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) {
+      if ($main::opt_gv) {
+        RunGV(TempName($main::next_tmpfile, "ps"), " &");
+      } elsif ($main::opt_evince) {
+        RunEvince(TempName($main::next_tmpfile, "pdf"), " &");
+      } elsif ($main::opt_web) {
+        RunWeb(TempName($main::next_tmpfile, "svg"));
+      }
+      $main::next_tmpfile++;
+    }
+    return 1;
+  }
+  if (m/^\s*$/) {
+    return 1;
+  }
+  print STDERR "Unknown command: try 'help'.\n";
+  return 1;
+}
+
+
+sub ProcessProfile {
+  my $total_count = shift;
+  my $orig_profile = shift;
+  my $symbols = shift;
+  my $focus = shift;
+  my $ignore = shift;
+
+  # Process current profile to account for various settings
+  my $profile = $orig_profile;
+  printf("Total: %s %s\n", Unparse($total_count), Units());
+  if ($focus ne '') {
+    $profile = FocusProfile($symbols, $profile, $focus);
+    my $focus_count = TotalProfile($profile);
+    printf("After focusing on '%s': %s %s of %s (%0.1f%%)\n",
+           $focus,
+           Unparse($focus_count), Units(),
+           Unparse($total_count), ($focus_count*100.0) / $total_count);
+  }
+  if ($ignore ne '') {
+    $profile = IgnoreProfile($symbols, $profile, $ignore);
+    my $ignore_count = TotalProfile($profile);
+    printf("After ignoring '%s': %s %s of %s (%0.1f%%)\n",
+           $ignore,
+           Unparse($ignore_count), Units(),
+           Unparse($total_count),
+           ($ignore_count*100.0) / $total_count);
+  }
+
+  return $profile;
+}
+
+sub InteractiveHelpMessage {
+  print STDERR <<ENDOFHELP;
+Interactive jeprof mode
+
+Commands:
+  gv
+  gv [focus] [-ignore1] [-ignore2]
+      Show graphical hierarchical display of current profile.  Without
+      any arguments, shows all samples in the profile.  With the optional
+      "focus" argument, restricts the samples shown to just those where
+      the "focus" regular expression matches a routine name on the stack
+      trace.
+
+  web
+  web [focus] [-ignore1] [-ignore2]
+      Like GV, but displays profile in your web browser instead of using
+      Ghostview. Works best if your web browser is already running.
+      To change the browser that gets used:
+      On Linux, set the /etc/alternatives/gnome-www-browser symlink.
+      On OS X, change the Finder association for SVG files.
+
+  list [routine_regexp] [-ignore1] [-ignore2]
+      Show source listing of routines whose names match "routine_regexp"
+
+  weblist [routine_regexp] [-ignore1] [-ignore2]
+     Displays a source listing of routines whose names match "routine_regexp"
+     in a web browser.  You can click on source lines to view the
+     corresponding disassembly.
+
+  top [--cum] [-ignore1] [-ignore2]
+  top20 [--cum] [-ignore1] [-ignore2]
+  top37 [--cum] [-ignore1] [-ignore2]
+      Show top lines ordered by flat profile count, or cumulative count
+      if --cum is specified.  If a number is present after 'top', the
+      top K routines will be shown (defaults to showing the top 10)
+
+  disasm [routine_regexp] [-ignore1] [-ignore2]
+      Show disassembly of routines whose names match "routine_regexp",
+      annotated with sample counts.
+
+  callgrind
+  callgrind [filename]
+      Generates callgrind file. If no filename is given, kcachegrind is called.
+
+  help - This listing
+  quit or ^D - End jeprof
+
+For commands that accept optional -ignore tags, samples where any routine in
+the stack trace matches the regular expression in any of the -ignore
+parameters will be ignored.
+
+Further pprof details are available at this location (or one similar):
+
+ /usr/doc/gperftools-$PPROF_VERSION/cpu_profiler.html
+ /usr/doc/gperftools-$PPROF_VERSION/heap_profiler.html
+
+ENDOFHELP
+}
+sub ParseInteractiveArgs {
+  my $args = shift;
+  my $focus = "";
+  my $ignore = "";
+  my @x = split(/ +/, $args);
+  foreach $a (@x) {
+    if ($a =~ m/^(--|-)lines$/) {
+      $main::opt_lines = 1;
+    } elsif ($a =~ m/^(--|-)cum$/) {
+      $main::opt_cum = 1;
+    } elsif ($a =~ m/^-(.*)/) {
+      $ignore .= (($ignore ne "") ? "|" : "" ) . $1;
+    } else {
+      $focus .= (($focus ne "") ? "|" : "" ) . $a;
+    }
+  }
+  if ($ignore ne "") {
+    print STDERR "Ignoring samples in call stacks that match '$ignore'\n";
+  }
+  return ($focus, $ignore);
+}
+
+##### Output code #####
+
+sub TempName {
+  my $fnum = shift;
+  my $ext = shift;
+  my $file = "$main::tmpfile_ps.$fnum.$ext";
+  $main::tempnames{$file} = 1;
+  return $file;
+}
+
+# Print profile data in packed binary format (64-bit) to standard out
+sub PrintProfileData {
+  my $profile = shift;
+
+  # print header (64-bit style)
+  # (zero) (header-size) (version) (sample-period) (zero)
+  print pack('L*', 0, 0, 3, 0, 0, 0, 1, 0, 0, 0);
+
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    if ($#addrs >= 0) {
+      my $depth = $#addrs + 1;
+      # int(foo / 2**32) is the only reliable way to get rid of bottom
+      # 32 bits on both 32- and 64-bit systems.
+      print pack('L*', $count & 0xFFFFFFFF, int($count / 2**32));
+      print pack('L*', $depth & 0xFFFFFFFF, int($depth / 2**32));
+
+      foreach my $full_addr (@addrs) {
+        my $addr = $full_addr;
+        $addr =~ s/0x0*//;  # strip off leading 0x, zeroes
+        if (length($addr) > 16) {
+          print STDERR "Invalid address in profile: $full_addr\n";
+          next;
+        }
+        my $low_addr = substr($addr, -8);       # get last 8 hex chars
+        my $high_addr = substr($addr, -16, 8);  # get up to 8 more hex chars
+        print pack('L*', hex('0x' . $low_addr), hex('0x' . $high_addr));
+      }
+    }
+  }
+}
+
+# Print symbols and profile data
+sub PrintSymbolizedProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $prog = shift;
+
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+
+  print '--- ', $symbol_marker, "\n";
+  if (defined($prog)) {
+    print 'binary=', $prog, "\n";
+  }
+  while (my ($pc, $name) = each(%{$symbols})) {
+    my $sep = ' ';
+    print '0x', $pc;
+    # We have a list of function names, which include the inlined
+    # calls.  They are separated (and terminated) by --, which is
+    # illegal in function names.
+    for (my $j = 2; $j <= $#{$name}; $j += 3) {
+      print $sep, $name->[$j];
+      $sep = '--';
+    }
+    print "\n";
+  }
+  print '---', "\n";
+
+  my $profile_marker;
+  if ($main::profile_type eq 'heap') {
+    $HEAP_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } elsif ($main::profile_type eq 'growth') {
+    $GROWTH_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } elsif ($main::profile_type eq 'contention') {
+    $CONTENTION_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } else { # elsif ($main::profile_type eq 'cpu')
+    $PROFILE_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  }
+
+  print '--- ', $profile_marker, "\n";
+  if (defined($main::collected_profile)) {
+    # if used with remote fetch, simply dump the collected profile to output.
+    open(SRC, "<$main::collected_profile");
+    while (<SRC>) {
+      print $_;
+    }
+    close(SRC);
+  } else {
+    # --raw/http: For everything to work correctly for non-remote profiles, we
+    # would need to extend PrintProfileData() to handle all possible profile
+    # types, re-enable the code that is currently disabled in ReadCPUProfile()
+    # and FixCallerAddresses(), and remove the remote profile dumping code in
+    # the block above.
+    die "--raw/http: jeprof can only dump remote profiles for --raw\n";
+    # dump a cpu-format profile to standard out
+    PrintProfileData($profile);
+  }
+}
+
+# Print text output
+sub PrintText {
+  my $symbols = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $line_limit = shift;
+
+  my $total = TotalProfile($flat);
+
+  # Which profile to sort by?
+  my $s = $main::opt_cum ? $cumulative : $flat;
+
+  my $running_sum = 0;
+  my $lines = 0;
+  foreach my $k (sort { GetEntry($s, $b) <=> GetEntry($s, $a) || $a cmp $b }
+                 keys(%{$cumulative})) {
+    my $f = GetEntry($flat, $k);
+    my $c = GetEntry($cumulative, $k);
+    $running_sum += $f;
+
+    my $sym = $k;
+    if (exists($symbols->{$k})) {
+      $sym = $symbols->{$k}->[0] . " " . $symbols->{$k}->[1];
+      if ($main::opt_addresses) {
+        $sym = $k . " " . $sym;
+      }
+    }
+
+    if ($f != 0 || $c != 0) {
+      printf("%8s %6s %6s %8s %6s %s\n",
+             Unparse($f),
+             Percent($f, $total),
+             Percent($running_sum, $total),
+             Unparse($c),
+             Percent($c, $total),
+             $sym);
+    }
+    $lines++;
+    last if ($line_limit >= 0 && $lines >= $line_limit);
+  }
+}
+
+# Callgrind format has a compression for repeated function and file
+# names.  You show the name the first time, and just use its number
+# subsequently.  This can cut down the file to about a third or a
+# quarter of its uncompressed size.  $key and $val are the key/value
+# pair that would normally be printed by callgrind; $map is a map from
+# value to number.
+sub CompressedCGName {
+  my($key, $val, $map) = @_;
+  my $idx = $map->{$val};
+  # For very short keys, providing an index hurts rather than helps.
+  if (length($val) <= 3) {
+    return "$key=$val\n";
+  } elsif (defined($idx)) {
+    return "$key=($idx)\n";
+  } else {
+    # scalar(keys $map) gives the number of items in the map.
+    $idx = scalar(keys(%{$map})) + 1;
+    $map->{$val} = $idx;
+    return "$key=($idx) $val\n";
+  }
+}
+
+# Print the call graph in a way that's suiteable for callgrind.
+sub PrintCallgrind {
+  my $calls = shift;
+  my $filename;
+  my %filename_to_index_map;
+  my %fnname_to_index_map;
+
+  if ($main::opt_interactive) {
+    $filename = shift;
+    print STDERR "Writing callgrind file to '$filename'.\n"
+  } else {
+    $filename = "&STDOUT";
+  }
+  open(CG, ">$filename");
+  printf CG ("events: Hits\n\n");
+  foreach my $call ( map { $_->[0] }
+                     sort { $a->[1] cmp $b ->[1] ||
+                            $a->[2] <=> $b->[2] }
+                     map { /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/;
+                           [$_, $1, $2] }
+                     keys %$calls ) {
+    my $count = int($calls->{$call});
+    $call =~ /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/;
+    my ( $caller_file, $caller_line, $caller_function,
+         $callee_file, $callee_line, $callee_function ) =
+       ( $1, $2, $3, $5, $6, $7 );
+
+    # TODO(csilvers): for better compression, collect all the
+    # caller/callee_files and functions first, before printing
+    # anything, and only compress those referenced more than once.
+    printf CG CompressedCGName("fl", $caller_file, \%filename_to_index_map);
+    printf CG CompressedCGName("fn", $caller_function, \%fnname_to_index_map);
+    if (defined $6) {
+      printf CG CompressedCGName("cfl", $callee_file, \%filename_to_index_map);
+      printf CG CompressedCGName("cfn", $callee_function, \%fnname_to_index_map);
+      printf CG ("calls=$count $callee_line\n");
+    }
+    printf CG ("$caller_line $count\n\n");
+  }
+}
+
+# Print disassembly for all all routines that match $main::opt_disasm
+sub PrintDisassembly {
+  my $libs = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $disasm_opts = shift;
+
+  my $total = TotalProfile($flat);
+
+  foreach my $lib (@{$libs}) {
+    my $symbol_table = GetProcedureBoundaries($lib->[0], $disasm_opts);
+    my $offset = AddressSub($lib->[1], $lib->[3]);
+    foreach my $routine (sort ByName keys(%{$symbol_table})) {
+      my $start_addr = $symbol_table->{$routine}->[0];
+      my $end_addr = $symbol_table->{$routine}->[1];
+      # See if there are any samples in this routine
+      my $length = hex(AddressSub($end_addr, $start_addr));
+      my $addr = AddressAdd($start_addr, $offset);
+      for (my $i = 0; $i < $length; $i++) {
+        if (defined($cumulative->{$addr})) {
+          PrintDisassembledFunction($lib->[0], $offset,
+                                    $routine, $flat, $cumulative,
+                                    $start_addr, $end_addr, $total);
+          last;
+        }
+        $addr = AddressInc($addr);
+      }
+    }
+  }
+}
+
+# Return reference to array of tuples of the form:
+#       [start_address, filename, linenumber, instruction, limit_address]
+# E.g.,
+#       ["0x806c43d", "/foo/bar.cc", 131, "ret", "0x806c440"]
+sub Disassemble {
+  my $prog = shift;
+  my $offset = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+
+  my $objdump = $obj_tool_map{"objdump"};
+  my $cmd = ShellEscape($objdump, "-C", "-d", "-l", "--no-show-raw-insn",
+                        "--start-address=0x$start_addr",
+                        "--stop-address=0x$end_addr", $prog);
+  open(OBJDUMP, "$cmd |") || error("$cmd: $!\n");
+  my @result = ();
+  my $filename = "";
+  my $linenumber = -1;
+  my $last = ["", "", "", ""];
+  while (<OBJDUMP>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    chop;
+    if (m|\s*([^:\s]+):(\d+)\s*$|) {
+      # Location line of the form:
+      #   <filename>:<linenumber>
+      $filename = $1;
+      $linenumber = $2;
+    } elsif (m/^ +([0-9a-f]+):\s*(.*)/) {
+      # Disassembly line -- zero-extend address to full length
+      my $addr = HexExtend($1);
+      my $k = AddressAdd($addr, $offset);
+      $last->[4] = $k;   # Store ending address for previous instruction
+      $last = [$k, $filename, $linenumber, $2, $end_addr];
+      push(@result, $last);
+    }
+  }
+  close(OBJDUMP);
+  return @result;
+}
+
+# The input file should contain lines of the form /proc/maps-like
+# output (same format as expected from the profiles) or that looks
+# like hex addresses (like "0xDEADBEEF").  We will parse all
+# /proc/maps output, and for all the hex addresses, we will output
+# "short" symbol names, one per line, in the same order as the input.
+sub PrintSymbols {
+  my $maps_and_symbols_file = shift;
+
+  # ParseLibraries expects pcs to be in a set.  Fine by us...
+  my @pclist = ();   # pcs in sorted order
+  my $pcs = {};
+  my $map = "";
+  foreach my $line (<$maps_and_symbols_file>) {
+    $line =~ s/\r//g;    # turn windows-looking lines into unix-looking lines
+    if ($line =~ /\b(0x[0-9a-f]+)\b/i) {
+      push(@pclist, HexExtend($1));
+      $pcs->{$pclist[-1]} = 1;
+    } else {
+      $map .= $line;
+    }
+  }
+
+  my $libs = ParseLibraries($main::prog, $map, $pcs);
+  my $symbols = ExtractSymbols($libs, $pcs);
+
+  foreach my $pc (@pclist) {
+    # ->[0] is the shortname, ->[2] is the full name
+    print(($symbols->{$pc}->[0] || "??") . "\n");
+  }
+}
+
+
+# For sorting functions by name
+sub ByName {
+  return ShortFunctionName($a) cmp ShortFunctionName($b);
+}
+
+# Print source-listing for all all routines that match $list_opts
+sub PrintListing {
+  my $total = shift;
+  my $libs = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $list_opts = shift;
+  my $html = shift;
+
+  my $output = \*STDOUT;
+  my $fname = "";
+
+  if ($html) {
+    # Arrange to write the output to a temporary file
+    $fname = TempName($main::next_tmpfile, "html");
+    $main::next_tmpfile++;
+    if (!open(TEMP, ">$fname")) {
+      print STDERR "$fname: $!\n";
+      return;
+    }
+    $output = \*TEMP;
+    print $output HtmlListingHeader();
+    printf $output ("<div class=\"legend\">%s<br>Total: %s %s</div>\n",
+                    $main::prog, Unparse($total), Units());
+  }
+
+  my $listed = 0;
+  foreach my $lib (@{$libs}) {
+    my $symbol_table = GetProcedureBoundaries($lib->[0], $list_opts);
+    my $offset = AddressSub($lib->[1], $lib->[3]);
+    foreach my $routine (sort ByName keys(%{$symbol_table})) {
+      # Print if there are any samples in this routine
+      my $start_addr = $symbol_table->{$routine}->[0];
+      my $end_addr = $symbol_table->{$routine}->[1];
+      my $length = hex(AddressSub($end_addr, $start_addr));
+      my $addr = AddressAdd($start_addr, $offset);
+      for (my $i = 0; $i < $length; $i++) {
+        if (defined($cumulative->{$addr})) {
+          $listed += PrintSource(
+            $lib->[0], $offset,
+            $routine, $flat, $cumulative,
+            $start_addr, $end_addr,
+            $html,
+            $output);
+          last;
+        }
+        $addr = AddressInc($addr);
+      }
+    }
+  }
+
+  if ($html) {
+    if ($listed > 0) {
+      print $output HtmlListingFooter();
+      close($output);
+      RunWeb($fname);
+    } else {
+      close($output);
+      unlink($fname);
+    }
+  }
+}
+
+sub HtmlListingHeader {
+  return <<'EOF';
+<DOCTYPE html>
+<html>
+<head>
+<title>Pprof listing</title>
+<style type="text/css">
+body {
+  font-family: sans-serif;
+}
+h1 {
+  font-size: 1.5em;
+  margin-bottom: 4px;
+}
+.legend {
+  font-size: 1.25em;
+}
+.line {
+  color: #aaaaaa;
+}
+.nop {
+  color: #aaaaaa;
+}
+.unimportant {
+  color: #cccccc;
+}
+.disasmloc {
+  color: #000000;
+}
+.deadsrc {
+  cursor: pointer;
+}
+.deadsrc:hover {
+  background-color: #eeeeee;
+}
+.livesrc {
+  color: #0000ff;
+  cursor: pointer;
+}
+.livesrc:hover {
+  background-color: #eeeeee;
+}
+.asm {
+  color: #008800;
+  display: none;
+}
+</style>
+<script type="text/javascript">
+function jeprof_toggle_asm(e) {
+  var target;
+  if (!e) e = window.event;
+  if (e.target) target = e.target;
+  else if (e.srcElement) target = e.srcElement;
+
+  if (target) {
+    var asm = target.nextSibling;
+    if (asm && asm.className == "asm") {
+      asm.style.display = (asm.style.display == "block" ? "" : "block");
+      e.preventDefault();
+      return false;
+    }
+  }
+}
+</script>
+</head>
+<body>
+EOF
+}
+
+sub HtmlListingFooter {
+  return <<'EOF';
+</body>
+</html>
+EOF
+}
+
+sub HtmlEscape {
+  my $text = shift;
+  $text =~ s/&/&amp;/g;
+  $text =~ s/</&lt;/g;
+  $text =~ s/>/&gt;/g;
+  return $text;
+}
+
+# Returns the indentation of the line, if it has any non-whitespace
+# characters.  Otherwise, returns -1.
+sub Indentation {
+  my $line = shift;
+  if (m/^(\s*)\S/) {
+    return length($1);
+  } else {
+    return -1;
+  }
+}
+
+# If the symbol table contains inlining info, Disassemble() may tag an
+# instruction with a location inside an inlined function.  But for
+# source listings, we prefer to use the location in the function we
+# are listing.  So use MapToSymbols() to fetch full location
+# information for each instruction and then pick out the first
+# location from a location list (location list contains callers before
+# callees in case of inlining).
+#
+# After this routine has run, each entry in $instructions contains:
+#   [0] start address
+#   [1] filename for function we are listing
+#   [2] line number for function we are listing
+#   [3] disassembly
+#   [4] limit address
+#   [5] most specific filename (may be different from [1] due to inlining)
+#   [6] most specific line number (may be different from [2] due to inlining)
+sub GetTopLevelLineNumbers {
+  my ($lib, $offset, $instructions) = @_;
+  my $pcs = [];
+  for (my $i = 0; $i <= $#{$instructions}; $i++) {
+    push(@{$pcs}, $instructions->[$i]->[0]);
+  }
+  my $symbols = {};
+  MapToSymbols($lib, $offset, $pcs, $symbols);
+  for (my $i = 0; $i <= $#{$instructions}; $i++) {
+    my $e = $instructions->[$i];
+    push(@{$e}, $e->[1]);
+    push(@{$e}, $e->[2]);
+    my $addr = $e->[0];
+    my $sym = $symbols->{$addr};
+    if (defined($sym)) {
+      if ($#{$sym} >= 2 && $sym->[1] =~ m/^(.*):(\d+)$/) {
+        $e->[1] = $1;  # File name
+        $e->[2] = $2;  # Line number
+      }
+    }
+  }
+}
+
+# Print source-listing for one routine
+sub PrintSource {
+  my $prog = shift;
+  my $offset = shift;
+  my $routine = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+  my $html = shift;
+  my $output = shift;
+
+  # Disassemble all instructions (just to get line numbers)
+  my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr);
+  GetTopLevelLineNumbers($prog, $offset, \@instructions);
+
+  # Hack 1: assume that the first source file encountered in the
+  # disassembly contains the routine
+  my $filename = undef;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    if ($instructions[$i]->[2] >= 0) {
+      $filename = $instructions[$i]->[1];
+      last;
+    }
+  }
+  if (!defined($filename)) {
+    print STDERR "no filename found in $routine\n";
+    return 0;
+  }
+
+  # Hack 2: assume that the largest line number from $filename is the
+  # end of the procedure.  This is typically safe since if P1 contains
+  # an inlined call to P2, then P2 usually occurs earlier in the
+  # source file.  If this does not work, we might have to compute a
+  # density profile or just print all regions we find.
+  my $lastline = 0;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    my $f = $instructions[$i]->[1];
+    my $l = $instructions[$i]->[2];
+    if (($f eq $filename) && ($l > $lastline)) {
+      $lastline = $l;
+    }
+  }
+
+  # Hack 3: assume the first source location from "filename" is the start of
+  # the source code.
+  my $firstline = 1;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    if ($instructions[$i]->[1] eq $filename) {
+      $firstline = $instructions[$i]->[2];
+      last;
+    }
+  }
+
+  # Hack 4: Extend last line forward until its indentation is less than
+  # the indentation we saw on $firstline
+  my $oldlastline = $lastline;
+  {
+    if (!open(FILE, "<$filename")) {
+      print STDERR "$filename: $!\n";
+      return 0;
+    }
+    my $l = 0;
+    my $first_indentation = -1;
+    while (<FILE>) {
+      s/\r//g;         # turn windows-looking lines into unix-looking lines
+      $l++;
+      my $indent = Indentation($_);
+      if ($l >= $firstline) {
+        if ($first_indentation < 0 && $indent >= 0) {
+          $first_indentation = $indent;
+          last if ($first_indentation == 0);
+        }
+      }
+      if ($l >= $lastline && $indent >= 0) {
+        if ($indent >= $first_indentation) {
+          $lastline = $l+1;
+        } else {
+          last;
+        }
+      }
+    }
+    close(FILE);
+  }
+
+  # Assign all samples to the range $firstline,$lastline,
+  # Hack 4: If an instruction does not occur in the range, its samples
+  # are moved to the next instruction that occurs in the range.
+  my $samples1 = {};        # Map from line number to flat count
+  my $samples2 = {};        # Map from line number to cumulative count
+  my $running1 = 0;         # Unassigned flat counts
+  my $running2 = 0;         # Unassigned cumulative counts
+  my $total1 = 0;           # Total flat counts
+  my $total2 = 0;           # Total cumulative counts
+  my %disasm = ();          # Map from line number to disassembly
+  my $running_disasm = "";  # Unassigned disassembly
+  my $skip_marker = "---\n";
+  if ($html) {
+    $skip_marker = "";
+    for (my $l = $firstline; $l <= $lastline; $l++) {
+      $disasm{$l} = "";
+    }
+  }
+  my $last_dis_filename = '';
+  my $last_dis_linenum = -1;
+  my $last_touched_line = -1;  # To detect gaps in disassembly for a line
+  foreach my $e (@instructions) {
+    # Add up counts for all address that fall inside this instruction
+    my $c1 = 0;
+    my $c2 = 0;
+    for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) {
+      $c1 += GetEntry($flat, $a);
+      $c2 += GetEntry($cumulative, $a);
+    }
+
+    if ($html) {
+      my $dis = sprintf("      %6s %6s \t\t%8s: %s ",
+                        HtmlPrintNumber($c1),
+                        HtmlPrintNumber($c2),
+                        UnparseAddress($offset, $e->[0]),
+                        CleanDisassembly($e->[3]));
+
+      # Append the most specific source line associated with this instruction
+      if (length($dis) < 80) { $dis .= (' ' x (80 - length($dis))) };
+      $dis = HtmlEscape($dis);
+      my $f = $e->[5];
+      my $l = $e->[6];
+      if ($f ne $last_dis_filename) {
+        $dis .= sprintf("<span class=disasmloc>%s:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      } elsif ($l ne $last_dis_linenum) {
+        # De-emphasize the unchanged file name portion
+        $dis .= sprintf("<span class=unimportant>%s</span>" .
+                        "<span class=disasmloc>:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      } else {
+        # De-emphasize the entire location
+        $dis .= sprintf("<span class=unimportant>%s:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      }
+      $last_dis_filename = $f;
+      $last_dis_linenum = $l;
+      $running_disasm .= $dis;
+      $running_disasm .= "\n";
+    }
+
+    $running1 += $c1;
+    $running2 += $c2;
+    $total1 += $c1;
+    $total2 += $c2;
+    my $file = $e->[1];
+    my $line = $e->[2];
+    if (($file eq $filename) &&
+        ($line >= $firstline) &&
+        ($line <= $lastline)) {
+      # Assign all accumulated samples to this line
+      AddEntry($samples1, $line, $running1);
+      AddEntry($samples2, $line, $running2);
+      $running1 = 0;
+      $running2 = 0;
+      if ($html) {
+        if ($line != $last_touched_line && $disasm{$line} ne '') {
+          $disasm{$line} .= "\n";
+        }
+        $disasm{$line} .= $running_disasm;
+        $running_disasm = '';
+        $last_touched_line = $line;
+      }
+    }
+  }
+
+  # Assign any leftover samples to $lastline
+  AddEntry($samples1, $lastline, $running1);
+  AddEntry($samples2, $lastline, $running2);
+  if ($html) {
+    if ($lastline != $last_touched_line && $disasm{$lastline} ne '') {
+      $disasm{$lastline} .= "\n";
+    }
+    $disasm{$lastline} .= $running_disasm;
+  }
+
+  if ($html) {
+    printf $output (
+      "<h1>%s</h1>%s\n<pre onClick=\"jeprof_toggle_asm()\">\n" .
+      "Total:%6s %6s (flat / cumulative %s)\n",
+      HtmlEscape(ShortFunctionName($routine)),
+      HtmlEscape(CleanFileName($filename)),
+      Unparse($total1),
+      Unparse($total2),
+      Units());
+  } else {
+    printf $output (
+      "ROUTINE ====================== %s in %s\n" .
+      "%6s %6s Total %s (flat / cumulative)\n",
+      ShortFunctionName($routine),
+      CleanFileName($filename),
+      Unparse($total1),
+      Unparse($total2),
+      Units());
+  }
+  if (!open(FILE, "<$filename")) {
+    print STDERR "$filename: $!\n";
+    return 0;
+  }
+  my $l = 0;
+  while (<FILE>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    $l++;
+    if ($l >= $firstline - 5 &&
+        (($l <= $oldlastline + 5) || ($l <= $lastline))) {
+      chop;
+      my $text = $_;
+      if ($l == $firstline) { print $output $skip_marker; }
+      my $n1 = GetEntry($samples1, $l);
+      my $n2 = GetEntry($samples2, $l);
+      if ($html) {
+        # Emit a span that has one of the following classes:
+        #    livesrc -- has samples
+        #    deadsrc -- has disassembly, but with no samples
+        #    nop     -- has no matching disasembly
+        # Also emit an optional span containing disassembly.
+        my $dis = $disasm{$l};
+        my $asm = "";
+        if (defined($dis) && $dis ne '') {
+          $asm = "<span class=\"asm\">" . $dis . "</span>";
+        }
+        my $source_class = (($n1 + $n2 > 0)
+                            ? "livesrc"
+                            : (($asm ne "") ? "deadsrc" : "nop"));
+        printf $output (
+          "<span class=\"line\">%5d</span> " .
+          "<span class=\"%s\">%6s %6s %s</span>%s\n",
+          $l, $source_class,
+          HtmlPrintNumber($n1),
+          HtmlPrintNumber($n2),
+          HtmlEscape($text),
+          $asm);
+      } else {
+        printf $output(
+          "%6s %6s %4d: %s\n",
+          UnparseAlt($n1),
+          UnparseAlt($n2),
+          $l,
+          $text);
+      }
+      if ($l == $lastline)  { print $output $skip_marker; }
+    };
+  }
+  close(FILE);
+  if ($html) {
+    print $output "</pre>\n";
+  }
+  return 1;
+}
+
+# Return the source line for the specified file/linenumber.
+# Returns undef if not found.
+sub SourceLine {
+  my $file = shift;
+  my $line = shift;
+
+  # Look in cache
+  if (!defined($main::source_cache{$file})) {
+    if (100 < scalar keys(%main::source_cache)) {
+      # Clear the cache when it gets too big
+      $main::source_cache = ();
+    }
+
+    # Read all lines from the file
+    if (!open(FILE, "<$file")) {
+      print STDERR "$file: $!\n";
+      $main::source_cache{$file} = [];  # Cache the negative result
+      return undef;
+    }
+    my $lines = [];
+    push(@{$lines}, "");        # So we can use 1-based line numbers as indices
+    while (<FILE>) {
+      push(@{$lines}, $_);
+    }
+    close(FILE);
+
+    # Save the lines in the cache
+    $main::source_cache{$file} = $lines;
+  }
+
+  my $lines = $main::source_cache{$file};
+  if (($line < 0) || ($line > $#{$lines})) {
+    return undef;
+  } else {
+    return $lines->[$line];
+  }
+}
+
+# Print disassembly for one routine with interspersed source if available
+sub PrintDisassembledFunction {
+  my $prog = shift;
+  my $offset = shift;
+  my $routine = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+  my $total = shift;
+
+  # Disassemble all instructions
+  my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr);
+
+  # Make array of counts per instruction
+  my @flat_count = ();
+  my @cum_count = ();
+  my $flat_total = 0;
+  my $cum_total = 0;
+  foreach my $e (@instructions) {
+    # Add up counts for all address that fall inside this instruction
+    my $c1 = 0;
+    my $c2 = 0;
+    for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) {
+      $c1 += GetEntry($flat, $a);
+      $c2 += GetEntry($cumulative, $a);
+    }
+    push(@flat_count, $c1);
+    push(@cum_count, $c2);
+    $flat_total += $c1;
+    $cum_total += $c2;
+  }
+
+  # Print header with total counts
+  printf("ROUTINE ====================== %s\n" .
+         "%6s %6s %s (flat, cumulative) %.1f%% of total\n",
+         ShortFunctionName($routine),
+         Unparse($flat_total),
+         Unparse($cum_total),
+         Units(),
+         ($cum_total * 100.0) / $total);
+
+  # Process instructions in order
+  my $current_file = "";
+  for (my $i = 0; $i <= $#instructions; ) {
+    my $e = $instructions[$i];
+
+    # Print the new file name whenever we switch files
+    if ($e->[1] ne $current_file) {
+      $current_file = $e->[1];
+      my $fname = $current_file;
+      $fname =~ s|^\./||;   # Trim leading "./"
+
+      # Shorten long file names
+      if (length($fname) >= 58) {
+        $fname = "..." . substr($fname, -55);
+      }
+      printf("-------------------- %s\n", $fname);
+    }
+
+    # TODO: Compute range of lines to print together to deal with
+    # small reorderings.
+    my $first_line = $e->[2];
+    my $last_line = $first_line;
+    my %flat_sum = ();
+    my %cum_sum = ();
+    for (my $l = $first_line; $l <= $last_line; $l++) {
+      $flat_sum{$l} = 0;
+      $cum_sum{$l} = 0;
+    }
+
+    # Find run of instructions for this range of source lines
+    my $first_inst = $i;
+    while (($i <= $#instructions) &&
+           ($instructions[$i]->[2] >= $first_line) &&
+           ($instructions[$i]->[2] <= $last_line)) {
+      $e = $instructions[$i];
+      $flat_sum{$e->[2]} += $flat_count[$i];
+      $cum_sum{$e->[2]} += $cum_count[$i];
+      $i++;
+    }
+    my $last_inst = $i - 1;
+
+    # Print source lines
+    for (my $l = $first_line; $l <= $last_line; $l++) {
+      my $line = SourceLine($current_file, $l);
+      if (!defined($line)) {
+        $line = "?\n";
+        next;
+      } else {
+        $line =~ s/^\s+//;
+      }
+      printf("%6s %6s %5d: %s",
+             UnparseAlt($flat_sum{$l}),
+             UnparseAlt($cum_sum{$l}),
+             $l,
+             $line);
+    }
+
+    # Print disassembly
+    for (my $x = $first_inst; $x <= $last_inst; $x++) {
+      my $e = $instructions[$x];
+      printf("%6s %6s    %8s: %6s\n",
+             UnparseAlt($flat_count[$x]),
+             UnparseAlt($cum_count[$x]),
+             UnparseAddress($offset, $e->[0]),
+             CleanDisassembly($e->[3]));
+    }
+  }
+}
+
+# Print DOT graph
+sub PrintDot {
+  my $prog = shift;
+  my $symbols = shift;
+  my $raw = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $overall_total = shift;
+
+  # Get total
+  my $local_total = TotalProfile($flat);
+  my $nodelimit = int($main::opt_nodefraction * $local_total);
+  my $edgelimit = int($main::opt_edgefraction * $local_total);
+  my $nodecount = $main::opt_nodecount;
+
+  # Find nodes to include
+  my @list = (sort { abs(GetEntry($cumulative, $b)) <=>
+                     abs(GetEntry($cumulative, $a))
+                     || $a cmp $b }
+              keys(%{$cumulative}));
+  my $last = $nodecount - 1;
+  if ($last > $#list) {
+    $last = $#list;
+  }
+  while (($last >= 0) &&
+         (abs(GetEntry($cumulative, $list[$last])) <= $nodelimit)) {
+    $last--;
+  }
+  if ($last < 0) {
+    print STDERR "No nodes to print\n";
+    return 0;
+  }
+
+  if ($nodelimit > 0 || $edgelimit > 0) {
+    printf STDERR ("Dropping nodes with <= %s %s; edges with <= %s abs(%s)\n",
+                   Unparse($nodelimit), Units(),
+                   Unparse($edgelimit), Units());
+  }
+
+  # Open DOT output file
+  my $output;
+  my $escaped_dot = ShellEscape(@DOT);
+  my $escaped_ps2pdf = ShellEscape(@PS2PDF);
+  if ($main::opt_gv) {
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "ps"));
+    $output = "| $escaped_dot -Tps2 >$escaped_outfile";
+  } elsif ($main::opt_evince) {
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "pdf"));
+    $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - $escaped_outfile";
+  } elsif ($main::opt_ps) {
+    $output = "| $escaped_dot -Tps2";
+  } elsif ($main::opt_pdf) {
+    $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - -";
+  } elsif ($main::opt_web || $main::opt_svg) {
+    # We need to post-process the SVG, so write to a temporary file always.
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "svg"));
+    $output = "| $escaped_dot -Tsvg >$escaped_outfile";
+  } elsif ($main::opt_gif) {
+    $output = "| $escaped_dot -Tgif";
+  } else {
+    $output = ">&STDOUT";
+  }
+  open(DOT, $output) || error("$output: $!\n");
+
+  # Title
+  printf DOT ("digraph \"%s; %s %s\" {\n",
+              $prog,
+              Unparse($overall_total),
+              Units());
+  if ($main::opt_pdf) {
+    # The output is more printable if we set the page size for dot.
+    printf DOT ("size=\"8,11\"\n");
+  }
+  printf DOT ("node [width=0.375,height=0.25];\n");
+
+  # Print legend
+  printf DOT ("Legend [shape=box,fontsize=24,shape=plaintext," .
+              "label=\"%s\\l%s\\l%s\\l%s\\l%s\\l\"];\n",
+              $prog,
+              sprintf("Total %s: %s", Units(), Unparse($overall_total)),
+              sprintf("Focusing on: %s", Unparse($local_total)),
+              sprintf("Dropped nodes with <= %s abs(%s)",
+                      Unparse($nodelimit), Units()),
+              sprintf("Dropped edges with <= %s %s",
+                      Unparse($edgelimit), Units())
+              );
+
+  # Print nodes
+  my %node = ();
+  my $nextnode = 1;
+  foreach my $a (@list[0..$last]) {
+    # Pick font size
+    my $f = GetEntry($flat, $a);
+    my $c = GetEntry($cumulative, $a);
+
+    my $fs = 8;
+    if ($local_total > 0) {
+      $fs = 8 + (50.0 * sqrt(abs($f * 1.0 / $local_total)));
+    }
+
+    $node{$a} = $nextnode++;
+    my $sym = $a;
+    $sym =~ s/\s+/\\n/g;
+    $sym =~ s/::/\\n/g;
+
+    # Extra cumulative info to print for non-leaves
+    my $extra = "";
+    if ($f != $c) {
+      $extra = sprintf("\\rof %s (%s)",
+                       Unparse($c),
+                       Percent($c, $local_total));
+    }
+    my $style = "";
+    if ($main::opt_heapcheck) {
+      if ($f > 0) {
+        # make leak-causing nodes more visible (add a background)
+        $style = ",style=filled,fillcolor=gray"
+      } elsif ($f < 0) {
+        # make anti-leak-causing nodes (which almost never occur)
+        # stand out as well (triple border)
+        $style = ",peripheries=3"
+      }
+    }
+
+    printf DOT ("N%d [label=\"%s\\n%s (%s)%s\\r" .
+                "\",shape=box,fontsize=%.1f%s];\n",
+                $node{$a},
+                $sym,
+                Unparse($f),
+                Percent($f, $local_total),
+                $extra,
+                $fs,
+                $style,
+               );
+  }
+
+  # Get edges and counts per edge
+  my %edge = ();
+  my $n;
+  my $fullname_to_shortname_map = {};
+  FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map);
+  foreach my $k (keys(%{$raw})) {
+    # TODO: omit low %age edges
+    $n = $raw->{$k};
+    my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k);
+    for (my $i = 1; $i <= $#translated; $i++) {
+      my $src = $translated[$i];
+      my $dst = $translated[$i-1];
+      #next if ($src eq $dst);  # Avoid self-edges?
+      if (exists($node{$src}) && exists($node{$dst})) {
+        my $edge_label = "$src\001$dst";
+        if (!exists($edge{$edge_label})) {
+          $edge{$edge_label} = 0;
+        }
+        $edge{$edge_label} += $n;
+      }
+    }
+  }
+
+  # Print edges (process in order of decreasing counts)
+  my %indegree = ();   # Number of incoming edges added per node so far
+  my %outdegree = ();  # Number of outgoing edges added per node so far
+  foreach my $e (sort { $edge{$b} <=> $edge{$a} } keys(%edge)) {
+    my @x = split(/\001/, $e);
+    $n = $edge{$e};
+
+    # Initialize degree of kept incoming and outgoing edges if necessary
+    my $src = $x[0];
+    my $dst = $x[1];
+    if (!exists($outdegree{$src})) { $outdegree{$src} = 0; }
+    if (!exists($indegree{$dst})) { $indegree{$dst} = 0; }
+
+    my $keep;
+    if ($indegree{$dst} == 0) {
+      # Keep edge if needed for reachability
+      $keep = 1;
+    } elsif (abs($n) <= $edgelimit) {
+      # Drop if we are below --edgefraction
+      $keep = 0;
+    } elsif ($outdegree{$src} >= $main::opt_maxdegree ||
+             $indegree{$dst} >= $main::opt_maxdegree) {
+      # Keep limited number of in/out edges per node
+      $keep = 0;
+    } else {
+      $keep = 1;
+    }
+
+    if ($keep) {
+      $outdegree{$src}++;
+      $indegree{$dst}++;
+
+      # Compute line width based on edge count
+      my $fraction = abs($local_total ? (3 * ($n / $local_total)) : 0);
+      if ($fraction > 1) { $fraction = 1; }
+      my $w = $fraction * 2;
+      if ($w < 1 && ($main::opt_web || $main::opt_svg)) {
+        # SVG output treats line widths < 1 poorly.
+        $w = 1;
+      }
+
+      # Dot sometimes segfaults if given edge weights that are too large, so
+      # we cap the weights at a large value
+      my $edgeweight = abs($n) ** 0.7;
+      if ($edgeweight > 100000) { $edgeweight = 100000; }
+      $edgeweight = int($edgeweight);
+
+      my $style = sprintf("setlinewidth(%f)", $w);
+      if ($x[1] =~ m/\(inline\)/) {
+        $style .= ",dashed";
+      }
+
+      # Use a slightly squashed function of the edge count as the weight
+      printf DOT ("N%s -> N%s [label=%s, weight=%d, style=\"%s\"];\n",
+                  $node{$x[0]},
+                  $node{$x[1]},
+                  Unparse($n),
+                  $edgeweight,
+                  $style);
+    }
+  }
+
+  print DOT ("}\n");
+  close(DOT);
+
+  if ($main::opt_web || $main::opt_svg) {
+    # Rewrite SVG to be more usable inside web browser.
+    RewriteSvg(TempName($main::next_tmpfile, "svg"));
+  }
+
+  return 1;
+}
+
+sub RewriteSvg {
+  my $svgfile = shift;
+
+  open(SVG, $svgfile) || die "open temp svg: $!";
+  my @svg = <SVG>;
+  close(SVG);
+  unlink $svgfile;
+  my $svg = join('', @svg);
+
+  # Dot's SVG output is
+  #
+  #    <svg width="___" height="___"
+  #     viewBox="___" xmlns=...>
+  #    <g id="graph0" transform="...">
+  #    ...
+  #    </g>
+  #    </svg>
+  #
+  # Change it to
+  #
+  #    <svg width="100%" height="100%"
+  #     xmlns=...>
+  #    $svg_javascript
+  #    <g id="viewport" transform="translate(0,0)">
+  #    <g id="graph0" transform="...">
+  #    ...
+  #    </g>
+  #    </g>
+  #    </svg>
+
+  # Fix width, height; drop viewBox.
+  $svg =~ s/(?s)<svg width="[^"]+" height="[^"]+"(.*?)viewBox="[^"]+"/<svg width="100%" height="100%"$1/;
+
+  # Insert script, viewport <g> above first <g>
+  my $svg_javascript = SvgJavascript();
+  my $viewport = "<g id=\"viewport\" transform=\"translate(0,0)\">\n";
+  $svg =~ s/<g id="graph\d"/$svg_javascript$viewport$&/;
+
+  # Insert final </g> above </svg>.
+  $svg =~ s/(.*)(<\/svg>)/$1<\/g>$2/;
+  $svg =~ s/<g id="graph\d"(.*?)/<g id="viewport"$1/;
+
+  if ($main::opt_svg) {
+    # --svg: write to standard output.
+    print $svg;
+  } else {
+    # Write back to temporary file.
+    open(SVG, ">$svgfile") || die "open $svgfile: $!";
+    print SVG $svg;
+    close(SVG);
+  }
+}
+
+sub SvgJavascript {
+  return <<'EOF';
+<script type="text/ecmascript"><![CDATA[
+// SVGPan
+// http://www.cyberz.org/blog/2009/12/08/svgpan-a-javascript-svg-panzoomdrag-library/
+// Local modification: if(true || ...) below to force panning, never moving.
+
+/**
+ *  SVGPan library 1.2
+ * ====================
+ *
+ * Given an unique existing element with id "viewport", including the
+ * the library into any SVG adds the following capabilities:
+ *
+ *  - Mouse panning
+ *  - Mouse zooming (using the wheel)
+ *  - Object dargging
+ *
+ * Known issues:
+ *
+ *  - Zooming (while panning) on Safari has still some issues
+ *
+ * Releases:
+ *
+ * 1.2, Sat Mar 20 08:42:50 GMT 2010, Zeng Xiaohui
+ *	Fixed a bug with browser mouse handler interaction
+ *
+ * 1.1, Wed Feb  3 17:39:33 GMT 2010, Zeng Xiaohui
+ *	Updated the zoom code to support the mouse wheel on Safari/Chrome
+ *
+ * 1.0, Andrea Leofreddi
+ *	First release
+ *
+ * This code is licensed under the following BSD license:
+ *
+ * Copyright 2009-2010 Andrea Leofreddi <a.leofreddi@itcharm.com>. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are
+ * permitted provided that the following conditions are met:
+ *
+ *    1. Redistributions of source code must retain the above copyright notice, this list of
+ *       conditions and the following disclaimer.
+ *
+ *    2. Redistributions in binary form must reproduce the above copyright notice, this list
+ *       of conditions and the following disclaimer in the documentation and/or other materials
+ *       provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Andrea Leofreddi ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+ * FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL Andrea Leofreddi OR
+ * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * The views and conclusions contained in the software and documentation are those of the
+ * authors and should not be interpreted as representing official policies, either expressed
+ * or implied, of Andrea Leofreddi.
+ */
+
+var root = document.documentElement;
+
+var state = 'none', stateTarget, stateOrigin, stateTf;
+
+setupHandlers(root);
+
+/**
+ * Register handlers
+ */
+function setupHandlers(root){
+	setAttributes(root, {
+		"onmouseup" : "add(evt)",
+		"onmousedown" : "handleMouseDown(evt)",
+		"onmousemove" : "handleMouseMove(evt)",
+		"onmouseup" : "handleMouseUp(evt)",
+		//"onmouseout" : "handleMouseUp(evt)", // Decomment this to stop the pan functionality when dragging out of the SVG element
+	});
+
+	if(navigator.userAgent.toLowerCase().indexOf('webkit') >= 0)
+		window.addEventListener('mousewheel', handleMouseWheel, false); // Chrome/Safari
+	else
+		window.addEventListener('DOMMouseScroll', handleMouseWheel, false); // Others
+
+	var g = svgDoc.getElementById("svg");
+	g.width = "100%";
+	g.height = "100%";
+}
+
+/**
+ * Instance an SVGPoint object with given event coordinates.
+ */
+function getEventPoint(evt) {
+	var p = root.createSVGPoint();
+
+	p.x = evt.clientX;
+	p.y = evt.clientY;
+
+	return p;
+}
+
+/**
+ * Sets the current transform matrix of an element.
+ */
+function setCTM(element, matrix) {
+	var s = "matrix(" + matrix.a + "," + matrix.b + "," + matrix.c + "," + matrix.d + "," + matrix.e + "," + matrix.f + ")";
+
+	element.setAttribute("transform", s);
+}
+
+/**
+ * Dumps a matrix to a string (useful for debug).
+ */
+function dumpMatrix(matrix) {
+	var s = "[ " + matrix.a + ", " + matrix.c + ", " + matrix.e + "\n  " + matrix.b + ", " + matrix.d + ", " + matrix.f + "\n  0, 0, 1 ]";
+
+	return s;
+}
+
+/**
+ * Sets attributes of an element.
+ */
+function setAttributes(element, attributes){
+	for (i in attributes)
+		element.setAttributeNS(null, i, attributes[i]);
+}
+
+/**
+ * Handle mouse move event.
+ */
+function handleMouseWheel(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var delta;
+
+	if(evt.wheelDelta)
+		delta = evt.wheelDelta / 3600; // Chrome/Safari
+	else
+		delta = evt.detail / -90; // Mozilla
+
+	var z = 1 + delta; // Zoom factor: 0.9/1.1
+
+	var g = svgDoc.getElementById("viewport");
+
+	var p = getEventPoint(evt);
+
+	p = p.matrixTransform(g.getCTM().inverse());
+
+	// Compute new scale matrix in current mouse position
+	var k = root.createSVGMatrix().translate(p.x, p.y).scale(z).translate(-p.x, -p.y);
+
+        setCTM(g, g.getCTM().multiply(k));
+
+	stateTf = stateTf.multiply(k.inverse());
+}
+
+/**
+ * Handle mouse move event.
+ */
+function handleMouseMove(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var g = svgDoc.getElementById("viewport");
+
+	if(state == 'pan') {
+		// Pan mode
+		var p = getEventPoint(evt).matrixTransform(stateTf);
+
+		setCTM(g, stateTf.inverse().translate(p.x - stateOrigin.x, p.y - stateOrigin.y));
+	} else if(state == 'move') {
+		// Move mode
+		var p = getEventPoint(evt).matrixTransform(g.getCTM().inverse());
+
+		setCTM(stateTarget, root.createSVGMatrix().translate(p.x - stateOrigin.x, p.y - stateOrigin.y).multiply(g.getCTM().inverse()).multiply(stateTarget.getCTM()));
+
+		stateOrigin = p;
+	}
+}
+
+/**
+ * Handle click event.
+ */
+function handleMouseDown(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var g = svgDoc.getElementById("viewport");
+
+	if(true || evt.target.tagName == "svg") {
+		// Pan mode
+		state = 'pan';
+
+		stateTf = g.getCTM().inverse();
+
+		stateOrigin = getEventPoint(evt).matrixTransform(stateTf);
+	} else {
+		// Move mode
+		state = 'move';
+
+		stateTarget = evt.target;
+
+		stateTf = g.getCTM().inverse();
+
+		stateOrigin = getEventPoint(evt).matrixTransform(stateTf);
+	}
+}
+
+/**
+ * Handle mouse button release event.
+ */
+function handleMouseUp(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	if(state == 'pan' || state == 'move') {
+		// Quit pan mode
+		state = '';
+	}
+}
+
+]]></script>
+EOF
+}
+
+# Provides a map from fullname to shortname for cases where the
+# shortname is ambiguous.  The symlist has both the fullname and
+# shortname for all symbols, which is usually fine, but sometimes --
+# such as overloaded functions -- two different fullnames can map to
+# the same shortname.  In that case, we use the address of the
+# function to disambiguate the two.  This function fills in a map that
+# maps fullnames to modified shortnames in such cases.  If a fullname
+# is not present in the map, the 'normal' shortname provided by the
+# symlist is the appropriate one to use.
+sub FillFullnameToShortnameMap {
+  my $symbols = shift;
+  my $fullname_to_shortname_map = shift;
+  my $shortnames_seen_once = {};
+  my $shortnames_seen_more_than_once = {};
+
+  foreach my $symlist (values(%{$symbols})) {
+    # TODO(csilvers): deal with inlined symbols too.
+    my $shortname = $symlist->[0];
+    my $fullname = $symlist->[2];
+    if ($fullname !~ /<[0-9a-fA-F]+>$/) {  # fullname doesn't end in an address
+      next;       # the only collisions we care about are when addresses differ
+    }
+    if (defined($shortnames_seen_once->{$shortname}) &&
+        $shortnames_seen_once->{$shortname} ne $fullname) {
+      $shortnames_seen_more_than_once->{$shortname} = 1;
+    } else {
+      $shortnames_seen_once->{$shortname} = $fullname;
+    }
+  }
+
+  foreach my $symlist (values(%{$symbols})) {
+    my $shortname = $symlist->[0];
+    my $fullname = $symlist->[2];
+    # TODO(csilvers): take in a list of addresses we care about, and only
+    # store in the map if $symlist->[1] is in that list.  Saves space.
+    next if defined($fullname_to_shortname_map->{$fullname});
+    if (defined($shortnames_seen_more_than_once->{$shortname})) {
+      if ($fullname =~ /<0*([^>]*)>$/) {   # fullname has address at end of it
+        $fullname_to_shortname_map->{$fullname} = "$shortname\@$1";
+      }
+    }
+  }
+}
+
+# Return a small number that identifies the argument.
+# Multiple calls with the same argument will return the same number.
+# Calls with different arguments will return different numbers.
+sub ShortIdFor {
+  my $key = shift;
+  my $id = $main::uniqueid{$key};
+  if (!defined($id)) {
+    $id = keys(%main::uniqueid) + 1;
+    $main::uniqueid{$key} = $id;
+  }
+  return $id;
+}
+
+# Translate a stack of addresses into a stack of symbols
+sub TranslateStack {
+  my $symbols = shift;
+  my $fullname_to_shortname_map = shift;
+  my $k = shift;
+
+  my @addrs = split(/\n/, $k);
+  my @result = ();
+  for (my $i = 0; $i <= $#addrs; $i++) {
+    my $a = $addrs[$i];
+
+    # Skip large addresses since they sometimes show up as fake entries on RH9
+    if (length($a) > 8 && $a gt "7fffffffffffffff") {
+      next;
+    }
+
+    if ($main::opt_disasm || $main::opt_list) {
+      # We want just the address for the key
+      push(@result, $a);
+      next;
+    }
+
+    my $symlist = $symbols->{$a};
+    if (!defined($symlist)) {
+      $symlist = [$a, "", $a];
+    }
+
+    # We can have a sequence of symbols for a particular entry
+    # (more than one symbol in the case of inlining).  Callers
+    # come before callees in symlist, so walk backwards since
+    # the translated stack should contain callees before callers.
+    for (my $j = $#{$symlist}; $j >= 2; $j -= 3) {
+      my $func = $symlist->[$j-2];
+      my $fileline = $symlist->[$j-1];
+      my $fullfunc = $symlist->[$j];
+      if (defined($fullname_to_shortname_map->{$fullfunc})) {
+        $func = $fullname_to_shortname_map->{$fullfunc};
+      }
+      if ($j > 2) {
+        $func = "$func (inline)";
+      }
+
+      # Do not merge nodes corresponding to Callback::Run since that
+      # causes confusing cycles in dot display.  Instead, we synthesize
+      # a unique name for this frame per caller.
+      if ($func =~ m/Callback.*::Run$/) {
+        my $caller = ($i > 0) ? $addrs[$i-1] : 0;
+        $func = "Run#" . ShortIdFor($caller);
+      }
+
+      if ($main::opt_addresses) {
+        push(@result, "$a $func $fileline");
+      } elsif ($main::opt_lines) {
+        if ($func eq '??' && $fileline eq '??:0') {
+          push(@result, "$a");
+        } else {
+          push(@result, "$func $fileline");
+        }
+      } elsif ($main::opt_functions) {
+        if ($func eq '??') {
+          push(@result, "$a");
+        } else {
+          push(@result, $func);
+        }
+      } elsif ($main::opt_files) {
+        if ($fileline eq '??:0' || $fileline eq '') {
+          push(@result, "$a");
+        } else {
+          my $f = $fileline;
+          $f =~ s/:\d+$//;
+          push(@result, $f);
+        }
+      } else {
+        push(@result, $a);
+        last;  # Do not print inlined info
+      }
+    }
+  }
+
+  # print join(",", @addrs), " => ", join(",", @result), "\n";
+  return @result;
+}
+
+# Generate percent string for a number and a total
+sub Percent {
+  my $num = shift;
+  my $tot = shift;
+  if ($tot != 0) {
+    return sprintf("%.1f%%", $num * 100.0 / $tot);
+  } else {
+    return ($num == 0) ? "nan" : (($num > 0) ? "+inf" : "-inf");
+  }
+}
+
+# Generate pretty-printed form of number
+sub Unparse {
+  my $num = shift;
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    if ($main::opt_inuse_objects || $main::opt_alloc_objects) {
+      return sprintf("%d", $num);
+    } else {
+      if ($main::opt_show_bytes) {
+        return sprintf("%d", $num);
+      } else {
+        return sprintf("%.1f", $num / 1048576.0);
+      }
+    }
+  } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) {
+    return sprintf("%.3f", $num / 1e9); # Convert nanoseconds to seconds
+  } else {
+    return sprintf("%d", $num);
+  }
+}
+
+# Alternate pretty-printed form: 0 maps to "."
+sub UnparseAlt {
+  my $num = shift;
+  if ($num == 0) {
+    return ".";
+  } else {
+    return Unparse($num);
+  }
+}
+
+# Alternate pretty-printed form: 0 maps to ""
+sub HtmlPrintNumber {
+  my $num = shift;
+  if ($num == 0) {
+    return "";
+  } else {
+    return Unparse($num);
+  }
+}
+
+# Return output units
+sub Units {
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    if ($main::opt_inuse_objects || $main::opt_alloc_objects) {
+      return "objects";
+    } else {
+      if ($main::opt_show_bytes) {
+        return "B";
+      } else {
+        return "MB";
+      }
+    }
+  } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) {
+    return "seconds";
+  } else {
+    return "samples";
+  }
+}
+
+##### Profile manipulation code #####
+
+# Generate flattened profile:
+# If count is charged to stack [a,b,c,d], in generated profile,
+# it will be charged to [a]
+sub FlatProfile {
+  my $profile = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    if ($#addrs >= 0) {
+      AddEntry($result, $addrs[0], $count);
+    }
+  }
+  return $result;
+}
+
+# Generate cumulative profile:
+# If count is charged to stack [a,b,c,d], in generated profile,
+# it will be charged to [a], [b], [c], [d]
+sub CumulativeProfile {
+  my $profile = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    foreach my $a (@addrs) {
+      AddEntry($result, $a, $count);
+    }
+  }
+  return $result;
+}
+
+# If the second-youngest PC on the stack is always the same, returns
+# that pc.  Otherwise, returns undef.
+sub IsSecondPcAlwaysTheSame {
+  my $profile = shift;
+
+  my $second_pc = undef;
+  foreach my $k (keys(%{$profile})) {
+    my @addrs = split(/\n/, $k);
+    if ($#addrs < 1) {
+      return undef;
+    }
+    if (not defined $second_pc) {
+      $second_pc = $addrs[1];
+    } else {
+      if ($second_pc ne $addrs[1]) {
+        return undef;
+      }
+    }
+  }
+  return $second_pc;
+}
+
+sub ExtractSymbolNameInlineStack {
+  my $symbols = shift;
+  my $address = shift;
+
+  my @stack = ();
+
+  if (exists $symbols->{$address}) {
+    my @localinlinestack = @{$symbols->{$address}};
+    for (my $i = $#localinlinestack; $i > 0; $i-=3) {
+      my $file = $localinlinestack[$i-1];
+      my $fn = $localinlinestack[$i-0];
+
+      if ($file eq "?" || $file eq ":0") {
+        $file = "??:0";
+      }
+      if ($fn eq '??') {
+        # If we can't get the symbol name, at least use the file information.
+        $fn = $file;
+      }
+      my $suffix = "[inline]";
+      if ($i == 2) {
+        $suffix = "";
+      }
+      push (@stack, $fn.$suffix);
+    }
+  }
+  else {
+    # If we can't get a symbol name, at least fill in the address.
+    push (@stack, $address);
+  }
+
+  return @stack;
+}
+
+sub ExtractSymbolLocation {
+  my $symbols = shift;
+  my $address = shift;
+  # 'addr2line' outputs "??:0" for unknown locations; we do the
+  # same to be consistent.
+  my $location = "??:0:unknown";
+  if (exists $symbols->{$address}) {
+    my $file = $symbols->{$address}->[1];
+    if ($file eq "?") {
+      $file = "??:0"
+    }
+    $location = $file . ":" . $symbols->{$address}->[0];
+  }
+  return $location;
+}
+
+# Extracts a graph of calls.
+sub ExtractCalls {
+  my $symbols = shift;
+  my $profile = shift;
+
+  my $calls = {};
+  while( my ($stack_trace, $count) = each %$profile ) {
+    my @address = split(/\n/, $stack_trace);
+    my $destination = ExtractSymbolLocation($symbols, $address[0]);
+    AddEntry($calls, $destination, $count);
+    for (my $i = 1; $i <= $#address; $i++) {
+      my $source = ExtractSymbolLocation($symbols, $address[$i]);
+      my $call = "$source -> $destination";
+      AddEntry($calls, $call, $count);
+      $destination = $source;
+    }
+  }
+
+  return $calls;
+}
+
+sub FilterFrames {
+  my $symbols = shift;
+  my $profile = shift;
+
+  if ($main::opt_retain eq '' && $main::opt_exclude eq '') {
+    return $profile;
+  }
+
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my @path = ();
+    foreach my $a (@addrs) {
+      my $sym;
+      if (exists($symbols->{$a})) {
+        $sym = $symbols->{$a}->[0];
+      } else {
+        $sym = $a;
+      }
+      if ($main::opt_retain ne '' && $sym !~ m/$main::opt_retain/) {
+        next;
+      }
+      if ($main::opt_exclude ne '' && $sym =~ m/$main::opt_exclude/) {
+        next;
+      }
+      push(@path, $a);
+    }
+    if (scalar(@path) > 0) {
+      my $reduced_path = join("\n", @path);
+      AddEntry($result, $reduced_path, $count);
+    }
+  }
+
+  return $result;
+}
+
+sub PrintCollapsedStacks {
+  my $symbols = shift;
+  my $profile = shift;
+
+  while (my ($stack_trace, $count) = each %$profile) {
+    my @address = split(/\n/, $stack_trace);
+    my @names = reverse ( map { ExtractSymbolNameInlineStack($symbols, $_) } @address );
+    printf("%s %d\n", join(";", @names), $count);
+  }
+}
+
+sub RemoveUninterestingFrames {
+  my $symbols = shift;
+  my $profile = shift;
+
+  # List of function names to skip
+  my %skip = ();
+  my $skip_regexp = 'NOMATCH';
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    foreach my $name ('je_calloc',
+                      'cfree',
+                      'je_malloc',
+                      'newImpl',
+                      'void* newImpl',
+                      'je_free',
+                      'je_memalign',
+                      'je_posix_memalign',
+                      'je_aligned_alloc',
+                      'pvalloc',
+                      'je_valloc',
+                      'je_realloc',
+                      'je_mallocx',
+                      'je_rallocx',
+                      'je_xallocx',
+                      'je_dallocx',
+                      'je_sdallocx',
+                      'je_sdallocx_noflags',
+                      'tc_calloc',
+                      'tc_cfree',
+                      'tc_malloc',
+                      'tc_free',
+                      'tc_memalign',
+                      'tc_posix_memalign',
+                      'tc_pvalloc',
+                      'tc_valloc',
+                      'tc_realloc',
+                      'tc_new',
+                      'tc_delete',
+                      'tc_newarray',
+                      'tc_deletearray',
+                      'tc_new_nothrow',
+                      'tc_newarray_nothrow',
+                      'do_malloc',
+                      '::do_malloc',   # new name -- got moved to an unnamed ns
+                      '::do_malloc_or_cpp_alloc',
+                      'DoSampledAllocation',
+                      'simple_alloc::allocate',
+                      '__malloc_alloc_template::allocate',
+                      '__builtin_delete',
+                      '__builtin_new',
+                      '__builtin_vec_delete',
+                      '__builtin_vec_new',
+                      'operator new',
+                      'operator new[]',
+                      # The entry to our memory-allocation routines on OS X
+                      'malloc_zone_malloc',
+                      'malloc_zone_calloc',
+                      'malloc_zone_valloc',
+                      'malloc_zone_realloc',
+                      'malloc_zone_memalign',
+                      'malloc_zone_free',
+                      # These mark the beginning/end of our custom sections
+                      '__start_google_malloc',
+                      '__stop_google_malloc',
+                      '__start_malloc_hook',
+                      '__stop_malloc_hook') {
+      $skip{$name} = 1;
+      $skip{"_" . $name} = 1;   # Mach (OS X) adds a _ prefix to everything
+    }
+    # TODO: Remove TCMalloc once everything has been
+    # moved into the tcmalloc:: namespace and we have flushed
+    # old code out of the system.
+    $skip_regexp = "TCMalloc|^tcmalloc::";
+  } elsif ($main::profile_type eq 'contention') {
+    foreach my $vname ('base::RecordLockProfileData',
+                       'base::SubmitMutexProfileData',
+                       'base::SubmitSpinLockProfileData',
+                       'Mutex::Unlock',
+                       'Mutex::UnlockSlow',
+                       'Mutex::ReaderUnlock',
+                       'MutexLock::~MutexLock',
+                       'SpinLock::Unlock',
+                       'SpinLock::SlowUnlock',
+                       'SpinLockHolder::~SpinLockHolder') {
+      $skip{$vname} = 1;
+    }
+  } elsif ($main::profile_type eq 'cpu') {
+    # Drop signal handlers used for CPU profile collection
+    # TODO(dpeng): this should not be necessary; it's taken
+    # care of by the general 2nd-pc mechanism below.
+    foreach my $name ('ProfileData::Add',           # historical
+                      'ProfileData::prof_handler',  # historical
+                      'CpuProfiler::prof_handler',
+                      '__FRAME_END__',
+                      '__pthread_sighandler',
+                      '__restore') {
+      $skip{$name} = 1;
+    }
+  } else {
+    # Nothing skipped for unknown types
+  }
+
+  if ($main::profile_type eq 'cpu') {
+    # If all the second-youngest program counters are the same,
+    # this STRONGLY suggests that it is an artifact of measurement,
+    # i.e., stack frames pushed by the CPU profiler signal handler.
+    # Hence, we delete them.
+    # (The topmost PC is read from the signal structure, not from
+    # the stack, so it does not get involved.)
+    while (my $second_pc = IsSecondPcAlwaysTheSame($profile)) {
+      my $result = {};
+      my $func = '';
+      if (exists($symbols->{$second_pc})) {
+        $second_pc = $symbols->{$second_pc}->[0];
+      }
+      print STDERR "Removing $second_pc from all stack traces.\n";
+      foreach my $k (keys(%{$profile})) {
+        my $count = $profile->{$k};
+        my @addrs = split(/\n/, $k);
+        splice @addrs, 1, 1;
+        my $reduced_path = join("\n", @addrs);
+        AddEntry($result, $reduced_path, $count);
+      }
+      $profile = $result;
+    }
+  }
+
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my @path = ();
+    foreach my $a (@addrs) {
+      if (exists($symbols->{$a})) {
+        my $func = $symbols->{$a}->[0];
+        if ($skip{$func} || ($func =~ m/$skip_regexp/)) {
+          # Throw away the portion of the backtrace seen so far, under the
+          # assumption that previous frames were for functions internal to the
+          # allocator.
+          @path = ();
+          next;
+        }
+      }
+      push(@path, $a);
+    }
+    my $reduced_path = join("\n", @path);
+    AddEntry($result, $reduced_path, $count);
+  }
+
+  $result = FilterFrames($symbols, $result);
+
+  return $result;
+}
+
+# Reduce profile to granularity given by user
+sub ReduceProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $result = {};
+  my $fullname_to_shortname_map = {};
+  FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map);
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k);
+    my @path = ();
+    my %seen = ();
+    $seen{''} = 1;      # So that empty keys are skipped
+    foreach my $e (@translated) {
+      # To avoid double-counting due to recursion, skip a stack-trace
+      # entry if it has already been seen
+      if (!$seen{$e}) {
+        $seen{$e} = 1;
+        push(@path, $e);
+      }
+    }
+    my $reduced_path = join("\n", @path);
+    AddEntry($result, $reduced_path, $count);
+  }
+  return $result;
+}
+
+# Does the specified symbol array match the regexp?
+sub SymbolMatches {
+  my $sym = shift;
+  my $re = shift;
+  if (defined($sym)) {
+    for (my $i = 0; $i < $#{$sym}; $i += 3) {
+      if ($sym->[$i] =~ m/$re/ || $sym->[$i+1] =~ m/$re/) {
+        return 1;
+      }
+    }
+  }
+  return 0;
+}
+
+# Focus only on paths involving specified regexps
+sub FocusProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $focus = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    foreach my $a (@addrs) {
+      # Reply if it matches either the address/shortname/fileline
+      if (($a =~ m/$focus/) || SymbolMatches($symbols->{$a}, $focus)) {
+        AddEntry($result, $k, $count);
+        last;
+      }
+    }
+  }
+  return $result;
+}
+
+# Focus only on paths not involving specified regexps
+sub IgnoreProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $ignore = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my $matched = 0;
+    foreach my $a (@addrs) {
+      # Reply if it matches either the address/shortname/fileline
+      if (($a =~ m/$ignore/) || SymbolMatches($symbols->{$a}, $ignore)) {
+        $matched = 1;
+        last;
+      }
+    }
+    if (!$matched) {
+      AddEntry($result, $k, $count);
+    }
+  }
+  return $result;
+}
+
+# Get total count in profile
+sub TotalProfile {
+  my $profile = shift;
+  my $result = 0;
+  foreach my $k (keys(%{$profile})) {
+    $result += $profile->{$k};
+  }
+  return $result;
+}
+
+# Add A to B
+sub AddProfile {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  # add all keys in A
+  foreach my $k (keys(%{$A})) {
+    my $v = $A->{$k};
+    AddEntry($R, $k, $v);
+  }
+  # add all keys in B
+  foreach my $k (keys(%{$B})) {
+    my $v = $B->{$k};
+    AddEntry($R, $k, $v);
+  }
+  return $R;
+}
+
+# Merges symbol maps
+sub MergeSymbols {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  foreach my $k (keys(%{$A})) {
+    $R->{$k} = $A->{$k};
+  }
+  if (defined($B)) {
+    foreach my $k (keys(%{$B})) {
+      $R->{$k} = $B->{$k};
+    }
+  }
+  return $R;
+}
+
+
+# Add A to B
+sub AddPcs {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  # add all keys in A
+  foreach my $k (keys(%{$A})) {
+    $R->{$k} = 1
+  }
+  # add all keys in B
+  foreach my $k (keys(%{$B})) {
+    $R->{$k} = 1
+  }
+  return $R;
+}
+
+# Subtract B from A
+sub SubtractProfile {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  foreach my $k (keys(%{$A})) {
+    my $v = $A->{$k} - GetEntry($B, $k);
+    if ($v < 0 && $main::opt_drop_negative) {
+      $v = 0;
+    }
+    AddEntry($R, $k, $v);
+  }
+  if (!$main::opt_drop_negative) {
+    # Take care of when subtracted profile has more entries
+    foreach my $k (keys(%{$B})) {
+      if (!exists($A->{$k})) {
+        AddEntry($R, $k, 0 - $B->{$k});
+      }
+    }
+  }
+  return $R;
+}
+
+# Get entry from profile; zero if not present
+sub GetEntry {
+  my $profile = shift;
+  my $k = shift;
+  if (exists($profile->{$k})) {
+    return $profile->{$k};
+  } else {
+    return 0;
+  }
+}
+
+# Add entry to specified profile
+sub AddEntry {
+  my $profile = shift;
+  my $k = shift;
+  my $n = shift;
+  if (!exists($profile->{$k})) {
+    $profile->{$k} = 0;
+  }
+  $profile->{$k} += $n;
+}
+
+# Add a stack of entries to specified profile, and add them to the $pcs
+# list.
+sub AddEntries {
+  my $profile = shift;
+  my $pcs = shift;
+  my $stack = shift;
+  my $count = shift;
+  my @k = ();
+
+  foreach my $e (split(/\s+/, $stack)) {
+    my $pc = HexExtend($e);
+    $pcs->{$pc} = 1;
+    push @k, $pc;
+  }
+  AddEntry($profile, (join "\n", @k), $count);
+}
+
+##### Code to profile a server dynamically #####
+
+sub CheckSymbolPage {
+  my $url = SymbolPageURL();
+  my $command = ShellEscape(@URL_FETCHER, $url);
+  open(SYMBOL, "$command |") or error($command);
+  my $line = <SYMBOL>;
+  $line =~ s/\r//g;         # turn windows-looking lines into unix-looking lines
+  close(SYMBOL);
+  unless (defined($line)) {
+    error("$url doesn't exist\n");
+  }
+
+  if ($line =~ /^num_symbols:\s+(\d+)$/) {
+    if ($1 == 0) {
+      error("Stripped binary. No symbols available.\n");
+    }
+  } else {
+    error("Failed to get the number of symbols from $url\n");
+  }
+}
+
+sub IsProfileURL {
+  my $profile_name = shift;
+  if (-f $profile_name) {
+    printf STDERR "Using local file $profile_name.\n";
+    return 0;
+  }
+  return 1;
+}
+
+sub ParseProfileURL {
+  my $profile_name = shift;
+
+  if (!defined($profile_name) || $profile_name eq "") {
+    return ();
+  }
+
+  # Split profile URL - matches all non-empty strings, so no test.
+  $profile_name =~ m,^(https?://)?([^/]+)(.*?)(/|$PROFILES)?$,;
+
+  my $proto = $1 || "http://";
+  my $hostport = $2;
+  my $prefix = $3;
+  my $profile = $4 || "/";
+
+  my $host = $hostport;
+  $host =~ s/:.*//;
+
+  my $baseurl = "$proto$hostport$prefix";
+  return ($host, $baseurl, $profile);
+}
+
+# We fetch symbols from the first profile argument.
+sub SymbolPageURL {
+  my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]);
+  return "$baseURL$SYMBOL_PAGE";
+}
+
+sub FetchProgramName() {
+  my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]);
+  my $url = "$baseURL$PROGRAM_NAME_PAGE";
+  my $command_line = ShellEscape(@URL_FETCHER, $url);
+  open(CMDLINE, "$command_line |") or error($command_line);
+  my $cmdline = <CMDLINE>;
+  $cmdline =~ s/\r//g;   # turn windows-looking lines into unix-looking lines
+  close(CMDLINE);
+  error("Failed to get program name from $url\n") unless defined($cmdline);
+  $cmdline =~ s/\x00.+//;  # Remove argv[1] and latters.
+  $cmdline =~ s!\n!!g;  # Remove LFs.
+  return $cmdline;
+}
+
+# Gee, curl's -L (--location) option isn't reliable at least
+# with its 7.12.3 version.  Curl will forget to post data if
+# there is a redirection.  This function is a workaround for
+# curl.  Redirection happens on borg hosts.
+sub ResolveRedirectionForCurl {
+  my $url = shift;
+  my $command_line = ShellEscape(@URL_FETCHER, "--head", $url);
+  open(CMDLINE, "$command_line |") or error($command_line);
+  while (<CMDLINE>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    if (/^Location: (.*)/) {
+      $url = $1;
+    }
+  }
+  close(CMDLINE);
+  return $url;
+}
+
+# Add a timeout flat to URL_FETCHER.  Returns a new list.
+sub AddFetchTimeout {
+  my $timeout = shift;
+  my @fetcher = @_;
+  if (defined($timeout)) {
+    if (join(" ", @fetcher) =~ m/\bcurl -s/) {
+      push(@fetcher, "--max-time", sprintf("%d", $timeout));
+    } elsif (join(" ", @fetcher) =~ m/\brpcget\b/) {
+      push(@fetcher, sprintf("--deadline=%d", $timeout));
+    }
+  }
+  return @fetcher;
+}
+
+# Reads a symbol map from the file handle name given as $1, returning
+# the resulting symbol map.  Also processes variables relating to symbols.
+# Currently, the only variable processed is 'binary=<value>' which updates
+# $main::prog to have the correct program name.
+sub ReadSymbols {
+  my $in = shift;
+  my $map = {};
+  while (<$in>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    # Removes all the leading zeroes from the symbols, see comment below.
+    if (m/^0x0*([0-9a-f]+)\s+(.+)/) {
+      $map->{$1} = $2;
+    } elsif (m/^---/) {
+      last;
+    } elsif (m/^([a-z][^=]*)=(.*)$/ ) {
+      my ($variable, $value) = ($1, $2);
+      for ($variable, $value) {
+        s/^\s+//;
+        s/\s+$//;
+      }
+      if ($variable eq "binary") {
+        if ($main::prog ne $UNKNOWN_BINARY && $main::prog ne $value) {
+          printf STDERR ("Warning: Mismatched binary name '%s', using '%s'.\n",
+                         $main::prog, $value);
+        }
+        $main::prog = $value;
+      } else {
+        printf STDERR ("Ignoring unknown variable in symbols list: " .
+            "'%s' = '%s'\n", $variable, $value);
+      }
+    }
+  }
+  return $map;
+}
+
+sub URLEncode {
+  my $str = shift;
+  $str =~ s/([^A-Za-z0-9\-_.!~*'()])/ sprintf "%%%02x", ord $1 /eg;
+  return $str;
+}
+
+sub AppendSymbolFilterParams {
+  my $url = shift;
+  my @params = ();
+  if ($main::opt_retain ne '') {
+    push(@params, sprintf("retain=%s", URLEncode($main::opt_retain)));
+  }
+  if ($main::opt_exclude ne '') {
+    push(@params, sprintf("exclude=%s", URLEncode($main::opt_exclude)));
+  }
+  if (scalar @params > 0) {
+    $url = sprintf("%s?%s", $url, join("&", @params));
+  }
+  return $url;
+}
+
+# Fetches and processes symbols to prepare them for use in the profile output
+# code.  If the optional 'symbol_map' arg is not given, fetches symbols from
+# $SYMBOL_PAGE for all PC values found in profile.  Otherwise, the raw symbols
+# are assumed to have already been fetched into 'symbol_map' and are simply
+# extracted and processed.
+sub FetchSymbols {
+  my $pcset = shift;
+  my $symbol_map = shift;
+
+  my %seen = ();
+  my @pcs = grep { !$seen{$_}++ } keys(%$pcset);  # uniq
+
+  if (!defined($symbol_map)) {
+    my $post_data = join("+", sort((map {"0x" . "$_"} @pcs)));
+
+    open(POSTFILE, ">$main::tmpfile_sym");
+    print POSTFILE $post_data;
+    close(POSTFILE);
+
+    my $url = SymbolPageURL();
+
+    my $command_line;
+    if (join(" ", @URL_FETCHER) =~ m/\bcurl -s/) {
+      $url = ResolveRedirectionForCurl($url);
+      $url = AppendSymbolFilterParams($url);
+      $command_line = ShellEscape(@URL_FETCHER, "-d", "\@$main::tmpfile_sym",
+                                  $url);
+    } else {
+      $url = AppendSymbolFilterParams($url);
+      $command_line = (ShellEscape(@URL_FETCHER, "--post", $url)
+                       . " < " . ShellEscape($main::tmpfile_sym));
+    }
+    # We use c++filt in case $SYMBOL_PAGE gives us mangled symbols.
+    my $escaped_cppfilt = ShellEscape($obj_tool_map{"c++filt"});
+    open(SYMBOL, "$command_line | $escaped_cppfilt |") or error($command_line);
+    $symbol_map = ReadSymbols(*SYMBOL{IO});
+    close(SYMBOL);
+  }
+
+  my $symbols = {};
+  foreach my $pc (@pcs) {
+    my $fullname;
+    # For 64 bits binaries, symbols are extracted with 8 leading zeroes.
+    # Then /symbol reads the long symbols in as uint64, and outputs
+    # the result with a "0x%08llx" format which get rid of the zeroes.
+    # By removing all the leading zeroes in both $pc and the symbols from
+    # /symbol, the symbols match and are retrievable from the map.
+    my $shortpc = $pc;
+    $shortpc =~ s/^0*//;
+    # Each line may have a list of names, which includes the function
+    # and also other functions it has inlined.  They are separated (in
+    # PrintSymbolizedProfile), by --, which is illegal in function names.
+    my $fullnames;
+    if (defined($symbol_map->{$shortpc})) {
+      $fullnames = $symbol_map->{$shortpc};
+    } else {
+      $fullnames = "0x" . $pc;  # Just use addresses
+    }
+    my $sym = [];
+    $symbols->{$pc} = $sym;
+    foreach my $fullname (split("--", $fullnames)) {
+      my $name = ShortFunctionName($fullname);
+      push(@{$sym}, $name, "?", $fullname);
+    }
+  }
+  return $symbols;
+}
+
+sub BaseName {
+  my $file_name = shift;
+  $file_name =~ s!^.*/!!;  # Remove directory name
+  return $file_name;
+}
+
+sub MakeProfileBaseName {
+  my ($binary_name, $profile_name) = @_;
+  my ($host, $baseURL, $path) = ParseProfileURL($profile_name);
+  my $binary_shortname = BaseName($binary_name);
+  return sprintf("%s.%s.%s",
+                 $binary_shortname, $main::op_time, $host);
+}
+
+sub FetchDynamicProfile {
+  my $binary_name = shift;
+  my $profile_name = shift;
+  my $fetch_name_only = shift;
+  my $encourage_patience = shift;
+
+  if (!IsProfileURL($profile_name)) {
+    return $profile_name;
+  } else {
+    my ($host, $baseURL, $path) = ParseProfileURL($profile_name);
+    if ($path eq "" || $path eq "/") {
+      # Missing type specifier defaults to cpu-profile
+      $path = $PROFILE_PAGE;
+    }
+
+    my $profile_file = MakeProfileBaseName($binary_name, $profile_name);
+
+    my $url = "$baseURL$path";
+    my $fetch_timeout = undef;
+    if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE/) {
+      if ($path =~ m/[?]/) {
+        $url .= "&";
+      } else {
+        $url .= "?";
+      }
+      $url .= sprintf("seconds=%d", $main::opt_seconds);
+      $fetch_timeout = $main::opt_seconds * 1.01 + 60;
+      # Set $profile_type for consumption by PrintSymbolizedProfile.
+      $main::profile_type = 'cpu';
+    } else {
+      # For non-CPU profiles, we add a type-extension to
+      # the target profile file name.
+      my $suffix = $path;
+      $suffix =~ s,/,.,g;
+      $profile_file .= $suffix;
+      # Set $profile_type for consumption by PrintSymbolizedProfile.
+      if ($path =~ m/$HEAP_PAGE/) {
+        $main::profile_type = 'heap';
+      } elsif ($path =~ m/$GROWTH_PAGE/) {
+        $main::profile_type = 'growth';
+      } elsif ($path =~ m/$CONTENTION_PAGE/) {
+        $main::profile_type = 'contention';
+      }
+    }
+
+    my $profile_dir = $ENV{"JEPROF_TMPDIR"} || ($ENV{HOME} . "/jeprof");
+    if (! -d $profile_dir) {
+      mkdir($profile_dir)
+          || die("Unable to create profile directory $profile_dir: $!\n");
+    }
+    my $tmp_profile = "$profile_dir/.tmp.$profile_file";
+    my $real_profile = "$profile_dir/$profile_file";
+
+    if ($fetch_name_only > 0) {
+      return $real_profile;
+    }
+
+    my @fetcher = AddFetchTimeout($fetch_timeout, @URL_FETCHER);
+    my $cmd = ShellEscape(@fetcher, $url) . " > " . ShellEscape($tmp_profile);
+    if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE|$CENSUSPROFILE_PAGE/){
+      print STDERR "Gathering CPU profile from $url for $main::opt_seconds seconds to\n  ${real_profile}\n";
+      if ($encourage_patience) {
+        print STDERR "Be patient...\n";
+      }
+    } else {
+      print STDERR "Fetching $path profile from $url to\n  ${real_profile}\n";
+    }
+
+    (system($cmd) == 0) || error("Failed to get profile: $cmd: $!\n");
+    (system("mv", $tmp_profile, $real_profile) == 0) || error("Unable to rename profile\n");
+    print STDERR "Wrote profile to $real_profile\n";
+    $main::collected_profile = $real_profile;
+    return $main::collected_profile;
+  }
+}
+
+# Collect profiles in parallel
+sub FetchDynamicProfiles {
+  my $items = scalar(@main::pfile_args);
+  my $levels = log($items) / log(2);
+
+  if ($items == 1) {
+    $main::profile_files[0] = FetchDynamicProfile($main::prog, $main::pfile_args[0], 0, 1);
+  } else {
+    # math rounding issues
+    if ((2 ** $levels) < $items) {
+     $levels++;
+    }
+    my $count = scalar(@main::pfile_args);
+    for (my $i = 0; $i < $count; $i++) {
+      $main::profile_files[$i] = FetchDynamicProfile($main::prog, $main::pfile_args[$i], 1, 0);
+    }
+    print STDERR "Fetching $count profiles, Be patient...\n";
+    FetchDynamicProfilesRecurse($levels, 0, 0);
+    $main::collected_profile = join(" \\\n    ", @main::profile_files);
+  }
+}
+
+# Recursively fork a process to get enough processes
+# collecting profiles
+sub FetchDynamicProfilesRecurse {
+  my $maxlevel = shift;
+  my $level = shift;
+  my $position = shift;
+
+  if (my $pid = fork()) {
+    $position = 0 | ($position << 1);
+    TryCollectProfile($maxlevel, $level, $position);
+    wait;
+  } else {
+    $position = 1 | ($position << 1);
+    TryCollectProfile($maxlevel, $level, $position);
+    cleanup();
+    exit(0);
+  }
+}
+
+# Collect a single profile
+sub TryCollectProfile {
+  my $maxlevel = shift;
+  my $level = shift;
+  my $position = shift;
+
+  if ($level >= ($maxlevel - 1)) {
+    if ($position < scalar(@main::pfile_args)) {
+      FetchDynamicProfile($main::prog, $main::pfile_args[$position], 0, 0);
+    }
+  } else {
+    FetchDynamicProfilesRecurse($maxlevel, $level+1, $position);
+  }
+}
+
+##### Parsing code #####
+
+# Provide a small streaming-read module to handle very large
+# cpu-profile files.  Stream in chunks along a sliding window.
+# Provides an interface to get one 'slot', correctly handling
+# endian-ness differences.  A slot is one 32-bit or 64-bit word
+# (depending on the input profile).  We tell endianness and bit-size
+# for the profile by looking at the first 8 bytes: in cpu profiles,
+# the second slot is always 3 (we'll accept anything that's not 0).
+BEGIN {
+  package CpuProfileStream;
+
+  sub new {
+    my ($class, $file, $fname) = @_;
+    my $self = { file        => $file,
+                 base        => 0,
+                 stride      => 512 * 1024,   # must be a multiple of bitsize/8
+                 slots       => [],
+                 unpack_code => "",           # N for big-endian, V for little
+                 perl_is_64bit => 1,          # matters if profile is 64-bit
+    };
+    bless $self, $class;
+    # Let unittests adjust the stride
+    if ($main::opt_test_stride > 0) {
+      $self->{stride} = $main::opt_test_stride;
+    }
+    # Read the first two slots to figure out bitsize and endianness.
+    my $slots = $self->{slots};
+    my $str;
+    read($self->{file}, $str, 8);
+    # Set the global $address_length based on what we see here.
+    # 8 is 32-bit (8 hexadecimal chars); 16 is 64-bit (16 hexadecimal chars).
+    $address_length = ($str eq (chr(0)x8)) ? 16 : 8;
+    if ($address_length == 8) {
+      if (substr($str, 6, 2) eq chr(0)x2) {
+        $self->{unpack_code} = 'V';  # Little-endian.
+      } elsif (substr($str, 4, 2) eq chr(0)x2) {
+        $self->{unpack_code} = 'N';  # Big-endian
+      } else {
+        ::error("$fname: header size >= 2**16\n");
+      }
+      @$slots = unpack($self->{unpack_code} . "*", $str);
+    } else {
+      # If we're a 64-bit profile, check if we're a 64-bit-capable
+      # perl.  Otherwise, each slot will be represented as a float
+      # instead of an int64, losing precision and making all the
+      # 64-bit addresses wrong.  We won't complain yet, but will
+      # later if we ever see a value that doesn't fit in 32 bits.
+      my $has_q = 0;
+      eval { $has_q = pack("Q", "1") ? 1 : 1; };
+      if (!$has_q) {
+        $self->{perl_is_64bit} = 0;
+      }
+      read($self->{file}, $str, 8);
+      if (substr($str, 4, 4) eq chr(0)x4) {
+        # We'd love to use 'Q', but it's a) not universal, b) not endian-proof.
+        $self->{unpack_code} = 'V';  # Little-endian.
+      } elsif (substr($str, 0, 4) eq chr(0)x4) {
+        $self->{unpack_code} = 'N';  # Big-endian
+      } else {
+        ::error("$fname: header size >= 2**32\n");
+      }
+      my @pair = unpack($self->{unpack_code} . "*", $str);
+      # Since we know one of the pair is 0, it's fine to just add them.
+      @$slots = (0, $pair[0] + $pair[1]);
+    }
+    return $self;
+  }
+
+  # Load more data when we access slots->get(X) which is not yet in memory.
+  sub overflow {
+    my ($self) = @_;
+    my $slots = $self->{slots};
+    $self->{base} += $#$slots + 1;   # skip over data we're replacing
+    my $str;
+    read($self->{file}, $str, $self->{stride});
+    if ($address_length == 8) {      # the 32-bit case
+      # This is the easy case: unpack provides 32-bit unpacking primitives.
+      @$slots = unpack($self->{unpack_code} . "*", $str);
+    } else {
+      # We need to unpack 32 bits at a time and combine.
+      my @b32_values = unpack($self->{unpack_code} . "*", $str);
+      my @b64_values = ();
+      for (my $i = 0; $i < $#b32_values; $i += 2) {
+        # TODO(csilvers): if this is a 32-bit perl, the math below
+        #    could end up in a too-large int, which perl will promote
+        #    to a double, losing necessary precision.  Deal with that.
+        #    Right now, we just die.
+        my ($lo, $hi) = ($b32_values[$i], $b32_values[$i+1]);
+        if ($self->{unpack_code} eq 'N') {    # big-endian
+          ($lo, $hi) = ($hi, $lo);
+        }
+        my $value = $lo + $hi * (2**32);
+        if (!$self->{perl_is_64bit} &&   # check value is exactly represented
+            (($value % (2**32)) != $lo || int($value / (2**32)) != $hi)) {
+          ::error("Need a 64-bit perl to process this 64-bit profile.\n");
+        }
+        push(@b64_values, $value);
+      }
+      @$slots = @b64_values;
+    }
+  }
+
+  # Access the i-th long in the file (logically), or -1 at EOF.
+  sub get {
+    my ($self, $idx) = @_;
+    my $slots = $self->{slots};
+    while ($#$slots >= 0) {
+      if ($idx < $self->{base}) {
+        # The only time we expect a reference to $slots[$i - something]
+        # after referencing $slots[$i] is reading the very first header.
+        # Since $stride > |header|, that shouldn't cause any lookback
+        # errors.  And everything after the header is sequential.
+        print STDERR "Unexpected look-back reading CPU profile";
+        return -1;   # shrug, don't know what better to return
+      } elsif ($idx > $self->{base} + $#$slots) {
+        $self->overflow();
+      } else {
+        return $slots->[$idx - $self->{base}];
+      }
+    }
+    # If we get here, $slots is [], which means we've reached EOF
+    return -1;  # unique since slots is supposed to hold unsigned numbers
+  }
+}
+
+# Reads the top, 'header' section of a profile, and returns the last
+# line of the header, commonly called a 'header line'.  The header
+# section of a profile consists of zero or more 'command' lines that
+# are instructions to jeprof, which jeprof executes when reading the
+# header.  All 'command' lines start with a %.  After the command
+# lines is the 'header line', which is a profile-specific line that
+# indicates what type of profile it is, and perhaps other global
+# information about the profile.  For instance, here's a header line
+# for a heap profile:
+#   heap profile:     53:    38236 [  5525:  1284029] @ heapprofile
+# For historical reasons, the CPU profile does not contain a text-
+# readable header line.  If the profile looks like a CPU profile,
+# this function returns "".  If no header line could be found, this
+# function returns undef.
+#
+# The following commands are recognized:
+#   %warn -- emit the rest of this line to stderr, prefixed by 'WARNING:'
+#
+# The input file should be in binmode.
+sub ReadProfileHeader {
+  local *PROFILE = shift;
+  my $firstchar = "";
+  my $line = "";
+  read(PROFILE, $firstchar, 1);
+  seek(PROFILE, -1, 1);                    # unread the firstchar
+  if ($firstchar !~ /[[:print:]]/) {       # is not a text character
+    return "";
+  }
+  while (defined($line = <PROFILE>)) {
+    $line =~ s/\r//g;   # turn windows-looking lines into unix-looking lines
+    if ($line =~ /^%warn\s+(.*)/) {        # 'warn' command
+      # Note this matches both '%warn blah\n' and '%warn\n'.
+      print STDERR "WARNING: $1\n";        # print the rest of the line
+    } elsif ($line =~ /^%/) {
+      print STDERR "Ignoring unknown command from profile header: $line";
+    } else {
+      # End of commands, must be the header line.
+      return $line;
+    }
+  }
+  return undef;     # got to EOF without seeing a header line
+}
+
+sub IsSymbolizedProfileFile {
+  my $file_name = shift;
+  if (!(-e $file_name) || !(-r $file_name)) {
+    return 0;
+  }
+  # Check if the file contains a symbol-section marker.
+  open(TFILE, "<$file_name");
+  binmode TFILE;
+  my $firstline = ReadProfileHeader(*TFILE);
+  close(TFILE);
+  if (!$firstline) {
+    return 0;
+  }
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+  return $firstline =~ /^--- *$symbol_marker/;
+}
+
+# Parse profile generated by common/profiler.cc and return a reference
+# to a map:
+#      $result->{version}     Version number of profile file
+#      $result->{period}      Sampling period (in microseconds)
+#      $result->{profile}     Profile object
+#      $result->{threads}     Map of thread IDs to profile objects
+#      $result->{map}         Memory map info from profile
+#      $result->{pcs}         Hash of all PC values seen, key is hex address
+sub ReadProfile {
+  my $prog = shift;
+  my $fname = shift;
+  my $result;            # return value
+
+  $CONTENTION_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $contention_marker = $&;
+  $GROWTH_PAGE  =~ m,[^/]+$,;    # matches everything after the last slash
+  my $growth_marker = $&;
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+  $PROFILE_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $profile_marker = $&;
+  $HEAP_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $heap_marker = $&;
+
+  # Look at first line to see if it is a heap or a CPU profile.
+  # CPU profile may start with no header at all, and just binary data
+  # (starting with \0\0\0\0) -- in that case, don't try to read the
+  # whole firstline, since it may be gigabytes(!) of data.
+  open(PROFILE, "<$fname") || error("$fname: $!\n");
+  binmode PROFILE;      # New perls do UTF-8 processing
+  my $header = ReadProfileHeader(*PROFILE);
+  if (!defined($header)) {   # means "at EOF"
+    error("Profile is empty.\n");
+  }
+
+  my $symbols;
+  if ($header =~ m/^--- *$symbol_marker/o) {
+    # Verify that the user asked for a symbolized profile
+    if (!$main::use_symbolized_profile) {
+      # we have both a binary and symbolized profiles, abort
+      error("FATAL ERROR: Symbolized profile\n   $fname\ncannot be used with " .
+            "a binary arg. Try again without passing\n   $prog\n");
+    }
+    # Read the symbol section of the symbolized profile file.
+    $symbols = ReadSymbols(*PROFILE{IO});
+    # Read the next line to get the header for the remaining profile.
+    $header = ReadProfileHeader(*PROFILE) || "";
+  }
+
+  if ($header =~ m/^--- *($heap_marker|$growth_marker)/o) {
+    # Skip "--- ..." line for profile types that have their own headers.
+    $header = ReadProfileHeader(*PROFILE) || "";
+  }
+
+  $main::profile_type = '';
+
+  if ($header =~ m/^heap profile:.*$growth_marker/o) {
+    $main::profile_type = 'growth';
+    $result =  ReadHeapProfile($prog, *PROFILE, $header);
+  } elsif ($header =~ m/^heap profile:/) {
+    $main::profile_type = 'heap';
+    $result =  ReadHeapProfile($prog, *PROFILE, $header);
+  } elsif ($header =~ m/^heap/) {
+    $main::profile_type = 'heap';
+    $result = ReadThreadedHeapProfile($prog, $fname, $header);
+  } elsif ($header =~ m/^--- *$contention_marker/o) {
+    $main::profile_type = 'contention';
+    $result = ReadSynchProfile($prog, *PROFILE);
+  } elsif ($header =~ m/^--- *Stacks:/) {
+    print STDERR
+      "Old format contention profile: mistakenly reports " .
+      "condition variable signals as lock contentions.\n";
+    $main::profile_type = 'contention';
+    $result = ReadSynchProfile($prog, *PROFILE);
+  } elsif ($header =~ m/^--- *$profile_marker/) {
+    # the binary cpu profile data starts immediately after this line
+    $main::profile_type = 'cpu';
+    $result = ReadCPUProfile($prog, $fname, *PROFILE);
+  } else {
+    if (defined($symbols)) {
+      # a symbolized profile contains a format we don't recognize, bail out
+      error("$fname: Cannot recognize profile section after symbols.\n");
+    }
+    # no ascii header present -- must be a CPU profile
+    $main::profile_type

