#!/usr/bin/env perl
=head1 NAME
sizeme_store.pl - process and store the raw data stream from Devel::SizeMe
=head1 SYNOPSIS
sizeme_store.pl [--text] [--dot=sizeme.dot] [--db=sizeme.db]
Typically used with Devel::SizeMe via the C<SIZEME> env var:
export SIZEME='|./sizeme_store.pl --text'
export SIZEME='|./sizeme_store.pl --dot=sizeme.dot'
export SIZEME='|./sizeme_store.pl --db=sizeme.db'
=head1 DESCRIPTION
Reads the raw memory data from Devel::SizeMe and processes the tree
via a stack, propagating data such as totals, up the tree nodes
as the data streams through. Output completed nodes in the request formats.
The --text output is similar to the textual representation output by the module
when the SIZEME env var is set to an empty string.
The --dot output is suitable for feeding to Graphviz. (On OSX the Graphviz
application will be started automatically.)
The --db output is a SQLite database. The db schema is very subject to change.
This output is destined to be the primary one. The other output types will
probably become separate programs that read the db.
=head1 TODO
Current implementation is all very alpha and rather hackish.
Refactor to separate the core code into a module.
Move the output formats into separate modules, which should probably read from
the db so the db becomes the canonical source of data.
Import constants from XS.
=cut
# Needs to be generalized to support pluggable output formats.
# Actually it needs to be split so sizeme_store.pl only does the store
# and another program drives the output with plugins.
# Making nodes into (lightweight fast) objects would be smart.
# Tests would be even smarter!
#
# When working on this code it's important to have a sense of the flow.
# Specifically the way that depth drives the completion of nodes.
# It's a depth-first stream processing machine, which only ever holds
# a single stack of the currently incomplete nodes, which is always the same as
# the current depth. I.e., when a node of depth N arrives, all nodes >N are
# popped off the stack and 'completed', each rippling data up to its parent.
use strict;
use warnings;
use autodie;
use DBI qw(looks_like_number);
use DBD::SQLite;
use JSON::XS;
use Devel::Dwarn;
use HTML::Entities qw(encode_entities);;
use Data::Dumper;
use Getopt::Long;
use Carp qw(carp croak confess);
# XXX import these from the XS code
use constant NPtype_NAME => 0x01;
use constant NPtype_LINK => 0x02;
use constant NPtype_SV => 0x03;
use constant NPtype_MAGIC => 0x04;
use constant NPtype_OP => 0x05;
use constant NPattr_LEAFSIZE => 0x00;
use constant NPattr_NAME => 0x01;
use constant NPattr_PADFAKE => 0x02;
use constant NPattr_PADNAME => 0x03;
use constant NPattr_PADTMP => 0x04;
use constant NPattr_NOTE => 0x05;
use constant NPattr_PRE_ATTR => 0x06;
my @attr_type_name = (qw(size NAME PADFAKE my PADTMP NOTE PREATTR)); # XXX get from XS in some way
GetOptions(
'text!' => \my $opt_text,
'dot=s' => \my $opt_dot,
'db=s' => \my $opt_db,
'verbose|v!' => \my $opt_verbose,
'debug|d!' => \my $opt_debug,
'showid!' => \my $opt_showid,
) or exit 1;
$| = 1 if $opt_debug;
my $run_size = 0;
my $total_size = 0;
my $j = JSON::XS->new->ascii->pretty(0);
my ($dbh, $node_ins_sth);
if ($opt_db) {
$dbh = DBI->connect("dbi:SQLite:dbname=$opt_db","","", {
RaiseError => 1, PrintError => 0, AutoCommit => 0
});
$dbh->do("PRAGMA synchronous = OFF");
}
my @stack;
my %seqn2node;
my $dotnode = sub {
my $name = encode_entities(shift);
$name =~ s/"/\\"/g;
return '"'.$name.'"';
};
my $dot_fh;
sub fmt_size {
my $size = shift;
my $kb = $size / 1024;
return $size if $kb < 5;
return sprintf "%.1fKb", $kb if $kb < 1000;
return sprintf "%.1fMb", $kb/1024;
}
sub enter_node {
my $x = shift;
warn ">> enter_node $x->{id}\n" if $opt_debug;
my $parent = $stack[-1];
if ($parent) {
if ($x->{name} eq 'AVelem' and $parent->{name} eq 'SV(PVAV)') {
my $index = $x->{attr}{+NPattr_NOTE}{i};
#Dwarn $x->{attr};
#Dwarn $index;
# If node is an AVelem of a CvPADLIST propagate pad name to AVelem
if (@stack >= 4 and (my $cvpl = $stack[-4])->{name} eq 'CvPADLIST') {
my $padnames = $cvpl->{_cached}{padnames} ||= do {
my @names = @{ $cvpl->{attr}{+NPattr_PADNAME} || []};
$_ = "my(".($_||'').")" for @names;
$names[0] = '@_';
\@names;
};
$x->{name} = (defined $index and $padnames->[$index]) || "?";
$x->{name} =~ s/my\(SVs_PADTMP\)/PADTMP/; # XXX hack for neatness
}
else {
$x->{name} = "[$index]" if defined $index;
}
}
}
return $x;
}
sub leave_node {
my $x = shift;
confess unless defined $x->{id};
warn "<< leave_node $x->{id}\n" if $opt_debug;
delete $seqn2node{$x->{id}};
my $self_size = 0; $self_size += $_ for values %{$x->{leaves}};
$x->{self_size} = $self_size;
if ($x->{name} eq 'AVelem') {
my $index = $x->{attr}{+NPattr_NOTE}{i};
$x->{name} = "[$index]" if defined $index;
}
my $parent = $stack[-1];
if ($parent) {
# link to parent
$x->{parent_id} = $parent->{id};
# accumulate into parent
$parent->{kids_node_count} += 1 + ($x->{kids_node_count}||0);
$parent->{kids_size} += $self_size + $x->{kids_size};
push @{$parent->{child_id}}, $x->{id};
}
else {
$x->{kids_node_count} ||= 0;
}
# output
# ...
if ($opt_dot) {
printf "// n%d parent=%s(type=%s)\n", $x->{id},
$parent ? $parent->{id} : "",
$parent ? $parent->{type} : ""
if 0;
if ($x->{type} != NPtype_LINK) {
my $name = $x->{title} ? "\"$x->{title}\" $x->{name}" : $x->{name};
if ($x->{kids_size}) {
$name .= sprintf " %s+%s=%s", fmt_size($x->{self_size}), fmt_size($x->{kids_size}), fmt_size($x->{self_size}+$x->{kids_size});
}
else {
$name .= sprintf " +%s", fmt_size($x->{self_size});
}
$name .= " #$x->{id}" if $opt_showid;
my @node_attr = (
sprintf("label=%s", $dotnode->($name)),
"id=$x->{id}",
);
printf $dot_fh qq{n%d [ %s ];\n}, $x->{id}, join(",", @node_attr);
}
else { # NPtype_LINK
my @kids = @{$x->{child_id}||[]};
die "panic: NPtype_LINK has more than one child: @kids"
if @kids > 1;
for my $child_id (@kids) { # wouldn't work right