[truncated 1061424 bytes]

ready patch attempt

patch

redis-check-rdb likely remains stuck in a busy-poll loop. A diagnosis report and patch proposal were created locally.

state: readycreated: 2026-05-10 13:20 UTCvalidation: ready
Published session

Prompt

## Plan Pass

You are planning a fixer patch before any edits happen.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. The original pre-edit snapshot is available at `./source` if you need to inspect it. For interpreter processes, plan from the script/application entrypoint evidence first and include the runtime only as a second investigation target unless the evidence proves a runtime bug.

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround. Inspect the relevant code, nearby callers, project contribution docs, and local helper/compat APIs, but do not edit files in this pass.

Return a short markdown plan with these exact sections:

## Problem
## Evidence Confidence
## Proposed Subject
## Patch Plan
## Risks
## Validation

Classify `## Evidence Confidence` as exactly one of `reproduced`, `observed`, or `inferred`. Use `inferred` only for a no-patch diagnosis/report plan unless you can name the extra evidence you will collect before editing; inferred source patches are blocked by Fixer because they are not pull-request-ready. For `observed` source-patch plans, plan to say in the final `## Issue Connection` that the failure was observed by Fixer and not independently reproduced. The plan must explain how the proposed code change addresses the observed issue evidence, call out any prior Fixer patch that should be improved or replaced, reject awkward control flow such as avoidable `goto` if there is a cleaner bounded alternative, name any local helper APIs or maintainer conventions the patch should follow, and keep the intended maintainer-facing explanation clear enough that someone unfamiliar with the local complaint wording can still follow the fix. In `## Validation`, name the reproducible configure/build/test entrypoint you will try from the workspace root before any focused leaf compile or smoke check, and include one bounded independent reproduction attempt for the collected failure signal when it is safe and cheap. Do not plan to claim `reproduced` unless that reproduction command or test can actually show the failure.

## Patch Pass

You are working on a bounded fixer proposal.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. Produce the smallest reasonable patch for the target repository, keep the change upstreamable, prefer the clearest control flow available, and do not keep avoidable `goto` when a simpler structure would read better. Before introducing new file, process, allocation, locking, networking, or platform APIs, inspect nearby code and project contribution docs for existing helpers or compatibility wrappers and use those local patterns unless you can explain why they do not fit. Validate from a reproducible workspace-root entrypoint before falling back to focused leaf commands; if a build or test cannot run, report the exact command, the exact blocker, and any narrower check you ran instead. During validation, also try one bounded independent reproduction of the collected failure signal when it is safe and cheap, such as a failing test, smoke command, perf/strace comparison, or before/after runtime check. Only use `reproduced` if that command or test actually reproduced the failure; otherwise keep `observed` and report the reproduction blocker. The final explanation must connect the observed issue evidence to the actual code change, not just paraphrase the diff. Write like a maintainer is going to read the patch mail cold: explain the bug in plain language, define subsystem-specific jargon the first time you need it, and make the causal story obvious. Explicitly classify evidence confidence as `reproduced`, `observed`, or `inferred`: `reproduced` means you reproduced the failure locally; `observed` means Fixer has direct crash/log/trace evidence but you did not independently reproduce it; `inferred` means the source patch is not pull-request-ready, so do not leave a source diff unless you first gather stronger observed/reproduced evidence; otherwise return a no-patch diagnosis/report. For any source-changing `observed` patch, say explicitly in `## Issue Connection` that the failure was observed by Fixer and not independently reproduced. If you introduce non-obvious state translation, index remapping, or backend split logic, add a short source comment that explains the invariant being preserved.

Start by explaining the likely root cause from the collected perf, strace, and /proc evidence. If you cannot land a safe patch, leave a diagnosis that is strong enough for an upstream bug report.

A previous Fixer patch attempt already exists for this issue. It was generated by Fixer `0.145.0`. Review that patch before changing code, improve it instead of starting blind, and clean up anything awkward or underexplained. In particular, remove avoidable `goto`, tighten the explanation of what the patch is doing, and make the resulting diff feel ready for upstream git review.
- Prior patch: `./prior-best.patch`
- Prior published session: `./prior-best-session.md`

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround. 

Keep the change narrowly scoped and summarize validation clearly.

In every authoring pass, your final response must start with `Subject: <single-line git commit subject>` and then include these markdown sections exactly:

## Commit Message
A short upstream-friendly explanation of what changed and why. Write it in plain language that a maintainer can follow without local complaint context. If you use subsystem jargon, define it immediately.

## Evidence Confidence
Exactly one word: `reproduced`, `observed`, or `inferred`. Use `reproduced` only when you reproduced the failure locally with a command or test, and include that command/test in `## Validation`. Use `observed` when Fixer has direct crash/log/trace evidence but you did not independently reproduce it. If `## Git Add Paths` lists source files for an `observed` patch, `## Issue Connection` must explicitly say the failure was observed by Fixer and not independently reproduced. Use `inferred` for profiler/strace/indirect evidence; inferred responses may be no-patch diagnoses or reports, but inferred source patches are not pull-request-ready until stronger evidence is gathered.

## Issue Connection
Write this as maintainer-facing patch mail, not as local Fixer notes. Cover four things explicitly in readable sentences: the user-visible symptom or the exact collected signal, the code-level cause or the cautious inference from evidence, the specific change you made, and the expected effect. Do not invent a reproducer, command line, crash, or user-visible failure that is not present in the evidence bundle. If the evidence is direct-but-not-reproduced, say it was observed by Fixer and not independently reproduced. If the evidence is indirect and you did not gather stronger evidence, do not leave a source diff; write a no-patch diagnosis/report instead. Include an explicit effect sentence such as `The expected effect is ...`, `This should reduce ...`, or `This prevents ...` for source patches. If the logic is non-obvious in code, mention that you added a short explanatory comment.

## Git Add Paths
List the repo-relative paths that belong in the final patch, one per line. Use `None` only when you intentionally made no source changes. Include intentionally new files, and do not list generated build artifacts.

## Validation
List the checks you ran, or say clearly that you could not run them. Include the independent reproduction command/test and result when `## Evidence Confidence` is `reproduced`; if reproduction was attempted but blocked, name the exact blocker and keep confidence at `observed` or `inferred`.

Before editing, read the plan at `./plan-output.txt` and follow it unless the code proves part of it wrong. If you change course, say so explicitly in the final write-up instead of silently drifting from the plan.

## Review Pass 1

You are reviewing a freshly generated fixer patch.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. Review the first patch pass. The original pre-edit snapshot is available at `./source` for diffing.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround.

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout. The latest author response is at `./patch-output.txt`. Inspect the current code and changed paths like a strict code reviewer. Focus on correctness, regressions, maintainability, awkward control flow such as avoidable `goto`, missing validation, weak or non-gittable commit message text, and explanations that fail to connect the observed issue evidence to the code change. Also review the maintainer experience: the patch mail should be easy to accept upstream, the user-visible bug should be explained in plain language, subsystem-specific jargon should be defined when first used, and any non-obvious state translation or index remapping in code should have a short explanatory comment. Reject patches that omit `## Evidence Confidence`, use anything other than `reproduced`, `observed`, or `inferred`, or sound more certain than that evidence level allows. Reject patches that claim `reproduced` unless `## Validation` names the concrete reproduction command or test and its result. Reject patches that invent a reproducer, command line, crash, or user-visible failure not present in the evidence bundle; indirect profiler or strace evidence may justify a cautious mitigation, but the author must say it is indirect instead of presenting it as a confirmed upstream bug. Reject patches that introduce generic libc/std APIs when nearby code or project docs provide a local compat/helper API for the same job, unless the author explicitly justifies the exception. Reject validation that only reports a leaf object/syntax build when a project-level configure/build/test entrypoint exists and was not attempted; if a full build cannot run, the author must show the exact failed command and blocker.

Do not apply code changes in this pass.

Return a short markdown review report. The first non-empty line must be exactly one of:

RESULT: ok
RESULT: fix-needed

If you choose `RESULT: fix-needed`, add a `## Findings` section with concrete, actionable items.

Response

## Plan Pass

## Problem
The bundle labels the process as `redis-check-rdb`, but the embedded command line, status, and thread names show `redis-server` sleeping in `epoll_wait`. The strace pattern is `epoll_wait(..., ~100ms)` followed by `/proc/self/stat` reads, with reported CPU around `0.3%`, which matches normal Redis idle cron/RSS sampling rather than a demonstrated busy loop.

## Evidence Confidence
inferred

## Proposed Subject
No source patch: diagnose idle Redis cron sampling evidence mismatch

## Patch Plan
Do not edit source in this pass. The evidence is profiler/trace-only and internally mismatched, so a Redis upstream source patch would be speculative.

Reject the prior Fixer patch: it adds generated `deps/jemalloc/bin/*` files and does not address the observed Redis event-loop/RSS-sampling trace.

If stronger evidence is needed before editing, collect a bounded local trace of an idle built `src/redis-server` and compare CPU plus syscall cadence. Only consider a patch if that shows abnormal CPU or a tight loop beyond Redis’ normal `serverCron()` timing.

Conventions/helpers found: `aeMain()`/`aeProcessEvents()` drive the event loop, `serverCron()` and `run_with_period()` schedule periodic work, and `zmalloc_get_rss()`/`get_proc_stat_ll()` are the local `/proc/self/stat` RSS helpers. Contribution docs checked: `CONTRIBUTING.md`, `README.md`, and `tests/README.md`.

## Risks
Changing cron or memory-stat sampling without stronger evidence could degrade `INFO memory`, allocator diagnostics, or normal maintenance behavior. Changing `redis-check-rdb` would not address the sampled process shown in the evidence.

## Validation
Before any focused check, try from the workspace root: `make`, then `make check` if available. If either fails due dependencies, generated files, or environment limits, record the exact command and failure.

Bounded independent evidence attempt: after `make`, run a short local idle `src/redis-server` smoke/trace, such as a `timeout`-bounded `strace` of `epoll_wait`, `openat`, `read`, and `close`, and check whether it merely reproduces the normal 100 ms cron `/proc/self/stat` cadence at low CPU. Do not claim `reproduced` unless the command demonstrates the reported runaway behavior.

## Patch Pass

Subject: diagnose redis idle cron sampling trace

## Commit Message
No source change is proposed. The collected perf, strace, and `/proc` evidence points at an idle `redis-server` event loop doing normal cron memory sampling, not at a demonstrated `redis-check-rdb` busy-loop bug.

## Evidence Confidence
inferred

## Issue Connection
The likely root cause from the collected evidence is normal Redis server cron activity. The report is labeled `redis-check-rdb`, but the command line is `/usr/bin/redis-server 127.0.0.1:6379`, the sampled process is sleeping in `do_epoll_wait`, CPU is about `0.3%`, and strace shows `epoll_wait` timing out around every 100 ms followed by reads of `/proc/self/stat`.

In Redis, `aeMain()` and `aeProcessEvents()` drive the event loop, `serverCron()` is the periodic maintenance callback, `run_with_period(100)` schedules 100 ms work, and `zmalloc_get_rss()` uses `get_proc_stat_ll()` to read `/proc/self/stat`. That matches the trace, so the evidence is indirect and internally mismatched rather than a confirmed upstream defect.

I made no source change and rejected the prior Fixer patch because it adds generated `deps/jemalloc/bin/*` files and does not address the observed Redis event-loop/RSS-sampling path. Contribution/style docs checked: `CONTRIBUTING.md`, `README.md`, and `tests/README.md`.

The expected effect is that no speculative upstream diff is produced until stronger evidence shows abnormal CPU use or a real tight loop distinct from normal idle cron sampling.

## Git Add Paths
None

## Validation
`make` from the workspace root completed successfully.

`make check` from the workspace root failed before running tests because the Redis test harness could not find a free port: `Can't find a non busy port in the 21079-21110 range.`

Bounded independent reproduction was attempted but blocked. `timeout 5s strace -f -ttT -e trace=epoll_wait,openat,read,close ./src/redis-server --port 0 --save '' --appendonly no --daemonize no` failed with `PTRACE_TRACEME: Operation not permitted` and `ptrace(PTRACE_SEIZE, 5): Operation not permitted`.

A local server smoke run could not stay running for comparison: `--port 0` exits because Redis is configured to listen nowhere, Unix socket startup fails with `setsockopt SO_REUSEADDR: Operation not permitted`, and TCP bind fails with `unable to bind socket, errno: 1`. Confidence remains `inferred`; no source diff was left.

## Review Pass 1

RESULT: ok

No blocking findings.

The workspace has no source diff, and `## Git Add Paths` correctly lists `None`. The author kept `## Evidence Confidence` as `inferred`, which matches the profiler/strace-only evidence and the internal mismatch between the `redis-check-rdb` artifact label and the sampled `redis-server` command line.

The response rejects the prior generated jemalloc-bin patch for the right reason: it does not address the observed Redis event-loop/RSS-sampling path. The maintainer-facing explanation is cautious, names the relevant Redis helpers/conventions (`aeMain`, `aeProcessEvents`, `serverCron`, `run_with_period`, `zmalloc_get_rss`, `get_proc_stat_ll`), and does not invent a confirmed bug or reproducer.

Validation is acceptable: project-level `make` was attempted and passed; `make check` was attempted and reports the exact port-range blocker; independent reproduction attempts and sandbox blockers are explicitly named.

Diff

--- a/deps/jemalloc/bin/jemalloc-config	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jemalloc-config	2026-05-10 17:18:05.797913375 +0400
@@ -0,0 +1,83 @@
+#!/bin/sh
+
+usage() {
+	cat <<EOF
+Usage:
+  /usr/local/bin/jemalloc-config <option>
+Options:
+  --help | -h  : Print usage.
+  --version    : Print jemalloc version.
+  --revision   : Print shared library revision number.
+  --config     : Print configure options used to build jemalloc.
+  --prefix     : Print installation directory prefix.
+  --bindir     : Print binary installation directory.
+  --datadir    : Print data installation directory.
+  --includedir : Print include installation directory.
+  --libdir     : Print library installation directory.
+  --mandir     : Print manual page installation directory.
+  --cc         : Print compiler used to build jemalloc.
+  --cflags     : Print compiler flags used to build jemalloc.
+  --cppflags   : Print preprocessor flags used to build jemalloc.
+  --cxxflags   : Print C++ compiler flags used to build jemalloc.
+  --ldflags    : Print library flags used to build jemalloc.
+  --libs       : Print libraries jemalloc was linked against.
+EOF
+}
+
+prefix="/usr/local"
+exec_prefix="/usr/local"
+
+case "$1" in
+--help | -h)
+	usage
+	exit 0
+	;;
+--version)
+	echo "5.3.0-0-g0"
+	;;
+--revision)
+	echo "2"
+	;;
+--config)
+	echo "--disable-cxx --with-version=5.3.0-0-g0 --with-lg-quantum=3 --disable-cache-oblivious --with-jemalloc-prefix=je_ CFLAGS= LDFLAGS="
+	;;
+--prefix)
+	echo "/usr/local"
+	;;
+--bindir)
+	echo "/usr/local/bin"
+	;;
+--datadir)
+	echo "/usr/local/share"
+	;;
+--includedir)
+	echo "/usr/local/include"
+	;;
+--libdir)
+	echo "/usr/local/lib"
+	;;
+--mandir)
+	echo "/usr/local/share/man"
+	;;
+--cc)
+	echo "gcc"
+	;;
+--cflags)
+	echo "-std=gnu11 -Wall -Wextra -Wsign-compare -Wundef -Wno-format-zero-length -Wpointer-arith -Wno-missing-braces -Wno-missing-field-initializers -Wno-missing-attributes -pipe -g3 -fvisibility=hidden -Wimplicit-fallthrough -O3 -funroll-loops"
+	;;
+--cppflags)
+	echo "-D_GNU_SOURCE -D_REENTRANT"
+	;;
+--cxxflags)
+	echo ""
+	;;
+--ldflags)
+	echo " "
+	;;
+--libs)
+	echo "-lm  -pthread"
+	;;
+*)
+	usage
+	exit 1
+esac
--- a/deps/jemalloc/bin/jemalloc.sh	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jemalloc.sh	2026-05-10 17:18:05.803227947 +0400
@@ -0,0 +1,9 @@
+#!/bin/sh
+
+prefix=/usr/local
+exec_prefix=/usr/local
+libdir=${exec_prefix}/lib
+
+LD_PRELOAD=${libdir}/libjemalloc.so.2
+export LD_PRELOAD
+exec "$@"
--- a/deps/jemalloc/bin/jeprof	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jeprof	2026-05-10 17:18:05.811227841 +0400
@@ -0,0 +1,5723 @@
+#! /usr/bin/env perl
+
+# Copyright (c) 1998-2007, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+# ---
+# Program for printing the profile generated by common/profiler.cc,
+# or by the heap profiler (common/debugallocation.cc)
+#
+# The profile contains a sequence of entries of the form:
+#       <count> <stack trace>
+# This program parses the profile, and generates user-readable
+# output.
+#
+# Examples:
+#
+# % tools/jeprof "program" "profile"
+#   Enters "interactive" mode
+#
+# % tools/jeprof --text "program" "profile"
+#   Generates one line per procedure
+#
+# % tools/jeprof --gv "program" "profile"
+#   Generates annotated call-graph and displays via "gv"
+#
+# % tools/jeprof --gv --focus=Mutex "program" "profile"
+#   Restrict to code paths that involve an entry that matches "Mutex"
+#
+# % tools/jeprof --gv --focus=Mutex --ignore=string "program" "profile"
+#   Restrict to code paths that involve an entry that matches "Mutex"
+#   and does not match "string"
+#
+# % tools/jeprof --list=IBF_CheckDocid "program" "profile"
+#   Generates disassembly listing of all routines with at least one
+#   sample that match the --list=<regexp> pattern.  The listing is
+#   annotated with the flat and cumulative sample counts at each line.
+#
+# % tools/jeprof --disasm=IBF_CheckDocid "program" "profile"
+#   Generates disassembly listing of all routines with at least one
+#   sample that match the --disasm=<regexp> pattern.  The listing is
+#   annotated with the flat and cumulative sample counts at each PC value.
+#
+# TODO: Use color to indicate files?
+
+use strict;
+use warnings;
+use Getopt::Long;
+use Cwd;
+
+my $JEPROF_VERSION = "5.3.0-0-g0";
+my $PPROF_VERSION = "2.0";
+
+# These are the object tools we use which can come from a
+# user-specified location using --tools, from the JEPROF_TOOLS
+# environment variable, or from the environment.
+my %obj_tool_map = (
+  "objdump" => "objdump",
+  "nm" => "nm",
+  "addr2line" => "addr2line",
+  "c++filt" => "c++filt",
+  ## ConfigureObjTools may add architecture-specific entries:
+  #"nm_pdb" => "nm-pdb",       # for reading windows (PDB-format) executables
+  #"addr2line_pdb" => "addr2line-pdb",                                # ditto
+  #"otool" => "otool",         # equivalent of objdump on OS X
+);
+# NOTE: these are lists, so you can put in commandline flags if you want.
+my @DOT = ("dot");          # leave non-absolute, since it may be in /usr/local
+my @GV = ("gv");
+my @EVINCE = ("evince");    # could also be xpdf or perhaps acroread
+my @KCACHEGRIND = ("kcachegrind");
+my @PS2PDF = ("ps2pdf");
+# These are used for dynamic profiles
+my @URL_FETCHER = ("curl", "-s", "--fail");
+
+# These are the web pages that servers need to support for dynamic profiles
+my $HEAP_PAGE = "/pprof/heap";
+my $PROFILE_PAGE = "/pprof/profile";   # must support cgi-param "?seconds=#"
+my $PMUPROFILE_PAGE = "/pprof/pmuprofile(?:\\?.*)?"; # must support cgi-param
+                                                # ?seconds=#&event=x&period=n
+my $GROWTH_PAGE = "/pprof/growth";
+my $CONTENTION_PAGE = "/pprof/contention";
+my $WALL_PAGE = "/pprof/wall(?:\\?.*)?";  # accepts options like namefilter
+my $FILTEREDPROFILE_PAGE = "/pprof/filteredprofile(?:\\?.*)?";
+my $CENSUSPROFILE_PAGE = "/pprof/censusprofile(?:\\?.*)?"; # must support cgi-param
+                                                       # "?seconds=#",
+                                                       # "?tags_regexp=#" and
+                                                       # "?type=#".
+my $SYMBOL_PAGE = "/pprof/symbol";     # must support symbol lookup via POST
+my $PROGRAM_NAME_PAGE = "/pprof/cmdline";
+
+# These are the web pages that can be named on the command line.
+# All the alternatives must begin with /.
+my $PROFILES = "($HEAP_PAGE|$PROFILE_PAGE|$PMUPROFILE_PAGE|" .
+               "$GROWTH_PAGE|$CONTENTION_PAGE|$WALL_PAGE|" .
+               "$FILTEREDPROFILE_PAGE|$CENSUSPROFILE_PAGE)";
+
+# default binary name
+my $UNKNOWN_BINARY = "(unknown)";
+
+# There is a pervasive dependency on the length (in hex characters,
+# i.e., nibbles) of an address, distinguishing between 32-bit and
+# 64-bit profiles.  To err on the safe size, default to 64-bit here:
+my $address_length = 16;
+
+my $dev_null = "/dev/null";
+if (! -e $dev_null && $^O =~ /MSWin/) {    # $^O is the OS perl was built for
+  $dev_null = "nul";
+}
+
+# A list of paths to search for shared object files
+my @prefix_list = ();
+
+# Special routine name that should not have any symbols.
+# Used as separator to parse "addr2line -i" output.
+my $sep_symbol = '_fini';
+my $sep_address = undef;
+
+##### Argument parsing #####
+
+sub usage_string {
+  return <<EOF;
+Usage:
+jeprof [options] <program> <profiles>
+   <profiles> is a space separated list of profile names.
+jeprof [options] <symbolized-profiles>
+   <symbolized-profiles> is a list of profile files where each file contains
+   the necessary symbol mappings  as well as profile data (likely generated
+   with --raw).
+jeprof [options] <profile>
+   <profile> is a remote form.  Symbols are obtained from host:port$SYMBOL_PAGE
+
+   Each name can be:
+   /path/to/profile        - a path to a profile file
+   host:port[/<service>]   - a location of a service to get profile from
+
+   The /<service> can be $HEAP_PAGE, $PROFILE_PAGE, /pprof/pmuprofile,
+                         $GROWTH_PAGE, $CONTENTION_PAGE, /pprof/wall,
+                         $CENSUSPROFILE_PAGE, or /pprof/filteredprofile.
+   For instance:
+     jeprof http://myserver.com:80$HEAP_PAGE
+   If /<service> is omitted, the service defaults to $PROFILE_PAGE (cpu profiling).
+jeprof --symbols <program>
+   Maps addresses to symbol names.  In this mode, stdin should be a
+   list of library mappings, in the same format as is found in the heap-
+   and cpu-profile files (this loosely matches that of /proc/self/maps
+   on linux), followed by a list of hex addresses to map, one per line.
+
+   For more help with querying remote servers, including how to add the
+   necessary server-side support code, see this filename (or one like it):
+
+   /usr/doc/gperftools-$PPROF_VERSION/pprof_remote_servers.html
+
+Options:
+   --cum               Sort by cumulative data
+   --base=<base>       Subtract <base> from <profile> before display
+   --interactive       Run in interactive mode (interactive "help" gives help) [default]
+   --seconds=<n>       Length of time for dynamic profiles [default=30 secs]
+   --add_lib=<file>    Read additional symbols and line info from the given library
+   --lib_prefix=<dir>  Comma separated list of library path prefixes
+
+Reporting Granularity:
+   --addresses         Report at address level
+   --lines             Report at source line level
+   --functions         Report at function level [default]
+   --files             Report at source file level
+
+Output type:
+   --text              Generate text report
+   --callgrind         Generate callgrind format to stdout
+   --gv                Generate Postscript and display
+   --evince            Generate PDF and display
+   --web               Generate SVG and display
+   --list=<regexp>     Generate source listing of matching routines
+   --disasm=<regexp>   Generate disassembly of matching routines
+   --symbols           Print demangled symbol names found at given addresses
+   --dot               Generate DOT file to stdout
+   --ps                Generate Postcript to stdout
+   --pdf               Generate PDF to stdout
+   --svg               Generate SVG to stdout
+   --gif               Generate GIF to stdout
+   --raw               Generate symbolized jeprof data (useful with remote fetch)
+   --collapsed         Generate collapsed stacks for building flame graphs
+                       (see http://www.brendangregg.com/flamegraphs.html)
+
+Heap-Profile Options:
+   --inuse_space       Display in-use (mega)bytes [default]
+   --inuse_objects     Display in-use objects
+   --alloc_space       Display allocated (mega)bytes
+   --alloc_objects     Display allocated objects
+   --show_bytes        Display space in bytes
+   --drop_negative     Ignore negative differences
+
+Contention-profile options:
+   --total_delay       Display total delay at each region [default]
+   --contentions       Display number of delays at each region
+   --mean_delay        Display mean delay at each region
+
+Call-graph Options:
+   --nodecount=<n>     Show at most so many nodes [default=80]
+   --nodefraction=<f>  Hide nodes below <f>*total [default=.005]
+   --edgefraction=<f>  Hide edges below <f>*total [default=.001]
+   --maxdegree=<n>     Max incoming/outgoing edges per node [default=8]
+   --focus=<regexp>    Focus on backtraces with nodes matching <regexp>
+   --thread=<n>        Show profile for thread <n>
+   --ignore=<regexp>   Ignore backtraces with nodes matching <regexp>
+   --scale=<n>         Set GV scaling [default=0]
+   --heapcheck         Make nodes with non-0 object counts
+                       (i.e. direct leak generators) more visible
+   --retain=<regexp>   Retain only nodes that match <regexp>
+   --exclude=<regexp>  Exclude all nodes that match <regexp>
+
+Miscellaneous:
+   --tools=<prefix or binary:fullpath>[,...]   \$PATH for object tool pathnames
+   --test              Run unit tests
+   --help              This message
+   --version           Version information
+   --debug-syms-by-id  (Linux only) Find debug symbol files by build ID as well as by name
+
+Environment Variables:
+   JEPROF_TMPDIR        Profiles directory. Defaults to \$HOME/jeprof
+   JEPROF_TOOLS         Prefix for object tools pathnames
+
+Examples:
+
+jeprof /bin/ls ls.prof
+                       Enters "interactive" mode
+jeprof --text /bin/ls ls.prof
+                       Outputs one line per procedure
+jeprof --web /bin/ls ls.prof
+                       Displays annotated call-graph in web browser
+jeprof --gv /bin/ls ls.prof
+                       Displays annotated call-graph via 'gv'
+jeprof --gv --focus=Mutex /bin/ls ls.prof
+                       Restricts to code paths including a .*Mutex.* entry
+jeprof --gv --focus=Mutex --ignore=string /bin/ls ls.prof
+                       Code paths including Mutex but not string
+jeprof --list=getdir /bin/ls ls.prof
+                       (Per-line) annotated source listing for getdir()
+jeprof --disasm=getdir /bin/ls ls.prof
+                       (Per-PC) annotated disassembly for getdir()
+
+jeprof http://localhost:1234/
+                       Enters "interactive" mode
+jeprof --text localhost:1234
+                       Outputs one line per procedure for localhost:1234
+jeprof --raw localhost:1234 > ./local.raw
+jeprof --text ./local.raw
+                       Fetches a remote profile for later analysis and then
+                       analyzes it in text mode.
+EOF
+}
+
+sub version_string {
+  return <<EOF
+jeprof (part of jemalloc $JEPROF_VERSION)
+based on pprof (part of gperftools $PPROF_VERSION)
+
+Copyright 1998-2007 Google Inc.
+
+This is BSD licensed software; see the source for copying conditions
+and license information.
+There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A
+PARTICULAR PURPOSE.
+EOF
+}
+
+sub usage {
+  my $msg = shift;
+  print STDERR "$msg\n\n";
+  print STDERR usage_string();
+  print STDERR "\nFATAL ERROR: $msg\n";    # just as a reminder
+  exit(1);
+}
+
+sub Init() {
+  # Setup tmp-file name and handler to clean it up.
+  # We do this in the very beginning so that we can use
+  # error() and cleanup() function anytime here after.
+  $main::tmpfile_sym = "/tmp/jeprof$$.sym";
+  $main::tmpfile_ps = "/tmp/jeprof$$";
+  $main::next_tmpfile = 0;
+  $SIG{'INT'} = \&sighandler;
+
+  # Cache from filename/linenumber to source code
+  $main::source_cache = ();
+
+  $main::opt_help = 0;
+  $main::opt_version = 0;
+
+  $main::opt_cum = 0;
+  $main::opt_base = '';
+  $main::opt_addresses = 0;
+  $main::opt_lines = 0;
+  $main::opt_functions = 0;
+  $main::opt_files = 0;
+  $main::opt_lib_prefix = "";
+
+  $main::opt_text = 0;
+  $main::opt_callgrind = 0;
+  $main::opt_list = "";
+  $main::opt_disasm = "";
+  $main::opt_symbols = 0;
+  $main::opt_gv = 0;
+  $main::opt_evince = 0;
+  $main::opt_web = 0;
+  $main::opt_dot = 0;
+  $main::opt_ps = 0;
+  $main::opt_pdf = 0;
+  $main::opt_gif = 0;
+  $main::opt_svg = 0;
+  $main::opt_raw = 0;
+  $main::opt_collapsed = 0;
+
+  $main::opt_nodecount = 80;
+  $main::opt_nodefraction = 0.005;
+  $main::opt_edgefraction = 0.001;
+  $main::opt_maxdegree = 8;
+  $main::opt_focus = '';
+  $main::opt_thread = undef;
+  $main::opt_ignore = '';
+  $main::opt_scale = 0;
+  $main::opt_heapcheck = 0;
+  $main::opt_retain = '';
+  $main::opt_exclude = '';
+  $main::opt_seconds = 30;
+  $main::opt_lib = "";
+
+  $main::opt_inuse_space   = 0;
+  $main::opt_inuse_objects = 0;
+  $main::opt_alloc_space   = 0;
+  $main::opt_alloc_objects = 0;
+  $main::opt_show_bytes    = 0;
+  $main::opt_drop_negative = 0;
+  $main::opt_interactive   = 0;
+
+  $main::opt_total_delay = 0;
+  $main::opt_contentions = 0;
+  $main::opt_mean_delay = 0;
+
+  $main::opt_tools   = "";
+  $main::opt_debug   = 0;
+  $main::opt_test    = 0;
+  $main::opt_debug_syms_by_id = 0;
+
+  # These are undocumented flags used only by unittests.
+  $main::opt_test_stride = 0;
+
+  # Are we using $SYMBOL_PAGE?
+  $main::use_symbol_page = 0;
+
+  # Files returned by TempName.
+  %main::tempnames = ();
+
+  # Type of profile we are dealing with
+  # Supported types:
+  #     cpu
+  #     heap
+  #     growth
+  #     contention
+  $main::profile_type = '';     # Empty type means "unknown"
+
+  GetOptions("help!"          => \$main::opt_help,
+             "version!"       => \$main::opt_version,
+             "cum!"           => \$main::opt_cum,
+             "base=s"         => \$main::opt_base,
+             "seconds=i"      => \$main::opt_seconds,
+             "add_lib=s"      => \$main::opt_lib,
+             "lib_prefix=s"   => \$main::opt_lib_prefix,
+             "functions!"     => \$main::opt_functions,
+             "lines!"         => \$main::opt_lines,
+             "addresses!"     => \$main::opt_addresses,
+             "files!"         => \$main::opt_files,
+             "text!"          => \$main::opt_text,
+             "callgrind!"     => \$main::opt_callgrind,
+             "list=s"         => \$main::opt_list,
+             "disasm=s"       => \$main::opt_disasm,
+             "symbols!"       => \$main::opt_symbols,
+             "gv!"            => \$main::opt_gv,
+             "evince!"        => \$main::opt_evince,
+             "web!"           => \$main::opt_web,
+             "dot!"           => \$main::opt_dot,
+             "ps!"            => \$main::opt_ps,
+             "pdf!"           => \$main::opt_pdf,
+             "svg!"           => \$main::opt_svg,
+             "gif!"           => \$main::opt_gif,
+             "raw!"           => \$main::opt_raw,
+             "collapsed!"     => \$main::opt_collapsed,
+             "interactive!"   => \$main::opt_interactive,
+             "nodecount=i"    => \$main::opt_nodecount,
+             "nodefraction=f" => \$main::opt_nodefraction,
+             "edgefraction=f" => \$main::opt_edgefraction,
+             "maxdegree=i"    => \$main::opt_maxdegree,
+             "focus=s"        => \$main::opt_focus,
+             "thread=s"       => \$main::opt_thread,
+             "ignore=s"       => \$main::opt_ignore,
+             "scale=i"        => \$main::opt_scale,
+             "heapcheck"      => \$main::opt_heapcheck,
+             "retain=s"       => \$main::opt_retain,
+             "exclude=s"      => \$main::opt_exclude,
+             "inuse_space!"   => \$main::opt_inuse_space,
+             "inuse_objects!" => \$main::opt_inuse_objects,
+             "alloc_space!"   => \$main::opt_alloc_space,
+             "alloc_objects!" => \$main::opt_alloc_objects,
+             "show_bytes!"    => \$main::opt_show_bytes,
+             "drop_negative!" => \$main::opt_drop_negative,
+             "total_delay!"   => \$main::opt_total_delay,
+             "contentions!"   => \$main::opt_contentions,
+             "mean_delay!"    => \$main::opt_mean_delay,
+             "tools=s"        => \$main::opt_tools,
+             "test!"          => \$main::opt_test,
+             "debug!"         => \$main::opt_debug,
+             "debug-syms-by-id!" => \$main::opt_debug_syms_by_id,
+             # Undocumented flags used only by unittests:
+             "test_stride=i"  => \$main::opt_test_stride,
+      ) || usage("Invalid option(s)");
+
+  # Deal with the standard --help and --version
+  if ($main::opt_help) {
+    print usage_string();
+    exit(0);
+  }
+
+  if ($main::opt_version) {
+    print version_string();
+    exit(0);
+  }
+
+  # Disassembly/listing/symbols mode requires address-level info
+  if ($main::opt_disasm || $main::opt_list || $main::opt_symbols) {
+    $main::opt_functions = 0;
+    $main::opt_lines = 0;
+    $main::opt_addresses = 1;
+    $main::opt_files = 0;
+  }
+
+  # Check heap-profiling flags
+  if ($main::opt_inuse_space +
+      $main::opt_inuse_objects +
+      $main::opt_alloc_space +
+      $main::opt_alloc_objects > 1) {
+    usage("Specify at most on of --inuse/--alloc options");
+  }
+
+  # Check output granularities
+  my $grains =
+      $main::opt_functions +
+      $main::opt_lines +
+      $main::opt_addresses +
+      $main::opt_files +
+      0;
+  if ($grains > 1) {
+    usage("Only specify one output granularity option");
+  }
+  if ($grains == 0) {
+    $main::opt_functions = 1;
+  }
+
+  # Check output modes
+  my $modes =
+      $main::opt_text +
+      $main::opt_callgrind +
+      ($main::opt_list eq '' ? 0 : 1) +
+      ($main::opt_disasm eq '' ? 0 : 1) +
+      ($main::opt_symbols == 0 ? 0 : 1) +
+      $main::opt_gv +
+      $main::opt_evince +
+      $main::opt_web +
+      $main::opt_dot +
+      $main::opt_ps +
+      $main::opt_pdf +
+      $main::opt_svg +
+      $main::opt_gif +
+      $main::opt_raw +
+      $main::opt_collapsed +
+      $main::opt_interactive +
+      0;
+  if ($modes > 1) {
+    usage("Only specify one output mode");
+  }
+  if ($modes == 0) {
+    if (-t STDOUT) {  # If STDOUT is a tty, activate interactive mode
+      $main::opt_interactive = 1;
+    } else {
+      $main::opt_text = 1;
+    }
+  }
+
+  if ($main::opt_test) {
+    RunUnitTests();
+    # Should not return
+    exit(1);
+  }
+
+  # Binary name and profile arguments list
+  $main::prog = "";
+  @main::pfile_args = ();
+
+  # Remote profiling without a binary (using $SYMBOL_PAGE instead)
+  if (@ARGV > 0) {
+    if (IsProfileURL($ARGV[0])) {
+      $main::use_symbol_page = 1;
+    } elsif (IsSymbolizedProfileFile($ARGV[0])) {
+      $main::use_symbolized_profile = 1;
+      $main::prog = $UNKNOWN_BINARY;  # will be set later from the profile file
+    }
+  }
+
+  if ($main::use_symbol_page || $main::use_symbolized_profile) {
+    # We don't need a binary!
+    my %disabled = ('--lines' => $main::opt_lines,
+                    '--disasm' => $main::opt_disasm);
+    for my $option (keys %disabled) {
+      usage("$option cannot be used without a binary") if $disabled{$option};
+    }
+    # Set $main::prog later...
+    scalar(@ARGV) || usage("Did not specify profile file");
+  } elsif ($main::opt_symbols) {
+    # --symbols needs a binary-name (to run nm on, etc) but not profiles
+    $main::prog = shift(@ARGV) || usage("Did not specify program");
+  } else {
+    $main::prog = shift(@ARGV) || usage("Did not specify program");
+    scalar(@ARGV) || usage("Did not specify profile file");
+  }
+
+  # Parse profile file/location arguments
+  foreach my $farg (@ARGV) {
+    if ($farg =~ m/(.*)\@([0-9]+)(|\/.*)$/ ) {
+      my $machine = $1;
+      my $num_machines = $2;
+      my $path = $3;
+      for (my $i = 0; $i < $num_machines; $i++) {
+        unshift(@main::pfile_args, "$i.$machine$path");
+      }
+    } else {
+      unshift(@main::pfile_args, $farg);
+    }
+  }
+
+  if ($main::use_symbol_page) {
+    unless (IsProfileURL($main::pfile_args[0])) {
+      error("The first profile should be a remote form to use $SYMBOL_PAGE\n");
+    }
+    CheckSymbolPage();
+    $main::prog = FetchProgramName();
+  } elsif (!$main::use_symbolized_profile) {  # may not need objtools!
+    ConfigureObjTools($main::prog)
+  }
+
+  # Break the opt_lib_prefix into the prefix_list array
+  @prefix_list = split (',', $main::opt_lib_prefix);
+
+  # Remove trailing / from the prefixes, in the list to prevent
+  # searching things like /my/path//lib/mylib.so
+  foreach (@prefix_list) {
+    s|/+$||;
+  }
+
+  # Flag to prevent us from trying over and over to use
+  #  elfutils if it's not installed (used only with
+  #  --debug-syms-by-id option).
+  $main::gave_up_on_elfutils = 0;
+}
+
+sub FilterAndPrint {
+  my ($profile, $symbols, $libs, $thread) = @_;
+
+  # Get total data in profile
+  my $total = TotalProfile($profile);
+
+  # Remove uniniteresting stack items
+  $profile = RemoveUninterestingFrames($symbols, $profile);
+
+  # Focus?
+  if ($main::opt_focus ne '') {
+    $profile = FocusProfile($symbols, $profile, $main::opt_focus);
+  }
+
+  # Ignore?
+  if ($main::opt_ignore ne '') {
+    $profile = IgnoreProfile($symbols, $profile, $main::opt_ignore);
+  }
+
+  my $calls = ExtractCalls($symbols, $profile);
+
+  # Reduce profiles to required output granularity, and also clean
+  # each stack trace so a given entry exists at most once.
+  my $reduced = ReduceProfile($symbols, $profile);
+
+  # Get derived profiles
+  my $flat = FlatProfile($reduced);
+  my $cumulative = CumulativeProfile($reduced);
+
+  # Print
+  if (!$main::opt_interactive) {
+    if ($main::opt_disasm) {
+      PrintDisassembly($libs, $flat, $cumulative, $main::opt_disasm);
+    } elsif ($main::opt_list) {
+      PrintListing($total, $libs, $flat, $cumulative, $main::opt_list, 0);
+    } elsif ($main::opt_text) {
+      # Make sure the output is empty when have nothing to report
+      # (only matters when --heapcheck is given but we must be
+      # compatible with old branches that did not pass --heapcheck always):
+      if ($total != 0) {
+        printf("Total%s: %s %s\n",
+               (defined($thread) ? " (t$thread)" : ""),
+               Unparse($total), Units());
+      }
+      PrintText($symbols, $flat, $cumulative, -1);
+    } elsif ($main::opt_raw) {
+      PrintSymbolizedProfile($symbols, $profile, $main::prog);
+    } elsif ($main::opt_collapsed) {
+      PrintCollapsedStacks($symbols, $profile);
+    } elsif ($main::opt_callgrind) {
+      PrintCallgrind($calls);
+    } else {
+      if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) {
+        if ($main::opt_gv) {
+          RunGV(TempName($main::next_tmpfile, "ps"), "");
+        } elsif ($main::opt_evince) {
+          RunEvince(TempName($main::next_tmpfile, "pdf"), "");
+        } elsif ($main::opt_web) {
+          my $tmp = TempName($main::next_tmpfile, "svg");
+          RunWeb($tmp);
+          # The command we run might hand the file name off
+          # to an already running browser instance and then exit.
+          # Normally, we'd remove $tmp on exit (right now),
+          # but fork a child to remove $tmp a little later, so that the
+          # browser has time to load it first.
+          delete $main::tempnames{$tmp};
+          if (fork() == 0) {
+            sleep 5;
+            unlink($tmp);
+            exit(0);
+          }
+        }
+      } else {
+        cleanup();
+        exit(1);
+      }
+    }
+  } else {
+    InteractiveMode($profile, $symbols, $libs, $total);
+  }
+}
+
+sub Main() {
+  Init();
+  $main::collected_profile = undef;
+  @main::profile_files = ();
+  $main::op_time = time();
+
+  # Printing symbols is special and requires a lot less info that most.
+  if ($main::opt_symbols) {
+    PrintSymbols(*STDIN);   # Get /proc/maps and symbols output from stdin
+    return;
+  }
+
+  # Fetch all profile data
+  FetchDynamicProfiles();
+
+  # this will hold symbols that we read from the profile files
+  my $symbol_map = {};
+
+  # Read one profile, pick the last item on the list
+  my $data = ReadProfile($main::prog, pop(@main::profile_files));
+  my $profile = $data->{profile};
+  my $pcs = $data->{pcs};
+  my $libs = $data->{libs};   # Info about main program and shared libraries
+  $symbol_map = MergeSymbols($symbol_map, $data->{symbols});
+
+  # Add additional profiles, if available.
+  if (scalar(@main::profile_files) > 0) {
+    foreach my $pname (@main::profile_files) {
+      my $data2 = ReadProfile($main::prog, $pname);
+      $profile = AddProfile($profile, $data2->{profile});
+      $pcs = AddPcs($pcs, $data2->{pcs});
+      $symbol_map = MergeSymbols($symbol_map, $data2->{symbols});
+    }
+  }
+
+  # Subtract base from profile, if specified
+  if ($main::opt_base ne '') {
+    my $base = ReadProfile($main::prog, $main::opt_base);
+    $profile = SubtractProfile($profile, $base->{profile});
+    $pcs = AddPcs($pcs, $base->{pcs});
+    $symbol_map = MergeSymbols($symbol_map, $base->{symbols});
+  }
+
+  # Collect symbols
+  my $symbols;
+  if ($main::use_symbolized_profile) {
+    $symbols = FetchSymbols($pcs, $symbol_map);
+  } elsif ($main::use_symbol_page) {
+    $symbols = FetchSymbols($pcs);
+  } else {
+    # TODO(csilvers): $libs uses the /proc/self/maps data from profile1,
+    # which may differ from the data from subsequent profiles, especially
+    # if they were run on different machines.  Use appropriate libs for
+    # each pc somehow.
+    $symbols = ExtractSymbols($libs, $pcs);
+  }
+
+  if (!defined($main::opt_thread)) {
+    FilterAndPrint($profile, $symbols, $libs);
+  }
+  if (defined($data->{threads})) {
+    foreach my $thread (sort { $a <=> $b } keys(%{$data->{threads}})) {
+      if (defined($main::opt_thread) &&
+          ($main::opt_thread eq '*' || $main::opt_thread == $thread)) {
+        my $thread_profile = $data->{threads}{$thread};
+        FilterAndPrint($thread_profile, $symbols, $libs, $thread);
+      }
+    }
+  }
+
+  cleanup();
+  exit(0);
+}
+
+##### Entry Point #####
+
+Main();
+
+# Temporary code to detect if we're running on a Goobuntu system.
+# These systems don't have the right stuff installed for the special
+# Readline libraries to work, so as a temporary workaround, we default
+# to using the normal stdio code, rather than the fancier readline-based
+# code
+sub ReadlineMightFail {
+  if (-e '/lib/libtermcap.so.2') {
+    return 0;  # libtermcap exists, so readline should be okay
+  } else {
+    return 1;
+  }
+}
+
+sub RunGV {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  if (!system(ShellEscape(@GV, "--version") . " >$dev_null 2>&1")) {
+    # Options using double dash are supported by this gv version.
+    # Also, turn on noantialias to better handle bug in gv for
+    # postscript files with large dimensions.
+    # TODO: Maybe we should not pass the --noantialias flag
+    # if the gv version is known to work properly without the flag.
+    system(ShellEscape(@GV, "--scale=$main::opt_scale", "--noantialias", $fname)
+           . $bg);
+  } else {
+    # Old gv version - only supports options that use single dash.
+    print STDERR ShellEscape(@GV, "-scale", $main::opt_scale) . "\n";
+    system(ShellEscape(@GV, "-scale", "$main::opt_scale", $fname) . $bg);
+  }
+}
+
+sub RunEvince {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  system(ShellEscape(@EVINCE, $fname) . $bg);
+}
+
+sub RunWeb {
+  my $fname = shift;
+  print STDERR "Loading web page file:///$fname\n";
+
+  if (`uname` =~ /Darwin/) {
+    # OS X: open will use standard preference for SVG files.
+    system("/usr/bin/open", $fname);
+    return;
+  }
+
+  # Some kind of Unix; try generic symlinks, then specific browsers.
+  # (Stop once we find one.)
+  # Works best if the browser is already running.
+  my @alt = (
+    "/etc/alternatives/gnome-www-browser",
+    "/etc/alternatives/x-www-browser",
+    "google-chrome",
+    "firefox",
+  );
+  foreach my $b (@alt) {
+    if (system($b, $fname) == 0) {
+      return;
+    }
+  }
+
+  print STDERR "Could not load web browser.\n";
+}
+
+sub RunKcachegrind {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  print STDERR "Starting '@KCACHEGRIND " . $fname . $bg . "'\n";
+  system(ShellEscape(@KCACHEGRIND, $fname) . $bg);
+}
+
+
+##### Interactive helper routines #####
+
+sub InteractiveMode {
+  $| = 1;  # Make output unbuffered for interactive mode
+  my ($orig_profile, $symbols, $libs, $total) = @_;
+
+  print STDERR "Welcome to jeprof!  For help, type 'help'.\n";
+
+  # Use ReadLine if it's installed and input comes from a console.
+  if ( -t STDIN &&
+       !ReadlineMightFail() &&
+       defined(eval {require Term::ReadLine}) ) {
+    my $term = new Term::ReadLine 'jeprof';
+    while ( defined ($_ = $term->readline('(jeprof) '))) {
+      $term->addhistory($_) if /\S/;
+      if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) {
+        last;    # exit when we get an interactive command to quit
+      }
+    }
+  } else {       # don't have readline
+    while (1) {
+      print STDERR "(jeprof) ";
+      $_ = <STDIN>;
+      last if ! defined $_ ;
+      s/\r//g;         # turn windows-looking lines into unix-looking lines
+
+      # Save some flags that might be reset by InteractiveCommand()
+      my $save_opt_lines = $main::opt_lines;
+
+      if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) {
+        last;    # exit when we get an interactive command to quit
+      }
+
+      # Restore flags
+      $main::opt_lines = $save_opt_lines;
+    }
+  }
+}
+
+# Takes two args: orig profile, and command to run.
+# Returns 1 if we should keep going, or 0 if we were asked to quit
+sub InteractiveCommand {
+  my($orig_profile, $symbols, $libs, $total, $command) = @_;
+  $_ = $command;                # just to make future m//'s easier
+  if (!defined($_)) {
+    print STDERR "\n";
+    return 0;
+  }
+  if (m/^\s*quit/) {
+    return 0;
+  }
+  if (m/^\s*help/) {
+    InteractiveHelpMessage();
+    return 1;
+  }
+  # Clear all the mode options -- mode is controlled by "$command"
+  $main::opt_text = 0;
+  $main::opt_callgrind = 0;
+  $main::opt_disasm = 0;
+  $main::opt_list = 0;
+  $main::opt_gv = 0;
+  $main::opt_evince = 0;
+  $main::opt_cum = 0;
+
+  if (m/^\s*(text|top)(\d*)\s*(.*)/) {
+    $main::opt_text = 1;
+
+    my $line_limit = ($2 ne "") ? int($2) : 10;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($3);
+
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintText($symbols, $flat, $cumulative, $line_limit);
+    return 1;
+  }
+  if (m/^\s*callgrind\s*([^ \n]*)/) {
+    $main::opt_callgrind = 1;
+
+    # Get derived profiles
+    my $calls = ExtractCalls($symbols, $orig_profile);
+    my $filename = $1;
+    if ( $1 eq '' ) {
+      $filename = TempName($main::next_tmpfile, "callgrind");
+    }
+    PrintCallgrind($calls, $filename);
+    if ( $1 eq '' ) {
+      RunKcachegrind($filename, " & ");
+      $main::next_tmpfile++;
+    }
+
+    return 1;
+  }
+  if (m/^\s*(web)?list\s*(.+)/) {
+    my $html = (defined($1) && ($1 eq "web"));
+    $main::opt_list = 1;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($2);
+
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintListing($total, $libs, $flat, $cumulative, $routine, $html);
+    return 1;
+  }
+  if (m/^\s*disasm\s*(.+)/) {
+    $main::opt_disasm = 1;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($1);
+
+    # Process current profile to account for various settings
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintDisassembly($libs, $flat, $cumulative, $routine);
+    return 1;
+  }
+  if (m/^\s*(gv|web|evince)\s*(.*)/) {
+    $main::opt_gv = 0;
+    $main::opt_evince = 0;
+    $main::opt_web = 0;
+    if ($1 eq "gv") {
+      $main::opt_gv = 1;
+    } elsif ($1 eq "evince") {
+      $main::opt_evince = 1;
+    } elsif ($1 eq "web") {
+      $main::opt_web = 1;
+    }
+
+    my $focus;
+    my $ignore;
+    ($focus, $ignore) = ParseInteractiveArgs($2);
+
+    # Process current profile to account for various settings
+    my $profile = ProcessProfile($total, $orig_profile, $symbols,
+                                 $focus, $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) {
+      if ($main::opt_gv) {
+        RunGV(TempName($main::next_tmpfile, "ps"), " &");
+      } elsif ($main::opt_evince) {
+        RunEvince(TempName($main::next_tmpfile, "pdf"), " &");
+      } elsif ($main::opt_web) {
+        RunWeb(TempName($main::next_tmpfile, "svg"));
+      }
+      $main::next_tmpfile++;
+    }
+    return 1;
+  }
+  if (m/^\s*$/) {
+    return 1;
+  }
+  print STDERR "Unknown command: try 'help'.\n";
+  return 1;
+}
+
+
+sub ProcessProfile {
+  my $total_count = shift;
+  my $orig_profile = shift;
+  my $symbols = shift;
+  my $focus = shift;
+  my $ignore = shift;
+
+  # Process current profile to account for various settings
+  my $profile = $orig_profile;
+  printf("Total: %s %s\n", Unparse($total_count), Units());
+  if ($focus ne '') {
+    $profile = FocusProfile($symbols, $profile, $focus);
+    my $focus_count = TotalProfile($profile);
+    printf("After focusing on '%s': %s %s of %s (%0.1f%%)\n",
+           $focus,
+           Unparse($focus_count), Units(),
+           Unparse($total_count), ($focus_count*100.0) / $total_count);
+  }
+  if ($ignore ne '') {
+    $profile = IgnoreProfile($symbols, $profile, $ignore);
+    my $ignore_count = TotalProfile($profile);
+    printf("After ignoring '%s': %s %s of %s (%0.1f%%)\n",
+           $ignore,
+           Unparse($ignore_count), Units(),
+           Unparse($total_count),
+           ($ignore_count*100.0) / $total_count);
+  }
+
+  return $profile;
+}
+
+sub InteractiveHelpMessage {
+  print STDERR <<ENDOFHELP;
+Interactive jeprof mode
+
+Commands:
+  gv
+  gv [focus] [-ignore1] [-ignore2]
+      Show graphical hierarchical display of current profile.  Without
+      any arguments, shows all samples in the profile.  With the optional
+      "focus" argument, restricts the samples shown to just those where
+      the "focus" regular expression matches a routine name on the stack
+      trace.
+
+  web
+  web [focus] [-ignore1] [-ignore2]
+      Like GV, but displays profile in your web browser instead of using
+      Ghostview. Works best if your web browser is already running.
+      To change the browser that gets used:
+      On Linux, set the /etc/alternatives/gnome-www-browser symlink.
+      On OS X, change the Finder association for SVG files.
+
+  list [routine_regexp] [-ignore1] [-ignore2]
+      Show source listing of routines whose names match "routine_regexp"
+
+  weblist [routine_regexp] [-ignore1] [-ignore2]
+     Displays a source listing of routines whose names match "routine_regexp"
+     in a web browser.  You can click on source lines to view the
+     corresponding disassembly.
+
+  top [--cum] [-ignore1] [-ignore2]
+  top20 [--cum] [-ignore1] [-ignore2]
+  top37 [--cum] [-ignore1] [-ignore2]
+      Show top lines ordered by flat profile count, or cumulative count
+      if --cum is specified.  If a number is present after 'top', the
+      top K routines will be shown (defaults to showing the top 10)
+
+  disasm [routine_regexp] [-ignore1] [-ignore2]
+      Show disassembly of routines whose names match "routine_regexp",
+      annotated with sample counts.
+
+  callgrind
+  callgrind [filename]
+      Generates callgrind file. If no filename is given, kcachegrind is called.
+
+  help - This listing
+  quit or ^D - End jeprof
+
+For commands that accept optional -ignore tags, samples where any routine in
+the stack trace matches the regular expression in any of the -ignore
+parameters will be ignored.
+
+Further pprof details are available at this location (or one similar):
+
+ /usr/doc/gperftools-$PPROF_VERSION/cpu_profiler.html
+ /usr/doc/gperftools-$PPROF_VERSION/heap_profiler.html
+
+ENDOFHELP
+}
+sub ParseInteractiveArgs {
+  my $args = shift;
+  my $focus = "";
+  my $ignore = "";
+  my @x = split(/ +/, $args);
+  foreach $a (@x) {
+    if ($a =~ m/^(--|-)lines$/) {
+      $main::opt_lines = 1;
+    } elsif ($a =~ m/^(--|-)cum$/) {
+      $main::opt_cum = 1;
+    } elsif ($a =~ m/^-(.*)/) {
+      $ignore .= (($ignore ne "") ? "|" : "" ) . $1;
+    } else {
+      $focus .= (($focus ne "") ? "|" : "" ) . $a;
+    }
+  }
+  if ($ignore ne "") {
+    print STDERR "Ignoring samples in call stacks that match '$ignore'\n";
+  }
+  return ($focus, $ignore);
+}
+
+##### Output code #####
+
+sub TempName {
+  my $fnum = shift;
+  my $ext = shift;
+  my $file = "$main::tmpfile_ps.$fnum.$ext";
+  $main::tempnames{$file} = 1;
+  return $file;
+}
+
+# Print profile data in packed binary format (64-bit) to standard out
+sub PrintProfileData {
+  my $profile = shift;
+
+  # print header (64-bit style)
+  # (zero) (header-size) (version) (sample-period) (zero)
+  print pack('L*', 0, 0, 3, 0, 0, 0, 1, 0, 0, 0);
+
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    if ($#addrs >= 0) {
+      my $depth = $#addrs + 1;
+      # int(foo / 2**32) is the only reliable way to get rid of bottom
+      # 32 bits on both 32- and 64-bit systems.
+      print pack('L*', $count & 0xFFFFFFFF, int($count / 2**32));
+      print pack('L*', $depth & 0xFFFFFFFF, int($depth / 2**32));
+
+      foreach my $full_addr (@addrs) {
+        my $addr = $full_addr;
+        $addr =~ s/0x0*//;  # strip off leading 0x, zeroes
+        if (length($addr) > 16) {
+          print STDERR "Invalid address in profile: $full_addr\n";
+          next;
+        }
+        my $low_addr = substr($addr, -8);       # get last 8 hex chars
+        my $high_addr = substr($addr, -16, 8);  # get up to 8 more hex chars
+        print pack('L*', hex('0x' . $low_addr), hex('0x' . $high_addr));
+      }
+    }
+  }
+}
+
+# Print symbols and profile data
+sub PrintSymbolizedProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $prog = shift;
+
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+
+  print '--- ', $symbol_marker, "\n";
+  if (defined($prog)) {
+    print 'binary=', $prog, "\n";
+  }
+  while (my ($pc, $name) = each(%{$symbols})) {
+    my $sep = ' ';
+    print '0x', $pc;
+    # We have a list of function names, which include the inlined
+    # calls.  They are separated (and terminated) by --, which is
+    # illegal in function names.
+    for (my $j = 2; $j <= $#{$name}; $j += 3) {
+      print $sep, $name->[$j];
+      $sep = '--';
+    }
+    print "\n";
+  }
+  print '---', "\n";
+
+  my $profile_marker;
+  if ($main::profile_type eq 'heap') {
+    $HEAP_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } elsif ($main::profile_type eq 'growth') {
+    $GROWTH_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } elsif ($main::profile_type eq 'contention') {
+    $CONTENTION_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } else { # elsif ($main::profile_type eq 'cpu')
+    $PROFILE_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  }
+
+  print '--- ', $profile_marker, "\n";
+  if (defined($main::collected_profile)) {
+    # if used with remote fetch, simply dump the collected profile to output.
+    open(SRC, "<$main::collected_profile");
+    while (<SRC>) {
+      print $_;
+    }
+    close(SRC);
+  } else {
+    # --raw/http: For everything to work correctly for non-remote profiles, we
+    # would need to extend PrintProfileData() to handle all possible profile
+    # types, re-enable the code that is currently disabled in ReadCPUProfile()
+    # and FixCallerAddresses(), and remove the remote profile dumping code in
+    # the block above.
+    die "--raw/http: jeprof can only dump remote profiles for --raw\n";
+    # dump a cpu-format profile to standard out
+    PrintProfileData($profile);
+  }
+}
+
+# Print text output
+sub PrintText {
+  my $symbols = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $line_limit = shift;
+
+  my $total = TotalProfile($flat);
+
+  # Which profile to sort by?
+  my $s = $main::opt_cum ? $cumulative : $flat;
+
+  my $running_sum = 0;
+  my $lines = 0;
+  foreach my $k (sort { GetEntry($s, $b) <=> GetEntry($s, $a) || $a cmp $b }
+                 keys(%{$cumulative})) {
+    my $f = GetEntry($flat, $k);
+    my $c = GetEntry($cumulative, $k);
+    $running_sum += $f;
+
+    my $sym = $k;
+    if (exists($symbols->{$k})) {
+      $sym = $symbols->{$k}->[0] . " " . $symbols->{$k}->[1];
+      if ($main::opt_addresses) {
+        $sym = $k . " " . $sym;
+      }
+    }
+
+    if ($f != 0 || $c != 0) {
+      printf("%8s %6s %6s %8s %6s %s\n",
+             Unparse($f),
+             Percent($f, $total),
+             Percent($running_sum, $total),
+             Unparse($c),
+             Percent($c, $total),
+             $sym);
+    }
+    $lines++;
+    last if ($line_limit >= 0 && $lines >= $line_limit);
+  }
+}
+
+# Callgrind format has a compression for repeated function and file
+# names.  You show the name the first time, and just use its number
+# subsequently.  This can cut down the file to about a third or a
+# quarter of its uncompressed size.  $key and $val are the key/value
+# pair that would normally be printed by callgrind; $map is a map from
+# value to number.
+sub CompressedCGName {
+  my($key, $val, $map) = @_;
+  my $idx = $map->{$val};
+  # For very short keys, providing an index hurts rather than helps.
+  if (length($val) <= 3) {
+    return "$key=$val\n";
+  } elsif (defined($idx)) {
+    return "$key=($idx)\n";
+  } else {
+    # scalar(keys $map) gives the number of items in the map.
+    $idx = scalar(keys(%{$map})) + 1;
+    $map->{$val} = $idx;
+    return "$key=($idx) $val\n";
+  }
+}
+
+# Print the call graph in a way that's suiteable for callgrind.
+sub PrintCallgrind {
+  my $calls = shift;
+  my $filename;
+  my %filename_to_index_map;
+  my %fnname_to_index_map;
+
+  if ($main::opt_interactive) {
+    $filename = shift;
+    print STDERR "Writing callgrind file to '$filename'.\n"
+  } else {
+    $filename = "&STDOUT";
+  }
+  open(CG, ">$filename");
+  printf CG ("events: Hits\n\n");
+  foreach my $call ( map { $_->[0] }
+                     sort { $a->[1] cmp $b ->[1] ||
+                            $a->[2] <=> $b->[2] }
+                     map { /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/;
+                           [$_, $1, $2] }
+                     keys %$calls ) {
+    my $count = int($calls->{$call});
+    $call =~ /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/;
+    my ( $caller_file, $caller_line, $caller_function,
+         $callee_file, $callee_line, $callee_function ) =
+       ( $1, $2, $3, $5, $6, $7 );
+
+    # TODO(csilvers): for better compression, collect all the
+    # caller/callee_files and functions first, before printing
+    # anything, and only compress those referenced more than once.
+    printf CG CompressedCGName("fl", $caller_file, \%filename_to_index_map);
+    printf CG CompressedCGName("fn", $caller_function, \%fnname_to_index_map);
+    if (defined $6) {
+      printf CG CompressedCGName("cfl", $callee_file, \%filename_to_index_map);
+      printf CG CompressedCGName("cfn", $callee_function, \%fnname_to_index_map);
+      printf CG ("calls=$count $callee_line\n");
+    }
+    printf CG ("$caller_line $count\n\n");
+  }
+}
+
+# Print disassembly for all all routines that match $main::opt_disasm
+sub PrintDisassembly {
+  my $libs = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $disasm_opts = shift;
+
+  my $total = TotalProfile($flat);
+
+  foreach my $lib (@{$libs}) {
+    my $symbol_table = GetProcedureBoundaries($lib->[0], $disasm_opts);
+    my $offset = AddressSub($lib->[1], $lib->[3]);
+    foreach my $routine (sort ByName keys(%{$symbol_table})) {
+      my $start_addr = $symbol_table->{$routine}->[0];
+      my $end_addr = $symbol_table->{$routine}->[1];
+      # See if there are any samples in this routine
+      my $length = hex(AddressSub($end_addr, $start_addr));
+      my $addr = AddressAdd($start_addr, $offset);
+      for (my $i = 0; $i < $length; $i++) {
+        if (defined($cumulative->{$addr})) {
+          PrintDisassembledFunction($lib->[0], $offset,
+                                    $routine, $flat, $cumulative,
+                                    $start_addr, $end_addr, $total);
+          last;
+        }
+        $addr = AddressInc($addr);
+      }
+    }
+  }
+}
+
+# Return reference to array of tuples of the form:
+#       [start_address, filename, linenumber, instruction, limit_address]
+# E.g.,
+#       ["0x806c43d", "/foo/bar.cc", 131, "ret", "0x806c440"]
+sub Disassemble {
+  my $prog = shift;
+  my $offset = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+
+  my $objdump = $obj_tool_map{"objdump"};
+  my $cmd = ShellEscape($objdump, "-C", "-d", "-l", "--no-show-raw-insn",
+                        "--start-address=0x$start_addr",
+                        "--stop-address=0x$end_addr", $prog);
+  open(OBJDUMP, "$cmd |") || error("$cmd: $!\n");
+  my @result = ();
+  my $filename = "";
+  my $linenumber = -1;
+  my $last = ["", "", "", ""];
+  while (<OBJDUMP>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    chop;
+    if (m|\s*([^:\s]+):(\d+)\s*$|) {
+      # Location line of the form:
+      #   <filename>:<linenumber>
+      $filename = $1;
+      $linenumber = $2;
+    } elsif (m/^ +([0-9a-f]+):\s*(.*)/) {
+      # Disassembly line -- zero-extend address to full length
+      my $addr = HexExtend($1);
+      my $k = AddressAdd($addr, $offset);
+      $last->[4] = $k;   # Store ending address for previous instruction
+      $last = [$k, $filename, $linenumber, $2, $end_addr];
+      push(@result, $last);
+    }
+  }
+  close(OBJDUMP);
+  return @result;
+}
+
+# The input file should contain lines of the form /proc/maps-like
+# output (same format as expected from the profiles) or that looks
+# like hex addresses (like "0xDEADBEEF").  We will parse all
+# /proc/maps output, and for all the hex addresses, we will output
+# "short" symbol names, one per line, in the same order as the input.
+sub PrintSymbols {
+  my $maps_and_symbols_file = shift;
+
+  # ParseLibraries expects pcs to be in a set.  Fine by us...
+  my @pclist = ();   # pcs in sorted order
+  my $pcs = {};
+  my $map = "";
+  foreach my $line (<$maps_and_symbols_file>) {
+    $line =~ s/\r//g;    # turn windows-looking lines into unix-looking lines
+    if ($line =~ /\b(0x[0-9a-f]+)\b/i) {
+      push(@pclist, HexExtend($1));
+      $pcs->{$pclist[-1]} = 1;
+    } else {
+      $map .= $line;
+    }
+  }
+
+  my $libs = ParseLibraries($main::prog, $map, $pcs);
+  my $symbols = ExtractSymbols($libs, $pcs);
+
+  foreach my $pc (@pclist) {
+    # ->[0] is the shortname, ->[2] is the full name
+    print(($symbols->{$pc}->[0] || "??") . "\n");
+  }
+}
+
+
+# For sorting functions by name
+sub ByName {
+  return ShortFunctionName($a) cmp ShortFunctionName($b);
+}
+
+# Print source-listing for all all routines that match $list_opts
+sub PrintListing {
+  my $total = shift;
+  my $libs = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $list_opts = shift;
+  my $html = shift;
+
+  my $output = \*STDOUT;
+  my $fname = "";
+
+  if ($html) {
+    # Arrange to write the output to a temporary file
+    $fname = TempName($main::next_tmpfile, "html");
+    $main::next_tmpfile++;
+    if (!open(TEMP, ">$fname")) {
+      print STDERR "$fname: $!\n";
+      return;
+    }
+    $output = \*TEMP;
+    print $output HtmlListingHeader();
+    printf $output ("<div class=\"legend\">%s<br>Total: %s %s</div>\n",
+                    $main::prog, Unparse($total), Units());
+  }
+
+  my $listed = 0;
+  foreach my $lib (@{$libs}) {
+    my $symbol_table = GetProcedureBoundaries($lib->[0], $list_opts);
+    my $offset = AddressSub($lib->[1], $lib->[3]);
+    foreach my $routine (sort ByName keys(%{$symbol_table})) {
+      # Print if there are any samples in this routine
+      my $start_addr = $symbol_table->{$routine}->[0];
+      my $end_addr = $symbol_table->{$routine}->[1];
+      my $length = hex(AddressSub($end_addr, $start_addr));
+      my $addr = AddressAdd($start_addr, $offset);
+      for (my $i = 0; $i < $length; $i++) {
+        if (defined($cumulative->{$addr})) {
+          $listed += PrintSource(
+            $lib->[0], $offset,
+            $routine, $flat, $cumulative,
+            $start_addr, $end_addr,
+            $html,
+            $output);
+          last;
+        }
+        $addr = AddressInc($addr);
+      }
+    }
+  }
+
+  if ($html) {
+    if ($listed > 0) {
+      print $output HtmlListingFooter();
+      close($output);
+      RunWeb($fname);
+    } else {
+      close($output);
+      unlink($fname);
+    }
+  }
+}
+
+sub HtmlListingHeader {
+  return <<'EOF';
+<DOCTYPE html>
+<html>
+<head>
+<title>Pprof listing</title>
+<style type="text/css">
+body {
+  font-family: sans-serif;
+}
+h1 {
+  font-size: 1.5em;
+  margin-bottom: 4px;
+}
+.legend {
+  font-size: 1.25em;
+}
+.line {
+  color: #aaaaaa;
+}
+.nop {
+  color: #aaaaaa;
+}
+.unimportant {
+  color: #cccccc;
+}
+.disasmloc {
+  color: #000000;
+}
+.deadsrc {
+  cursor: pointer;
+}
+.deadsrc:hover {
+  background-color: #eeeeee;
+}
+.livesrc {
+  color: #0000ff;
+  cursor: pointer;
+}
+.livesrc:hover {
+  background-color: #eeeeee;
+}
+.asm {
+  color: #008800;
+  display: none;
+}
+</style>
+<script type="text/javascript">
+function jeprof_toggle_asm(e) {
+  var target;
+  if (!e) e = window.event;
+  if (e.target) target = e.target;
+  else if (e.srcElement) target = e.srcElement;
+
+  if (target) {
+    var asm = target.nextSibling;
+    if (asm && asm.className == "asm") {
+      asm.style.display = (asm.style.display == "block" ? "" : "block");
+      e.preventDefault();
+      return false;
+    }
+  }
+}
+</script>
+</head>
+<body>
+EOF
+}
+
+sub HtmlListingFooter {
+  return <<'EOF';
+</body>
+</html>
+EOF
+}
+
+sub HtmlEscape {
+  my $text = shift;
+  $text =~ s/&/&amp;/g;
+  $text =~ s/</&lt;/g;
+  $text =~ s/>/&gt;/g;
+  return $text;
+}
+
+# Returns the indentation of the line, if it has any non-whitespace
+# characters.  Otherwise, returns -1.
+sub Indentation {
+  my $line = shift;
+  if (m/^(\s*)\S/) {
+    return length($1);
+  } else {
+    return -1;
+  }
+}
+
+# If the symbol table contains inlining info, Disassemble() may tag an
+# instruction with a location inside an inlined function.  But for
+# source listings, we prefer to use the location in the function we
+# are listing.  So use MapToSymbols() to fetch full location
+# information for each instruction and then pick out the first
+# location from a location list (location list contains callers before
+# callees in case of inlining).
+#
+# After this routine has run, each entry in $instructions contains:
+#   [0] start address
+#   [1] filename for function we are listing
+#   [2] line number for function we are listing
+#   [3] disassembly
+#   [4] limit address
+#   [5] most specific filename (may be different from [1] due to inlining)
+#   [6] most specific line number (may be different from [2] due to inlining)
+sub GetTopLevelLineNumbers {
+  my ($lib, $offset, $instructions) = @_;
+  my $pcs = [];
+  for (my $i = 0; $i <= $#{$instructions}; $i++) {
+    push(@{$pcs}, $instructions->[$i]->[0]);
+  }
+  my $symbols = {};
+  MapToSymbols($lib, $offset, $pcs, $symbols);
+  for (my $i = 0; $i <= $#{$instructions}; $i++) {
+    my $e = $instructions->[$i];
+    push(@{$e}, $e->[1]);
+    push(@{$e}, $e->[2]);
+    my $addr = $e->[0];
+    my $sym = $symbols->{$addr};
+    if (defined($sym)) {
+      if ($#{$sym} >= 2 && $sym->[1] =~ m/^(.*):(\d+)$/) {
+        $e->[1] = $1;  # File name
+        $e->[2] = $2;  # Line number
+      }
+    }
+  }
+}
+
+# Print source-listing for one routine
+sub PrintSource {
+  my $prog = shift;
+  my $offset = shift;
+  my $routine = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+  my $html = shift;
+  my $output = shift;
+
+  # Disassemble all instructions (just to get line numbers)
+  my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr);
+  GetTopLevelLineNumbers($prog, $offset, \@instructions);
+
+  # Hack 1: assume that the first source file encountered in the
+  # disassembly contains the routine
+  my $filename = undef;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    if ($instructions[$i]->[2] >= 0) {
+      $filename = $instructions[$i]->[1];
+      last;
+    }
+  }
+  if (!defined($filename)) {
+    print STDERR "no filename found in $routine\n";
+    return 0;
+  }
+
+  # Hack 2: assume that the largest line number from $filename is the
+  # end of the procedure.  This is typically safe since if P1 contains
+  # an inlined call to P2, then P2 usually occurs earlier in the
+  # source file.  If this does not work, we might have to compute a
+  # density profile or just print all regions we find.
+  my $lastline = 0;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    my $f = $instructions[$i]->[1];
+    my $l = $instructions[$i]->[2];
+    if (($f eq $filename) && ($l > $lastline)) {
+      $lastline = $l;
+    }
+  }
+
+  # Hack 3: assume the first source location from "filename" is the start of
+  # the source code.
+  my $firstline = 1;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    if ($instructions[$i]->[1] eq $filename) {
+      $firstline = $instructions[$i]->[2];
+      last;
+    }
+  }
+
+  # Hack 4: Extend last line forward until its indentation is less than
+  # the indentation we saw on $firstline
+  my $oldlastline = $lastline;
+  {
+    if (!open(FILE, "<$filename")) {
+      print STDERR "$filename: $!\n";
+      return 0;
+    }
+    my $l = 0;
+    my $first_indentation = -1;
+    while (<FILE>) {
+      s/\r//g;         # turn windows-looking lines into unix-looking lines
+      $l++;
+      my $indent = Indentation($_);
+      if ($l >= $firstline) {
+        if ($first_indentation < 0 && $indent >= 0) {
+          $first_indentation = $indent;
+          last if ($first_indentation == 0);
+        }
+      }
+      if ($l >= $lastline && $indent >= 0) {
+        if ($indent >= $first_indentation) {
+          $lastline = $l+1;
+        } else {
+          last;
+        }
+      }
+    }
+    close(FILE);
+  }
+
+  # Assign all samples to the range $firstline,$lastline,
+  # Hack 4: If an instruction does not occur in the range, its samples
+  # are moved to the next instruction that occurs in the range.
+  my $samples1 = {};        # Map from line number to flat count
+  my $samples2 = {};        # Map from line number to cumulative count
+  my $running1 = 0;         # Unassigned flat counts
+  my $running2 = 0;         # Unassigned cumulative counts
+  my $total1 = 0;           # Total flat counts
+  my $total2 = 0;           # Total cumulative counts
+  my %disasm = ();          # Map from line number to disassembly
+  my $running_disasm = "";  # Unassigned disassembly
+  my $skip_marker = "---\n";
+  if ($html) {
+    $skip_marker = "";
+    for (my $l = $firstline; $l <= $lastline; $l++) {
+      $disasm{$l} = "";
+    }
+  }
+  my $last_dis_filename = '';
+  my $last_dis_linenum = -1;
+  my $last_touched_line = -1;  # To detect gaps in disassembly for a line
+  foreach my $e (@instructions) {
+    # Add up counts for all address that fall inside this instruction
+    my $c1 = 0;
+    my $c2 = 0;
+    for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) {
+      $c1 += GetEntry($flat, $a);
+      $c2 += GetEntry($cumulative, $a);
+    }
+
+    if ($html) {
+      my $dis = sprintf("      %6s %6s \t\t%8s: %s ",
+                        HtmlPrintNumber($c1),
+                        HtmlPrintNumber($c2),
+                        UnparseAddress($offset, $e->[0]),
+                        CleanDisassembly($e->[3]));
+
+      # Append the most specific source line associated with this instruction
+      if (length($dis) < 80) { $dis .= (' ' x (80 - length($dis))) };
+      $dis = HtmlEscape($dis);
+      my $f = $e->[5];
+      my $l = $e->[6];
+      if ($f ne $last_dis_filename) {
+        $dis .= sprintf("<span class=disasmloc>%s:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      } elsif ($l ne $last_dis_linenum) {
+        # De-emphasize the unchanged file name portion
+        $dis .= sprintf("<span class=unimportant>%s</span>" .
+                        "<span class=disasmloc>:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      } else {
+        # De-emphasize the entire location
+        $dis .= sprintf("<span class=unimportant>%s:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      }
+      $last_dis_filename = $f;
+      $last_dis_linenum = $l;
+      $running_disasm .= $dis;
+      $running_disasm .= "\n";
+    }
+
+    $running1 += $c1;
+    $running2 += $c2;
+    $total1 += $c1;
+    $total2 += $c2;
+    my $file = $e->[1];
+    my $line = $e->[2];
+    if (($file eq $filename) &&
+        ($line >= $firstline) &&
+        ($line <= $lastline)) {
+      # Assign all accumulated samples to this line
+      AddEntry($samples1, $line, $running1);
+      AddEntry($samples2, $line, $running2);
+      $running1 = 0;
+      $running2 = 0;
+      if ($html) {
+        if ($line != $last_touched_line && $disasm{$line} ne '') {
+          $disasm{$line} .= "\n";
+        }
+        $disasm{$line} .= $running_disasm;
+        $running_disasm = '';
+        $last_touched_line = $line;
+      }
+    }
+  }
+
+  # Assign any leftover samples to $lastline
+  AddEntry($samples1, $lastline, $running1);
+  AddEntry($samples2, $lastline, $running2);
+  if ($html) {
+    if ($lastline != $last_touched_line && $disasm{$lastline} ne '') {
+      $disasm{$lastline} .= "\n";
+    }
+    $disasm{$lastline} .= $running_disasm;
+  }
+
+  if ($html) {
+    printf $output (
+      "<h1>%s</h1>%s\n<pre onClick=\"jeprof_toggle_asm()\">\n" .
+      "Total:%6s %6s (flat / cumulative %s)\n",
+      HtmlEscape(ShortFunctionName($routine)),
+      HtmlEscape(CleanFileName($filename)),
+      Unparse($total1),
+      Unparse($total2),
+      Units());
+  } else {
+    printf $output (
+      "ROUTINE ====================== %s in %s\n" .
+      "%6s %6s Total %s (flat / cumulative)\n",
+      ShortFunctionName($routine),
+      CleanFileName($filename),
+      Unparse($total1),
+      Unparse($total2),
+      Units());
+  }
+  if (!open(FILE, "<$filename")) {
+    print STDERR "$filename: $!\n";
+    return 0;
+  }
+  my $l = 0;
+  while (<FILE>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    $l++;
+    if ($l >= $firstline - 5 &&
+        (($l <= $oldlastline + 5) || ($l <= $lastline))) {
+      chop;
+      my $text = $_;
+      if ($l == $firstline) { print $output $skip_marker; }
+      my $n1 = GetEntry($samples1, $l);
+      my $n2 = GetEntry($samples2, $l);
+      if ($html) {
+        # Emit a span that has one of the following classes:
+        #    livesrc -- has samples
+        #    deadsrc -- has disassembly, but with no samples
+        #    nop     -- has no matching disasembly
+        # Also emit an optional span containing disassembly.
+        my $dis = $disasm{$l};
+        my $asm = "";
+        if (defined($dis) && $dis ne '') {
+          $asm = "<span class=\"asm\">" . $dis . "</span>";
+        }
+        my $source_class = (($n1 + $n2 > 0)
+                            ? "livesrc"
+                            : (($asm ne "") ? "deadsrc" : "nop"));
+        printf $output (
+          "<span class=\"line\">%5d</span> " .
+          "<span class=\"%s\">%6s %6s %s</span>%s\n",
+          $l, $source_class,
+          HtmlPrintNumber($n1),
+          HtmlPrintNumber($n2),
+          HtmlEscape($text),
+          $asm);
+      } else {
+        printf $output(
+          "%6s %6s %4d: %s\n",
+          UnparseAlt($n1),
+          UnparseAlt($n2),
+          $l,
+          $text);
+      }
+      if ($l == $lastline)  { print $output $skip_marker; }
+    };
+  }
+  close(FILE);
+  if ($html) {
+    print $output "</pre>\n";
+  }
+  return 1;
+}
+
+# Return the source line for the specified file/linenumber.
+# Returns undef if not found.
+sub SourceLine {
+  my $file = shift;
+  my $line = shift;
+
+  # Look in cache
+  if (!defined($main::source_cache{$file})) {
+    if (100 < scalar keys(%main::source_cache)) {
+      # Clear the cache when it gets too big
+      $main::source_cache = ();
+    }
+
+    # Read all lines from the file
+    if (!open(FILE, "<$file")) {
+      print STDERR "$file: $!\n";
+      $main::source_cache{$file} = [];  # Cache the negative result
+      return undef;
+    }
+    my $lines = [];
+    push(@{$lines}, "");        # So we can use 1-based line numbers as indices
+    while (<FILE>) {
+      push(@{$lines}, $_);
+    }
+    close(FILE);
+
+    # Save the lines in the cache
+    $main::source_cache{$file} = $lines;
+  }
+
+  my $lines = $main::source_cache{$file};
+  if (($line < 0) || ($line > $#{$lines})) {
+    return undef;
+  } else {
+    return $lines->[$line];
+  }
+}
+
+# Print disassembly for one routine with interspersed source if available
+sub PrintDisassembledFunction {
+  my $prog = shift;
+  my $offset = shift;
+  my $routine = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+  my $total = shift;
+
+  # Disassemble all instructions
+  my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr);
+
+  # Make array of counts per instruction
+  my @flat_count = ();
+  my @cum_count = ();
+  my $flat_total = 0;
+  my $cum_total = 0;
+  foreach my $e (@instructions) {
+    # Add up counts for all address that fall inside this instruction
+    my $c1 = 0;
+    my $c2 = 0;
+    for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) {
+      $c1 += GetEntry($flat, $a);
+      $c2 += GetEntry($cumulative, $a);
+    }
+    push(@flat_count, $c1);
+    push(@cum_count, $c2);
+    $flat_total += $c1;
+    $cum_total += $c2;
+  }
+
+  # Print header with total counts
+  printf("ROUTINE ====================== %s\n" .
+         "%6s %6s %s (flat, cumulative) %.1f%% of total\n",
+         ShortFunctionName($routine),
+         Unparse($flat_total),
+         Unparse($cum_total),
+         Units(),
+         ($cum_total * 100.0) / $total);
+
+  # Process instructions in order
+  my $current_file = "";
+  for (my $i = 0; $i <= $#instructions; ) {
+    my $e = $instructions[$i];
+
+    # Print the new file name whenever we switch files
+    if ($e->[1] ne $current_file) {
+      $current_file = $e->[1];
+      my $fname = $current_file;
+      $fname =~ s|^\./||;   # Trim leading "./"
+
+      # Shorten long file names
+      if (length($fname) >= 58) {
+        $fname = "..." . substr($fname, -55);
+      }
+      printf("-------------------- %s\n", $fname);
+    }
+
+    # TODO: Compute range of lines to print together to deal with
+    # small reorderings.
+    my $first_line = $e->[2];
+    my $last_line = $first_line;
+    my %flat_sum = ();
+    my %cum_sum = ();
+    for (my $l = $first_line; $l <= $last_line; $l++) {
+      $flat_sum{$l} = 0;
+      $cum_sum{$l} = 0;
+    }
+
+    # Find run of instructions for this range of source lines
+    my $first_inst = $i;
+    while (($i <= $#instructions) &&
+           ($instructions[$i]->[2] >= $first_line) &&
+           ($instructions[$i]->[2] <= $last_line)) {
+      $e = $instructions[$i];
+      $flat_sum{$e->[2]} += $flat_count[$i];
+      $cum_sum{$e->[2]} += $cum_count[$i];
+      $i++;
+    }
+    my $last_inst = $i - 1;
+
+    # Print source lines
+    for (my $l = $first_line; $l <= $last_line; $l++) {
+      my $line = SourceLine($current_file, $l);
+      if (!defined($line)) {
+        $line = "?\n";
+        next;
+      } else {
+        $line =~ s/^\s+//;
+      }
+      printf("%6s %6s %5d: %s",
+             UnparseAlt($flat_sum{$l}),
+             UnparseAlt($cum_sum{$l}),
+             $l,
+             $line);
+    }
+
+    # Print disassembly
+    for (my $x = $first_inst; $x <= $last_inst; $x++) {
+      my $e = $instructions[$x];
+      printf("%6s %6s    %8s: %6s\n",
+             UnparseAlt($flat_count[$x]),
+             UnparseAlt($cum_count[$x]),
+             UnparseAddress($offset, $e->[0]),
+             CleanDisassembly($e->[3]));
+    }
+  }
+}
+
+# Print DOT graph
+sub PrintDot {
+  my $prog = shift;
+  my $symbols = shift;
+  my $raw = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $overall_total = shift;
+
+  # Get total
+  my $local_total = TotalProfile($flat);
+  my $nodelimit = int($main::opt_nodefraction * $local_total);
+  my $edgelimit = int($main::opt_edgefraction * $local_total);
+  my $nodecount = $main::opt_nodecount;
+
+  # Find nodes to include
+  my @list = (sort { abs(GetEntry($cumulative, $b)) <=>
+                     abs(GetEntry($cumulative, $a))
+                     || $a cmp $b }
+              keys(%{$cumulative}));
+  my $last = $nodecount - 1;
+  if ($last > $#list) {
+    $last = $#list;
+  }
+  while (($last >= 0) &&
+         (abs(GetEntry($cumulative, $list[$last])) <= $nodelimit)) {
+    $last--;
+  }
+  if ($last < 0) {
+    print STDERR "No nodes to print\n";
+    return 0;
+  }
+
+  if ($nodelimit > 0 || $edgelimit > 0) {
+    printf STDERR ("Dropping nodes with <= %s %s; edges with <= %s abs(%s)\n",
+                   Unparse($nodelimit), Units(),
+                   Unparse($edgelimit), Units());
+  }
+
+  # Open DOT output file
+  my $output;
+  my $escaped_dot = ShellEscape(@DOT);
+  my $escaped_ps2pdf = ShellEscape(@PS2PDF);
+  if ($main::opt_gv) {
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "ps"));
+    $output = "| $escaped_dot -Tps2 >$escaped_outfile";
+  } elsif ($main::opt_evince) {
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "pdf"));
+    $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - $escaped_outfile";
+  } elsif ($main::opt_ps) {
+    $output = "| $escaped_dot -Tps2";
+  } elsif ($main::opt_pdf) {
+    $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - -";
+  } elsif ($main::opt_web || $main::opt_svg) {
+    # We need to post-process the SVG, so write to a temporary file always.
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "svg"));
+    $output = "| $escaped_dot -Tsvg >$escaped_outfile";
+  } elsif ($main::opt_gif) {
+    $output = "| $escaped_dot -Tgif";
+  } else {
+    $output = ">&STDOUT";
+  }
+  open(DOT, $output) || error("$output: $!\n");
+
+  # Title
+  printf DOT ("digraph \"%s; %s %s\" {\n",
+              $prog,
+              Unparse($overall_total),
+              Units());
+  if ($main::opt_pdf) {
+    # The output is more printable if we set the page size for dot.
+    printf DOT ("size=\"8,11\"\n");
+  }
+  printf DOT ("node [width=0.375,height=0.25];\n");
+
+  # Print legend
+  printf DOT ("Legend [shape=box,fontsize=24,shape=plaintext," .
+              "label=\"%s\\l%s\\l%s\\l%s\\l%s\\l\"];\n",
+              $prog,
+              sprintf("Total %s: %s", Units(), Unparse($overall_total)),
+              sprintf("Focusing on: %s", Unparse($local_total)),
+              sprintf("Dropped nodes with <= %s abs(%s)",
+                      Unparse($nodelimit), Units()),
+              sprintf("Dropped edges with <= %s %s",
+                      Unparse($edgelimit), Units())
+              );
+
+  # Print nodes
+  my %node = ();
+  my $nextnode = 1;
+  foreach my $a (@list[0..$last]) {
+    # Pick font size
+    my $f = GetEntry($flat, $a);
+    my $c = GetEntry($cumulative, $a);
+
+    my $fs = 8;
+    if ($local_total > 0) {
+      $fs = 8 + (50.0 * sqrt(abs($f * 1.0 / $local_total)));
+    }
+
+    $node{$a} = $nextnode++;
+    my $sym = $a;
+    $sym =~ s/\s+/\\n/g;
+    $sym =~ s/::/\\n/g;
+
+    # Extra cumulative info to print for non-leaves
+    my $extra = "";
+    if ($f != $c) {
+      $extra = sprintf("\\rof %s (%s)",
+                       Unparse($c),
+                       Percent($c, $local_total));
+    }
+    my $style = "";
+    if ($main::opt_heapcheck) {
+      if ($f > 0) {
+        # make leak-causing nodes more visible (add a background)
+        $style = ",style=filled,fillcolor=gray"
+      } elsif ($f < 0) {
+        # make anti-leak-causing nodes (which almost never occur)
+        # stand out as well (triple border)
+        $style = ",peripheries=3"
+      }
+    }
+
+    printf DOT ("N%d [label=\"%s\\n%s (%s)%s\\r" .
+                "\",shape=box,fontsize=%.1f%s];\n",
+                $node{$a},
+                $sym,
+                Unparse($f),
+                Percent($f, $local_total),
+                $extra,
+                $fs,
+                $style,
+               );
+  }
+
+  # Get edges and counts per edge
+  my %edge = ();
+  my $n;
+  my $fullname_to_shortname_map = {};
+  FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map);
+  foreach my $k (keys(%{$raw})) {
+    # TODO: omit low %age edges
+    $n = $raw->{$k};
+    my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k);
+    for (my $i = 1; $i <= $#translated; $i++) {
+      my $src = $translated[$i];
+      my $dst = $translated[$i-1];
+      #next if ($src eq $dst);  # Avoid self-edges?
+      if (exists($node{$src}) && exists($node{$dst})) {
+        my $edge_label = "$src\001$dst";
+        if (!exists($edge{$edge_label})) {
+          $edge{$edge_label} = 0;
+        }
+        $edge{$edge_label} += $n;
+      }
+    }
+  }
+
+  # Print edges (process in order of decreasing counts)
+  my %indegree = ();   # Number of incoming edges added per node so far
+  my %outdegree = ();  # Number of outgoing edges added per node so far
+  foreach my $e (sort { $edge{$b} <=> $edge{$a} } keys(%edge)) {
+    my @x = split(/\001/, $e);
+    $n = $edge{$e};
+
+    # Initialize degree of kept incoming and outgoing edges if necessary
+    my $src = $x[0];
+    my $dst = $x[1];
+    if (!exists($outdegree{$src})) { $outdegree{$src} = 0; }
+    if (!exists($indegree{$dst})) { $indegree{$dst} = 0; }
+
+    my $keep;
+    if ($indegree{$dst} == 0) {
+      # Keep edge if needed for reachability
+      $keep = 1;
+    } elsif (abs($n) <= $edgelimit) {
+      # Drop if we are below --edgefraction
+      $keep = 0;
+    } elsif ($outdegree{$src} >= $main::opt_maxdegree ||
+             $indegree{$dst} >= $main::opt_maxdegree) {
+      # Keep limited number of in/out edges per node
+      $keep = 0;
+    } else {
+      $keep = 1;
+    }
+
+    if ($keep) {
+      $outdegree{$src}++;
+      $indegree{$dst}++;
+
+      # Compute line width based on edge count
+      my $fraction = abs($local_total ? (3 * ($n / $local_total)) : 0);
+      if ($fraction > 1) { $fraction = 1; }
+      my $w = $fraction * 2;
+      if ($w < 1 && ($main::opt_web || $main::opt_svg)) {
+        # SVG output treats line widths < 1 poorly.
+        $w = 1;
+      }
+
+      # Dot sometimes segfaults if given edge weights that are too large, so
+      # we cap the weights at a large value
+      my $edgeweight = abs($n) ** 0.7;
+      if ($edgeweight > 100000) { $edgeweight = 100000; }
+      $edgeweight = int($edgeweight);
+
+      my $style = sprintf("setlinewidth(%f)", $w);
+      if ($x[1] =~ m/\(inline\)/) {
+        $style .= ",dashed";
+      }
+
+      # Use a slightly squashed function of the edge count as the weight
+      printf DOT ("N%s -> N%s [label=%s, weight=%d, style=\"%s\"];\n",
+                  $node{$x[0]},
+                  $node{$x[1]},
+                  Unparse($n),
+                  $edgeweight,
+                  $style);
+    }
+  }
+
+  print DOT ("}\n");
+  close(DOT);
+
+  if ($main::opt_web || $main::opt_svg) {
+    # Rewrite SVG to be more usable inside web browser.
+    RewriteSvg(TempName($main::next_tmpfile, "svg"));
+  }
+
+  return 1;
+}
+
+sub RewriteSvg {
+  my $svgfile = shift;
+
+  open(SVG, $svgfile) || die "open temp svg: $!";
+  my @svg = <SVG>;
+  close(SVG);
+  unlink $svgfile;
+  my $svg = join('', @svg);
+
+  # Dot's SVG output is
+  #
+  #    <svg width="___" height="___"
+  #     viewBox="___" xmlns=...>
+  #    <g id="graph0" transform="...">
+  #    ...
+  #    </g>
+  #    </svg>
+  #
+  # Change it to
+  #
+  #    <svg width="100%" height="100%"
+  #     xmlns=...>
+  #    $svg_javascript
+  #    <g id="viewport" transform="translate(0,0)">
+  #    <g id="graph0" transform="...">
+  #    ...
+  #    </g>
+  #    </g>
+  #    </svg>
+
+  # Fix width, height; drop viewBox.
+  $svg =~ s/(?s)<svg width="[^"]+" height="[^"]+"(.*?)viewBox="[^"]+"/<svg width="100%" height="100%"$1/;
+
+  # Insert script, viewport <g> above first <g>
+  my $svg_javascript = SvgJavascript();
+  my $viewport = "<g id=\"viewport\" transform=\"translate(0,0)\">\n";
+  $svg =~ s/<g id="graph\d"/$svg_javascript$viewport$&/;
+
+  # Insert final </g> above </svg>.
+  $svg =~ s/(.*)(<\/svg>)/$1<\/g>$2/;
+  $svg =~ s/<g id="graph\d"(.*?)/<g id="viewport"$1/;
+
+  if ($main::opt_svg) {
+    # --svg: write to standard output.
+    print $svg;
+  } else {
+    # Write back to temporary file.
+    open(SVG, ">$svgfile") || die "open $svgfile: $!";
+    print SVG $svg;
+    close(SVG);
+  }
+}
+
+sub SvgJavascript {
+  return <<'EOF';
+<script type="text/ecmascript"><![CDATA[
+// SVGPan
+// http://www.cyberz.org/blog/2009/12/08/svgpan-a-javascript-svg-panzoomdrag-library/
+// Local modification: if(true || ...) below to force panning, never moving.
+
+/**
+ *  SVGPan library 1.2
+ * ====================
+ *
+ * Given an unique existing element with id "viewport", including the
+ * the library into any SVG adds the following capabilities:
+ *
+ *  - Mouse panning
+ *  - Mouse zooming (using the wheel)
+ *  - Object dargging
+ *
+ * Known issues:
+ *
+ *  - Zooming (while panning) on Safari has still some issues
+ *
+ * Releases:
+ *
+ * 1.2, Sat Mar 20 08:42:50 GMT 2010, Zeng Xiaohui
+ *	Fixed a bug with browser mouse handler interaction
+ *
+ * 1.1, Wed Feb  3 17:39:33 GMT 2010, Zeng Xiaohui
+ *	Updated the zoom code to support the mouse wheel on Safari/Chrome
+ *
+ * 1.0, Andrea Leofreddi
+ *	First release
+ *
+ * This code is licensed under the following BSD license:
+ *
+ * Copyright 2009-2010 Andrea Leofreddi <a.leofreddi@itcharm.com>. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are
+ * permitted provided that the following conditions are met:
+ *
+ *    1. Redistributions of source code must retain the above copyright notice, this list of
+ *       conditions and the following disclaimer.
+ *
+ *    2. Redistributions in binary form must reproduce the above copyright notice, this list
+ *       of conditions and the following disclaimer in the documentation and/or other materials
+ *       provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Andrea Leofreddi ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+ * FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL Andrea Leofreddi OR
+ * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * The views and conclusions contained in the software and documentation are those of the
+ * authors and should not be interpreted as representing official policies, either expressed
+ * or implied, of Andrea Leofreddi.
+ */
+
+var root = document.documentElement;
+
+var state = 'none', stateTarget, stateOrigin, stateTf;
+
+setupHandlers(root);
+
+/**
+ * Register handlers
+ */
+function setupHandlers(root){
+	setAttributes(root, {
+		"onmouseup" : "add(evt)",
+		"onmousedown" : "handleMouseDown(evt)",
+		"onmousemove" : "handleMouseMove(evt)",
+		"onmouseup" : "handleMouseUp(evt)",
+		//"onmouseout" : "handleMouseUp(evt)", // Decomment this to stop the pan functionality when dragging out of the SVG element
+	});
+
+	if(navigator.userAgent.toLowerCase().indexOf('webkit') >= 0)
+		window.addEventListener('mousewheel', handleMouseWheel, false); // Chrome/Safari
+	else
+		window.addEventListener('DOMMouseScroll', handleMouseWheel, false); // Others
+
+	var g = svgDoc.getElementById("svg");
+	g.width = "100%";
+	g.height = "100%";
+}
+
+/**
+ * Instance an SVGPoint object with given event coordinates.
+ */
+function getEventPoint(evt) {
+	var p = root.createSVGPoint();
+
+	p.x = evt.clientX;
+	p.y = evt.clientY;
+
+	return p;
+}
+
+/**
+ * Sets the current transform matrix of an element.
+ */
+function setCTM(element, matrix) {
+	var s = "matrix(" + matrix.a + "," + matrix.b + "," + matrix.c + "," + matrix.d + "," + matrix.e + "," + matrix.f + ")";
+
+	element.setAttribute("transform", s);
+}
+
+/**
+ * Dumps a matrix to a string (useful for debug).
+ */
+function dumpMatrix(matrix) {
+	var s = "[ " + matrix.a + ", " + matrix.c + ", " + matrix.e + "\n  " + matrix.b + ", " + matrix.d + ", " + matrix.f + "\n  0, 0, 1 ]";
+
+	return s;
+}
+
+/**
+ * Sets attributes of an element.
+ */
+function setAttributes(element, attributes){
+	for (i in attributes)
+		element.setAttributeNS(null, i, attributes[i]);
+}
+
+/**
+ * Handle mouse move event.
+ */
+function handleMouseWheel(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var delta;
+
+	if(evt.wheelDelta)
+		delta = evt.wheelDelta / 3600; // Chrome/Safari
+	else
+		delta = evt.detail / -90; // Mozilla
+
+	var z = 1 + delta; // Zoom factor: 0.9/1.1
+
+	var g = svgDoc.getElementById("viewport");
+
+	var p = getEventPoint(evt);
+
+	p = p.matrixTransform(g.getCTM().inverse());
+
+	// Compute new scale matrix in current mouse position
+	var k = root.createSVGMatrix().translate(p.x, p.y).scale(z).translate(-p.x, -p.y);
+
+        setCTM(g, g.getCTM().multiply(k));
+
+	stateTf = stateTf.multiply(k.inverse());
+}
+
+/**
+ * Handle mouse move event.
+ */
+function handleMouseMove(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var g = svgDoc.getElementById("viewport");
+
+	if(state == 'pan') {
+		// Pan mode
+		var p = getEventPoint(evt).matrixTransform(stateTf);
+
+		setCTM(g, stateTf.inverse().translate(p.x - stateOrigin.x, p.y - stateOrigin.y));
+	} else if(state == 'move') {
+		// Move mode
+		var p = getEventPoint(evt).matrixTransform(g.getCTM().inverse());
+
+		setCTM(stateTarget, root.createSVGMatrix().translate(p.x - stateOrigin.x, p.y - stateOrigin.y).multiply(g.getCTM().inverse()).multiply(stateTarget.getCTM()));
+
+		stateOrigin = p;
+	}
+}
+
+/**
+ * Handle click event.
+ */
+function handleMouseDown(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var g = svgDoc.getElementById("viewport");
+
+	if(true || evt.target.tagName == "svg") {
+		// Pan mode
+		state = 'pan';
+
+		stateTf = g.getCTM().inverse();
+
+		stateOrigin = getEventPoint(evt).matrixTransform(stateTf);
+	} else {
+		// Move mode
+		state = 'move';
+
+		stateTarget = evt.target;
+
+		stateTf = g.getCTM().inverse();
+
+		stateOrigin = getEventPoint(evt).matrixTransform(stateTf);
+	}
+}
+
+/**
+ * Handle mouse button release event.
+ */
+function handleMouseUp(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	if(state == 'pan' || state == 'move') {
+		// Quit pan mode
+		state = '';
+	}
+}
+
+]]></script>
+EOF
+}
+
+# Provides a map from fullname to shortname for cases where the
+# shortname is ambiguous.  The symlist has both the fullname and
+# shortname for all symbols, which is usually fine, but sometimes --
+# such as overloaded functions -- two different fullnames can map to
+# the same shortname.  In that case, we use the address of the
+# function to disambiguate the two.  This function fills in a map that
+# maps fullnames to modified shortnames in such cases.  If a fullname
+# is not present in the map, the 'normal' shortname provided by the
+# symlist is the appropriate one to use.
+sub FillFullnameToShortnameMap {
+  my $symbols = shift;
+  my $fullname_to_shortname_map = shift;
+  my $shortnames_seen_once = {};
+  my $shortnames_seen_more_than_once = {};
+
+  foreach my $symlist (values(%{$symbols})) {
+    # TODO(csilvers): deal with inlined symbols too.
+    my $shortname = $symlist->[0];
+    my $fullname = $symlist->[2];
+    if ($fullname !~ /<[0-9a-fA-F]+>$/) {  # fullname doesn't end in an address
+      next;       # the only collisions we care about are when addresses differ
+    }
+    if (defined($shortnames_seen_once->{$shortname}) &&
+        $shortnames_seen_once->{$shortname} ne $fullname) {
+      $shortnames_seen_more_than_once->{$shortname} = 1;
+    } else {
+      $shortnames_seen_once->{$shortname} = $fullname;
+    }
+  }
+
+  foreach my $symlist (values(%{$symbols})) {
+    my $shortname = $symlist->[0];
+    my $fullname = $symlist->[2];
+    # TODO(csilvers): take in a list of addresses we care about, and only
+    # store in the map if $symlist->[1] is in that list.  Saves space.
+    next if defined($fullname_to_shortname_map->{$fullname});
+    if (defined($shortnames_seen_more_than_once->{$shortname})) {
+      if ($fullname =~ /<0*([^>]*)>$/) {   # fullname has address at end of it
+        $fullname_to_shortname_map->{$fullname} = "$shortname\@$1";
+      }
+    }
+  }
+}
+
+# Return a small number that identifies the argument.
+# Multiple calls with the same argument will return the same number.
+# Calls with different arguments will return different numbers.
+sub ShortIdFor {
+  my $key = shift;
+  my $id = $main::uniqueid{$key};
+  if (!defined($id)) {
+    $id = keys(%main::uniqueid) + 1;
+    $main::uniqueid{$key} = $id;
+  }
+  return $id;
+}
+
+# Translate a stack of addresses into a stack of symbols
+sub TranslateStack {
+  my $symbols = shift;
+  my $fullname_to_shortname_map = shift;
+  my $k = shift;
+
+  my @addrs = split(/\n/, $k);
+  my @result = ();
+  for (my $i = 0; $i <= $#addrs; $i++) {
+    my $a = $addrs[$i];
+
+    # Skip large addresses since they sometimes show up as fake entries on RH9
+    if (length($a) > 8 && $a gt "7fffffffffffffff") {
+      next;
+    }
+
+    if ($main::opt_disasm || $main::opt_list) {
+      # We want just the address for the key
+      push(@result, $a);
+      next;
+    }
+
+    my $symlist = $symbols->{$a};
+    if (!defined($symlist)) {
+      $symlist = [$a, "", $a];
+    }
+
+    # We can have a sequence of symbols for a particular entry
+    # (more than one symbol in the case of inlining).  Callers
+    # come before callees in symlist, so walk backwards since
+    # the translated stack should contain callees before callers.
+    for (my $j = $#{$symlist}; $j >= 2; $j -= 3) {
+      my $func = $symlist->[$j-2];
+      my $fileline = $symlist->[$j-1];
+      my $fullfunc = $symlist->[$j];
+      if (defined($fullname_to_shortname_map->{$fullfunc})) {
+        $func = $fullname_to_shortname_map->{$fullfunc};
+      }
+      if ($j > 2) {
+        $func = "$func (inline)";
+      }
+
+      # Do not merge nodes corresponding to Callback::Run since that
+      # causes confusing cycles in dot display.  Instead, we synthesize
+      # a unique name for this frame per caller.
+      if ($func =~ m/Callback.*::Run$/) {
+        my $caller = ($i > 0) ? $addrs[$i-1] : 0;
+        $func = "Run#" . ShortIdFor($caller);
+      }
+
+      if ($main::opt_addresses) {
+        push(@result, "$a $func $fileline");
+      } elsif ($main::opt_lines) {
+        if ($func eq '??' && $fileline eq '??:0') {
+          push(@result, "$a");
+        } else {
+          push(@result, "$func $fileline");
+        }
+      } elsif ($main::opt_functions) {
+        if ($func eq '??') {
+          push(@result, "$a");
+        } else {
+          push(@result, $func);
+        }
+      } elsif ($main::opt_files) {
+        if ($fileline eq '??:0' || $fileline eq '') {
+          push(@result, "$a");
+        } else {
+          my $f = $fileline;
+          $f =~ s/:\d+$//;
+          push(@result, $f);
+        }
+      } else {
+        push(@result, $a);
+        last;  # Do not print inlined info
+      }
+    }
+  }
+
+  # print join(",", @addrs), " => ", join(",", @result), "\n";
+  return @result;
+}
+
+# Generate percent string for a number and a total
+sub Percent {
+  my $num = shift;
+  my $tot = shift;
+  if ($tot != 0) {
+    return sprintf("%.1f%%", $num * 100.0 / $tot);
+  } else {
+    return ($num == 0) ? "nan" : (($num > 0) ? "+inf" : "-inf");
+  }
+}
+
+# Generate pretty-printed form of number
+sub Unparse {
+  my $num = shift;
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    if ($main::opt_inuse_objects || $main::opt_alloc_objects) {
+      return sprintf("%d", $num);
+    } else {
+      if ($main::opt_show_bytes) {
+        return sprintf("%d", $num);
+      } else {
+        return sprintf("%.1f", $num / 1048576.0);
+      }
+    }
+  } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) {
+    return sprintf("%.3f", $num / 1e9); # Convert nanoseconds to seconds
+  } else {
+    return sprintf("%d", $num);
+  }
+}
+
+# Alternate pretty-printed form: 0 maps to "."
+sub UnparseAlt {
+  my $num = shift;
+  if ($num == 0) {
+    return ".";
+  } else {
+    return Unparse($num);
+  }
+}
+
+# Alternate pretty-printed form: 0 maps to ""
+sub HtmlPrintNumber {
+  my $num = shift;
+  if ($num == 0) {
+    return "";
+  } else {
+    return Unparse($num);
+  }
+}
+
+# Return output units
+sub Units {
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    if ($main::opt_inuse_objects || $main::opt_alloc_objects) {
+      return "objects";
+    } else {
+      if ($main::opt_show_bytes) {
+        return "B";
+      } else {
+        return "MB";
+      }
+    }
+  } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) {
+    return "seconds";
+  } else {
+    return "samples";
+  }
+}
+
+##### Profile manipulation code #####
+
+# Generate flattened profile:
+# If count is charged to stack [a,b,c,d], in generated profile,
+# it will be charged to [a]
+sub FlatProfile {
+  my $profile = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    if ($#addrs >= 0) {
+      AddEntry($result, $addrs[0], $count);
+    }
+  }
+  return $result;
+}
+
+# Generate cumulative profile:
+# If count is charged to stack [a,b,c,d], in generated profile,
+# it will be charged to [a], [b], [c], [d]
+sub CumulativeProfile {
+  my $profile = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    foreach my $a (@addrs) {
+      AddEntry($result, $a, $count);
+    }
+  }
+  return $result;
+}
+
+# If the second-youngest PC on the stack is always the same, returns
+# that pc.  Otherwise, returns undef.
+sub IsSecondPcAlwaysTheSame {
+  my $profile = shift;
+
+  my $second_pc = undef;
+  foreach my $k (keys(%{$profile})) {
+    my @addrs = split(/\n/, $k);
+    if ($#addrs < 1) {
+      return undef;
+    }
+    if (not defined $second_pc) {
+      $second_pc = $addrs[1];
+    } else {
+      if ($second_pc ne $addrs[1]) {
+        return undef;
+      }
+    }
+  }
+  return $second_pc;
+}
+
+sub ExtractSymbolNameInlineStack {
+  my $symbols = shift;
+  my $address = shift;
+
+  my @stack = ();
+
+  if (exists $symbols->{$address}) {
+    my @localinlinestack = @{$symbols->{$address}};
+    for (my $i = $#localinlinestack; $i > 0; $i-=3) {
+      my $file = $localinlinestack[$i-1];
+      my $fn = $localinlinestack[$i-0];
+
+      if ($file eq "?" || $file eq ":0") {
+        $file = "??:0";
+      }
+      if ($fn eq '??') {
+        # If we can't get the symbol name, at least use the file information.
+        $fn = $file;
+      }
+      my $suffix = "[inline]";
+      if ($i == 2) {
+        $suffix = "";
+      }
+      push (@stack, $fn.$suffix);
+    }
+  }
+  else {
+    # If we can't get a symbol name, at least fill in the address.
+    push (@stack, $address);
+  }
+
+  return @stack;
+}
+
+sub ExtractSymbolLocation {
+  my $symbols = shift;
+  my $address = shift;
+  # 'addr2line' outputs "??:0" for unknown locations; we do the
+  # same to be consistent.
+  my $location = "??:0:unknown";
+  if (exists $symbols->{$address}) {
+    my $file = $symbols->{$address}->[1];
+    if ($file eq "?") {
+      $file = "??:0"
+    }
+    $location = $file . ":" . $symbols->{$address}->[0];
+  }
+  return $location;
+}
+
+# Extracts a graph of calls.
+sub ExtractCalls {
+  my $symbols = shift;
+  my $profile = shift;
+
+  my $calls = {};
+  while( my ($stack_trace, $count) = each %$profile ) {
+    my @address = split(/\n/, $stack_trace);
+    my $destination = ExtractSymbolLocation($symbols, $address[0]);
+    AddEntry($calls, $destination, $count);
+    for (my $i = 1; $i <= $#address; $i++) {
+      my $source = ExtractSymbolLocation($symbols, $address[$i]);
+      my $call = "$source -> $destination";
+      AddEntry($calls, $call, $count);
+      $destination = $source;
+    }
+  }
+
+  return $calls;
+}
+
+sub FilterFrames {
+  my $symbols = shift;
+  my $profile = shift;
+
+  if ($main::opt_retain eq '' && $main::opt_exclude eq '') {
+    return $profile;
+  }
+
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my @path = ();
+    foreach my $a (@addrs) {
+      my $sym;
+      if (exists($symbols->{$a})) {
+        $sym = $symbols->{$a}->[0];
+      } else {
+        $sym = $a;
+      }
+      if ($main::opt_retain ne '' && $sym !~ m/$main::opt_retain/) {
+        next;
+      }
+      if ($main::opt_exclude ne '' && $sym =~ m/$main::opt_exclude/) {
+        next;
+      }
+      push(@path, $a);
+    }
+    if (scalar(@path) > 0) {
+      my $reduced_path = join("\n", @path);
+      AddEntry($result, $reduced_path, $count);
+    }
+  }
+
+  return $result;
+}
+
+sub PrintCollapsedStacks {
+  my $symbols = shift;
+  my $profile = shift;
+
+  while (my ($stack_trace, $count) = each %$profile) {
+    my @address = split(/\n/, $stack_trace);
+    my @names = reverse ( map { ExtractSymbolNameInlineStack($symbols, $_) } @address );
+    printf("%s %d\n", join(";", @names), $count);
+  }
+}
+
+sub RemoveUninterestingFrames {
+  my $symbols = shift;
+  my $profile = shift;
+
+  # List of function names to skip
+  my %skip = ();
+  my $skip_regexp = 'NOMATCH';
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    foreach my $name ('je_calloc',
+                      'cfree',
+                      'je_malloc',
+                      'newImpl',
+                      'void* newImpl',
+                      'je_free',
+                      'je_memalign',
+                      'je_posix_memalign',
+                      'je_aligned_alloc',
+                      'pvalloc',
+                      'je_valloc',
+                      'je_realloc',
+                      'je_mallocx',
+                      'je_rallocx',
+                      'je_xallocx',
+                      'je_dallocx',
+                      'je_sdallocx',
+                      'je_sdallocx_noflags',
+                      'tc_calloc',
+                      'tc_cfree',
+                      'tc_malloc',
+                      'tc_free',
+                      'tc_memalign',
+                      'tc_posix_memalign',
+                      'tc_pvalloc',
+                      'tc_valloc',
+                      'tc_realloc',
+                      'tc_new',
+                      'tc_delete',
+                      'tc_newarray',
+                      'tc_deletearray',
+                      'tc_new_nothrow',
+                      'tc_newarray_nothrow',
+                      'do_malloc',
+                      '::do_malloc',   # new name -- got moved to an unnamed ns
+                      '::do_malloc_or_cpp_alloc',
+                      'DoSampledAllocation',
+                      'simple_alloc::allocate',
+                      '__malloc_alloc_template::allocate',
+                      '__builtin_delete',
+                      '__builtin_new',
+                      '__builtin_vec_delete',
+                      '__builtin_vec_new',
+                      'operator new',
+                      'operator new[]',
+                      # The entry to our memory-allocation routines on OS X
+                      'malloc_zone_malloc',
+                      'malloc_zone_calloc',
+                      'malloc_zone_valloc',
+                      'malloc_zone_realloc',
+                      'malloc_zone_memalign',
+                      'malloc_zone_free',
+                      # These mark the beginning/end of our custom sections
+                      '__start_google_malloc',
+                      '__stop_google_malloc',
+                      '__start_malloc_hook',
+                      '__stop_malloc_hook') {
+      $skip{$name} = 1;
+      $skip{"_" . $name} = 1;   # Mach (OS X) adds a _ prefix to everything
+    }
+    # TODO: Remove TCMalloc once everything has been
+    # moved into the tcmalloc:: namespace and we have flushed
+    # old code out of the system.
+    $skip_regexp = "TCMalloc|^tcmalloc::";
+  } elsif ($main::profile_type eq 'contention') {
+    foreach my $vname ('base::RecordLockProfileData',
+                       'base::SubmitMutexProfileData',
+                       'base::SubmitSpinLockProfileData',
+                       'Mutex::Unlock',
+                       'Mutex::UnlockSlow',
+                       'Mutex::ReaderUnlock',
+                       'MutexLock::~MutexLock',
+                       'SpinLock::Unlock',
+                       'SpinLock::SlowUnlock',
+                       'SpinLockHolder::~SpinLockHolder') {
+      $skip{$vname} = 1;
+    }
+  } elsif ($main::profile_type eq 'cpu') {
+    # Drop signal handlers used for CPU profile collection
+    # TODO(dpeng): this should not be necessary; it's taken
+    # care of by the general 2nd-pc mechanism below.
+    foreach my $name ('ProfileData::Add',           # historical
+                      'ProfileData::prof_handler',  # historical
+                      'CpuProfiler::prof_handler',
+                      '__FRAME_END__',
+                      '__pthread_sighandler',
+                      '__restore') {
+      $skip{$name} = 1;
+    }
+  } else {
+    # Nothing skipped for unknown types
+  }
+
+  if ($main::profile_type eq 'cpu') {
+    # If all the second-youngest program counters are the same,
+    # this STRONGLY suggests that it is an artifact of measurement,
+    # i.e., stack frames pushed by the CPU profiler signal handler.
+    # Hence, we delete them.
+    # (The topmost PC is read from the signal structure, not from
+    # the stack, so it does not get involved.)
+    while (my $second_pc = IsSecondPcAlwaysTheSame($profile)) {
+      my $result = {};
+      my $func = '';
+      if (exists($symbols->{$second_pc})) {
+        $second_pc = $symbols->{$second_pc}->[0];
+      }
+      print STDERR "Removing $second_pc from all stack traces.\n";
+      foreach my $k (keys(%{$profile})) {
+        my $count = $profile->{$k};
+        my @addrs = split(/\n/, $k);
+        splice @addrs, 1, 1;
+        my $reduced_path = join("\n", @addrs);
+        AddEntry($result, $reduced_path, $count);
+      }
+      $profile = $result;
+    }
+  }
+
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my @path = ();
+    foreach my $a (@addrs) {
+      if (exists($symbols->{$a})) {
+        my $func = $symbols->{$a}->[0];
+        if ($skip{$func} || ($func =~ m/$skip_regexp/)) {
+          # Throw away the portion of the backtrace seen so far, under the
+          # assumption that previous frames were for functions internal to the
+          # allocator.
+          @path = ();
+          next;
+        }
+      }
+      push(@path, $a);
+    }
+    my $reduced_path = join("\n", @path);
+    AddEntry($result, $reduced_path, $count);
+  }
+
+  $result = FilterFrames($symbols, $result);
+
+  return $result;
+}
+
+# Reduce profile to granularity given by user
+sub ReduceProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $result = {};
+  my $fullname_to_shortname_map = {};
+  FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map);
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k);
+    my @path = ();
+    my %seen = ();
+    $seen{''} = 1;      # So that empty keys are skipped
+    foreach my $e (@translated) {
+      # To avoid double-counting due to recursion, skip a stack-trace
+      # entry if it has already been seen
+      if (!$seen{$e}) {
+        $seen{$e} = 1;
+        push(@path, $e);
+      }
+    }
+    my $reduced_path = join("\n", @path);
+    AddEntry($result, $reduced_path, $count);
+  }
+  return $result;
+}
+
+# Does the specified symbol array match the regexp?
+sub SymbolMatches {
+  my $sym = shift;
+  my $re = shift;
+  if (defined($sym)) {
+    for (my $i = 0; $i < $#{$sym}; $i += 3) {
+      if ($sym->[$i] =~ m/$re/ || $sym->[$i+1] =~ m/$re/) {
+        return 1;
+      }
+    }
+  }
+  return 0;
+}
+
+# Focus only on paths involving specified regexps
+sub FocusProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $focus = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    foreach my $a (@addrs) {
+      # Reply if it matches either the address/shortname/fileline
+      if (($a =~ m/$focus/) || SymbolMatches($symbols->{$a}, $focus)) {
+        AddEntry($result, $k, $count);
+        last;
+      }
+    }
+  }
+  return $result;
+}
+
+# Focus only on paths not involving specified regexps
+sub IgnoreProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $ignore = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my $matched = 0;
+    foreach my $a (@addrs) {
+      # Reply if it matches either the address/shortname/fileline
+      if (($a =~ m/$ignore/) || SymbolMatches($symbols->{$a}, $ignore)) {
+        $matched = 1;
+        last;
+      }
+    }
+    if (!$matched) {
+      AddEntry($result, $k, $count);
+    }
+  }
+  return $result;
+}
+
+# Get total count in profile
+sub TotalProfile {
+  my $profile = shift;
+  my $result = 0;
+  foreach my $k (keys(%{$profile})) {
+    $result += $profile->{$k};
+  }
+  return $result;
+}
+
+# Add A to B
+sub AddProfile {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  # add all keys in A
+  foreach my $k (keys(%{$A})) {
+    my $v = $A->{$k};
+    AddEntry($R, $k, $v);
+  }
+  # add all keys in B
+  foreach my $k (keys(%{$B})) {
+    my $v = $B->{$k};
+    AddEntry($R, $k, $v);
+  }
+  return $R;
+}
+
+# Merges symbol maps
+sub MergeSymbols {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  foreach my $k (keys(%{$A})) {
+    $R->{$k} = $A->{$k};
+  }
+  if (defined($B)) {
+    foreach my $k (keys(%{$B})) {
+      $R->{$k} = $B->{$k};
+    }
+  }
+  return $R;
+}
+
+
+# Add A to B
+sub AddPcs {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  # add all keys in A
+  foreach my $k (keys(%{$A})) {
+    $R->{$k} = 1
+  }
+  # add all keys in B
+  foreach my $k (keys(%{$B})) {
+    $R->{$k} = 1
+  }
+  return $R;
+}
+
+# Subtract B from A
+sub SubtractProfile {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  foreach my $k (keys(%{$A})) {
+    my $v = $A->{$k} - GetEntry($B, $k);
+    if ($v < 0 && $main::opt_drop_negative) {
+      $v = 0;
+    }
+    AddEntry($R, $k, $v);
+  }
+  if (!$main::opt_drop_negative) {
+    # Take care of when subtracted profile has more entries
+    foreach my $k (keys(%{$B})) {
+      if (!exists($A->{$k})) {
+        AddEntry($R, $k, 0 - $B->{$k});
+      }
+    }
+  }
+  return $R;
+}
+
+# Get entry from profile; zero if not present
+sub GetEntry {
+  my $profile = shift;
+  my $k = shift;
+  if (exists($profile->{$k})) {
+    return $profile->{$k};
+  } else {
+    return 0;
+  }
+}
+
+# Add entry to specified profile
+sub AddEntry {
+  my $profile = shift;
+  my $k = shift;
+  my $n = shift;
+  if (!exists($profile->{$k})) {
+    $profile->{$k} = 0;
+  }
+  $profile->{$k} += $n;
+}
+
+# Add a stack of entries to specified profile, and add them to the $pcs
+# list.
+sub AddEntries {
+  my $profile = shift;
+  my $pcs = shift;
+  my $stack = shift;
+  my $count = shift;
+  my @k = ();
+
+  foreach my $e (split(/\s+/, $stack)) {
+    my $pc = HexExtend($e);
+    $pcs->{$pc} = 1;
+    push @k, $pc;
+  }
+  AddEntry($profile, (join "\n", @k), $count);
+}
+
+##### Code to profile a server dynamically #####
+
+sub CheckSymbolPage {
+  my $url = SymbolPageURL();
+  my $command = ShellEscape(@URL_FETCHER, $url);
+  open(SYMBOL, "$command |") or error($command);
+  my $line = <SYMBOL>;
+  $line =~ s/\r//g;         # turn windows-looking lines into unix-looking lines
+  close(SYMBOL);
+  unless (defined($line)) {
+    error("$url doesn't exist\n");
+  }
+
+  if ($line =~ /^num_symbols:\s+(\d+)$/) {
+    if ($1 == 0) {
+      error("Stripped binary. No symbols available.\n");
+    }
+  } else {
+    error("Failed to get the number of symbols from $url\n");
+  }
+}
+
+sub IsProfileURL {
+  my $profile_name = shift;
+  if (-f $profile_name) {
+    printf STDERR "Using local file $profile_name.\n";
+    return 0;
+  }
+  return 1;
+}
+
+sub ParseProfileURL {
+  my $profile_name = shift;
+
+  if (!defined($profile_name) || $profile_name eq "") {
+    return ();
+  }
+
+  # Split profile URL - matches all non-empty strings, so no test.
+  $profile_name =~ m,^(https?://)?([^/]+)(.*?)(/|$PROFILES)?$,;
+
+  my $proto = $1 || "http://";
+  my $hostport = $2;
+  my $prefix = $3;
+  my $profile = $4 || "/";
+
+  my $host = $hostport;
+  $host =~ s/:.*//;
+
+  my $baseurl = "$proto$hostport$prefix";
+  return ($host, $baseurl, $profile);
+}
+
+# We fetch symbols from the first profile argument.
+sub SymbolPageURL {
+  my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]);
+  return "$baseURL$SYMBOL_PAGE";
+}
+
+sub FetchProgramName() {
+  my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]);
+  my $url = "$baseURL$PROGRAM_NAME_PAGE";
+  my $command_line = ShellEscape(@URL_FETCHER, $url);
+  open(CMDLINE, "$command_line |") or error($command_line);
+  my $cmdline = <CMDLINE>;
+  $cmdline =~ s/\r//g;   # turn windows-looking lines into unix-looking lines
+  close(CMDLINE);
+  error("Failed to get program name from $url\n") unless defined($cmdline);
+  $cmdline =~ s/\x00.+//;  # Remove argv[1] and latters.
+  $cmdline =~ s!\n!!g;  # Remove LFs.
+  return $cmdline;
+}
+
+# Gee, curl's -L (--location) option isn't reliable at least
+# with its 7.12.3 version.  Curl will forget to post data if
+# there is a redirection.  This function is a workaround for
+# curl.  Redirection happens on borg hosts.
+sub ResolveRedirectionForCurl {
+  my $url = shift;
+  my $command_line = ShellEscape(@URL_FETCHER, "--head", $url);
+  open(CMDLINE, "$command_line |") or error($command_line);
+  while (<CMDLINE>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    if (/^Location: (.*)/) {
+      $url = $1;
+    }
+  }
+  close(CMDLINE);
+  return $url;
+}
+
+# Add a timeout flat to URL_FETCHER.  Returns a new list.
+sub AddFetchTimeout {
+  my $timeout = shift;
+  my @fetcher = @_;
+  if (defined($timeout)) {
+    if (join(" ", @fetcher) =~ m/\bcurl -s/) {
+      push(@fetcher, "--max-time", sprintf("%d", $timeout));
+    } elsif (join(" ", @fetcher) =~ m/\brpcget\b/) {
+      push(@fetcher, sprintf("--deadline=%d", $timeout));
+    }
+  }
+  return @fetcher;
+}
+
+# Reads a symbol map from the file handle name given as $1, returning
+# the resulting symbol map.  Also processes variables relating to symbols.
+# Currently, the only variable processed is 'binary=<value>' which updates
+# $main::prog to have the correct program name.
+sub ReadSymbols {
+  my $in = shift;
+  my $map = {};
+  while (<$in>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    # Removes all the leading zeroes from the symbols, see comment below.
+    if (m/^0x0*([0-9a-f]+)\s+(.+)/) {
+      $map->{$1} = $2;
+    } elsif (m/^---/) {
+      last;
+    } elsif (m/^([a-z][^=]*)=(.*)$/ ) {
+      my ($variable, $value) = ($1, $2);
+      for ($variable, $value) {
+        s/^\s+//;
+        s/\s+$//;
+      }
+      if ($variable eq "binary") {
+        if ($main::prog ne $UNKNOWN_BINARY && $main::prog ne $value) {
+          printf STDERR ("Warning: Mismatched binary name '%s', using '%s'.\n",
+                         $main::prog, $value);
+        }
+        $main::prog = $value;
+      } else {
+        printf STDERR ("Ignoring unknown variable in symbols list: " .
+            "'%s' = '%s'\n", $variable, $value);
+      }
+    }
+  }
+  return $map;
+}
+
+sub URLEncode {
+  my $str = shift;
+  $str =~ s/([^A-Za-z0-9\-_.!~*'()])/ sprintf "%%%02x", ord $1 /eg;
+  return $str;
+}
+
+sub AppendSymbolFilterParams {
+  my $url = shift;
+  my @params = ();
+  if ($main::opt_retain ne '') {
+    push(@params, sprintf("retain=%s", URLEncode($main::opt_retain)));
+  }
+  if ($main::opt_exclude ne '') {
+    push(@params, sprintf("exclude=%s", URLEncode($main::opt_exclude)));
+  }
+  if (scalar @params > 0) {
+    $url = sprintf("%s?%s", $url, join("&", @params));
+  }
+  return $url;
+}
+
+# Fetches and processes symbols to prepare them for use in the profile output
+# code.  If the optional 'symbol_map' arg is not given, fetches symbols from
+# $SYMBOL_PAGE for all PC values found in profile.  Otherwise, the raw symbols
+# are assumed to have already been fetched into 'symbol_map' and are simply
+# extracted and processed.
+sub FetchSymbols {
+  my $pcset = shift;
+  my $symbol_map = shift;
+
+  my %seen = ();
+  my @pcs = grep { !$seen{$_}++ } keys(%$pcset);  # uniq
+
+  if (!defined($symbol_map)) {
+    my $post_data = join("+", sort((map {"0x" . "$_"} @pcs)));
+
+    open(POSTFILE, ">$main::tmpfile_sym");
+    print POSTFILE $post_data;
+    close(POSTFILE);
+
+    my $url = SymbolPageURL();
+
+    my $command_line;
+    if (join(" ", @URL_FETCHER) =~ m/\bcurl -s/) {
+      $url = ResolveRedirectionForCurl($url);
+      $url = AppendSymbolFilterParams($url);
+      $command_line = ShellEscape(@URL_FETCHER, "-d", "\@$main::tmpfile_sym",
+                                  $url);
+    } else {
+      $url = AppendSymbolFilterParams($url);
+      $command_line = (ShellEscape(@URL_FETCHER, "--post", $url)
+                       . " < " . ShellEscape($main::tmpfile_sym));
+    }
+    # We use c++filt in case $SYMBOL_PAGE gives us mangled symbols.
+    my $escaped_cppfilt = ShellEscape($obj_tool_map{"c++filt"});
+    open(SYMBOL, "$command_line | $escaped_cppfilt |") or error($command_line);
+    $symbol_map = ReadSymbols(*SYMBOL{IO});
+    close(SYMBOL);
+  }
+
+  my $symbols = {};
+  foreach my $pc (@pcs) {
+    my $fullname;
+    # For 64 bits binaries, symbols are extracted with 8 leading zeroes.
+    # Then /symbol reads the long symbols in as uint64, and outputs
+    # the result with a "0x%08llx" format which get rid of the zeroes.
+    # By removing all the leading zeroes in both $pc and the symbols from
+    # /symbol, the symbols match and are retrievable from the map.
+    my $shortpc = $pc;
+    $shortpc =~ s/^0*//;
+    # Each line may have a list of names, which includes the function
+    # and also other functions it has inlined.  They are separated (in
+    # PrintSymbolizedProfile), by --, which is illegal in function names.
+    my $fullnames;
+    if (defined($symbol_map->{$shortpc})) {
+      $fullnames = $symbol_map->{$shortpc};
+    } else {
+      $fullnames = "0x" . $pc;  # Just use addresses
+    }
+    my $sym = [];
+    $symbols->{$pc} = $sym;
+    foreach my $fullname (split("--", $fullnames)) {
+      my $name = ShortFunctionName($fullname);
+      push(@{$sym}, $name, "?", $fullname);
+    }
+  }
+  return $symbols;
+}
+
+sub BaseName {
+  my $file_name = shift;
+  $file_name =~ s!^.*/!!;  # Remove directory name
+  return $file_name;
+}
+
+sub MakeProfileBaseName {
+  my ($binary_name, $profile_name) = @_;
+  my ($host, $baseURL, $path) = ParseProfileURL($profile_name);
+  my $binary_shortname = BaseName($binary_name);
+  return sprintf("%s.%s.%s",
+                 $binary_shortname, $main::op_time, $host);
+}
+
+sub FetchDynamicProfile {
+  my $binary_name = shift;
+  my $profile_name = shift;
+  my $fetch_name_only = shift;
+  my $encourage_patience = shift;
+
+  if (!IsProfileURL($profile_name)) {
+    return $profile_name;
+  } else {
+    my ($host, $baseURL, $path) = ParseProfileURL($profile_name);
+    if ($path eq "" || $path eq "/") {
+      # Missing type specifier defaults to cpu-profile
+      $path = $PROFILE_PAGE;
+    }
+
+    my $profile_file = MakeProfileBaseName($binary_name, $profile_name);
+
+    my $url = "$baseURL$path";
+    my $fetch_timeout = undef;
+    if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE/) {
+      if ($path =~ m/[?]/) {
+        $url .= "&";
+      } else {
+        $url .= "?";
+      }
+      $url .= sprintf("seconds=%d", $main::opt_seconds);
+      $fetch_timeout = $main::opt_seconds * 1.01 + 60;
+      # Set $profile_type for consumption by PrintSymbolizedProfile.
+      $main::profile_type = 'cpu';
+    } else {
+      # For non-CPU profiles, we add a type-extension to
+      # the target profile file name.
+      my $suffix = $path;
+      $suffix =~ s,/,.,g;
+      $profile_file .= $suffix;
+      # Set $profile_type for consumption by PrintSymbolizedProfile.
+      if ($path =~ m/$HEAP_PAGE/) {
+        $main::profile_type = 'heap';
+      } elsif ($path =~ m/$GROWTH_PAGE/) {
+        $main::profile_type = 'growth';
+      } elsif ($path =~ m/$CONTENTION_PAGE/) {
+        $main::profile_type = 'contention';
+      }
+    }
+
+    my $profile_dir = $ENV{"JEPROF_TMPDIR"} || ($ENV{HOME} . "/jeprof");
+    if (! -d $profile_dir) {
+      mkdir($profile_dir)
+          || die("Unable to create profile directory $profile_dir: $!\n");
+    }
+    my $tmp_profile = "$profile_dir/.tmp.$profile_file";
+    my $real_profile = "$profile_dir/$profile_file";
+
+    if ($fetch_name_only > 0) {
+      return $real_profile;
+    }
+
+    my @fetcher = AddFetchTimeout($fetch_timeout, @URL_FETCHER);
+    my $cmd = ShellEscape(@fetcher, $url) . " > " . ShellEscape($tmp_profile);
+    if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE|$CENSUSPROFILE_PAGE/){
+      print STDERR "Gathering CPU profile from $url for $main::opt_seconds seconds to\n  ${real_profile}\n";
+      if ($encourage_patience) {
+        print STDERR "Be patient...\n";
+      }
+    } else {
+      print STDERR "Fetching $path profile from $url to\n  ${real_profile}\n";
+    }
+
+    (system($cmd) == 0) || error("Failed to get profile: $cmd: $!\n");
+    (system("mv", $tmp_profile, $real_profile) == 0) || error("Unable to rename profile\n");
+    print STDERR "Wrote profile to $real_profile\n";
+    $main::collected_profile = $real_profile;
+    return $main::collected_profile;
+  }
+}
+
+# Collect profiles in parallel
+sub FetchDynamicProfiles {
+  my $items = scalar(@main::pfile_args);
+  my $levels = log($items) / log(2);
+
+  if ($items == 1) {
+    $main::profile_files[0] = FetchDynamicProfile($main::prog, $main::pfile_args[0], 0, 1);
+  } else {
+    # math rounding issues
+    if ((2 ** $levels) < $items) {
+     $levels++;
+    }
+    my $count = scalar(@main::pfile_args);
+    for (my $i = 0; $i < $count; $i++) {
+      $main::profile_files[$i] = FetchDynamicProfile($main::prog, $main::pfile_args[$i], 1, 0);
+    }
+    print STDERR "Fetching $count profiles, Be patient...\n";
+    FetchDynamicProfilesRecurse($levels, 0, 0);
+    $main::collected_profile = join(" \\\n    ", @main::profile_files);
+  }
+}
+
+# Recursively fork a process to get enough processes
+# collecting profiles
+sub FetchDynamicProfilesRecurse {
+  my $maxlevel = shift;
+  my $level = shift;
+  my $position = shift;
+
+  if (my $pid = fork()) {
+    $position = 0 | ($position << 1);
+    TryCollectProfile($maxlevel, $level, $position);
+    wait;
+  } else {
+    $position = 1 | ($position << 1);
+    TryCollectProfile($maxlevel, $level, $position);
+    cleanup();
+    exit(0);
+  }
+}
+
+# Collect a single profile
+sub TryCollectProfile {
+  my $maxlevel = shift;
+  my $level = shift;
+  my $position = shift;
+
+  if ($level >= ($maxlevel - 1)) {
+    if ($position < scalar(@main::pfile_args)) {
+      FetchDynamicProfile($main::prog, $main::pfile_args[$position], 0, 0);
+    }
+  } else {
+    FetchDynamicProfilesRecurse($maxlevel, $level+1, $position);
+  }
+}
+
+##### Parsing code #####
+
+# Provide a small streaming-read module to handle very large
+# cpu-profile files.  Stream in chunks along a sliding window.
+# Provides an interface to get one 'slot', correctly handling
+# endian-ness differences.  A slot is one 32-bit or 64-bit word
+# (depending on the input profile).  We tell endianness and bit-size
+# for the profile by looking at the first 8 bytes: in cpu profiles,
+# the second slot is always 3 (we'll accept anything that's not 0).
+BEGIN {
+  package CpuProfileStream;
+
+  sub new {
+    my ($class, $file, $fname) = @_;
+    my $self = { file        => $file,
+                 base        => 0,
+                 stride      => 512 * 1024,   # must be a multiple of bitsize/8
+                 slots       => [],
+                 unpack_code => "",           # N for big-endian, V for little
+                 perl_is_64bit => 1,          # matters if profile is 64-bit
+    };
+    bless $self, $class;
+    # Let unittests adjust the stride
+    if ($main::opt_test_stride > 0) {
+      $self->{stride} = $main::opt_test_stride;
+    }
+    # Read the first two slots to figure out bitsize and endianness.
+    my $slots = $self->{slots};
+    my $str;
+    read($self->{file}, $str, 8);
+    # Set the global $address_length based on what we see here.
+    # 8 is 32-bit (8 hexadecimal chars); 16 is 64-bit (16 hexadecimal chars).
+    $address_length = ($str eq (chr(0)x8)) ? 16 : 8;
+    if ($address_length == 8) {
+      if (substr($str, 6, 2) eq chr(0)x2) {
+        $self->{unpack_code} = 'V';  # Little-endian.
+      } elsif (substr($str, 4, 2) eq chr(0)x2) {
+        $self->{unpack_code} = 'N';  # Big-endian
+      } else {
+        ::error("$fname: header size >= 2**16\n");
+      }
+      @$slots = unpack($self->{unpack_code} . "*", $str);
+    } else {
+      # If we're a 64-bit profile, check if we're a 64-bit-capable
+      # perl.  Otherwise, each slot will be represented as a float
+      # instead of an int64, losing precision and making all the
+      # 64-bit addresses wrong.  We won't complain yet, but will
+      # later if we ever see a value that doesn't fit in 32 bits.
+      my $has_q = 0;
+      eval { $has_q = pack("Q", "1") ? 1 : 1; };
+      if (!$has_q) {
+        $self->{perl_is_64bit} = 0;
+      }
+      read($self->{file}, $str, 8);
+      if (substr($str, 4, 4) eq chr(0)x4) {
+        # We'd love to use 'Q', but it's a) not universal, b) not endian-proof.
+        $self->{unpack_code} = 'V';  # Little-endian.
+      } elsif (substr($str, 0, 4) eq chr(0)x4) {
+        $self->{unpack_code} = 'N';  # Big-endian
+      } else {
+        ::error("$fname: header size >= 2**32\n");
+      }
+      my @pair = unpack($self->{unpack_code} . "*", $str);
+      # Since we know one of the pair is 0, it's fine to just add them.
+      @$slots = (0, $pair[0] + $pair[1]);
+    }
+    return $self;
+  }
+
+  # Load more data when we access slots->get(X) which is not yet in memory.
+  sub overflow {
+    my ($self) = @_;
+    my $slots = $self->{slots};
+    $self->{base} += $#$slots + 1;   # skip over data we're replacing
+    my $str;
+    read($self->{file}, $str, $self->{stride});
+    if ($address_length == 8) {      # the 32-bit case
+      # This is the easy case: unpack provides 32-bit unpacking primitives.
+      @$slots = unpack($self->{unpack_code} . "*", $str);
+    } else {
+      # We need to unpack 32 bits at a time and combine.
+      my @b32_values = unpack($self->{unpack_code} . "*", $str);
+      my @b64_values = ();
+      for (my $i = 0; $i < $#b32_values; $i += 2) {
+        # TODO(csilvers): if this is a 32-bit perl, the math below
+        #    could end up in a too-large int, which perl will promote
+        #    to a double, losing necessary precision.  Deal with that.
+        #    Right now, we just die.
+        my ($lo, $hi) = ($b32_values[$i], $b32_values[$i+1]);
+        if ($self->{unpack_code} eq 'N') {    # big-endian
+          ($lo, $hi) = ($hi, $lo);
+        }
+        my $value = $lo + $hi * (2**32);
+        if (!$self->{perl_is_64bit} &&   # check value is exactly represented
+            (($value % (2**32)) != $lo || int($value / (2**32)) != $hi)) {
+          ::error("Need a 64-bit perl to process this 64-bit profile.\n");
+        }
+        push(@b64_values, $value);
+      }
+      @$slots = @b64_values;
+    }
+  }
+
+  # Access the i-th long in the file (logically), or -1 at EOF.
+  sub get {
+    my ($self, $idx) = @_;
+    my $slots = $self->{slots};
+    while ($#$slots >= 0) {
+      if ($idx < $self->{base}) {
+        # The only time we expect a reference to $slots[$i - something]
+        # after referencing $slots[$i] is reading the very first header.
+        # Since $stride > |header|, that shouldn't cause any lookback
+        # errors.  And everything after the header is sequential.
+        print STDERR "Unexpected look-back reading CPU profile";
+        return -1;   # shrug, don't know what better to return
+      } elsif ($idx > $self->{base} + $#$slots) {
+        $self->overflow();
+      } else {
+        return $slots->[$idx - $self->{base}];
+      }
+    }
+    # If we get here, $slots is [], which means we've reached EOF
+    return -1;  # unique since slots is supposed to hold unsigned numbers
+  }
+}
+
+# Reads the top, 'header' section of a profile, and returns the last
+# line of the header, commonly called a 'header line'.  The header
+# section of a profile consists of zero or more 'command' lines that
+# are instructions to jeprof, which jeprof executes when reading the
+# header.  All 'command' lines start with a %.  After the command
+# lines is the 'header line', which is a profile-specific line that
+# indicates what type of profile it is, and perhaps other global
+# information about the profile.  For instance, here's a header line
+# for a heap profile:
+#   heap profile:     53:    38236 [  5525:  1284029] @ heapprofile
+# For historical reasons, the CPU profile does not contain a text-
+# readable header line.  If the profile looks like a CPU profile,
+# this function returns "".  If no header line could be found, this
+# function returns undef.
+#
+# The following commands are recognized:
+#   %warn -- emit the rest of this line to stderr, prefixed by 'WARNING:'
+#
+# The input file should be in binmode.
+sub ReadProfileHeader {
+  local *PROFILE = shift;
+  my $firstchar = "";
+  my $line = "";
+  read(PROFILE, $firstchar, 1);
+  seek(PROFILE, -1, 1);                    # unread the firstchar
+  if ($firstchar !~ /[[:print:]]/) {       # is not a text character
+    return "";
+  }
+  while (defined($line = <PROFILE>)) {
+    $line =~ s/\r//g;   # turn windows-looking lines into unix-looking lines
+    if ($line =~ /^%warn\s+(.*)/) {        # 'warn' command
+      # Note this matches both '%warn blah\n' and '%warn\n'.
+      print STDERR "WARNING: $1\n";        # print the rest of the line
+    } elsif ($line =~ /^%/) {
+      print STDERR "Ignoring unknown command from profile header: $line";
+    } else {
+      # End of commands, must be the header line.
+      return $line;
+    }
+  }
+  return undef;     # got to EOF without seeing a header line
+}
+
+sub IsSymbolizedProfileFile {
+  my $file_name = shift;
+  if (!(-e $file_name) || !(-r $file_name)) {
+    return 0;
+  }
+  # Check if the file contains a symbol-section marker.
+  open(TFILE, "<$file_name");
+  binmode TFILE;
+  my $firstline = ReadProfileHeader(*TFILE);
+  close(TFILE);
+  if (!$firstline) {
+    return 0;
+  }
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+  return $firstline =~ /^--- *$symbol_marker/;
+}
+
+# Parse profile generated by common/profiler.cc and return a reference
+# to a map:
+#      $result->{version}     Version number of profile file
+#      $result->{period}      Sampling period (in microseconds)
+#      $result->{profile}     Profile object
+#      $result->{threads}     Map of thread IDs to profile objects
+#      $result->{map}         Memory map info from profile
+#      $result->{pcs}         Hash of all PC values seen, key is hex address
+sub ReadProfile {
+  my $prog = shift;
+  my $fname = shift;
+  my $result;            # return value
+
+  $CONTENTION_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $contention_marker = $&;
+  $GROWTH_PAGE  =~ m,[^/]+$,;    # matches everything after the last slash
+  my $growth_marker = $&;
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+  $PROFILE_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $profile_marker = $&;
+  $HEAP_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $heap_marker = $&;
+
+  # Look at first line to see if it is a heap or a CPU profile.
+  # CPU profile may start with no header at all, and just binary data
+  # (starting with \0\0\0\0) -- in that case, don't try to read the
+  # whole firstline, since it may be gigabytes(!) of data.
+  open(PROFILE, "<$fname") || error("$fname: $!\n");
+  binmode PROFILE;      # New perls do UTF-8 processing
+  my $header = ReadProfileHeader(*PROFILE);
+  if (!defined($header)) {   # means "at EOF"
+    error("Profile is empty.\n");
+  }
+
+  my $symbols;
+  if ($header =~ m/^--- *$symbol_marker/o) {
+    # Verify that the user asked for a symbolized profile
+    if (!$main::use_symbolized_profile) {
+      # we have both a binary and symbolized profiles, abort
+      error("FATAL ERROR: Symbolized profile\n   $fname\ncannot be used with " .
+            "a binary arg. Try again without passing\n   $prog\n");
+    }
+    # Read the symbol section of the symbolized profile file.
+    $symbols = ReadSymbols(*PROFILE{IO});
+    # Read the next line to get the header for the remaining profile.
+    $header = ReadProfileHeader(*PROFILE) || "";
+  }
+
+  if ($header =~ m/^--- *($heap_marker|$growth_marker)/o) {
+    # Skip "--- ..." line for profile types that have their own headers.
+    $header = ReadProfileHeader(*PROFILE) || "";
+  }
+
+  $main::profile_type = '';
+
+  if ($header =~ m/^heap profile:.*$growth_marker/o) {
+    $main::profile_type = 'growth';
+    $result =  ReadHeapProfile($prog, *PROFILE, $header);
+  } elsif ($header =~ m/^heap profile:/) {
+    $main::profile_type = 'heap';
+    $result =  ReadHeapProfile($prog, *PROFILE, $header);
+  } elsif ($header =~ m/^heap/) {
+    $main::profile_type = 'heap';
+    $result = ReadThreadedHeapProfile($prog, $fname, $header);
+  } elsif ($header =~ m/^--- *$contention_marker/o) {
+    $main::profile_type = 'contention';
+    $result = ReadSynchProfile($prog, *PROFILE);
+  } elsif ($header =~ m/^--- *Stacks:/) {
+    print STDERR
+      "Old format contention profile: mistakenly reports " .
+      "condition variable signals as lock contentions.\n";
+    $main::profile_type = 'contention';
+    $result = ReadSynchProfile($prog, *PROFILE);
+  } elsif ($header =~ m/^--- *$profile_marker/) {
+    # the binary cpu profile data starts immediately after this line
+    $main::profile_type = 'cpu';
+    $result = ReadCPUProfile($prog, $fname, *PROFILE);
+  } else {
+    if (defined($symbols)) {
+      # a symbolized profile contains a format we don't recognize, bail out
+      error("$fname: Cannot recognize profile section after symbols.\n");
+    }
+    # no ascii header present -- must be a CPU profile
+    $main::profile_type

[truncated 1061424 bytes]

ready patch attempt

patch

redis-check-rdb likely remains stuck in a busy-poll loop. A diagnosis report and patch proposal were created locally.

state: readycreated: 2026-05-10 04:07 UTCvalidation: ready
Published session

Prompt

## Plan Pass

You are planning a fixer patch before any edits happen.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. The original pre-edit snapshot is available at `./source` if you need to inspect it. For interpreter processes, plan from the script/application entrypoint evidence first and include the runtime only as a second investigation target unless the evidence proves a runtime bug.

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround. Inspect the relevant code, nearby callers, project contribution docs, and local helper/compat APIs, but do not edit files in this pass.

Return a short markdown plan with these exact sections:

## Problem
## Evidence Confidence
## Proposed Subject
## Patch Plan
## Risks
## Validation

Classify `## Evidence Confidence` as exactly one of `reproduced`, `observed`, or `inferred`. Use `inferred` only for a no-patch diagnosis/report plan unless you can name the extra evidence you will collect before editing; inferred source patches are blocked by Fixer because they are not pull-request-ready. For `observed` source-patch plans, plan to say in the final `## Issue Connection` that the failure was observed by Fixer and not independently reproduced. The plan must explain how the proposed code change addresses the observed issue evidence, call out any prior Fixer patch that should be improved or replaced, reject awkward control flow such as avoidable `goto` if there is a cleaner bounded alternative, name any local helper APIs or maintainer conventions the patch should follow, and keep the intended maintainer-facing explanation clear enough that someone unfamiliar with the local complaint wording can still follow the fix. In `## Validation`, name the reproducible configure/build/test entrypoint you will try from the workspace root before any focused leaf compile or smoke check, and include one bounded independent reproduction attempt for the collected failure signal when it is safe and cheap. Do not plan to claim `reproduced` unless that reproduction command or test can actually show the failure.

## Patch Pass

You are working on a bounded fixer proposal.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. Produce the smallest reasonable patch for the target repository, keep the change upstreamable, prefer the clearest control flow available, and do not keep avoidable `goto` when a simpler structure would read better. Before introducing new file, process, allocation, locking, networking, or platform APIs, inspect nearby code and project contribution docs for existing helpers or compatibility wrappers and use those local patterns unless you can explain why they do not fit. Validate from a reproducible workspace-root entrypoint before falling back to focused leaf commands; if a build or test cannot run, report the exact command, the exact blocker, and any narrower check you ran instead. During validation, also try one bounded independent reproduction of the collected failure signal when it is safe and cheap, such as a failing test, smoke command, perf/strace comparison, or before/after runtime check. Only use `reproduced` if that command or test actually reproduced the failure; otherwise keep `observed` and report the reproduction blocker. The final explanation must connect the observed issue evidence to the actual code change, not just paraphrase the diff. Write like a maintainer is going to read the patch mail cold: explain the bug in plain language, define subsystem-specific jargon the first time you need it, and make the causal story obvious. Explicitly classify evidence confidence as `reproduced`, `observed`, or `inferred`: `reproduced` means you reproduced the failure locally; `observed` means Fixer has direct crash/log/trace evidence but you did not independently reproduce it; `inferred` means the source patch is not pull-request-ready, so do not leave a source diff unless you first gather stronger observed/reproduced evidence; otherwise return a no-patch diagnosis/report. For any source-changing `observed` patch, say explicitly in `## Issue Connection` that the failure was observed by Fixer and not independently reproduced. If you introduce non-obvious state translation, index remapping, or backend split logic, add a short source comment that explains the invariant being preserved.

Start by explaining the likely root cause from the collected perf, strace, and /proc evidence. If you cannot land a safe patch, leave a diagnosis that is strong enough for an upstream bug report.

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround. 

Keep the change narrowly scoped and summarize validation clearly.

In every authoring pass, your final response must start with `Subject: <single-line git commit subject>` and then include these markdown sections exactly:

## Commit Message
A short upstream-friendly explanation of what changed and why. Write it in plain language that a maintainer can follow without local complaint context. If you use subsystem jargon, define it immediately.

## Evidence Confidence
Exactly one word: `reproduced`, `observed`, or `inferred`. Use `reproduced` only when you reproduced the failure locally with a command or test, and include that command/test in `## Validation`. Use `observed` when Fixer has direct crash/log/trace evidence but you did not independently reproduce it. If `## Git Add Paths` lists source files for an `observed` patch, `## Issue Connection` must explicitly say the failure was observed by Fixer and not independently reproduced. Use `inferred` for profiler/strace/indirect evidence; inferred responses may be no-patch diagnoses or reports, but inferred source patches are not pull-request-ready until stronger evidence is gathered.

## Issue Connection
Write this as maintainer-facing patch mail, not as local Fixer notes. Cover four things explicitly in readable sentences: the user-visible symptom or the exact collected signal, the code-level cause or the cautious inference from evidence, the specific change you made, and the expected effect. Do not invent a reproducer, command line, crash, or user-visible failure that is not present in the evidence bundle. If the evidence is direct-but-not-reproduced, say it was observed by Fixer and not independently reproduced. If the evidence is indirect and you did not gather stronger evidence, do not leave a source diff; write a no-patch diagnosis/report instead. Include an explicit effect sentence such as `The expected effect is ...`, `This should reduce ...`, or `This prevents ...` for source patches. If the logic is non-obvious in code, mention that you added a short explanatory comment.

## Git Add Paths
List the repo-relative paths that belong in the final patch, one per line. Use `None` only when you intentionally made no source changes. Include intentionally new files, and do not list generated build artifacts.

## Validation
List the checks you ran, or say clearly that you could not run them. Include the independent reproduction command/test and result when `## Evidence Confidence` is `reproduced`; if reproduction was attempted but blocked, name the exact blocker and keep confidence at `observed` or `inferred`.

Before editing, read the plan at `./plan-output.txt` and follow it unless the code proves part of it wrong. If you change course, say so explicitly in the final write-up instead of silently drifting from the plan.

## Review Pass 1

You are reviewing a freshly generated fixer patch.

Read the evidence bundle at `./evidence.json`. The prepared workspace is `./workspace` and it was acquired via `debian-source`. Review the first patch pass. The original pre-edit snapshot is available at `./source` for diffing.

Upstream-style expectation: before planning or editing, check for contribution/style docs (`CONTRIBUTING`, `HACKING`, `README-hacking`, `README.md`, `docs/`, `dev-docs/`) and scan the touched subsystem for local helpers. If the project has wrappers for file IO, path-relative IO, process spawning, memory allocation, logging, locking, or platform compatibility, prefer those wrappers over generic libc/std APIs. Do not invent a reproducer or user-visible failure that is not in the evidence bundle; if the evidence is profiler-only or indirect, describe it as a targeted mitigation or stop with a diagnosis instead of presenting a speculative patch as a confirmed bug fix. In the plan and final validation, name any such helper, convention, or evidence limit you found, or say that no relevant local helper was found. Treat this as a `redis` upstream patch, not just a Debian-local workaround.

Validation expectation: try the project-level build/test entrypoint from the workspace root before reporting only a focused leaf compile. Detected candidate(s): `make` and, if available, `make check`. If the project-level command fails because dependencies or generated files are missing, include the exact command and failure reason in `## Validation`, then run the narrowest relevant compile/test that is still reproducible from a clean checkout. The latest author response is at `./patch-output.txt`. Inspect the current code and changed paths like a strict code reviewer. Focus on correctness, regressions, maintainability, awkward control flow such as avoidable `goto`, missing validation, weak or non-gittable commit message text, and explanations that fail to connect the observed issue evidence to the code change. Also review the maintainer experience: the patch mail should be easy to accept upstream, the user-visible bug should be explained in plain language, subsystem-specific jargon should be defined when first used, and any non-obvious state translation or index remapping in code should have a short explanatory comment. Reject patches that omit `## Evidence Confidence`, use anything other than `reproduced`, `observed`, or `inferred`, or sound more certain than that evidence level allows. Reject patches that claim `reproduced` unless `## Validation` names the concrete reproduction command or test and its result. Reject patches that invent a reproducer, command line, crash, or user-visible failure not present in the evidence bundle; indirect profiler or strace evidence may justify a cautious mitigation, but the author must say it is indirect instead of presenting it as a confirmed upstream bug. Reject patches that introduce generic libc/std APIs when nearby code or project docs provide a local compat/helper API for the same job, unless the author explicitly justifies the exception. Reject validation that only reports a leaf object/syntax build when a project-level configure/build/test entrypoint exists and was not attempted; if a full build cannot run, the author must show the exact failed command and blocker.

Do not apply code changes in this pass.

Return a short markdown review report. The first non-empty line must be exactly one of:

RESULT: ok
RESULT: fix-needed

If you choose `RESULT: fix-needed`, add a `## Findings` section with concrete, actionable items.

Response

## Plan Pass

## Problem
The bundle reports `redis-check-rdb` as “stuck in a likely busy poll loop,” but the embedded evidence shows a `redis-server` process in `S (sleeping)` state with `wchan=do_epoll_wait`, about `0.3%` CPU, and `epoll_wait(..., 100)` repeatedly timing out before `openat/read/close` of `/proc/self/stat`. In Redis 8.6.2 this matches normal idle server cron behavior: `serverCron()` runs at default `hz=10`, `cronUpdateMemoryStats()` runs under `run_with_period(100)`, and `zmalloc_get_rss()` reads `/proc/self/stat`.

## Evidence Confidence
inferred

## Proposed Subject
No upstream source patch: diagnose the evidence as normal idle Redis server cron/RSS sampling, not a demonstrated `redis-check-rdb` busy-loop bug.

## Patch Plan
Do not edit files in this pass. The available evidence is profiler-only and internally mismatched: package artifact says `redis-check-rdb`, while command line, status, and thread names identify `redis-server`.

If extra evidence is required before any source patch, collect a bounded independent trace of an idle locally built `src/redis-server` and confirm whether it shows the same `epoll_wait` plus `/proc/self/stat` cadence at low CPU. A real patch should only follow if that reproduction shows abnormal CPU or an avoidable tight loop.

Relevant local conventions/helpers found:
`aeMain()`/`aeProcessEvents()` drive the event loop, `serverCron()` and `run_with_period()` throttle periodic work, and `zmalloc_get_rss()`/`get_proc_stat_ll()` are the existing platform helper path for RSS sampling. A patch should use those helpers rather than generic replacement IO. No prior Fixer patch is present in the workspace diff or Debian patch set, so there is nothing to improve or replace.

Avoid awkward control flow such as adding `goto` around cron sampling; if future evidence proves over-sampling, prefer a bounded local throttle or reuse of cached stats in `cronUpdateMemoryStats()`.

## Risks
Changing the cron/RSS path without stronger evidence could reduce freshness of `INFO memory` metrics, fragmentation reporting, or allocator diagnostics. Changing `redis-check-rdb` would not address the observed trace because the sampled process is actually `redis-server`.

## Validation
Before reporting any source patch, try project-level validation from the workspace root: `make`, then `make check` if available. If those fail due missing dependencies or generated files, record the exact command and failure.

For the independent evidence check, use a locally built idle `src/redis-server` with a short `strace`/CPU sample and compare whether the same 100 ms `epoll_wait` plus `/proc/self/stat` pattern appears at low CPU. Do not claim `reproduced` unless that command demonstrates the reported runaway behavior, not just normal idle cron activity.

## Patch Pass

Subject: diagnose redis idle cron sampling trace

## Commit Message
No source change is proposed. The collected signal points at an idle `redis-server` event loop, not a `redis-check-rdb` bug: Redis wakes its cron timer about every 100 ms and samples memory through the existing `/proc/self/stat` helper.

## Evidence Confidence
inferred

## Issue Connection
The collected signal is a profiler/strace-style report labeled as `redis-check-rdb`, but the embedded `/proc` status, command line, and thread names identify the sampled process as `redis-server`. The trace shows `epoll_wait` timing out around every 100 ms, followed by reads of `/proc/self/stat`, while the process is sleeping and using about `0.3%` CPU.

The likely root cause is normal Redis server cron activity, not a runaway checker loop. In Redis, `serverCron()` is the periodic maintenance callback, `run_with_period(100)` schedules 100 ms work, and `cronUpdateMemoryStats()` calls `zmalloc_get_rss()`, which uses `get_proc_stat_ll()` to read `/proc/self/stat`.

I followed the saved plan and made no source change because the evidence is indirect and internally mismatched. The relevant local helpers/conventions are `aeMain()`/`aeProcessEvents()` for the event loop, `serverCron()`/`run_with_period()` for periodic work, and `zmalloc_get_rss()`/`get_proc_stat_ll()` for RSS sampling. There is no prior Fixer source patch to improve or replace.

The expected effect is that no speculative upstream diff is produced until stronger evidence shows abnormal CPU use or an actual tight loop distinct from normal idle cron sampling.

## Git Add Paths
None

## Validation
`make` from the workspace root completed successfully.

`make check` from the workspace root failed before running tests because the Redis test harness could not find a free port:
`Can't find a non busy port in the 21079-21110 range.`

Independent reproduction attempts were blocked by the sandbox:
`strace` attempt failed with `PTRACE_TRACEME: Operation not permitted` and `ptrace(PTRACE_SEIZE, 7): Operation not permitted`.
A local idle `redis-server` smoke run also could not stay up because socket creation/bind is blocked: Unix socket failed with `setsockopt SO_REUSEADDR: Operation not permitted`, and TCP bind failed with `unable to bind socket, errno: 1`.

Because the failure signal could not be independently reproduced and the available evidence remains profiler/trace inference, confidence stays `inferred` and no source diff was left.

## Review Pass 1

RESULT: ok

No blocking findings.

The workspace has no source diff, `## Evidence Confidence` is valid (`inferred`), and the response correctly avoids presenting indirect profiler/strace evidence as a confirmed Redis bug. The diagnosis is consistent with the evidence mismatch: artifact label points at `redis-check-rdb`, while status/command/thread data show `redis-server` sleeping in `epoll_wait` with periodic `/proc/self/stat` reads.

Validation reporting is acceptable: project-level `make` was attempted and passed; `make check` was attempted and the exact harness blocker was reported; independent reproduction attempts and sandbox blockers were named. The response also identifies relevant Redis helpers/conventions (`serverCron`, `run_with_period`, `zmalloc_get_rss`, `get_proc_stat_ll`) and leaves `Git Add Paths` as `None`, matching the empty diff.

Diff

--- a/deps/jemalloc/bin/jemalloc-config	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jemalloc-config	2026-05-10 08:04:46.530382678 +0400
@@ -0,0 +1,83 @@
+#!/bin/sh
+
+usage() {
+	cat <<EOF
+Usage:
+  /usr/local/bin/jemalloc-config <option>
+Options:
+  --help | -h  : Print usage.
+  --version    : Print jemalloc version.
+  --revision   : Print shared library revision number.
+  --config     : Print configure options used to build jemalloc.
+  --prefix     : Print installation directory prefix.
+  --bindir     : Print binary installation directory.
+  --datadir    : Print data installation directory.
+  --includedir : Print include installation directory.
+  --libdir     : Print library installation directory.
+  --mandir     : Print manual page installation directory.
+  --cc         : Print compiler used to build jemalloc.
+  --cflags     : Print compiler flags used to build jemalloc.
+  --cppflags   : Print preprocessor flags used to build jemalloc.
+  --cxxflags   : Print C++ compiler flags used to build jemalloc.
+  --ldflags    : Print library flags used to build jemalloc.
+  --libs       : Print libraries jemalloc was linked against.
+EOF
+}
+
+prefix="/usr/local"
+exec_prefix="/usr/local"
+
+case "$1" in
+--help | -h)
+	usage
+	exit 0
+	;;
+--version)
+	echo "5.3.0-0-g0"
+	;;
+--revision)
+	echo "2"
+	;;
+--config)
+	echo "--disable-cxx --with-version=5.3.0-0-g0 --with-lg-quantum=3 --disable-cache-oblivious --with-jemalloc-prefix=je_ CFLAGS= LDFLAGS="
+	;;
+--prefix)
+	echo "/usr/local"
+	;;
+--bindir)
+	echo "/usr/local/bin"
+	;;
+--datadir)
+	echo "/usr/local/share"
+	;;
+--includedir)
+	echo "/usr/local/include"
+	;;
+--libdir)
+	echo "/usr/local/lib"
+	;;
+--mandir)
+	echo "/usr/local/share/man"
+	;;
+--cc)
+	echo "gcc"
+	;;
+--cflags)
+	echo "-std=gnu11 -Wall -Wextra -Wsign-compare -Wundef -Wno-format-zero-length -Wpointer-arith -Wno-missing-braces -Wno-missing-field-initializers -Wno-missing-attributes -pipe -g3 -fvisibility=hidden -Wimplicit-fallthrough -O3 -funroll-loops"
+	;;
+--cppflags)
+	echo "-D_GNU_SOURCE -D_REENTRANT"
+	;;
+--cxxflags)
+	echo ""
+	;;
+--ldflags)
+	echo " "
+	;;
+--libs)
+	echo "-lm  -pthread"
+	;;
+*)
+	usage
+	exit 1
+esac
--- a/deps/jemalloc/bin/jemalloc.sh	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jemalloc.sh	2026-05-10 08:04:46.535115161 +0400
@@ -0,0 +1,9 @@
+#!/bin/sh
+
+prefix=/usr/local
+exec_prefix=/usr/local
+libdir=${exec_prefix}/lib
+
+LD_PRELOAD=${libdir}/libjemalloc.so.2
+export LD_PRELOAD
+exec "$@"
--- a/deps/jemalloc/bin/jeprof	1970-01-01 04:00:00.000000000 +0400
+++ b/deps/jemalloc/bin/jeprof	2026-05-10 08:04:46.540707405 +0400
@@ -0,0 +1,5723 @@
+#! /usr/bin/env perl
+
+# Copyright (c) 1998-2007, Google Inc.
+# All rights reserved.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met:
+#
+#     * Redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer.
+#     * Redistributions in binary form must reproduce the above
+# copyright notice, this list of conditions and the following disclaimer
+# in the documentation and/or other materials provided with the
+# distribution.
+#     * Neither the name of Google Inc. nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+# ---
+# Program for printing the profile generated by common/profiler.cc,
+# or by the heap profiler (common/debugallocation.cc)
+#
+# The profile contains a sequence of entries of the form:
+#       <count> <stack trace>
+# This program parses the profile, and generates user-readable
+# output.
+#
+# Examples:
+#
+# % tools/jeprof "program" "profile"
+#   Enters "interactive" mode
+#
+# % tools/jeprof --text "program" "profile"
+#   Generates one line per procedure
+#
+# % tools/jeprof --gv "program" "profile"
+#   Generates annotated call-graph and displays via "gv"
+#
+# % tools/jeprof --gv --focus=Mutex "program" "profile"
+#   Restrict to code paths that involve an entry that matches "Mutex"
+#
+# % tools/jeprof --gv --focus=Mutex --ignore=string "program" "profile"
+#   Restrict to code paths that involve an entry that matches "Mutex"
+#   and does not match "string"
+#
+# % tools/jeprof --list=IBF_CheckDocid "program" "profile"
+#   Generates disassembly listing of all routines with at least one
+#   sample that match the --list=<regexp> pattern.  The listing is
+#   annotated with the flat and cumulative sample counts at each line.
+#
+# % tools/jeprof --disasm=IBF_CheckDocid "program" "profile"
+#   Generates disassembly listing of all routines with at least one
+#   sample that match the --disasm=<regexp> pattern.  The listing is
+#   annotated with the flat and cumulative sample counts at each PC value.
+#
+# TODO: Use color to indicate files?
+
+use strict;
+use warnings;
+use Getopt::Long;
+use Cwd;
+
+my $JEPROF_VERSION = "5.3.0-0-g0";
+my $PPROF_VERSION = "2.0";
+
+# These are the object tools we use which can come from a
+# user-specified location using --tools, from the JEPROF_TOOLS
+# environment variable, or from the environment.
+my %obj_tool_map = (
+  "objdump" => "objdump",
+  "nm" => "nm",
+  "addr2line" => "addr2line",
+  "c++filt" => "c++filt",
+  ## ConfigureObjTools may add architecture-specific entries:
+  #"nm_pdb" => "nm-pdb",       # for reading windows (PDB-format) executables
+  #"addr2line_pdb" => "addr2line-pdb",                                # ditto
+  #"otool" => "otool",         # equivalent of objdump on OS X
+);
+# NOTE: these are lists, so you can put in commandline flags if you want.
+my @DOT = ("dot");          # leave non-absolute, since it may be in /usr/local
+my @GV = ("gv");
+my @EVINCE = ("evince");    # could also be xpdf or perhaps acroread
+my @KCACHEGRIND = ("kcachegrind");
+my @PS2PDF = ("ps2pdf");
+# These are used for dynamic profiles
+my @URL_FETCHER = ("curl", "-s", "--fail");
+
+# These are the web pages that servers need to support for dynamic profiles
+my $HEAP_PAGE = "/pprof/heap";
+my $PROFILE_PAGE = "/pprof/profile";   # must support cgi-param "?seconds=#"
+my $PMUPROFILE_PAGE = "/pprof/pmuprofile(?:\\?.*)?"; # must support cgi-param
+                                                # ?seconds=#&event=x&period=n
+my $GROWTH_PAGE = "/pprof/growth";
+my $CONTENTION_PAGE = "/pprof/contention";
+my $WALL_PAGE = "/pprof/wall(?:\\?.*)?";  # accepts options like namefilter
+my $FILTEREDPROFILE_PAGE = "/pprof/filteredprofile(?:\\?.*)?";
+my $CENSUSPROFILE_PAGE = "/pprof/censusprofile(?:\\?.*)?"; # must support cgi-param
+                                                       # "?seconds=#",
+                                                       # "?tags_regexp=#" and
+                                                       # "?type=#".
+my $SYMBOL_PAGE = "/pprof/symbol";     # must support symbol lookup via POST
+my $PROGRAM_NAME_PAGE = "/pprof/cmdline";
+
+# These are the web pages that can be named on the command line.
+# All the alternatives must begin with /.
+my $PROFILES = "($HEAP_PAGE|$PROFILE_PAGE|$PMUPROFILE_PAGE|" .
+               "$GROWTH_PAGE|$CONTENTION_PAGE|$WALL_PAGE|" .
+               "$FILTEREDPROFILE_PAGE|$CENSUSPROFILE_PAGE)";
+
+# default binary name
+my $UNKNOWN_BINARY = "(unknown)";
+
+# There is a pervasive dependency on the length (in hex characters,
+# i.e., nibbles) of an address, distinguishing between 32-bit and
+# 64-bit profiles.  To err on the safe size, default to 64-bit here:
+my $address_length = 16;
+
+my $dev_null = "/dev/null";
+if (! -e $dev_null && $^O =~ /MSWin/) {    # $^O is the OS perl was built for
+  $dev_null = "nul";
+}
+
+# A list of paths to search for shared object files
+my @prefix_list = ();
+
+# Special routine name that should not have any symbols.
+# Used as separator to parse "addr2line -i" output.
+my $sep_symbol = '_fini';
+my $sep_address = undef;
+
+##### Argument parsing #####
+
+sub usage_string {
+  return <<EOF;
+Usage:
+jeprof [options] <program> <profiles>
+   <profiles> is a space separated list of profile names.
+jeprof [options] <symbolized-profiles>
+   <symbolized-profiles> is a list of profile files where each file contains
+   the necessary symbol mappings  as well as profile data (likely generated
+   with --raw).
+jeprof [options] <profile>
+   <profile> is a remote form.  Symbols are obtained from host:port$SYMBOL_PAGE
+
+   Each name can be:
+   /path/to/profile        - a path to a profile file
+   host:port[/<service>]   - a location of a service to get profile from
+
+   The /<service> can be $HEAP_PAGE, $PROFILE_PAGE, /pprof/pmuprofile,
+                         $GROWTH_PAGE, $CONTENTION_PAGE, /pprof/wall,
+                         $CENSUSPROFILE_PAGE, or /pprof/filteredprofile.
+   For instance:
+     jeprof http://myserver.com:80$HEAP_PAGE
+   If /<service> is omitted, the service defaults to $PROFILE_PAGE (cpu profiling).
+jeprof --symbols <program>
+   Maps addresses to symbol names.  In this mode, stdin should be a
+   list of library mappings, in the same format as is found in the heap-
+   and cpu-profile files (this loosely matches that of /proc/self/maps
+   on linux), followed by a list of hex addresses to map, one per line.
+
+   For more help with querying remote servers, including how to add the
+   necessary server-side support code, see this filename (or one like it):
+
+   /usr/doc/gperftools-$PPROF_VERSION/pprof_remote_servers.html
+
+Options:
+   --cum               Sort by cumulative data
+   --base=<base>       Subtract <base> from <profile> before display
+   --interactive       Run in interactive mode (interactive "help" gives help) [default]
+   --seconds=<n>       Length of time for dynamic profiles [default=30 secs]
+   --add_lib=<file>    Read additional symbols and line info from the given library
+   --lib_prefix=<dir>  Comma separated list of library path prefixes
+
+Reporting Granularity:
+   --addresses         Report at address level
+   --lines             Report at source line level
+   --functions         Report at function level [default]
+   --files             Report at source file level
+
+Output type:
+   --text              Generate text report
+   --callgrind         Generate callgrind format to stdout
+   --gv                Generate Postscript and display
+   --evince            Generate PDF and display
+   --web               Generate SVG and display
+   --list=<regexp>     Generate source listing of matching routines
+   --disasm=<regexp>   Generate disassembly of matching routines
+   --symbols           Print demangled symbol names found at given addresses
+   --dot               Generate DOT file to stdout
+   --ps                Generate Postcript to stdout
+   --pdf               Generate PDF to stdout
+   --svg               Generate SVG to stdout
+   --gif               Generate GIF to stdout
+   --raw               Generate symbolized jeprof data (useful with remote fetch)
+   --collapsed         Generate collapsed stacks for building flame graphs
+                       (see http://www.brendangregg.com/flamegraphs.html)
+
+Heap-Profile Options:
+   --inuse_space       Display in-use (mega)bytes [default]
+   --inuse_objects     Display in-use objects
+   --alloc_space       Display allocated (mega)bytes
+   --alloc_objects     Display allocated objects
+   --show_bytes        Display space in bytes
+   --drop_negative     Ignore negative differences
+
+Contention-profile options:
+   --total_delay       Display total delay at each region [default]
+   --contentions       Display number of delays at each region
+   --mean_delay        Display mean delay at each region
+
+Call-graph Options:
+   --nodecount=<n>     Show at most so many nodes [default=80]
+   --nodefraction=<f>  Hide nodes below <f>*total [default=.005]
+   --edgefraction=<f>  Hide edges below <f>*total [default=.001]
+   --maxdegree=<n>     Max incoming/outgoing edges per node [default=8]
+   --focus=<regexp>    Focus on backtraces with nodes matching <regexp>
+   --thread=<n>        Show profile for thread <n>
+   --ignore=<regexp>   Ignore backtraces with nodes matching <regexp>
+   --scale=<n>         Set GV scaling [default=0]
+   --heapcheck         Make nodes with non-0 object counts
+                       (i.e. direct leak generators) more visible
+   --retain=<regexp>   Retain only nodes that match <regexp>
+   --exclude=<regexp>  Exclude all nodes that match <regexp>
+
+Miscellaneous:
+   --tools=<prefix or binary:fullpath>[,...]   \$PATH for object tool pathnames
+   --test              Run unit tests
+   --help              This message
+   --version           Version information
+   --debug-syms-by-id  (Linux only) Find debug symbol files by build ID as well as by name
+
+Environment Variables:
+   JEPROF_TMPDIR        Profiles directory. Defaults to \$HOME/jeprof
+   JEPROF_TOOLS         Prefix for object tools pathnames
+
+Examples:
+
+jeprof /bin/ls ls.prof
+                       Enters "interactive" mode
+jeprof --text /bin/ls ls.prof
+                       Outputs one line per procedure
+jeprof --web /bin/ls ls.prof
+                       Displays annotated call-graph in web browser
+jeprof --gv /bin/ls ls.prof
+                       Displays annotated call-graph via 'gv'
+jeprof --gv --focus=Mutex /bin/ls ls.prof
+                       Restricts to code paths including a .*Mutex.* entry
+jeprof --gv --focus=Mutex --ignore=string /bin/ls ls.prof
+                       Code paths including Mutex but not string
+jeprof --list=getdir /bin/ls ls.prof
+                       (Per-line) annotated source listing for getdir()
+jeprof --disasm=getdir /bin/ls ls.prof
+                       (Per-PC) annotated disassembly for getdir()
+
+jeprof http://localhost:1234/
+                       Enters "interactive" mode
+jeprof --text localhost:1234
+                       Outputs one line per procedure for localhost:1234
+jeprof --raw localhost:1234 > ./local.raw
+jeprof --text ./local.raw
+                       Fetches a remote profile for later analysis and then
+                       analyzes it in text mode.
+EOF
+}
+
+sub version_string {
+  return <<EOF
+jeprof (part of jemalloc $JEPROF_VERSION)
+based on pprof (part of gperftools $PPROF_VERSION)
+
+Copyright 1998-2007 Google Inc.
+
+This is BSD licensed software; see the source for copying conditions
+and license information.
+There is NO warranty; not even for MERCHANTABILITY or FITNESS FOR A
+PARTICULAR PURPOSE.
+EOF
+}
+
+sub usage {
+  my $msg = shift;
+  print STDERR "$msg\n\n";
+  print STDERR usage_string();
+  print STDERR "\nFATAL ERROR: $msg\n";    # just as a reminder
+  exit(1);
+}
+
+sub Init() {
+  # Setup tmp-file name and handler to clean it up.
+  # We do this in the very beginning so that we can use
+  # error() and cleanup() function anytime here after.
+  $main::tmpfile_sym = "/tmp/jeprof$$.sym";
+  $main::tmpfile_ps = "/tmp/jeprof$$";
+  $main::next_tmpfile = 0;
+  $SIG{'INT'} = \&sighandler;
+
+  # Cache from filename/linenumber to source code
+  $main::source_cache = ();
+
+  $main::opt_help = 0;
+  $main::opt_version = 0;
+
+  $main::opt_cum = 0;
+  $main::opt_base = '';
+  $main::opt_addresses = 0;
+  $main::opt_lines = 0;
+  $main::opt_functions = 0;
+  $main::opt_files = 0;
+  $main::opt_lib_prefix = "";
+
+  $main::opt_text = 0;
+  $main::opt_callgrind = 0;
+  $main::opt_list = "";
+  $main::opt_disasm = "";
+  $main::opt_symbols = 0;
+  $main::opt_gv = 0;
+  $main::opt_evince = 0;
+  $main::opt_web = 0;
+  $main::opt_dot = 0;
+  $main::opt_ps = 0;
+  $main::opt_pdf = 0;
+  $main::opt_gif = 0;
+  $main::opt_svg = 0;
+  $main::opt_raw = 0;
+  $main::opt_collapsed = 0;
+
+  $main::opt_nodecount = 80;
+  $main::opt_nodefraction = 0.005;
+  $main::opt_edgefraction = 0.001;
+  $main::opt_maxdegree = 8;
+  $main::opt_focus = '';
+  $main::opt_thread = undef;
+  $main::opt_ignore = '';
+  $main::opt_scale = 0;
+  $main::opt_heapcheck = 0;
+  $main::opt_retain = '';
+  $main::opt_exclude = '';
+  $main::opt_seconds = 30;
+  $main::opt_lib = "";
+
+  $main::opt_inuse_space   = 0;
+  $main::opt_inuse_objects = 0;
+  $main::opt_alloc_space   = 0;
+  $main::opt_alloc_objects = 0;
+  $main::opt_show_bytes    = 0;
+  $main::opt_drop_negative = 0;
+  $main::opt_interactive   = 0;
+
+  $main::opt_total_delay = 0;
+  $main::opt_contentions = 0;
+  $main::opt_mean_delay = 0;
+
+  $main::opt_tools   = "";
+  $main::opt_debug   = 0;
+  $main::opt_test    = 0;
+  $main::opt_debug_syms_by_id = 0;
+
+  # These are undocumented flags used only by unittests.
+  $main::opt_test_stride = 0;
+
+  # Are we using $SYMBOL_PAGE?
+  $main::use_symbol_page = 0;
+
+  # Files returned by TempName.
+  %main::tempnames = ();
+
+  # Type of profile we are dealing with
+  # Supported types:
+  #     cpu
+  #     heap
+  #     growth
+  #     contention
+  $main::profile_type = '';     # Empty type means "unknown"
+
+  GetOptions("help!"          => \$main::opt_help,
+             "version!"       => \$main::opt_version,
+             "cum!"           => \$main::opt_cum,
+             "base=s"         => \$main::opt_base,
+             "seconds=i"      => \$main::opt_seconds,
+             "add_lib=s"      => \$main::opt_lib,
+             "lib_prefix=s"   => \$main::opt_lib_prefix,
+             "functions!"     => \$main::opt_functions,
+             "lines!"         => \$main::opt_lines,
+             "addresses!"     => \$main::opt_addresses,
+             "files!"         => \$main::opt_files,
+             "text!"          => \$main::opt_text,
+             "callgrind!"     => \$main::opt_callgrind,
+             "list=s"         => \$main::opt_list,
+             "disasm=s"       => \$main::opt_disasm,
+             "symbols!"       => \$main::opt_symbols,
+             "gv!"            => \$main::opt_gv,
+             "evince!"        => \$main::opt_evince,
+             "web!"           => \$main::opt_web,
+             "dot!"           => \$main::opt_dot,
+             "ps!"            => \$main::opt_ps,
+             "pdf!"           => \$main::opt_pdf,
+             "svg!"           => \$main::opt_svg,
+             "gif!"           => \$main::opt_gif,
+             "raw!"           => \$main::opt_raw,
+             "collapsed!"     => \$main::opt_collapsed,
+             "interactive!"   => \$main::opt_interactive,
+             "nodecount=i"    => \$main::opt_nodecount,
+             "nodefraction=f" => \$main::opt_nodefraction,
+             "edgefraction=f" => \$main::opt_edgefraction,
+             "maxdegree=i"    => \$main::opt_maxdegree,
+             "focus=s"        => \$main::opt_focus,
+             "thread=s"       => \$main::opt_thread,
+             "ignore=s"       => \$main::opt_ignore,
+             "scale=i"        => \$main::opt_scale,
+             "heapcheck"      => \$main::opt_heapcheck,
+             "retain=s"       => \$main::opt_retain,
+             "exclude=s"      => \$main::opt_exclude,
+             "inuse_space!"   => \$main::opt_inuse_space,
+             "inuse_objects!" => \$main::opt_inuse_objects,
+             "alloc_space!"   => \$main::opt_alloc_space,
+             "alloc_objects!" => \$main::opt_alloc_objects,
+             "show_bytes!"    => \$main::opt_show_bytes,
+             "drop_negative!" => \$main::opt_drop_negative,
+             "total_delay!"   => \$main::opt_total_delay,
+             "contentions!"   => \$main::opt_contentions,
+             "mean_delay!"    => \$main::opt_mean_delay,
+             "tools=s"        => \$main::opt_tools,
+             "test!"          => \$main::opt_test,
+             "debug!"         => \$main::opt_debug,
+             "debug-syms-by-id!" => \$main::opt_debug_syms_by_id,
+             # Undocumented flags used only by unittests:
+             "test_stride=i"  => \$main::opt_test_stride,
+      ) || usage("Invalid option(s)");
+
+  # Deal with the standard --help and --version
+  if ($main::opt_help) {
+    print usage_string();
+    exit(0);
+  }
+
+  if ($main::opt_version) {
+    print version_string();
+    exit(0);
+  }
+
+  # Disassembly/listing/symbols mode requires address-level info
+  if ($main::opt_disasm || $main::opt_list || $main::opt_symbols) {
+    $main::opt_functions = 0;
+    $main::opt_lines = 0;
+    $main::opt_addresses = 1;
+    $main::opt_files = 0;
+  }
+
+  # Check heap-profiling flags
+  if ($main::opt_inuse_space +
+      $main::opt_inuse_objects +
+      $main::opt_alloc_space +
+      $main::opt_alloc_objects > 1) {
+    usage("Specify at most on of --inuse/--alloc options");
+  }
+
+  # Check output granularities
+  my $grains =
+      $main::opt_functions +
+      $main::opt_lines +
+      $main::opt_addresses +
+      $main::opt_files +
+      0;
+  if ($grains > 1) {
+    usage("Only specify one output granularity option");
+  }
+  if ($grains == 0) {
+    $main::opt_functions = 1;
+  }
+
+  # Check output modes
+  my $modes =
+      $main::opt_text +
+      $main::opt_callgrind +
+      ($main::opt_list eq '' ? 0 : 1) +
+      ($main::opt_disasm eq '' ? 0 : 1) +
+      ($main::opt_symbols == 0 ? 0 : 1) +
+      $main::opt_gv +
+      $main::opt_evince +
+      $main::opt_web +
+      $main::opt_dot +
+      $main::opt_ps +
+      $main::opt_pdf +
+      $main::opt_svg +
+      $main::opt_gif +
+      $main::opt_raw +
+      $main::opt_collapsed +
+      $main::opt_interactive +
+      0;
+  if ($modes > 1) {
+    usage("Only specify one output mode");
+  }
+  if ($modes == 0) {
+    if (-t STDOUT) {  # If STDOUT is a tty, activate interactive mode
+      $main::opt_interactive = 1;
+    } else {
+      $main::opt_text = 1;
+    }
+  }
+
+  if ($main::opt_test) {
+    RunUnitTests();
+    # Should not return
+    exit(1);
+  }
+
+  # Binary name and profile arguments list
+  $main::prog = "";
+  @main::pfile_args = ();
+
+  # Remote profiling without a binary (using $SYMBOL_PAGE instead)
+  if (@ARGV > 0) {
+    if (IsProfileURL($ARGV[0])) {
+      $main::use_symbol_page = 1;
+    } elsif (IsSymbolizedProfileFile($ARGV[0])) {
+      $main::use_symbolized_profile = 1;
+      $main::prog = $UNKNOWN_BINARY;  # will be set later from the profile file
+    }
+  }
+
+  if ($main::use_symbol_page || $main::use_symbolized_profile) {
+    # We don't need a binary!
+    my %disabled = ('--lines' => $main::opt_lines,
+                    '--disasm' => $main::opt_disasm);
+    for my $option (keys %disabled) {
+      usage("$option cannot be used without a binary") if $disabled{$option};
+    }
+    # Set $main::prog later...
+    scalar(@ARGV) || usage("Did not specify profile file");
+  } elsif ($main::opt_symbols) {
+    # --symbols needs a binary-name (to run nm on, etc) but not profiles
+    $main::prog = shift(@ARGV) || usage("Did not specify program");
+  } else {
+    $main::prog = shift(@ARGV) || usage("Did not specify program");
+    scalar(@ARGV) || usage("Did not specify profile file");
+  }
+
+  # Parse profile file/location arguments
+  foreach my $farg (@ARGV) {
+    if ($farg =~ m/(.*)\@([0-9]+)(|\/.*)$/ ) {
+      my $machine = $1;
+      my $num_machines = $2;
+      my $path = $3;
+      for (my $i = 0; $i < $num_machines; $i++) {
+        unshift(@main::pfile_args, "$i.$machine$path");
+      }
+    } else {
+      unshift(@main::pfile_args, $farg);
+    }
+  }
+
+  if ($main::use_symbol_page) {
+    unless (IsProfileURL($main::pfile_args[0])) {
+      error("The first profile should be a remote form to use $SYMBOL_PAGE\n");
+    }
+    CheckSymbolPage();
+    $main::prog = FetchProgramName();
+  } elsif (!$main::use_symbolized_profile) {  # may not need objtools!
+    ConfigureObjTools($main::prog)
+  }
+
+  # Break the opt_lib_prefix into the prefix_list array
+  @prefix_list = split (',', $main::opt_lib_prefix);
+
+  # Remove trailing / from the prefixes, in the list to prevent
+  # searching things like /my/path//lib/mylib.so
+  foreach (@prefix_list) {
+    s|/+$||;
+  }
+
+  # Flag to prevent us from trying over and over to use
+  #  elfutils if it's not installed (used only with
+  #  --debug-syms-by-id option).
+  $main::gave_up_on_elfutils = 0;
+}
+
+sub FilterAndPrint {
+  my ($profile, $symbols, $libs, $thread) = @_;
+
+  # Get total data in profile
+  my $total = TotalProfile($profile);
+
+  # Remove uniniteresting stack items
+  $profile = RemoveUninterestingFrames($symbols, $profile);
+
+  # Focus?
+  if ($main::opt_focus ne '') {
+    $profile = FocusProfile($symbols, $profile, $main::opt_focus);
+  }
+
+  # Ignore?
+  if ($main::opt_ignore ne '') {
+    $profile = IgnoreProfile($symbols, $profile, $main::opt_ignore);
+  }
+
+  my $calls = ExtractCalls($symbols, $profile);
+
+  # Reduce profiles to required output granularity, and also clean
+  # each stack trace so a given entry exists at most once.
+  my $reduced = ReduceProfile($symbols, $profile);
+
+  # Get derived profiles
+  my $flat = FlatProfile($reduced);
+  my $cumulative = CumulativeProfile($reduced);
+
+  # Print
+  if (!$main::opt_interactive) {
+    if ($main::opt_disasm) {
+      PrintDisassembly($libs, $flat, $cumulative, $main::opt_disasm);
+    } elsif ($main::opt_list) {
+      PrintListing($total, $libs, $flat, $cumulative, $main::opt_list, 0);
+    } elsif ($main::opt_text) {
+      # Make sure the output is empty when have nothing to report
+      # (only matters when --heapcheck is given but we must be
+      # compatible with old branches that did not pass --heapcheck always):
+      if ($total != 0) {
+        printf("Total%s: %s %s\n",
+               (defined($thread) ? " (t$thread)" : ""),
+               Unparse($total), Units());
+      }
+      PrintText($symbols, $flat, $cumulative, -1);
+    } elsif ($main::opt_raw) {
+      PrintSymbolizedProfile($symbols, $profile, $main::prog);
+    } elsif ($main::opt_collapsed) {
+      PrintCollapsedStacks($symbols, $profile);
+    } elsif ($main::opt_callgrind) {
+      PrintCallgrind($calls);
+    } else {
+      if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) {
+        if ($main::opt_gv) {
+          RunGV(TempName($main::next_tmpfile, "ps"), "");
+        } elsif ($main::opt_evince) {
+          RunEvince(TempName($main::next_tmpfile, "pdf"), "");
+        } elsif ($main::opt_web) {
+          my $tmp = TempName($main::next_tmpfile, "svg");
+          RunWeb($tmp);
+          # The command we run might hand the file name off
+          # to an already running browser instance and then exit.
+          # Normally, we'd remove $tmp on exit (right now),
+          # but fork a child to remove $tmp a little later, so that the
+          # browser has time to load it first.
+          delete $main::tempnames{$tmp};
+          if (fork() == 0) {
+            sleep 5;
+            unlink($tmp);
+            exit(0);
+          }
+        }
+      } else {
+        cleanup();
+        exit(1);
+      }
+    }
+  } else {
+    InteractiveMode($profile, $symbols, $libs, $total);
+  }
+}
+
+sub Main() {
+  Init();
+  $main::collected_profile = undef;
+  @main::profile_files = ();
+  $main::op_time = time();
+
+  # Printing symbols is special and requires a lot less info that most.
+  if ($main::opt_symbols) {
+    PrintSymbols(*STDIN);   # Get /proc/maps and symbols output from stdin
+    return;
+  }
+
+  # Fetch all profile data
+  FetchDynamicProfiles();
+
+  # this will hold symbols that we read from the profile files
+  my $symbol_map = {};
+
+  # Read one profile, pick the last item on the list
+  my $data = ReadProfile($main::prog, pop(@main::profile_files));
+  my $profile = $data->{profile};
+  my $pcs = $data->{pcs};
+  my $libs = $data->{libs};   # Info about main program and shared libraries
+  $symbol_map = MergeSymbols($symbol_map, $data->{symbols});
+
+  # Add additional profiles, if available.
+  if (scalar(@main::profile_files) > 0) {
+    foreach my $pname (@main::profile_files) {
+      my $data2 = ReadProfile($main::prog, $pname);
+      $profile = AddProfile($profile, $data2->{profile});
+      $pcs = AddPcs($pcs, $data2->{pcs});
+      $symbol_map = MergeSymbols($symbol_map, $data2->{symbols});
+    }
+  }
+
+  # Subtract base from profile, if specified
+  if ($main::opt_base ne '') {
+    my $base = ReadProfile($main::prog, $main::opt_base);
+    $profile = SubtractProfile($profile, $base->{profile});
+    $pcs = AddPcs($pcs, $base->{pcs});
+    $symbol_map = MergeSymbols($symbol_map, $base->{symbols});
+  }
+
+  # Collect symbols
+  my $symbols;
+  if ($main::use_symbolized_profile) {
+    $symbols = FetchSymbols($pcs, $symbol_map);
+  } elsif ($main::use_symbol_page) {
+    $symbols = FetchSymbols($pcs);
+  } else {
+    # TODO(csilvers): $libs uses the /proc/self/maps data from profile1,
+    # which may differ from the data from subsequent profiles, especially
+    # if they were run on different machines.  Use appropriate libs for
+    # each pc somehow.
+    $symbols = ExtractSymbols($libs, $pcs);
+  }
+
+  if (!defined($main::opt_thread)) {
+    FilterAndPrint($profile, $symbols, $libs);
+  }
+  if (defined($data->{threads})) {
+    foreach my $thread (sort { $a <=> $b } keys(%{$data->{threads}})) {
+      if (defined($main::opt_thread) &&
+          ($main::opt_thread eq '*' || $main::opt_thread == $thread)) {
+        my $thread_profile = $data->{threads}{$thread};
+        FilterAndPrint($thread_profile, $symbols, $libs, $thread);
+      }
+    }
+  }
+
+  cleanup();
+  exit(0);
+}
+
+##### Entry Point #####
+
+Main();
+
+# Temporary code to detect if we're running on a Goobuntu system.
+# These systems don't have the right stuff installed for the special
+# Readline libraries to work, so as a temporary workaround, we default
+# to using the normal stdio code, rather than the fancier readline-based
+# code
+sub ReadlineMightFail {
+  if (-e '/lib/libtermcap.so.2') {
+    return 0;  # libtermcap exists, so readline should be okay
+  } else {
+    return 1;
+  }
+}
+
+sub RunGV {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  if (!system(ShellEscape(@GV, "--version") . " >$dev_null 2>&1")) {
+    # Options using double dash are supported by this gv version.
+    # Also, turn on noantialias to better handle bug in gv for
+    # postscript files with large dimensions.
+    # TODO: Maybe we should not pass the --noantialias flag
+    # if the gv version is known to work properly without the flag.
+    system(ShellEscape(@GV, "--scale=$main::opt_scale", "--noantialias", $fname)
+           . $bg);
+  } else {
+    # Old gv version - only supports options that use single dash.
+    print STDERR ShellEscape(@GV, "-scale", $main::opt_scale) . "\n";
+    system(ShellEscape(@GV, "-scale", "$main::opt_scale", $fname) . $bg);
+  }
+}
+
+sub RunEvince {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  system(ShellEscape(@EVINCE, $fname) . $bg);
+}
+
+sub RunWeb {
+  my $fname = shift;
+  print STDERR "Loading web page file:///$fname\n";
+
+  if (`uname` =~ /Darwin/) {
+    # OS X: open will use standard preference for SVG files.
+    system("/usr/bin/open", $fname);
+    return;
+  }
+
+  # Some kind of Unix; try generic symlinks, then specific browsers.
+  # (Stop once we find one.)
+  # Works best if the browser is already running.
+  my @alt = (
+    "/etc/alternatives/gnome-www-browser",
+    "/etc/alternatives/x-www-browser",
+    "google-chrome",
+    "firefox",
+  );
+  foreach my $b (@alt) {
+    if (system($b, $fname) == 0) {
+      return;
+    }
+  }
+
+  print STDERR "Could not load web browser.\n";
+}
+
+sub RunKcachegrind {
+  my $fname = shift;
+  my $bg = shift;       # "" or " &" if we should run in background
+  print STDERR "Starting '@KCACHEGRIND " . $fname . $bg . "'\n";
+  system(ShellEscape(@KCACHEGRIND, $fname) . $bg);
+}
+
+
+##### Interactive helper routines #####
+
+sub InteractiveMode {
+  $| = 1;  # Make output unbuffered for interactive mode
+  my ($orig_profile, $symbols, $libs, $total) = @_;
+
+  print STDERR "Welcome to jeprof!  For help, type 'help'.\n";
+
+  # Use ReadLine if it's installed and input comes from a console.
+  if ( -t STDIN &&
+       !ReadlineMightFail() &&
+       defined(eval {require Term::ReadLine}) ) {
+    my $term = new Term::ReadLine 'jeprof';
+    while ( defined ($_ = $term->readline('(jeprof) '))) {
+      $term->addhistory($_) if /\S/;
+      if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) {
+        last;    # exit when we get an interactive command to quit
+      }
+    }
+  } else {       # don't have readline
+    while (1) {
+      print STDERR "(jeprof) ";
+      $_ = <STDIN>;
+      last if ! defined $_ ;
+      s/\r//g;         # turn windows-looking lines into unix-looking lines
+
+      # Save some flags that might be reset by InteractiveCommand()
+      my $save_opt_lines = $main::opt_lines;
+
+      if (!InteractiveCommand($orig_profile, $symbols, $libs, $total, $_)) {
+        last;    # exit when we get an interactive command to quit
+      }
+
+      # Restore flags
+      $main::opt_lines = $save_opt_lines;
+    }
+  }
+}
+
+# Takes two args: orig profile, and command to run.
+# Returns 1 if we should keep going, or 0 if we were asked to quit
+sub InteractiveCommand {
+  my($orig_profile, $symbols, $libs, $total, $command) = @_;
+  $_ = $command;                # just to make future m//'s easier
+  if (!defined($_)) {
+    print STDERR "\n";
+    return 0;
+  }
+  if (m/^\s*quit/) {
+    return 0;
+  }
+  if (m/^\s*help/) {
+    InteractiveHelpMessage();
+    return 1;
+  }
+  # Clear all the mode options -- mode is controlled by "$command"
+  $main::opt_text = 0;
+  $main::opt_callgrind = 0;
+  $main::opt_disasm = 0;
+  $main::opt_list = 0;
+  $main::opt_gv = 0;
+  $main::opt_evince = 0;
+  $main::opt_cum = 0;
+
+  if (m/^\s*(text|top)(\d*)\s*(.*)/) {
+    $main::opt_text = 1;
+
+    my $line_limit = ($2 ne "") ? int($2) : 10;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($3);
+
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintText($symbols, $flat, $cumulative, $line_limit);
+    return 1;
+  }
+  if (m/^\s*callgrind\s*([^ \n]*)/) {
+    $main::opt_callgrind = 1;
+
+    # Get derived profiles
+    my $calls = ExtractCalls($symbols, $orig_profile);
+    my $filename = $1;
+    if ( $1 eq '' ) {
+      $filename = TempName($main::next_tmpfile, "callgrind");
+    }
+    PrintCallgrind($calls, $filename);
+    if ( $1 eq '' ) {
+      RunKcachegrind($filename, " & ");
+      $main::next_tmpfile++;
+    }
+
+    return 1;
+  }
+  if (m/^\s*(web)?list\s*(.+)/) {
+    my $html = (defined($1) && ($1 eq "web"));
+    $main::opt_list = 1;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($2);
+
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintListing($total, $libs, $flat, $cumulative, $routine, $html);
+    return 1;
+  }
+  if (m/^\s*disasm\s*(.+)/) {
+    $main::opt_disasm = 1;
+
+    my $routine;
+    my $ignore;
+    ($routine, $ignore) = ParseInteractiveArgs($1);
+
+    # Process current profile to account for various settings
+    my $profile = ProcessProfile($total, $orig_profile, $symbols, "", $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    PrintDisassembly($libs, $flat, $cumulative, $routine);
+    return 1;
+  }
+  if (m/^\s*(gv|web|evince)\s*(.*)/) {
+    $main::opt_gv = 0;
+    $main::opt_evince = 0;
+    $main::opt_web = 0;
+    if ($1 eq "gv") {
+      $main::opt_gv = 1;
+    } elsif ($1 eq "evince") {
+      $main::opt_evince = 1;
+    } elsif ($1 eq "web") {
+      $main::opt_web = 1;
+    }
+
+    my $focus;
+    my $ignore;
+    ($focus, $ignore) = ParseInteractiveArgs($2);
+
+    # Process current profile to account for various settings
+    my $profile = ProcessProfile($total, $orig_profile, $symbols,
+                                 $focus, $ignore);
+    my $reduced = ReduceProfile($symbols, $profile);
+
+    # Get derived profiles
+    my $flat = FlatProfile($reduced);
+    my $cumulative = CumulativeProfile($reduced);
+
+    if (PrintDot($main::prog, $symbols, $profile, $flat, $cumulative, $total)) {
+      if ($main::opt_gv) {
+        RunGV(TempName($main::next_tmpfile, "ps"), " &");
+      } elsif ($main::opt_evince) {
+        RunEvince(TempName($main::next_tmpfile, "pdf"), " &");
+      } elsif ($main::opt_web) {
+        RunWeb(TempName($main::next_tmpfile, "svg"));
+      }
+      $main::next_tmpfile++;
+    }
+    return 1;
+  }
+  if (m/^\s*$/) {
+    return 1;
+  }
+  print STDERR "Unknown command: try 'help'.\n";
+  return 1;
+}
+
+
+sub ProcessProfile {
+  my $total_count = shift;
+  my $orig_profile = shift;
+  my $symbols = shift;
+  my $focus = shift;
+  my $ignore = shift;
+
+  # Process current profile to account for various settings
+  my $profile = $orig_profile;
+  printf("Total: %s %s\n", Unparse($total_count), Units());
+  if ($focus ne '') {
+    $profile = FocusProfile($symbols, $profile, $focus);
+    my $focus_count = TotalProfile($profile);
+    printf("After focusing on '%s': %s %s of %s (%0.1f%%)\n",
+           $focus,
+           Unparse($focus_count), Units(),
+           Unparse($total_count), ($focus_count*100.0) / $total_count);
+  }
+  if ($ignore ne '') {
+    $profile = IgnoreProfile($symbols, $profile, $ignore);
+    my $ignore_count = TotalProfile($profile);
+    printf("After ignoring '%s': %s %s of %s (%0.1f%%)\n",
+           $ignore,
+           Unparse($ignore_count), Units(),
+           Unparse($total_count),
+           ($ignore_count*100.0) / $total_count);
+  }
+
+  return $profile;
+}
+
+sub InteractiveHelpMessage {
+  print STDERR <<ENDOFHELP;
+Interactive jeprof mode
+
+Commands:
+  gv
+  gv [focus] [-ignore1] [-ignore2]
+      Show graphical hierarchical display of current profile.  Without
+      any arguments, shows all samples in the profile.  With the optional
+      "focus" argument, restricts the samples shown to just those where
+      the "focus" regular expression matches a routine name on the stack
+      trace.
+
+  web
+  web [focus] [-ignore1] [-ignore2]
+      Like GV, but displays profile in your web browser instead of using
+      Ghostview. Works best if your web browser is already running.
+      To change the browser that gets used:
+      On Linux, set the /etc/alternatives/gnome-www-browser symlink.
+      On OS X, change the Finder association for SVG files.
+
+  list [routine_regexp] [-ignore1] [-ignore2]
+      Show source listing of routines whose names match "routine_regexp"
+
+  weblist [routine_regexp] [-ignore1] [-ignore2]
+     Displays a source listing of routines whose names match "routine_regexp"
+     in a web browser.  You can click on source lines to view the
+     corresponding disassembly.
+
+  top [--cum] [-ignore1] [-ignore2]
+  top20 [--cum] [-ignore1] [-ignore2]
+  top37 [--cum] [-ignore1] [-ignore2]
+      Show top lines ordered by flat profile count, or cumulative count
+      if --cum is specified.  If a number is present after 'top', the
+      top K routines will be shown (defaults to showing the top 10)
+
+  disasm [routine_regexp] [-ignore1] [-ignore2]
+      Show disassembly of routines whose names match "routine_regexp",
+      annotated with sample counts.
+
+  callgrind
+  callgrind [filename]
+      Generates callgrind file. If no filename is given, kcachegrind is called.
+
+  help - This listing
+  quit or ^D - End jeprof
+
+For commands that accept optional -ignore tags, samples where any routine in
+the stack trace matches the regular expression in any of the -ignore
+parameters will be ignored.
+
+Further pprof details are available at this location (or one similar):
+
+ /usr/doc/gperftools-$PPROF_VERSION/cpu_profiler.html
+ /usr/doc/gperftools-$PPROF_VERSION/heap_profiler.html
+
+ENDOFHELP
+}
+sub ParseInteractiveArgs {
+  my $args = shift;
+  my $focus = "";
+  my $ignore = "";
+  my @x = split(/ +/, $args);
+  foreach $a (@x) {
+    if ($a =~ m/^(--|-)lines$/) {
+      $main::opt_lines = 1;
+    } elsif ($a =~ m/^(--|-)cum$/) {
+      $main::opt_cum = 1;
+    } elsif ($a =~ m/^-(.*)/) {
+      $ignore .= (($ignore ne "") ? "|" : "" ) . $1;
+    } else {
+      $focus .= (($focus ne "") ? "|" : "" ) . $a;
+    }
+  }
+  if ($ignore ne "") {
+    print STDERR "Ignoring samples in call stacks that match '$ignore'\n";
+  }
+  return ($focus, $ignore);
+}
+
+##### Output code #####
+
+sub TempName {
+  my $fnum = shift;
+  my $ext = shift;
+  my $file = "$main::tmpfile_ps.$fnum.$ext";
+  $main::tempnames{$file} = 1;
+  return $file;
+}
+
+# Print profile data in packed binary format (64-bit) to standard out
+sub PrintProfileData {
+  my $profile = shift;
+
+  # print header (64-bit style)
+  # (zero) (header-size) (version) (sample-period) (zero)
+  print pack('L*', 0, 0, 3, 0, 0, 0, 1, 0, 0, 0);
+
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    if ($#addrs >= 0) {
+      my $depth = $#addrs + 1;
+      # int(foo / 2**32) is the only reliable way to get rid of bottom
+      # 32 bits on both 32- and 64-bit systems.
+      print pack('L*', $count & 0xFFFFFFFF, int($count / 2**32));
+      print pack('L*', $depth & 0xFFFFFFFF, int($depth / 2**32));
+
+      foreach my $full_addr (@addrs) {
+        my $addr = $full_addr;
+        $addr =~ s/0x0*//;  # strip off leading 0x, zeroes
+        if (length($addr) > 16) {
+          print STDERR "Invalid address in profile: $full_addr\n";
+          next;
+        }
+        my $low_addr = substr($addr, -8);       # get last 8 hex chars
+        my $high_addr = substr($addr, -16, 8);  # get up to 8 more hex chars
+        print pack('L*', hex('0x' . $low_addr), hex('0x' . $high_addr));
+      }
+    }
+  }
+}
+
+# Print symbols and profile data
+sub PrintSymbolizedProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $prog = shift;
+
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+
+  print '--- ', $symbol_marker, "\n";
+  if (defined($prog)) {
+    print 'binary=', $prog, "\n";
+  }
+  while (my ($pc, $name) = each(%{$symbols})) {
+    my $sep = ' ';
+    print '0x', $pc;
+    # We have a list of function names, which include the inlined
+    # calls.  They are separated (and terminated) by --, which is
+    # illegal in function names.
+    for (my $j = 2; $j <= $#{$name}; $j += 3) {
+      print $sep, $name->[$j];
+      $sep = '--';
+    }
+    print "\n";
+  }
+  print '---', "\n";
+
+  my $profile_marker;
+  if ($main::profile_type eq 'heap') {
+    $HEAP_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } elsif ($main::profile_type eq 'growth') {
+    $GROWTH_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } elsif ($main::profile_type eq 'contention') {
+    $CONTENTION_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  } else { # elsif ($main::profile_type eq 'cpu')
+    $PROFILE_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+    $profile_marker = $&;
+  }
+
+  print '--- ', $profile_marker, "\n";
+  if (defined($main::collected_profile)) {
+    # if used with remote fetch, simply dump the collected profile to output.
+    open(SRC, "<$main::collected_profile");
+    while (<SRC>) {
+      print $_;
+    }
+    close(SRC);
+  } else {
+    # --raw/http: For everything to work correctly for non-remote profiles, we
+    # would need to extend PrintProfileData() to handle all possible profile
+    # types, re-enable the code that is currently disabled in ReadCPUProfile()
+    # and FixCallerAddresses(), and remove the remote profile dumping code in
+    # the block above.
+    die "--raw/http: jeprof can only dump remote profiles for --raw\n";
+    # dump a cpu-format profile to standard out
+    PrintProfileData($profile);
+  }
+}
+
+# Print text output
+sub PrintText {
+  my $symbols = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $line_limit = shift;
+
+  my $total = TotalProfile($flat);
+
+  # Which profile to sort by?
+  my $s = $main::opt_cum ? $cumulative : $flat;
+
+  my $running_sum = 0;
+  my $lines = 0;
+  foreach my $k (sort { GetEntry($s, $b) <=> GetEntry($s, $a) || $a cmp $b }
+                 keys(%{$cumulative})) {
+    my $f = GetEntry($flat, $k);
+    my $c = GetEntry($cumulative, $k);
+    $running_sum += $f;
+
+    my $sym = $k;
+    if (exists($symbols->{$k})) {
+      $sym = $symbols->{$k}->[0] . " " . $symbols->{$k}->[1];
+      if ($main::opt_addresses) {
+        $sym = $k . " " . $sym;
+      }
+    }
+
+    if ($f != 0 || $c != 0) {
+      printf("%8s %6s %6s %8s %6s %s\n",
+             Unparse($f),
+             Percent($f, $total),
+             Percent($running_sum, $total),
+             Unparse($c),
+             Percent($c, $total),
+             $sym);
+    }
+    $lines++;
+    last if ($line_limit >= 0 && $lines >= $line_limit);
+  }
+}
+
+# Callgrind format has a compression for repeated function and file
+# names.  You show the name the first time, and just use its number
+# subsequently.  This can cut down the file to about a third or a
+# quarter of its uncompressed size.  $key and $val are the key/value
+# pair that would normally be printed by callgrind; $map is a map from
+# value to number.
+sub CompressedCGName {
+  my($key, $val, $map) = @_;
+  my $idx = $map->{$val};
+  # For very short keys, providing an index hurts rather than helps.
+  if (length($val) <= 3) {
+    return "$key=$val\n";
+  } elsif (defined($idx)) {
+    return "$key=($idx)\n";
+  } else {
+    # scalar(keys $map) gives the number of items in the map.
+    $idx = scalar(keys(%{$map})) + 1;
+    $map->{$val} = $idx;
+    return "$key=($idx) $val\n";
+  }
+}
+
+# Print the call graph in a way that's suiteable for callgrind.
+sub PrintCallgrind {
+  my $calls = shift;
+  my $filename;
+  my %filename_to_index_map;
+  my %fnname_to_index_map;
+
+  if ($main::opt_interactive) {
+    $filename = shift;
+    print STDERR "Writing callgrind file to '$filename'.\n"
+  } else {
+    $filename = "&STDOUT";
+  }
+  open(CG, ">$filename");
+  printf CG ("events: Hits\n\n");
+  foreach my $call ( map { $_->[0] }
+                     sort { $a->[1] cmp $b ->[1] ||
+                            $a->[2] <=> $b->[2] }
+                     map { /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/;
+                           [$_, $1, $2] }
+                     keys %$calls ) {
+    my $count = int($calls->{$call});
+    $call =~ /([^:]+):(\d+):([^ ]+)( -> ([^:]+):(\d+):(.+))?/;
+    my ( $caller_file, $caller_line, $caller_function,
+         $callee_file, $callee_line, $callee_function ) =
+       ( $1, $2, $3, $5, $6, $7 );
+
+    # TODO(csilvers): for better compression, collect all the
+    # caller/callee_files and functions first, before printing
+    # anything, and only compress those referenced more than once.
+    printf CG CompressedCGName("fl", $caller_file, \%filename_to_index_map);
+    printf CG CompressedCGName("fn", $caller_function, \%fnname_to_index_map);
+    if (defined $6) {
+      printf CG CompressedCGName("cfl", $callee_file, \%filename_to_index_map);
+      printf CG CompressedCGName("cfn", $callee_function, \%fnname_to_index_map);
+      printf CG ("calls=$count $callee_line\n");
+    }
+    printf CG ("$caller_line $count\n\n");
+  }
+}
+
+# Print disassembly for all all routines that match $main::opt_disasm
+sub PrintDisassembly {
+  my $libs = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $disasm_opts = shift;
+
+  my $total = TotalProfile($flat);
+
+  foreach my $lib (@{$libs}) {
+    my $symbol_table = GetProcedureBoundaries($lib->[0], $disasm_opts);
+    my $offset = AddressSub($lib->[1], $lib->[3]);
+    foreach my $routine (sort ByName keys(%{$symbol_table})) {
+      my $start_addr = $symbol_table->{$routine}->[0];
+      my $end_addr = $symbol_table->{$routine}->[1];
+      # See if there are any samples in this routine
+      my $length = hex(AddressSub($end_addr, $start_addr));
+      my $addr = AddressAdd($start_addr, $offset);
+      for (my $i = 0; $i < $length; $i++) {
+        if (defined($cumulative->{$addr})) {
+          PrintDisassembledFunction($lib->[0], $offset,
+                                    $routine, $flat, $cumulative,
+                                    $start_addr, $end_addr, $total);
+          last;
+        }
+        $addr = AddressInc($addr);
+      }
+    }
+  }
+}
+
+# Return reference to array of tuples of the form:
+#       [start_address, filename, linenumber, instruction, limit_address]
+# E.g.,
+#       ["0x806c43d", "/foo/bar.cc", 131, "ret", "0x806c440"]
+sub Disassemble {
+  my $prog = shift;
+  my $offset = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+
+  my $objdump = $obj_tool_map{"objdump"};
+  my $cmd = ShellEscape($objdump, "-C", "-d", "-l", "--no-show-raw-insn",
+                        "--start-address=0x$start_addr",
+                        "--stop-address=0x$end_addr", $prog);
+  open(OBJDUMP, "$cmd |") || error("$cmd: $!\n");
+  my @result = ();
+  my $filename = "";
+  my $linenumber = -1;
+  my $last = ["", "", "", ""];
+  while (<OBJDUMP>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    chop;
+    if (m|\s*([^:\s]+):(\d+)\s*$|) {
+      # Location line of the form:
+      #   <filename>:<linenumber>
+      $filename = $1;
+      $linenumber = $2;
+    } elsif (m/^ +([0-9a-f]+):\s*(.*)/) {
+      # Disassembly line -- zero-extend address to full length
+      my $addr = HexExtend($1);
+      my $k = AddressAdd($addr, $offset);
+      $last->[4] = $k;   # Store ending address for previous instruction
+      $last = [$k, $filename, $linenumber, $2, $end_addr];
+      push(@result, $last);
+    }
+  }
+  close(OBJDUMP);
+  return @result;
+}
+
+# The input file should contain lines of the form /proc/maps-like
+# output (same format as expected from the profiles) or that looks
+# like hex addresses (like "0xDEADBEEF").  We will parse all
+# /proc/maps output, and for all the hex addresses, we will output
+# "short" symbol names, one per line, in the same order as the input.
+sub PrintSymbols {
+  my $maps_and_symbols_file = shift;
+
+  # ParseLibraries expects pcs to be in a set.  Fine by us...
+  my @pclist = ();   # pcs in sorted order
+  my $pcs = {};
+  my $map = "";
+  foreach my $line (<$maps_and_symbols_file>) {
+    $line =~ s/\r//g;    # turn windows-looking lines into unix-looking lines
+    if ($line =~ /\b(0x[0-9a-f]+)\b/i) {
+      push(@pclist, HexExtend($1));
+      $pcs->{$pclist[-1]} = 1;
+    } else {
+      $map .= $line;
+    }
+  }
+
+  my $libs = ParseLibraries($main::prog, $map, $pcs);
+  my $symbols = ExtractSymbols($libs, $pcs);
+
+  foreach my $pc (@pclist) {
+    # ->[0] is the shortname, ->[2] is the full name
+    print(($symbols->{$pc}->[0] || "??") . "\n");
+  }
+}
+
+
+# For sorting functions by name
+sub ByName {
+  return ShortFunctionName($a) cmp ShortFunctionName($b);
+}
+
+# Print source-listing for all all routines that match $list_opts
+sub PrintListing {
+  my $total = shift;
+  my $libs = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $list_opts = shift;
+  my $html = shift;
+
+  my $output = \*STDOUT;
+  my $fname = "";
+
+  if ($html) {
+    # Arrange to write the output to a temporary file
+    $fname = TempName($main::next_tmpfile, "html");
+    $main::next_tmpfile++;
+    if (!open(TEMP, ">$fname")) {
+      print STDERR "$fname: $!\n";
+      return;
+    }
+    $output = \*TEMP;
+    print $output HtmlListingHeader();
+    printf $output ("<div class=\"legend\">%s<br>Total: %s %s</div>\n",
+                    $main::prog, Unparse($total), Units());
+  }
+
+  my $listed = 0;
+  foreach my $lib (@{$libs}) {
+    my $symbol_table = GetProcedureBoundaries($lib->[0], $list_opts);
+    my $offset = AddressSub($lib->[1], $lib->[3]);
+    foreach my $routine (sort ByName keys(%{$symbol_table})) {
+      # Print if there are any samples in this routine
+      my $start_addr = $symbol_table->{$routine}->[0];
+      my $end_addr = $symbol_table->{$routine}->[1];
+      my $length = hex(AddressSub($end_addr, $start_addr));
+      my $addr = AddressAdd($start_addr, $offset);
+      for (my $i = 0; $i < $length; $i++) {
+        if (defined($cumulative->{$addr})) {
+          $listed += PrintSource(
+            $lib->[0], $offset,
+            $routine, $flat, $cumulative,
+            $start_addr, $end_addr,
+            $html,
+            $output);
+          last;
+        }
+        $addr = AddressInc($addr);
+      }
+    }
+  }
+
+  if ($html) {
+    if ($listed > 0) {
+      print $output HtmlListingFooter();
+      close($output);
+      RunWeb($fname);
+    } else {
+      close($output);
+      unlink($fname);
+    }
+  }
+}
+
+sub HtmlListingHeader {
+  return <<'EOF';
+<DOCTYPE html>
+<html>
+<head>
+<title>Pprof listing</title>
+<style type="text/css">
+body {
+  font-family: sans-serif;
+}
+h1 {
+  font-size: 1.5em;
+  margin-bottom: 4px;
+}
+.legend {
+  font-size: 1.25em;
+}
+.line {
+  color: #aaaaaa;
+}
+.nop {
+  color: #aaaaaa;
+}
+.unimportant {
+  color: #cccccc;
+}
+.disasmloc {
+  color: #000000;
+}
+.deadsrc {
+  cursor: pointer;
+}
+.deadsrc:hover {
+  background-color: #eeeeee;
+}
+.livesrc {
+  color: #0000ff;
+  cursor: pointer;
+}
+.livesrc:hover {
+  background-color: #eeeeee;
+}
+.asm {
+  color: #008800;
+  display: none;
+}
+</style>
+<script type="text/javascript">
+function jeprof_toggle_asm(e) {
+  var target;
+  if (!e) e = window.event;
+  if (e.target) target = e.target;
+  else if (e.srcElement) target = e.srcElement;
+
+  if (target) {
+    var asm = target.nextSibling;
+    if (asm && asm.className == "asm") {
+      asm.style.display = (asm.style.display == "block" ? "" : "block");
+      e.preventDefault();
+      return false;
+    }
+  }
+}
+</script>
+</head>
+<body>
+EOF
+}
+
+sub HtmlListingFooter {
+  return <<'EOF';
+</body>
+</html>
+EOF
+}
+
+sub HtmlEscape {
+  my $text = shift;
+  $text =~ s/&/&amp;/g;
+  $text =~ s/</&lt;/g;
+  $text =~ s/>/&gt;/g;
+  return $text;
+}
+
+# Returns the indentation of the line, if it has any non-whitespace
+# characters.  Otherwise, returns -1.
+sub Indentation {
+  my $line = shift;
+  if (m/^(\s*)\S/) {
+    return length($1);
+  } else {
+    return -1;
+  }
+}
+
+# If the symbol table contains inlining info, Disassemble() may tag an
+# instruction with a location inside an inlined function.  But for
+# source listings, we prefer to use the location in the function we
+# are listing.  So use MapToSymbols() to fetch full location
+# information for each instruction and then pick out the first
+# location from a location list (location list contains callers before
+# callees in case of inlining).
+#
+# After this routine has run, each entry in $instructions contains:
+#   [0] start address
+#   [1] filename for function we are listing
+#   [2] line number for function we are listing
+#   [3] disassembly
+#   [4] limit address
+#   [5] most specific filename (may be different from [1] due to inlining)
+#   [6] most specific line number (may be different from [2] due to inlining)
+sub GetTopLevelLineNumbers {
+  my ($lib, $offset, $instructions) = @_;
+  my $pcs = [];
+  for (my $i = 0; $i <= $#{$instructions}; $i++) {
+    push(@{$pcs}, $instructions->[$i]->[0]);
+  }
+  my $symbols = {};
+  MapToSymbols($lib, $offset, $pcs, $symbols);
+  for (my $i = 0; $i <= $#{$instructions}; $i++) {
+    my $e = $instructions->[$i];
+    push(@{$e}, $e->[1]);
+    push(@{$e}, $e->[2]);
+    my $addr = $e->[0];
+    my $sym = $symbols->{$addr};
+    if (defined($sym)) {
+      if ($#{$sym} >= 2 && $sym->[1] =~ m/^(.*):(\d+)$/) {
+        $e->[1] = $1;  # File name
+        $e->[2] = $2;  # Line number
+      }
+    }
+  }
+}
+
+# Print source-listing for one routine
+sub PrintSource {
+  my $prog = shift;
+  my $offset = shift;
+  my $routine = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+  my $html = shift;
+  my $output = shift;
+
+  # Disassemble all instructions (just to get line numbers)
+  my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr);
+  GetTopLevelLineNumbers($prog, $offset, \@instructions);
+
+  # Hack 1: assume that the first source file encountered in the
+  # disassembly contains the routine
+  my $filename = undef;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    if ($instructions[$i]->[2] >= 0) {
+      $filename = $instructions[$i]->[1];
+      last;
+    }
+  }
+  if (!defined($filename)) {
+    print STDERR "no filename found in $routine\n";
+    return 0;
+  }
+
+  # Hack 2: assume that the largest line number from $filename is the
+  # end of the procedure.  This is typically safe since if P1 contains
+  # an inlined call to P2, then P2 usually occurs earlier in the
+  # source file.  If this does not work, we might have to compute a
+  # density profile or just print all regions we find.
+  my $lastline = 0;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    my $f = $instructions[$i]->[1];
+    my $l = $instructions[$i]->[2];
+    if (($f eq $filename) && ($l > $lastline)) {
+      $lastline = $l;
+    }
+  }
+
+  # Hack 3: assume the first source location from "filename" is the start of
+  # the source code.
+  my $firstline = 1;
+  for (my $i = 0; $i <= $#instructions; $i++) {
+    if ($instructions[$i]->[1] eq $filename) {
+      $firstline = $instructions[$i]->[2];
+      last;
+    }
+  }
+
+  # Hack 4: Extend last line forward until its indentation is less than
+  # the indentation we saw on $firstline
+  my $oldlastline = $lastline;
+  {
+    if (!open(FILE, "<$filename")) {
+      print STDERR "$filename: $!\n";
+      return 0;
+    }
+    my $l = 0;
+    my $first_indentation = -1;
+    while (<FILE>) {
+      s/\r//g;         # turn windows-looking lines into unix-looking lines
+      $l++;
+      my $indent = Indentation($_);
+      if ($l >= $firstline) {
+        if ($first_indentation < 0 && $indent >= 0) {
+          $first_indentation = $indent;
+          last if ($first_indentation == 0);
+        }
+      }
+      if ($l >= $lastline && $indent >= 0) {
+        if ($indent >= $first_indentation) {
+          $lastline = $l+1;
+        } else {
+          last;
+        }
+      }
+    }
+    close(FILE);
+  }
+
+  # Assign all samples to the range $firstline,$lastline,
+  # Hack 4: If an instruction does not occur in the range, its samples
+  # are moved to the next instruction that occurs in the range.
+  my $samples1 = {};        # Map from line number to flat count
+  my $samples2 = {};        # Map from line number to cumulative count
+  my $running1 = 0;         # Unassigned flat counts
+  my $running2 = 0;         # Unassigned cumulative counts
+  my $total1 = 0;           # Total flat counts
+  my $total2 = 0;           # Total cumulative counts
+  my %disasm = ();          # Map from line number to disassembly
+  my $running_disasm = "";  # Unassigned disassembly
+  my $skip_marker = "---\n";
+  if ($html) {
+    $skip_marker = "";
+    for (my $l = $firstline; $l <= $lastline; $l++) {
+      $disasm{$l} = "";
+    }
+  }
+  my $last_dis_filename = '';
+  my $last_dis_linenum = -1;
+  my $last_touched_line = -1;  # To detect gaps in disassembly for a line
+  foreach my $e (@instructions) {
+    # Add up counts for all address that fall inside this instruction
+    my $c1 = 0;
+    my $c2 = 0;
+    for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) {
+      $c1 += GetEntry($flat, $a);
+      $c2 += GetEntry($cumulative, $a);
+    }
+
+    if ($html) {
+      my $dis = sprintf("      %6s %6s \t\t%8s: %s ",
+                        HtmlPrintNumber($c1),
+                        HtmlPrintNumber($c2),
+                        UnparseAddress($offset, $e->[0]),
+                        CleanDisassembly($e->[3]));
+
+      # Append the most specific source line associated with this instruction
+      if (length($dis) < 80) { $dis .= (' ' x (80 - length($dis))) };
+      $dis = HtmlEscape($dis);
+      my $f = $e->[5];
+      my $l = $e->[6];
+      if ($f ne $last_dis_filename) {
+        $dis .= sprintf("<span class=disasmloc>%s:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      } elsif ($l ne $last_dis_linenum) {
+        # De-emphasize the unchanged file name portion
+        $dis .= sprintf("<span class=unimportant>%s</span>" .
+                        "<span class=disasmloc>:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      } else {
+        # De-emphasize the entire location
+        $dis .= sprintf("<span class=unimportant>%s:%d</span>",
+                        HtmlEscape(CleanFileName($f)), $l);
+      }
+      $last_dis_filename = $f;
+      $last_dis_linenum = $l;
+      $running_disasm .= $dis;
+      $running_disasm .= "\n";
+    }
+
+    $running1 += $c1;
+    $running2 += $c2;
+    $total1 += $c1;
+    $total2 += $c2;
+    my $file = $e->[1];
+    my $line = $e->[2];
+    if (($file eq $filename) &&
+        ($line >= $firstline) &&
+        ($line <= $lastline)) {
+      # Assign all accumulated samples to this line
+      AddEntry($samples1, $line, $running1);
+      AddEntry($samples2, $line, $running2);
+      $running1 = 0;
+      $running2 = 0;
+      if ($html) {
+        if ($line != $last_touched_line && $disasm{$line} ne '') {
+          $disasm{$line} .= "\n";
+        }
+        $disasm{$line} .= $running_disasm;
+        $running_disasm = '';
+        $last_touched_line = $line;
+      }
+    }
+  }
+
+  # Assign any leftover samples to $lastline
+  AddEntry($samples1, $lastline, $running1);
+  AddEntry($samples2, $lastline, $running2);
+  if ($html) {
+    if ($lastline != $last_touched_line && $disasm{$lastline} ne '') {
+      $disasm{$lastline} .= "\n";
+    }
+    $disasm{$lastline} .= $running_disasm;
+  }
+
+  if ($html) {
+    printf $output (
+      "<h1>%s</h1>%s\n<pre onClick=\"jeprof_toggle_asm()\">\n" .
+      "Total:%6s %6s (flat / cumulative %s)\n",
+      HtmlEscape(ShortFunctionName($routine)),
+      HtmlEscape(CleanFileName($filename)),
+      Unparse($total1),
+      Unparse($total2),
+      Units());
+  } else {
+    printf $output (
+      "ROUTINE ====================== %s in %s\n" .
+      "%6s %6s Total %s (flat / cumulative)\n",
+      ShortFunctionName($routine),
+      CleanFileName($filename),
+      Unparse($total1),
+      Unparse($total2),
+      Units());
+  }
+  if (!open(FILE, "<$filename")) {
+    print STDERR "$filename: $!\n";
+    return 0;
+  }
+  my $l = 0;
+  while (<FILE>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    $l++;
+    if ($l >= $firstline - 5 &&
+        (($l <= $oldlastline + 5) || ($l <= $lastline))) {
+      chop;
+      my $text = $_;
+      if ($l == $firstline) { print $output $skip_marker; }
+      my $n1 = GetEntry($samples1, $l);
+      my $n2 = GetEntry($samples2, $l);
+      if ($html) {
+        # Emit a span that has one of the following classes:
+        #    livesrc -- has samples
+        #    deadsrc -- has disassembly, but with no samples
+        #    nop     -- has no matching disasembly
+        # Also emit an optional span containing disassembly.
+        my $dis = $disasm{$l};
+        my $asm = "";
+        if (defined($dis) && $dis ne '') {
+          $asm = "<span class=\"asm\">" . $dis . "</span>";
+        }
+        my $source_class = (($n1 + $n2 > 0)
+                            ? "livesrc"
+                            : (($asm ne "") ? "deadsrc" : "nop"));
+        printf $output (
+          "<span class=\"line\">%5d</span> " .
+          "<span class=\"%s\">%6s %6s %s</span>%s\n",
+          $l, $source_class,
+          HtmlPrintNumber($n1),
+          HtmlPrintNumber($n2),
+          HtmlEscape($text),
+          $asm);
+      } else {
+        printf $output(
+          "%6s %6s %4d: %s\n",
+          UnparseAlt($n1),
+          UnparseAlt($n2),
+          $l,
+          $text);
+      }
+      if ($l == $lastline)  { print $output $skip_marker; }
+    };
+  }
+  close(FILE);
+  if ($html) {
+    print $output "</pre>\n";
+  }
+  return 1;
+}
+
+# Return the source line for the specified file/linenumber.
+# Returns undef if not found.
+sub SourceLine {
+  my $file = shift;
+  my $line = shift;
+
+  # Look in cache
+  if (!defined($main::source_cache{$file})) {
+    if (100 < scalar keys(%main::source_cache)) {
+      # Clear the cache when it gets too big
+      $main::source_cache = ();
+    }
+
+    # Read all lines from the file
+    if (!open(FILE, "<$file")) {
+      print STDERR "$file: $!\n";
+      $main::source_cache{$file} = [];  # Cache the negative result
+      return undef;
+    }
+    my $lines = [];
+    push(@{$lines}, "");        # So we can use 1-based line numbers as indices
+    while (<FILE>) {
+      push(@{$lines}, $_);
+    }
+    close(FILE);
+
+    # Save the lines in the cache
+    $main::source_cache{$file} = $lines;
+  }
+
+  my $lines = $main::source_cache{$file};
+  if (($line < 0) || ($line > $#{$lines})) {
+    return undef;
+  } else {
+    return $lines->[$line];
+  }
+}
+
+# Print disassembly for one routine with interspersed source if available
+sub PrintDisassembledFunction {
+  my $prog = shift;
+  my $offset = shift;
+  my $routine = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $start_addr = shift;
+  my $end_addr = shift;
+  my $total = shift;
+
+  # Disassemble all instructions
+  my @instructions = Disassemble($prog, $offset, $start_addr, $end_addr);
+
+  # Make array of counts per instruction
+  my @flat_count = ();
+  my @cum_count = ();
+  my $flat_total = 0;
+  my $cum_total = 0;
+  foreach my $e (@instructions) {
+    # Add up counts for all address that fall inside this instruction
+    my $c1 = 0;
+    my $c2 = 0;
+    for (my $a = $e->[0]; $a lt $e->[4]; $a = AddressInc($a)) {
+      $c1 += GetEntry($flat, $a);
+      $c2 += GetEntry($cumulative, $a);
+    }
+    push(@flat_count, $c1);
+    push(@cum_count, $c2);
+    $flat_total += $c1;
+    $cum_total += $c2;
+  }
+
+  # Print header with total counts
+  printf("ROUTINE ====================== %s\n" .
+         "%6s %6s %s (flat, cumulative) %.1f%% of total\n",
+         ShortFunctionName($routine),
+         Unparse($flat_total),
+         Unparse($cum_total),
+         Units(),
+         ($cum_total * 100.0) / $total);
+
+  # Process instructions in order
+  my $current_file = "";
+  for (my $i = 0; $i <= $#instructions; ) {
+    my $e = $instructions[$i];
+
+    # Print the new file name whenever we switch files
+    if ($e->[1] ne $current_file) {
+      $current_file = $e->[1];
+      my $fname = $current_file;
+      $fname =~ s|^\./||;   # Trim leading "./"
+
+      # Shorten long file names
+      if (length($fname) >= 58) {
+        $fname = "..." . substr($fname, -55);
+      }
+      printf("-------------------- %s\n", $fname);
+    }
+
+    # TODO: Compute range of lines to print together to deal with
+    # small reorderings.
+    my $first_line = $e->[2];
+    my $last_line = $first_line;
+    my %flat_sum = ();
+    my %cum_sum = ();
+    for (my $l = $first_line; $l <= $last_line; $l++) {
+      $flat_sum{$l} = 0;
+      $cum_sum{$l} = 0;
+    }
+
+    # Find run of instructions for this range of source lines
+    my $first_inst = $i;
+    while (($i <= $#instructions) &&
+           ($instructions[$i]->[2] >= $first_line) &&
+           ($instructions[$i]->[2] <= $last_line)) {
+      $e = $instructions[$i];
+      $flat_sum{$e->[2]} += $flat_count[$i];
+      $cum_sum{$e->[2]} += $cum_count[$i];
+      $i++;
+    }
+    my $last_inst = $i - 1;
+
+    # Print source lines
+    for (my $l = $first_line; $l <= $last_line; $l++) {
+      my $line = SourceLine($current_file, $l);
+      if (!defined($line)) {
+        $line = "?\n";
+        next;
+      } else {
+        $line =~ s/^\s+//;
+      }
+      printf("%6s %6s %5d: %s",
+             UnparseAlt($flat_sum{$l}),
+             UnparseAlt($cum_sum{$l}),
+             $l,
+             $line);
+    }
+
+    # Print disassembly
+    for (my $x = $first_inst; $x <= $last_inst; $x++) {
+      my $e = $instructions[$x];
+      printf("%6s %6s    %8s: %6s\n",
+             UnparseAlt($flat_count[$x]),
+             UnparseAlt($cum_count[$x]),
+             UnparseAddress($offset, $e->[0]),
+             CleanDisassembly($e->[3]));
+    }
+  }
+}
+
+# Print DOT graph
+sub PrintDot {
+  my $prog = shift;
+  my $symbols = shift;
+  my $raw = shift;
+  my $flat = shift;
+  my $cumulative = shift;
+  my $overall_total = shift;
+
+  # Get total
+  my $local_total = TotalProfile($flat);
+  my $nodelimit = int($main::opt_nodefraction * $local_total);
+  my $edgelimit = int($main::opt_edgefraction * $local_total);
+  my $nodecount = $main::opt_nodecount;
+
+  # Find nodes to include
+  my @list = (sort { abs(GetEntry($cumulative, $b)) <=>
+                     abs(GetEntry($cumulative, $a))
+                     || $a cmp $b }
+              keys(%{$cumulative}));
+  my $last = $nodecount - 1;
+  if ($last > $#list) {
+    $last = $#list;
+  }
+  while (($last >= 0) &&
+         (abs(GetEntry($cumulative, $list[$last])) <= $nodelimit)) {
+    $last--;
+  }
+  if ($last < 0) {
+    print STDERR "No nodes to print\n";
+    return 0;
+  }
+
+  if ($nodelimit > 0 || $edgelimit > 0) {
+    printf STDERR ("Dropping nodes with <= %s %s; edges with <= %s abs(%s)\n",
+                   Unparse($nodelimit), Units(),
+                   Unparse($edgelimit), Units());
+  }
+
+  # Open DOT output file
+  my $output;
+  my $escaped_dot = ShellEscape(@DOT);
+  my $escaped_ps2pdf = ShellEscape(@PS2PDF);
+  if ($main::opt_gv) {
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "ps"));
+    $output = "| $escaped_dot -Tps2 >$escaped_outfile";
+  } elsif ($main::opt_evince) {
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "pdf"));
+    $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - $escaped_outfile";
+  } elsif ($main::opt_ps) {
+    $output = "| $escaped_dot -Tps2";
+  } elsif ($main::opt_pdf) {
+    $output = "| $escaped_dot -Tps2 | $escaped_ps2pdf - -";
+  } elsif ($main::opt_web || $main::opt_svg) {
+    # We need to post-process the SVG, so write to a temporary file always.
+    my $escaped_outfile = ShellEscape(TempName($main::next_tmpfile, "svg"));
+    $output = "| $escaped_dot -Tsvg >$escaped_outfile";
+  } elsif ($main::opt_gif) {
+    $output = "| $escaped_dot -Tgif";
+  } else {
+    $output = ">&STDOUT";
+  }
+  open(DOT, $output) || error("$output: $!\n");
+
+  # Title
+  printf DOT ("digraph \"%s; %s %s\" {\n",
+              $prog,
+              Unparse($overall_total),
+              Units());
+  if ($main::opt_pdf) {
+    # The output is more printable if we set the page size for dot.
+    printf DOT ("size=\"8,11\"\n");
+  }
+  printf DOT ("node [width=0.375,height=0.25];\n");
+
+  # Print legend
+  printf DOT ("Legend [shape=box,fontsize=24,shape=plaintext," .
+              "label=\"%s\\l%s\\l%s\\l%s\\l%s\\l\"];\n",
+              $prog,
+              sprintf("Total %s: %s", Units(), Unparse($overall_total)),
+              sprintf("Focusing on: %s", Unparse($local_total)),
+              sprintf("Dropped nodes with <= %s abs(%s)",
+                      Unparse($nodelimit), Units()),
+              sprintf("Dropped edges with <= %s %s",
+                      Unparse($edgelimit), Units())
+              );
+
+  # Print nodes
+  my %node = ();
+  my $nextnode = 1;
+  foreach my $a (@list[0..$last]) {
+    # Pick font size
+    my $f = GetEntry($flat, $a);
+    my $c = GetEntry($cumulative, $a);
+
+    my $fs = 8;
+    if ($local_total > 0) {
+      $fs = 8 + (50.0 * sqrt(abs($f * 1.0 / $local_total)));
+    }
+
+    $node{$a} = $nextnode++;
+    my $sym = $a;
+    $sym =~ s/\s+/\\n/g;
+    $sym =~ s/::/\\n/g;
+
+    # Extra cumulative info to print for non-leaves
+    my $extra = "";
+    if ($f != $c) {
+      $extra = sprintf("\\rof %s (%s)",
+                       Unparse($c),
+                       Percent($c, $local_total));
+    }
+    my $style = "";
+    if ($main::opt_heapcheck) {
+      if ($f > 0) {
+        # make leak-causing nodes more visible (add a background)
+        $style = ",style=filled,fillcolor=gray"
+      } elsif ($f < 0) {
+        # make anti-leak-causing nodes (which almost never occur)
+        # stand out as well (triple border)
+        $style = ",peripheries=3"
+      }
+    }
+
+    printf DOT ("N%d [label=\"%s\\n%s (%s)%s\\r" .
+                "\",shape=box,fontsize=%.1f%s];\n",
+                $node{$a},
+                $sym,
+                Unparse($f),
+                Percent($f, $local_total),
+                $extra,
+                $fs,
+                $style,
+               );
+  }
+
+  # Get edges and counts per edge
+  my %edge = ();
+  my $n;
+  my $fullname_to_shortname_map = {};
+  FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map);
+  foreach my $k (keys(%{$raw})) {
+    # TODO: omit low %age edges
+    $n = $raw->{$k};
+    my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k);
+    for (my $i = 1; $i <= $#translated; $i++) {
+      my $src = $translated[$i];
+      my $dst = $translated[$i-1];
+      #next if ($src eq $dst);  # Avoid self-edges?
+      if (exists($node{$src}) && exists($node{$dst})) {
+        my $edge_label = "$src\001$dst";
+        if (!exists($edge{$edge_label})) {
+          $edge{$edge_label} = 0;
+        }
+        $edge{$edge_label} += $n;
+      }
+    }
+  }
+
+  # Print edges (process in order of decreasing counts)
+  my %indegree = ();   # Number of incoming edges added per node so far
+  my %outdegree = ();  # Number of outgoing edges added per node so far
+  foreach my $e (sort { $edge{$b} <=> $edge{$a} } keys(%edge)) {
+    my @x = split(/\001/, $e);
+    $n = $edge{$e};
+
+    # Initialize degree of kept incoming and outgoing edges if necessary
+    my $src = $x[0];
+    my $dst = $x[1];
+    if (!exists($outdegree{$src})) { $outdegree{$src} = 0; }
+    if (!exists($indegree{$dst})) { $indegree{$dst} = 0; }
+
+    my $keep;
+    if ($indegree{$dst} == 0) {
+      # Keep edge if needed for reachability
+      $keep = 1;
+    } elsif (abs($n) <= $edgelimit) {
+      # Drop if we are below --edgefraction
+      $keep = 0;
+    } elsif ($outdegree{$src} >= $main::opt_maxdegree ||
+             $indegree{$dst} >= $main::opt_maxdegree) {
+      # Keep limited number of in/out edges per node
+      $keep = 0;
+    } else {
+      $keep = 1;
+    }
+
+    if ($keep) {
+      $outdegree{$src}++;
+      $indegree{$dst}++;
+
+      # Compute line width based on edge count
+      my $fraction = abs($local_total ? (3 * ($n / $local_total)) : 0);
+      if ($fraction > 1) { $fraction = 1; }
+      my $w = $fraction * 2;
+      if ($w < 1 && ($main::opt_web || $main::opt_svg)) {
+        # SVG output treats line widths < 1 poorly.
+        $w = 1;
+      }
+
+      # Dot sometimes segfaults if given edge weights that are too large, so
+      # we cap the weights at a large value
+      my $edgeweight = abs($n) ** 0.7;
+      if ($edgeweight > 100000) { $edgeweight = 100000; }
+      $edgeweight = int($edgeweight);
+
+      my $style = sprintf("setlinewidth(%f)", $w);
+      if ($x[1] =~ m/\(inline\)/) {
+        $style .= ",dashed";
+      }
+
+      # Use a slightly squashed function of the edge count as the weight
+      printf DOT ("N%s -> N%s [label=%s, weight=%d, style=\"%s\"];\n",
+                  $node{$x[0]},
+                  $node{$x[1]},
+                  Unparse($n),
+                  $edgeweight,
+                  $style);
+    }
+  }
+
+  print DOT ("}\n");
+  close(DOT);
+
+  if ($main::opt_web || $main::opt_svg) {
+    # Rewrite SVG to be more usable inside web browser.
+    RewriteSvg(TempName($main::next_tmpfile, "svg"));
+  }
+
+  return 1;
+}
+
+sub RewriteSvg {
+  my $svgfile = shift;
+
+  open(SVG, $svgfile) || die "open temp svg: $!";
+  my @svg = <SVG>;
+  close(SVG);
+  unlink $svgfile;
+  my $svg = join('', @svg);
+
+  # Dot's SVG output is
+  #
+  #    <svg width="___" height="___"
+  #     viewBox="___" xmlns=...>
+  #    <g id="graph0" transform="...">
+  #    ...
+  #    </g>
+  #    </svg>
+  #
+  # Change it to
+  #
+  #    <svg width="100%" height="100%"
+  #     xmlns=...>
+  #    $svg_javascript
+  #    <g id="viewport" transform="translate(0,0)">
+  #    <g id="graph0" transform="...">
+  #    ...
+  #    </g>
+  #    </g>
+  #    </svg>
+
+  # Fix width, height; drop viewBox.
+  $svg =~ s/(?s)<svg width="[^"]+" height="[^"]+"(.*?)viewBox="[^"]+"/<svg width="100%" height="100%"$1/;
+
+  # Insert script, viewport <g> above first <g>
+  my $svg_javascript = SvgJavascript();
+  my $viewport = "<g id=\"viewport\" transform=\"translate(0,0)\">\n";
+  $svg =~ s/<g id="graph\d"/$svg_javascript$viewport$&/;
+
+  # Insert final </g> above </svg>.
+  $svg =~ s/(.*)(<\/svg>)/$1<\/g>$2/;
+  $svg =~ s/<g id="graph\d"(.*?)/<g id="viewport"$1/;
+
+  if ($main::opt_svg) {
+    # --svg: write to standard output.
+    print $svg;
+  } else {
+    # Write back to temporary file.
+    open(SVG, ">$svgfile") || die "open $svgfile: $!";
+    print SVG $svg;
+    close(SVG);
+  }
+}
+
+sub SvgJavascript {
+  return <<'EOF';
+<script type="text/ecmascript"><![CDATA[
+// SVGPan
+// http://www.cyberz.org/blog/2009/12/08/svgpan-a-javascript-svg-panzoomdrag-library/
+// Local modification: if(true || ...) below to force panning, never moving.
+
+/**
+ *  SVGPan library 1.2
+ * ====================
+ *
+ * Given an unique existing element with id "viewport", including the
+ * the library into any SVG adds the following capabilities:
+ *
+ *  - Mouse panning
+ *  - Mouse zooming (using the wheel)
+ *  - Object dargging
+ *
+ * Known issues:
+ *
+ *  - Zooming (while panning) on Safari has still some issues
+ *
+ * Releases:
+ *
+ * 1.2, Sat Mar 20 08:42:50 GMT 2010, Zeng Xiaohui
+ *	Fixed a bug with browser mouse handler interaction
+ *
+ * 1.1, Wed Feb  3 17:39:33 GMT 2010, Zeng Xiaohui
+ *	Updated the zoom code to support the mouse wheel on Safari/Chrome
+ *
+ * 1.0, Andrea Leofreddi
+ *	First release
+ *
+ * This code is licensed under the following BSD license:
+ *
+ * Copyright 2009-2010 Andrea Leofreddi <a.leofreddi@itcharm.com>. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without modification, are
+ * permitted provided that the following conditions are met:
+ *
+ *    1. Redistributions of source code must retain the above copyright notice, this list of
+ *       conditions and the following disclaimer.
+ *
+ *    2. Redistributions in binary form must reproduce the above copyright notice, this list
+ *       of conditions and the following disclaimer in the documentation and/or other materials
+ *       provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY Andrea Leofreddi ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
+ * FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL Andrea Leofreddi OR
+ * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
+ * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ * ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * The views and conclusions contained in the software and documentation are those of the
+ * authors and should not be interpreted as representing official policies, either expressed
+ * or implied, of Andrea Leofreddi.
+ */
+
+var root = document.documentElement;
+
+var state = 'none', stateTarget, stateOrigin, stateTf;
+
+setupHandlers(root);
+
+/**
+ * Register handlers
+ */
+function setupHandlers(root){
+	setAttributes(root, {
+		"onmouseup" : "add(evt)",
+		"onmousedown" : "handleMouseDown(evt)",
+		"onmousemove" : "handleMouseMove(evt)",
+		"onmouseup" : "handleMouseUp(evt)",
+		//"onmouseout" : "handleMouseUp(evt)", // Decomment this to stop the pan functionality when dragging out of the SVG element
+	});
+
+	if(navigator.userAgent.toLowerCase().indexOf('webkit') >= 0)
+		window.addEventListener('mousewheel', handleMouseWheel, false); // Chrome/Safari
+	else
+		window.addEventListener('DOMMouseScroll', handleMouseWheel, false); // Others
+
+	var g = svgDoc.getElementById("svg");
+	g.width = "100%";
+	g.height = "100%";
+}
+
+/**
+ * Instance an SVGPoint object with given event coordinates.
+ */
+function getEventPoint(evt) {
+	var p = root.createSVGPoint();
+
+	p.x = evt.clientX;
+	p.y = evt.clientY;
+
+	return p;
+}
+
+/**
+ * Sets the current transform matrix of an element.
+ */
+function setCTM(element, matrix) {
+	var s = "matrix(" + matrix.a + "," + matrix.b + "," + matrix.c + "," + matrix.d + "," + matrix.e + "," + matrix.f + ")";
+
+	element.setAttribute("transform", s);
+}
+
+/**
+ * Dumps a matrix to a string (useful for debug).
+ */
+function dumpMatrix(matrix) {
+	var s = "[ " + matrix.a + ", " + matrix.c + ", " + matrix.e + "\n  " + matrix.b + ", " + matrix.d + ", " + matrix.f + "\n  0, 0, 1 ]";
+
+	return s;
+}
+
+/**
+ * Sets attributes of an element.
+ */
+function setAttributes(element, attributes){
+	for (i in attributes)
+		element.setAttributeNS(null, i, attributes[i]);
+}
+
+/**
+ * Handle mouse move event.
+ */
+function handleMouseWheel(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var delta;
+
+	if(evt.wheelDelta)
+		delta = evt.wheelDelta / 3600; // Chrome/Safari
+	else
+		delta = evt.detail / -90; // Mozilla
+
+	var z = 1 + delta; // Zoom factor: 0.9/1.1
+
+	var g = svgDoc.getElementById("viewport");
+
+	var p = getEventPoint(evt);
+
+	p = p.matrixTransform(g.getCTM().inverse());
+
+	// Compute new scale matrix in current mouse position
+	var k = root.createSVGMatrix().translate(p.x, p.y).scale(z).translate(-p.x, -p.y);
+
+        setCTM(g, g.getCTM().multiply(k));
+
+	stateTf = stateTf.multiply(k.inverse());
+}
+
+/**
+ * Handle mouse move event.
+ */
+function handleMouseMove(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var g = svgDoc.getElementById("viewport");
+
+	if(state == 'pan') {
+		// Pan mode
+		var p = getEventPoint(evt).matrixTransform(stateTf);
+
+		setCTM(g, stateTf.inverse().translate(p.x - stateOrigin.x, p.y - stateOrigin.y));
+	} else if(state == 'move') {
+		// Move mode
+		var p = getEventPoint(evt).matrixTransform(g.getCTM().inverse());
+
+		setCTM(stateTarget, root.createSVGMatrix().translate(p.x - stateOrigin.x, p.y - stateOrigin.y).multiply(g.getCTM().inverse()).multiply(stateTarget.getCTM()));
+
+		stateOrigin = p;
+	}
+}
+
+/**
+ * Handle click event.
+ */
+function handleMouseDown(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	var g = svgDoc.getElementById("viewport");
+
+	if(true || evt.target.tagName == "svg") {
+		// Pan mode
+		state = 'pan';
+
+		stateTf = g.getCTM().inverse();
+
+		stateOrigin = getEventPoint(evt).matrixTransform(stateTf);
+	} else {
+		// Move mode
+		state = 'move';
+
+		stateTarget = evt.target;
+
+		stateTf = g.getCTM().inverse();
+
+		stateOrigin = getEventPoint(evt).matrixTransform(stateTf);
+	}
+}
+
+/**
+ * Handle mouse button release event.
+ */
+function handleMouseUp(evt) {
+	if(evt.preventDefault)
+		evt.preventDefault();
+
+	evt.returnValue = false;
+
+	var svgDoc = evt.target.ownerDocument;
+
+	if(state == 'pan' || state == 'move') {
+		// Quit pan mode
+		state = '';
+	}
+}
+
+]]></script>
+EOF
+}
+
+# Provides a map from fullname to shortname for cases where the
+# shortname is ambiguous.  The symlist has both the fullname and
+# shortname for all symbols, which is usually fine, but sometimes --
+# such as overloaded functions -- two different fullnames can map to
+# the same shortname.  In that case, we use the address of the
+# function to disambiguate the two.  This function fills in a map that
+# maps fullnames to modified shortnames in such cases.  If a fullname
+# is not present in the map, the 'normal' shortname provided by the
+# symlist is the appropriate one to use.
+sub FillFullnameToShortnameMap {
+  my $symbols = shift;
+  my $fullname_to_shortname_map = shift;
+  my $shortnames_seen_once = {};
+  my $shortnames_seen_more_than_once = {};
+
+  foreach my $symlist (values(%{$symbols})) {
+    # TODO(csilvers): deal with inlined symbols too.
+    my $shortname = $symlist->[0];
+    my $fullname = $symlist->[2];
+    if ($fullname !~ /<[0-9a-fA-F]+>$/) {  # fullname doesn't end in an address
+      next;       # the only collisions we care about are when addresses differ
+    }
+    if (defined($shortnames_seen_once->{$shortname}) &&
+        $shortnames_seen_once->{$shortname} ne $fullname) {
+      $shortnames_seen_more_than_once->{$shortname} = 1;
+    } else {
+      $shortnames_seen_once->{$shortname} = $fullname;
+    }
+  }
+
+  foreach my $symlist (values(%{$symbols})) {
+    my $shortname = $symlist->[0];
+    my $fullname = $symlist->[2];
+    # TODO(csilvers): take in a list of addresses we care about, and only
+    # store in the map if $symlist->[1] is in that list.  Saves space.
+    next if defined($fullname_to_shortname_map->{$fullname});
+    if (defined($shortnames_seen_more_than_once->{$shortname})) {
+      if ($fullname =~ /<0*([^>]*)>$/) {   # fullname has address at end of it
+        $fullname_to_shortname_map->{$fullname} = "$shortname\@$1";
+      }
+    }
+  }
+}
+
+# Return a small number that identifies the argument.
+# Multiple calls with the same argument will return the same number.
+# Calls with different arguments will return different numbers.
+sub ShortIdFor {
+  my $key = shift;
+  my $id = $main::uniqueid{$key};
+  if (!defined($id)) {
+    $id = keys(%main::uniqueid) + 1;
+    $main::uniqueid{$key} = $id;
+  }
+  return $id;
+}
+
+# Translate a stack of addresses into a stack of symbols
+sub TranslateStack {
+  my $symbols = shift;
+  my $fullname_to_shortname_map = shift;
+  my $k = shift;
+
+  my @addrs = split(/\n/, $k);
+  my @result = ();
+  for (my $i = 0; $i <= $#addrs; $i++) {
+    my $a = $addrs[$i];
+
+    # Skip large addresses since they sometimes show up as fake entries on RH9
+    if (length($a) > 8 && $a gt "7fffffffffffffff") {
+      next;
+    }
+
+    if ($main::opt_disasm || $main::opt_list) {
+      # We want just the address for the key
+      push(@result, $a);
+      next;
+    }
+
+    my $symlist = $symbols->{$a};
+    if (!defined($symlist)) {
+      $symlist = [$a, "", $a];
+    }
+
+    # We can have a sequence of symbols for a particular entry
+    # (more than one symbol in the case of inlining).  Callers
+    # come before callees in symlist, so walk backwards since
+    # the translated stack should contain callees before callers.
+    for (my $j = $#{$symlist}; $j >= 2; $j -= 3) {
+      my $func = $symlist->[$j-2];
+      my $fileline = $symlist->[$j-1];
+      my $fullfunc = $symlist->[$j];
+      if (defined($fullname_to_shortname_map->{$fullfunc})) {
+        $func = $fullname_to_shortname_map->{$fullfunc};
+      }
+      if ($j > 2) {
+        $func = "$func (inline)";
+      }
+
+      # Do not merge nodes corresponding to Callback::Run since that
+      # causes confusing cycles in dot display.  Instead, we synthesize
+      # a unique name for this frame per caller.
+      if ($func =~ m/Callback.*::Run$/) {
+        my $caller = ($i > 0) ? $addrs[$i-1] : 0;
+        $func = "Run#" . ShortIdFor($caller);
+      }
+
+      if ($main::opt_addresses) {
+        push(@result, "$a $func $fileline");
+      } elsif ($main::opt_lines) {
+        if ($func eq '??' && $fileline eq '??:0') {
+          push(@result, "$a");
+        } else {
+          push(@result, "$func $fileline");
+        }
+      } elsif ($main::opt_functions) {
+        if ($func eq '??') {
+          push(@result, "$a");
+        } else {
+          push(@result, $func);
+        }
+      } elsif ($main::opt_files) {
+        if ($fileline eq '??:0' || $fileline eq '') {
+          push(@result, "$a");
+        } else {
+          my $f = $fileline;
+          $f =~ s/:\d+$//;
+          push(@result, $f);
+        }
+      } else {
+        push(@result, $a);
+        last;  # Do not print inlined info
+      }
+    }
+  }
+
+  # print join(",", @addrs), " => ", join(",", @result), "\n";
+  return @result;
+}
+
+# Generate percent string for a number and a total
+sub Percent {
+  my $num = shift;
+  my $tot = shift;
+  if ($tot != 0) {
+    return sprintf("%.1f%%", $num * 100.0 / $tot);
+  } else {
+    return ($num == 0) ? "nan" : (($num > 0) ? "+inf" : "-inf");
+  }
+}
+
+# Generate pretty-printed form of number
+sub Unparse {
+  my $num = shift;
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    if ($main::opt_inuse_objects || $main::opt_alloc_objects) {
+      return sprintf("%d", $num);
+    } else {
+      if ($main::opt_show_bytes) {
+        return sprintf("%d", $num);
+      } else {
+        return sprintf("%.1f", $num / 1048576.0);
+      }
+    }
+  } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) {
+    return sprintf("%.3f", $num / 1e9); # Convert nanoseconds to seconds
+  } else {
+    return sprintf("%d", $num);
+  }
+}
+
+# Alternate pretty-printed form: 0 maps to "."
+sub UnparseAlt {
+  my $num = shift;
+  if ($num == 0) {
+    return ".";
+  } else {
+    return Unparse($num);
+  }
+}
+
+# Alternate pretty-printed form: 0 maps to ""
+sub HtmlPrintNumber {
+  my $num = shift;
+  if ($num == 0) {
+    return "";
+  } else {
+    return Unparse($num);
+  }
+}
+
+# Return output units
+sub Units {
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    if ($main::opt_inuse_objects || $main::opt_alloc_objects) {
+      return "objects";
+    } else {
+      if ($main::opt_show_bytes) {
+        return "B";
+      } else {
+        return "MB";
+      }
+    }
+  } elsif ($main::profile_type eq 'contention' && !$main::opt_contentions) {
+    return "seconds";
+  } else {
+    return "samples";
+  }
+}
+
+##### Profile manipulation code #####
+
+# Generate flattened profile:
+# If count is charged to stack [a,b,c,d], in generated profile,
+# it will be charged to [a]
+sub FlatProfile {
+  my $profile = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    if ($#addrs >= 0) {
+      AddEntry($result, $addrs[0], $count);
+    }
+  }
+  return $result;
+}
+
+# Generate cumulative profile:
+# If count is charged to stack [a,b,c,d], in generated profile,
+# it will be charged to [a], [b], [c], [d]
+sub CumulativeProfile {
+  my $profile = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    foreach my $a (@addrs) {
+      AddEntry($result, $a, $count);
+    }
+  }
+  return $result;
+}
+
+# If the second-youngest PC on the stack is always the same, returns
+# that pc.  Otherwise, returns undef.
+sub IsSecondPcAlwaysTheSame {
+  my $profile = shift;
+
+  my $second_pc = undef;
+  foreach my $k (keys(%{$profile})) {
+    my @addrs = split(/\n/, $k);
+    if ($#addrs < 1) {
+      return undef;
+    }
+    if (not defined $second_pc) {
+      $second_pc = $addrs[1];
+    } else {
+      if ($second_pc ne $addrs[1]) {
+        return undef;
+      }
+    }
+  }
+  return $second_pc;
+}
+
+sub ExtractSymbolNameInlineStack {
+  my $symbols = shift;
+  my $address = shift;
+
+  my @stack = ();
+
+  if (exists $symbols->{$address}) {
+    my @localinlinestack = @{$symbols->{$address}};
+    for (my $i = $#localinlinestack; $i > 0; $i-=3) {
+      my $file = $localinlinestack[$i-1];
+      my $fn = $localinlinestack[$i-0];
+
+      if ($file eq "?" || $file eq ":0") {
+        $file = "??:0";
+      }
+      if ($fn eq '??') {
+        # If we can't get the symbol name, at least use the file information.
+        $fn = $file;
+      }
+      my $suffix = "[inline]";
+      if ($i == 2) {
+        $suffix = "";
+      }
+      push (@stack, $fn.$suffix);
+    }
+  }
+  else {
+    # If we can't get a symbol name, at least fill in the address.
+    push (@stack, $address);
+  }
+
+  return @stack;
+}
+
+sub ExtractSymbolLocation {
+  my $symbols = shift;
+  my $address = shift;
+  # 'addr2line' outputs "??:0" for unknown locations; we do the
+  # same to be consistent.
+  my $location = "??:0:unknown";
+  if (exists $symbols->{$address}) {
+    my $file = $symbols->{$address}->[1];
+    if ($file eq "?") {
+      $file = "??:0"
+    }
+    $location = $file . ":" . $symbols->{$address}->[0];
+  }
+  return $location;
+}
+
+# Extracts a graph of calls.
+sub ExtractCalls {
+  my $symbols = shift;
+  my $profile = shift;
+
+  my $calls = {};
+  while( my ($stack_trace, $count) = each %$profile ) {
+    my @address = split(/\n/, $stack_trace);
+    my $destination = ExtractSymbolLocation($symbols, $address[0]);
+    AddEntry($calls, $destination, $count);
+    for (my $i = 1; $i <= $#address; $i++) {
+      my $source = ExtractSymbolLocation($symbols, $address[$i]);
+      my $call = "$source -> $destination";
+      AddEntry($calls, $call, $count);
+      $destination = $source;
+    }
+  }
+
+  return $calls;
+}
+
+sub FilterFrames {
+  my $symbols = shift;
+  my $profile = shift;
+
+  if ($main::opt_retain eq '' && $main::opt_exclude eq '') {
+    return $profile;
+  }
+
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my @path = ();
+    foreach my $a (@addrs) {
+      my $sym;
+      if (exists($symbols->{$a})) {
+        $sym = $symbols->{$a}->[0];
+      } else {
+        $sym = $a;
+      }
+      if ($main::opt_retain ne '' && $sym !~ m/$main::opt_retain/) {
+        next;
+      }
+      if ($main::opt_exclude ne '' && $sym =~ m/$main::opt_exclude/) {
+        next;
+      }
+      push(@path, $a);
+    }
+    if (scalar(@path) > 0) {
+      my $reduced_path = join("\n", @path);
+      AddEntry($result, $reduced_path, $count);
+    }
+  }
+
+  return $result;
+}
+
+sub PrintCollapsedStacks {
+  my $symbols = shift;
+  my $profile = shift;
+
+  while (my ($stack_trace, $count) = each %$profile) {
+    my @address = split(/\n/, $stack_trace);
+    my @names = reverse ( map { ExtractSymbolNameInlineStack($symbols, $_) } @address );
+    printf("%s %d\n", join(";", @names), $count);
+  }
+}
+
+sub RemoveUninterestingFrames {
+  my $symbols = shift;
+  my $profile = shift;
+
+  # List of function names to skip
+  my %skip = ();
+  my $skip_regexp = 'NOMATCH';
+  if ($main::profile_type eq 'heap' || $main::profile_type eq 'growth') {
+    foreach my $name ('je_calloc',
+                      'cfree',
+                      'je_malloc',
+                      'newImpl',
+                      'void* newImpl',
+                      'je_free',
+                      'je_memalign',
+                      'je_posix_memalign',
+                      'je_aligned_alloc',
+                      'pvalloc',
+                      'je_valloc',
+                      'je_realloc',
+                      'je_mallocx',
+                      'je_rallocx',
+                      'je_xallocx',
+                      'je_dallocx',
+                      'je_sdallocx',
+                      'je_sdallocx_noflags',
+                      'tc_calloc',
+                      'tc_cfree',
+                      'tc_malloc',
+                      'tc_free',
+                      'tc_memalign',
+                      'tc_posix_memalign',
+                      'tc_pvalloc',
+                      'tc_valloc',
+                      'tc_realloc',
+                      'tc_new',
+                      'tc_delete',
+                      'tc_newarray',
+                      'tc_deletearray',
+                      'tc_new_nothrow',
+                      'tc_newarray_nothrow',
+                      'do_malloc',
+                      '::do_malloc',   # new name -- got moved to an unnamed ns
+                      '::do_malloc_or_cpp_alloc',
+                      'DoSampledAllocation',
+                      'simple_alloc::allocate',
+                      '__malloc_alloc_template::allocate',
+                      '__builtin_delete',
+                      '__builtin_new',
+                      '__builtin_vec_delete',
+                      '__builtin_vec_new',
+                      'operator new',
+                      'operator new[]',
+                      # The entry to our memory-allocation routines on OS X
+                      'malloc_zone_malloc',
+                      'malloc_zone_calloc',
+                      'malloc_zone_valloc',
+                      'malloc_zone_realloc',
+                      'malloc_zone_memalign',
+                      'malloc_zone_free',
+                      # These mark the beginning/end of our custom sections
+                      '__start_google_malloc',
+                      '__stop_google_malloc',
+                      '__start_malloc_hook',
+                      '__stop_malloc_hook') {
+      $skip{$name} = 1;
+      $skip{"_" . $name} = 1;   # Mach (OS X) adds a _ prefix to everything
+    }
+    # TODO: Remove TCMalloc once everything has been
+    # moved into the tcmalloc:: namespace and we have flushed
+    # old code out of the system.
+    $skip_regexp = "TCMalloc|^tcmalloc::";
+  } elsif ($main::profile_type eq 'contention') {
+    foreach my $vname ('base::RecordLockProfileData',
+                       'base::SubmitMutexProfileData',
+                       'base::SubmitSpinLockProfileData',
+                       'Mutex::Unlock',
+                       'Mutex::UnlockSlow',
+                       'Mutex::ReaderUnlock',
+                       'MutexLock::~MutexLock',
+                       'SpinLock::Unlock',
+                       'SpinLock::SlowUnlock',
+                       'SpinLockHolder::~SpinLockHolder') {
+      $skip{$vname} = 1;
+    }
+  } elsif ($main::profile_type eq 'cpu') {
+    # Drop signal handlers used for CPU profile collection
+    # TODO(dpeng): this should not be necessary; it's taken
+    # care of by the general 2nd-pc mechanism below.
+    foreach my $name ('ProfileData::Add',           # historical
+                      'ProfileData::prof_handler',  # historical
+                      'CpuProfiler::prof_handler',
+                      '__FRAME_END__',
+                      '__pthread_sighandler',
+                      '__restore') {
+      $skip{$name} = 1;
+    }
+  } else {
+    # Nothing skipped for unknown types
+  }
+
+  if ($main::profile_type eq 'cpu') {
+    # If all the second-youngest program counters are the same,
+    # this STRONGLY suggests that it is an artifact of measurement,
+    # i.e., stack frames pushed by the CPU profiler signal handler.
+    # Hence, we delete them.
+    # (The topmost PC is read from the signal structure, not from
+    # the stack, so it does not get involved.)
+    while (my $second_pc = IsSecondPcAlwaysTheSame($profile)) {
+      my $result = {};
+      my $func = '';
+      if (exists($symbols->{$second_pc})) {
+        $second_pc = $symbols->{$second_pc}->[0];
+      }
+      print STDERR "Removing $second_pc from all stack traces.\n";
+      foreach my $k (keys(%{$profile})) {
+        my $count = $profile->{$k};
+        my @addrs = split(/\n/, $k);
+        splice @addrs, 1, 1;
+        my $reduced_path = join("\n", @addrs);
+        AddEntry($result, $reduced_path, $count);
+      }
+      $profile = $result;
+    }
+  }
+
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my @path = ();
+    foreach my $a (@addrs) {
+      if (exists($symbols->{$a})) {
+        my $func = $symbols->{$a}->[0];
+        if ($skip{$func} || ($func =~ m/$skip_regexp/)) {
+          # Throw away the portion of the backtrace seen so far, under the
+          # assumption that previous frames were for functions internal to the
+          # allocator.
+          @path = ();
+          next;
+        }
+      }
+      push(@path, $a);
+    }
+    my $reduced_path = join("\n", @path);
+    AddEntry($result, $reduced_path, $count);
+  }
+
+  $result = FilterFrames($symbols, $result);
+
+  return $result;
+}
+
+# Reduce profile to granularity given by user
+sub ReduceProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $result = {};
+  my $fullname_to_shortname_map = {};
+  FillFullnameToShortnameMap($symbols, $fullname_to_shortname_map);
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @translated = TranslateStack($symbols, $fullname_to_shortname_map, $k);
+    my @path = ();
+    my %seen = ();
+    $seen{''} = 1;      # So that empty keys are skipped
+    foreach my $e (@translated) {
+      # To avoid double-counting due to recursion, skip a stack-trace
+      # entry if it has already been seen
+      if (!$seen{$e}) {
+        $seen{$e} = 1;
+        push(@path, $e);
+      }
+    }
+    my $reduced_path = join("\n", @path);
+    AddEntry($result, $reduced_path, $count);
+  }
+  return $result;
+}
+
+# Does the specified symbol array match the regexp?
+sub SymbolMatches {
+  my $sym = shift;
+  my $re = shift;
+  if (defined($sym)) {
+    for (my $i = 0; $i < $#{$sym}; $i += 3) {
+      if ($sym->[$i] =~ m/$re/ || $sym->[$i+1] =~ m/$re/) {
+        return 1;
+      }
+    }
+  }
+  return 0;
+}
+
+# Focus only on paths involving specified regexps
+sub FocusProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $focus = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    foreach my $a (@addrs) {
+      # Reply if it matches either the address/shortname/fileline
+      if (($a =~ m/$focus/) || SymbolMatches($symbols->{$a}, $focus)) {
+        AddEntry($result, $k, $count);
+        last;
+      }
+    }
+  }
+  return $result;
+}
+
+# Focus only on paths not involving specified regexps
+sub IgnoreProfile {
+  my $symbols = shift;
+  my $profile = shift;
+  my $ignore = shift;
+  my $result = {};
+  foreach my $k (keys(%{$profile})) {
+    my $count = $profile->{$k};
+    my @addrs = split(/\n/, $k);
+    my $matched = 0;
+    foreach my $a (@addrs) {
+      # Reply if it matches either the address/shortname/fileline
+      if (($a =~ m/$ignore/) || SymbolMatches($symbols->{$a}, $ignore)) {
+        $matched = 1;
+        last;
+      }
+    }
+    if (!$matched) {
+      AddEntry($result, $k, $count);
+    }
+  }
+  return $result;
+}
+
+# Get total count in profile
+sub TotalProfile {
+  my $profile = shift;
+  my $result = 0;
+  foreach my $k (keys(%{$profile})) {
+    $result += $profile->{$k};
+  }
+  return $result;
+}
+
+# Add A to B
+sub AddProfile {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  # add all keys in A
+  foreach my $k (keys(%{$A})) {
+    my $v = $A->{$k};
+    AddEntry($R, $k, $v);
+  }
+  # add all keys in B
+  foreach my $k (keys(%{$B})) {
+    my $v = $B->{$k};
+    AddEntry($R, $k, $v);
+  }
+  return $R;
+}
+
+# Merges symbol maps
+sub MergeSymbols {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  foreach my $k (keys(%{$A})) {
+    $R->{$k} = $A->{$k};
+  }
+  if (defined($B)) {
+    foreach my $k (keys(%{$B})) {
+      $R->{$k} = $B->{$k};
+    }
+  }
+  return $R;
+}
+
+
+# Add A to B
+sub AddPcs {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  # add all keys in A
+  foreach my $k (keys(%{$A})) {
+    $R->{$k} = 1
+  }
+  # add all keys in B
+  foreach my $k (keys(%{$B})) {
+    $R->{$k} = 1
+  }
+  return $R;
+}
+
+# Subtract B from A
+sub SubtractProfile {
+  my $A = shift;
+  my $B = shift;
+
+  my $R = {};
+  foreach my $k (keys(%{$A})) {
+    my $v = $A->{$k} - GetEntry($B, $k);
+    if ($v < 0 && $main::opt_drop_negative) {
+      $v = 0;
+    }
+    AddEntry($R, $k, $v);
+  }
+  if (!$main::opt_drop_negative) {
+    # Take care of when subtracted profile has more entries
+    foreach my $k (keys(%{$B})) {
+      if (!exists($A->{$k})) {
+        AddEntry($R, $k, 0 - $B->{$k});
+      }
+    }
+  }
+  return $R;
+}
+
+# Get entry from profile; zero if not present
+sub GetEntry {
+  my $profile = shift;
+  my $k = shift;
+  if (exists($profile->{$k})) {
+    return $profile->{$k};
+  } else {
+    return 0;
+  }
+}
+
+# Add entry to specified profile
+sub AddEntry {
+  my $profile = shift;
+  my $k = shift;
+  my $n = shift;
+  if (!exists($profile->{$k})) {
+    $profile->{$k} = 0;
+  }
+  $profile->{$k} += $n;
+}
+
+# Add a stack of entries to specified profile, and add them to the $pcs
+# list.
+sub AddEntries {
+  my $profile = shift;
+  my $pcs = shift;
+  my $stack = shift;
+  my $count = shift;
+  my @k = ();
+
+  foreach my $e (split(/\s+/, $stack)) {
+    my $pc = HexExtend($e);
+    $pcs->{$pc} = 1;
+    push @k, $pc;
+  }
+  AddEntry($profile, (join "\n", @k), $count);
+}
+
+##### Code to profile a server dynamically #####
+
+sub CheckSymbolPage {
+  my $url = SymbolPageURL();
+  my $command = ShellEscape(@URL_FETCHER, $url);
+  open(SYMBOL, "$command |") or error($command);
+  my $line = <SYMBOL>;
+  $line =~ s/\r//g;         # turn windows-looking lines into unix-looking lines
+  close(SYMBOL);
+  unless (defined($line)) {
+    error("$url doesn't exist\n");
+  }
+
+  if ($line =~ /^num_symbols:\s+(\d+)$/) {
+    if ($1 == 0) {
+      error("Stripped binary. No symbols available.\n");
+    }
+  } else {
+    error("Failed to get the number of symbols from $url\n");
+  }
+}
+
+sub IsProfileURL {
+  my $profile_name = shift;
+  if (-f $profile_name) {
+    printf STDERR "Using local file $profile_name.\n";
+    return 0;
+  }
+  return 1;
+}
+
+sub ParseProfileURL {
+  my $profile_name = shift;
+
+  if (!defined($profile_name) || $profile_name eq "") {
+    return ();
+  }
+
+  # Split profile URL - matches all non-empty strings, so no test.
+  $profile_name =~ m,^(https?://)?([^/]+)(.*?)(/|$PROFILES)?$,;
+
+  my $proto = $1 || "http://";
+  my $hostport = $2;
+  my $prefix = $3;
+  my $profile = $4 || "/";
+
+  my $host = $hostport;
+  $host =~ s/:.*//;
+
+  my $baseurl = "$proto$hostport$prefix";
+  return ($host, $baseurl, $profile);
+}
+
+# We fetch symbols from the first profile argument.
+sub SymbolPageURL {
+  my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]);
+  return "$baseURL$SYMBOL_PAGE";
+}
+
+sub FetchProgramName() {
+  my ($host, $baseURL, $path) = ParseProfileURL($main::pfile_args[0]);
+  my $url = "$baseURL$PROGRAM_NAME_PAGE";
+  my $command_line = ShellEscape(@URL_FETCHER, $url);
+  open(CMDLINE, "$command_line |") or error($command_line);
+  my $cmdline = <CMDLINE>;
+  $cmdline =~ s/\r//g;   # turn windows-looking lines into unix-looking lines
+  close(CMDLINE);
+  error("Failed to get program name from $url\n") unless defined($cmdline);
+  $cmdline =~ s/\x00.+//;  # Remove argv[1] and latters.
+  $cmdline =~ s!\n!!g;  # Remove LFs.
+  return $cmdline;
+}
+
+# Gee, curl's -L (--location) option isn't reliable at least
+# with its 7.12.3 version.  Curl will forget to post data if
+# there is a redirection.  This function is a workaround for
+# curl.  Redirection happens on borg hosts.
+sub ResolveRedirectionForCurl {
+  my $url = shift;
+  my $command_line = ShellEscape(@URL_FETCHER, "--head", $url);
+  open(CMDLINE, "$command_line |") or error($command_line);
+  while (<CMDLINE>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    if (/^Location: (.*)/) {
+      $url = $1;
+    }
+  }
+  close(CMDLINE);
+  return $url;
+}
+
+# Add a timeout flat to URL_FETCHER.  Returns a new list.
+sub AddFetchTimeout {
+  my $timeout = shift;
+  my @fetcher = @_;
+  if (defined($timeout)) {
+    if (join(" ", @fetcher) =~ m/\bcurl -s/) {
+      push(@fetcher, "--max-time", sprintf("%d", $timeout));
+    } elsif (join(" ", @fetcher) =~ m/\brpcget\b/) {
+      push(@fetcher, sprintf("--deadline=%d", $timeout));
+    }
+  }
+  return @fetcher;
+}
+
+# Reads a symbol map from the file handle name given as $1, returning
+# the resulting symbol map.  Also processes variables relating to symbols.
+# Currently, the only variable processed is 'binary=<value>' which updates
+# $main::prog to have the correct program name.
+sub ReadSymbols {
+  my $in = shift;
+  my $map = {};
+  while (<$in>) {
+    s/\r//g;         # turn windows-looking lines into unix-looking lines
+    # Removes all the leading zeroes from the symbols, see comment below.
+    if (m/^0x0*([0-9a-f]+)\s+(.+)/) {
+      $map->{$1} = $2;
+    } elsif (m/^---/) {
+      last;
+    } elsif (m/^([a-z][^=]*)=(.*)$/ ) {
+      my ($variable, $value) = ($1, $2);
+      for ($variable, $value) {
+        s/^\s+//;
+        s/\s+$//;
+      }
+      if ($variable eq "binary") {
+        if ($main::prog ne $UNKNOWN_BINARY && $main::prog ne $value) {
+          printf STDERR ("Warning: Mismatched binary name '%s', using '%s'.\n",
+                         $main::prog, $value);
+        }
+        $main::prog = $value;
+      } else {
+        printf STDERR ("Ignoring unknown variable in symbols list: " .
+            "'%s' = '%s'\n", $variable, $value);
+      }
+    }
+  }
+  return $map;
+}
+
+sub URLEncode {
+  my $str = shift;
+  $str =~ s/([^A-Za-z0-9\-_.!~*'()])/ sprintf "%%%02x", ord $1 /eg;
+  return $str;
+}
+
+sub AppendSymbolFilterParams {
+  my $url = shift;
+  my @params = ();
+  if ($main::opt_retain ne '') {
+    push(@params, sprintf("retain=%s", URLEncode($main::opt_retain)));
+  }
+  if ($main::opt_exclude ne '') {
+    push(@params, sprintf("exclude=%s", URLEncode($main::opt_exclude)));
+  }
+  if (scalar @params > 0) {
+    $url = sprintf("%s?%s", $url, join("&", @params));
+  }
+  return $url;
+}
+
+# Fetches and processes symbols to prepare them for use in the profile output
+# code.  If the optional 'symbol_map' arg is not given, fetches symbols from
+# $SYMBOL_PAGE for all PC values found in profile.  Otherwise, the raw symbols
+# are assumed to have already been fetched into 'symbol_map' and are simply
+# extracted and processed.
+sub FetchSymbols {
+  my $pcset = shift;
+  my $symbol_map = shift;
+
+  my %seen = ();
+  my @pcs = grep { !$seen{$_}++ } keys(%$pcset);  # uniq
+
+  if (!defined($symbol_map)) {
+    my $post_data = join("+", sort((map {"0x" . "$_"} @pcs)));
+
+    open(POSTFILE, ">$main::tmpfile_sym");
+    print POSTFILE $post_data;
+    close(POSTFILE);
+
+    my $url = SymbolPageURL();
+
+    my $command_line;
+    if (join(" ", @URL_FETCHER) =~ m/\bcurl -s/) {
+      $url = ResolveRedirectionForCurl($url);
+      $url = AppendSymbolFilterParams($url);
+      $command_line = ShellEscape(@URL_FETCHER, "-d", "\@$main::tmpfile_sym",
+                                  $url);
+    } else {
+      $url = AppendSymbolFilterParams($url);
+      $command_line = (ShellEscape(@URL_FETCHER, "--post", $url)
+                       . " < " . ShellEscape($main::tmpfile_sym));
+    }
+    # We use c++filt in case $SYMBOL_PAGE gives us mangled symbols.
+    my $escaped_cppfilt = ShellEscape($obj_tool_map{"c++filt"});
+    open(SYMBOL, "$command_line | $escaped_cppfilt |") or error($command_line);
+    $symbol_map = ReadSymbols(*SYMBOL{IO});
+    close(SYMBOL);
+  }
+
+  my $symbols = {};
+  foreach my $pc (@pcs) {
+    my $fullname;
+    # For 64 bits binaries, symbols are extracted with 8 leading zeroes.
+    # Then /symbol reads the long symbols in as uint64, and outputs
+    # the result with a "0x%08llx" format which get rid of the zeroes.
+    # By removing all the leading zeroes in both $pc and the symbols from
+    # /symbol, the symbols match and are retrievable from the map.
+    my $shortpc = $pc;
+    $shortpc =~ s/^0*//;
+    # Each line may have a list of names, which includes the function
+    # and also other functions it has inlined.  They are separated (in
+    # PrintSymbolizedProfile), by --, which is illegal in function names.
+    my $fullnames;
+    if (defined($symbol_map->{$shortpc})) {
+      $fullnames = $symbol_map->{$shortpc};
+    } else {
+      $fullnames = "0x" . $pc;  # Just use addresses
+    }
+    my $sym = [];
+    $symbols->{$pc} = $sym;
+    foreach my $fullname (split("--", $fullnames)) {
+      my $name = ShortFunctionName($fullname);
+      push(@{$sym}, $name, "?", $fullname);
+    }
+  }
+  return $symbols;
+}
+
+sub BaseName {
+  my $file_name = shift;
+  $file_name =~ s!^.*/!!;  # Remove directory name
+  return $file_name;
+}
+
+sub MakeProfileBaseName {
+  my ($binary_name, $profile_name) = @_;
+  my ($host, $baseURL, $path) = ParseProfileURL($profile_name);
+  my $binary_shortname = BaseName($binary_name);
+  return sprintf("%s.%s.%s",
+                 $binary_shortname, $main::op_time, $host);
+}
+
+sub FetchDynamicProfile {
+  my $binary_name = shift;
+  my $profile_name = shift;
+  my $fetch_name_only = shift;
+  my $encourage_patience = shift;
+
+  if (!IsProfileURL($profile_name)) {
+    return $profile_name;
+  } else {
+    my ($host, $baseURL, $path) = ParseProfileURL($profile_name);
+    if ($path eq "" || $path eq "/") {
+      # Missing type specifier defaults to cpu-profile
+      $path = $PROFILE_PAGE;
+    }
+
+    my $profile_file = MakeProfileBaseName($binary_name, $profile_name);
+
+    my $url = "$baseURL$path";
+    my $fetch_timeout = undef;
+    if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE/) {
+      if ($path =~ m/[?]/) {
+        $url .= "&";
+      } else {
+        $url .= "?";
+      }
+      $url .= sprintf("seconds=%d", $main::opt_seconds);
+      $fetch_timeout = $main::opt_seconds * 1.01 + 60;
+      # Set $profile_type for consumption by PrintSymbolizedProfile.
+      $main::profile_type = 'cpu';
+    } else {
+      # For non-CPU profiles, we add a type-extension to
+      # the target profile file name.
+      my $suffix = $path;
+      $suffix =~ s,/,.,g;
+      $profile_file .= $suffix;
+      # Set $profile_type for consumption by PrintSymbolizedProfile.
+      if ($path =~ m/$HEAP_PAGE/) {
+        $main::profile_type = 'heap';
+      } elsif ($path =~ m/$GROWTH_PAGE/) {
+        $main::profile_type = 'growth';
+      } elsif ($path =~ m/$CONTENTION_PAGE/) {
+        $main::profile_type = 'contention';
+      }
+    }
+
+    my $profile_dir = $ENV{"JEPROF_TMPDIR"} || ($ENV{HOME} . "/jeprof");
+    if (! -d $profile_dir) {
+      mkdir($profile_dir)
+          || die("Unable to create profile directory $profile_dir: $!\n");
+    }
+    my $tmp_profile = "$profile_dir/.tmp.$profile_file";
+    my $real_profile = "$profile_dir/$profile_file";
+
+    if ($fetch_name_only > 0) {
+      return $real_profile;
+    }
+
+    my @fetcher = AddFetchTimeout($fetch_timeout, @URL_FETCHER);
+    my $cmd = ShellEscape(@fetcher, $url) . " > " . ShellEscape($tmp_profile);
+    if ($path =~ m/$PROFILE_PAGE|$PMUPROFILE_PAGE|$CENSUSPROFILE_PAGE/){
+      print STDERR "Gathering CPU profile from $url for $main::opt_seconds seconds to\n  ${real_profile}\n";
+      if ($encourage_patience) {
+        print STDERR "Be patient...\n";
+      }
+    } else {
+      print STDERR "Fetching $path profile from $url to\n  ${real_profile}\n";
+    }
+
+    (system($cmd) == 0) || error("Failed to get profile: $cmd: $!\n");
+    (system("mv", $tmp_profile, $real_profile) == 0) || error("Unable to rename profile\n");
+    print STDERR "Wrote profile to $real_profile\n";
+    $main::collected_profile = $real_profile;
+    return $main::collected_profile;
+  }
+}
+
+# Collect profiles in parallel
+sub FetchDynamicProfiles {
+  my $items = scalar(@main::pfile_args);
+  my $levels = log($items) / log(2);
+
+  if ($items == 1) {
+    $main::profile_files[0] = FetchDynamicProfile($main::prog, $main::pfile_args[0], 0, 1);
+  } else {
+    # math rounding issues
+    if ((2 ** $levels) < $items) {
+     $levels++;
+    }
+    my $count = scalar(@main::pfile_args);
+    for (my $i = 0; $i < $count; $i++) {
+      $main::profile_files[$i] = FetchDynamicProfile($main::prog, $main::pfile_args[$i], 1, 0);
+    }
+    print STDERR "Fetching $count profiles, Be patient...\n";
+    FetchDynamicProfilesRecurse($levels, 0, 0);
+    $main::collected_profile = join(" \\\n    ", @main::profile_files);
+  }
+}
+
+# Recursively fork a process to get enough processes
+# collecting profiles
+sub FetchDynamicProfilesRecurse {
+  my $maxlevel = shift;
+  my $level = shift;
+  my $position = shift;
+
+  if (my $pid = fork()) {
+    $position = 0 | ($position << 1);
+    TryCollectProfile($maxlevel, $level, $position);
+    wait;
+  } else {
+    $position = 1 | ($position << 1);
+    TryCollectProfile($maxlevel, $level, $position);
+    cleanup();
+    exit(0);
+  }
+}
+
+# Collect a single profile
+sub TryCollectProfile {
+  my $maxlevel = shift;
+  my $level = shift;
+  my $position = shift;
+
+  if ($level >= ($maxlevel - 1)) {
+    if ($position < scalar(@main::pfile_args)) {
+      FetchDynamicProfile($main::prog, $main::pfile_args[$position], 0, 0);
+    }
+  } else {
+    FetchDynamicProfilesRecurse($maxlevel, $level+1, $position);
+  }
+}
+
+##### Parsing code #####
+
+# Provide a small streaming-read module to handle very large
+# cpu-profile files.  Stream in chunks along a sliding window.
+# Provides an interface to get one 'slot', correctly handling
+# endian-ness differences.  A slot is one 32-bit or 64-bit word
+# (depending on the input profile).  We tell endianness and bit-size
+# for the profile by looking at the first 8 bytes: in cpu profiles,
+# the second slot is always 3 (we'll accept anything that's not 0).
+BEGIN {
+  package CpuProfileStream;
+
+  sub new {
+    my ($class, $file, $fname) = @_;
+    my $self = { file        => $file,
+                 base        => 0,
+                 stride      => 512 * 1024,   # must be a multiple of bitsize/8
+                 slots       => [],
+                 unpack_code => "",           # N for big-endian, V for little
+                 perl_is_64bit => 1,          # matters if profile is 64-bit
+    };
+    bless $self, $class;
+    # Let unittests adjust the stride
+    if ($main::opt_test_stride > 0) {
+      $self->{stride} = $main::opt_test_stride;
+    }
+    # Read the first two slots to figure out bitsize and endianness.
+    my $slots = $self->{slots};
+    my $str;
+    read($self->{file}, $str, 8);
+    # Set the global $address_length based on what we see here.
+    # 8 is 32-bit (8 hexadecimal chars); 16 is 64-bit (16 hexadecimal chars).
+    $address_length = ($str eq (chr(0)x8)) ? 16 : 8;
+    if ($address_length == 8) {
+      if (substr($str, 6, 2) eq chr(0)x2) {
+        $self->{unpack_code} = 'V';  # Little-endian.
+      } elsif (substr($str, 4, 2) eq chr(0)x2) {
+        $self->{unpack_code} = 'N';  # Big-endian
+      } else {
+        ::error("$fname: header size >= 2**16\n");
+      }
+      @$slots = unpack($self->{unpack_code} . "*", $str);
+    } else {
+      # If we're a 64-bit profile, check if we're a 64-bit-capable
+      # perl.  Otherwise, each slot will be represented as a float
+      # instead of an int64, losing precision and making all the
+      # 64-bit addresses wrong.  We won't complain yet, but will
+      # later if we ever see a value that doesn't fit in 32 bits.
+      my $has_q = 0;
+      eval { $has_q = pack("Q", "1") ? 1 : 1; };
+      if (!$has_q) {
+        $self->{perl_is_64bit} = 0;
+      }
+      read($self->{file}, $str, 8);
+      if (substr($str, 4, 4) eq chr(0)x4) {
+        # We'd love to use 'Q', but it's a) not universal, b) not endian-proof.
+        $self->{unpack_code} = 'V';  # Little-endian.
+      } elsif (substr($str, 0, 4) eq chr(0)x4) {
+        $self->{unpack_code} = 'N';  # Big-endian
+      } else {
+        ::error("$fname: header size >= 2**32\n");
+      }
+      my @pair = unpack($self->{unpack_code} . "*", $str);
+      # Since we know one of the pair is 0, it's fine to just add them.
+      @$slots = (0, $pair[0] + $pair[1]);
+    }
+    return $self;
+  }
+
+  # Load more data when we access slots->get(X) which is not yet in memory.
+  sub overflow {
+    my ($self) = @_;
+    my $slots = $self->{slots};
+    $self->{base} += $#$slots + 1;   # skip over data we're replacing
+    my $str;
+    read($self->{file}, $str, $self->{stride});
+    if ($address_length == 8) {      # the 32-bit case
+      # This is the easy case: unpack provides 32-bit unpacking primitives.
+      @$slots = unpack($self->{unpack_code} . "*", $str);
+    } else {
+      # We need to unpack 32 bits at a time and combine.
+      my @b32_values = unpack($self->{unpack_code} . "*", $str);
+      my @b64_values = ();
+      for (my $i = 0; $i < $#b32_values; $i += 2) {
+        # TODO(csilvers): if this is a 32-bit perl, the math below
+        #    could end up in a too-large int, which perl will promote
+        #    to a double, losing necessary precision.  Deal with that.
+        #    Right now, we just die.
+        my ($lo, $hi) = ($b32_values[$i], $b32_values[$i+1]);
+        if ($self->{unpack_code} eq 'N') {    # big-endian
+          ($lo, $hi) = ($hi, $lo);
+        }
+        my $value = $lo + $hi * (2**32);
+        if (!$self->{perl_is_64bit} &&   # check value is exactly represented
+            (($value % (2**32)) != $lo || int($value / (2**32)) != $hi)) {
+          ::error("Need a 64-bit perl to process this 64-bit profile.\n");
+        }
+        push(@b64_values, $value);
+      }
+      @$slots = @b64_values;
+    }
+  }
+
+  # Access the i-th long in the file (logically), or -1 at EOF.
+  sub get {
+    my ($self, $idx) = @_;
+    my $slots = $self->{slots};
+    while ($#$slots >= 0) {
+      if ($idx < $self->{base}) {
+        # The only time we expect a reference to $slots[$i - something]
+        # after referencing $slots[$i] is reading the very first header.
+        # Since $stride > |header|, that shouldn't cause any lookback
+        # errors.  And everything after the header is sequential.
+        print STDERR "Unexpected look-back reading CPU profile";
+        return -1;   # shrug, don't know what better to return
+      } elsif ($idx > $self->{base} + $#$slots) {
+        $self->overflow();
+      } else {
+        return $slots->[$idx - $self->{base}];
+      }
+    }
+    # If we get here, $slots is [], which means we've reached EOF
+    return -1;  # unique since slots is supposed to hold unsigned numbers
+  }
+}
+
+# Reads the top, 'header' section of a profile, and returns the last
+# line of the header, commonly called a 'header line'.  The header
+# section of a profile consists of zero or more 'command' lines that
+# are instructions to jeprof, which jeprof executes when reading the
+# header.  All 'command' lines start with a %.  After the command
+# lines is the 'header line', which is a profile-specific line that
+# indicates what type of profile it is, and perhaps other global
+# information about the profile.  For instance, here's a header line
+# for a heap profile:
+#   heap profile:     53:    38236 [  5525:  1284029] @ heapprofile
+# For historical reasons, the CPU profile does not contain a text-
+# readable header line.  If the profile looks like a CPU profile,
+# this function returns "".  If no header line could be found, this
+# function returns undef.
+#
+# The following commands are recognized:
+#   %warn -- emit the rest of this line to stderr, prefixed by 'WARNING:'
+#
+# The input file should be in binmode.
+sub ReadProfileHeader {
+  local *PROFILE = shift;
+  my $firstchar = "";
+  my $line = "";
+  read(PROFILE, $firstchar, 1);
+  seek(PROFILE, -1, 1);                    # unread the firstchar
+  if ($firstchar !~ /[[:print:]]/) {       # is not a text character
+    return "";
+  }
+  while (defined($line = <PROFILE>)) {
+    $line =~ s/\r//g;   # turn windows-looking lines into unix-looking lines
+    if ($line =~ /^%warn\s+(.*)/) {        # 'warn' command
+      # Note this matches both '%warn blah\n' and '%warn\n'.
+      print STDERR "WARNING: $1\n";        # print the rest of the line
+    } elsif ($line =~ /^%/) {
+      print STDERR "Ignoring unknown command from profile header: $line";
+    } else {
+      # End of commands, must be the header line.
+      return $line;
+    }
+  }
+  return undef;     # got to EOF without seeing a header line
+}
+
+sub IsSymbolizedProfileFile {
+  my $file_name = shift;
+  if (!(-e $file_name) || !(-r $file_name)) {
+    return 0;
+  }
+  # Check if the file contains a symbol-section marker.
+  open(TFILE, "<$file_name");
+  binmode TFILE;
+  my $firstline = ReadProfileHeader(*TFILE);
+  close(TFILE);
+  if (!$firstline) {
+    return 0;
+  }
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+  return $firstline =~ /^--- *$symbol_marker/;
+}
+
+# Parse profile generated by common/profiler.cc and return a reference
+# to a map:
+#      $result->{version}     Version number of profile file
+#      $result->{period}      Sampling period (in microseconds)
+#      $result->{profile}     Profile object
+#      $result->{threads}     Map of thread IDs to profile objects
+#      $result->{map}         Memory map info from profile
+#      $result->{pcs}         Hash of all PC values seen, key is hex address
+sub ReadProfile {
+  my $prog = shift;
+  my $fname = shift;
+  my $result;            # return value
+
+  $CONTENTION_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $contention_marker = $&;
+  $GROWTH_PAGE  =~ m,[^/]+$,;    # matches everything after the last slash
+  my $growth_marker = $&;
+  $SYMBOL_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $symbol_marker = $&;
+  $PROFILE_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $profile_marker = $&;
+  $HEAP_PAGE =~ m,[^/]+$,;    # matches everything after the last slash
+  my $heap_marker = $&;
+
+  # Look at first line to see if it is a heap or a CPU profile.
+  # CPU profile may start with no header at all, and just binary data
+  # (starting with \0\0\0\0) -- in that case, don't try to read the
+  # whole firstline, since it may be gigabytes(!) of data.
+  open(PROFILE, "<$fname") || error("$fname: $!\n");
+  binmode PROFILE;      # New perls do UTF-8 processing
+  my $header = ReadProfileHeader(*PROFILE);
+  if (!defined($header)) {   # means "at EOF"
+    error("Profile is empty.\n");
+  }
+
+  my $symbols;
+  if ($header =~ m/^--- *$symbol_marker/o) {
+    # Verify that the user asked for a symbolized profile
+    if (!$main::use_symbolized_profile) {
+      # we have both a binary and symbolized profiles, abort
+      error("FATAL ERROR: Symbolized profile\n   $fname\ncannot be used with " .
+            "a binary arg. Try again without passing\n   $prog\n");
+    }
+    # Read the symbol section of the symbolized profile file.
+    $symbols = ReadSymbols(*PROFILE{IO});
+    # Read the next line to get the header for the remaining profile.
+    $header = ReadProfileHeader(*PROFILE) || "";
+  }
+
+  if ($header =~ m/^--- *($heap_marker|$growth_marker)/o) {
+    # Skip "--- ..." line for profile types that have their own headers.
+    $header = ReadProfileHeader(*PROFILE) || "";
+  }
+
+  $main::profile_type = '';
+
+  if ($header =~ m/^heap profile:.*$growth_marker/o) {
+    $main::profile_type = 'growth';
+    $result =  ReadHeapProfile($prog, *PROFILE, $header);
+  } elsif ($header =~ m/^heap profile:/) {
+    $main::profile_type = 'heap';
+    $result =  ReadHeapProfile($prog, *PROFILE, $header);
+  } elsif ($header =~ m/^heap/) {
+    $main::profile_type = 'heap';
+    $result = ReadThreadedHeapProfile($prog, $fname, $header);
+  } elsif ($header =~ m/^--- *$contention_marker/o) {
+    $main::profile_type = 'contention';
+    $result = ReadSynchProfile($prog, *PROFILE);
+  } elsif ($header =~ m/^--- *Stacks:/) {
+    print STDERR
+      "Old format contention profile: mistakenly reports " .
+      "condition variable signals as lock contentions.\n";
+    $main::profile_type = 'contention';
+    $result = ReadSynchProfile($prog, *PROFILE);
+  } elsif ($header =~ m/^--- *$profile_marker/) {
+    # the binary cpu profile data starts immediately after this line
+    $main::profile_type = 'cpu';
+    $result = ReadCPUProfile($prog, $fname, *PROFILE);
+  } else {
+    if (defined($symbols)) {
+      # a symbolized profile contains a format we don't recognize, bail out
+      error("$fname: Cannot recognize profile section after symbols.\n");
+    }
+    # no ascii header present -- must be a CPU profile
+    $main::profile_type

[truncated 1061424 bytes]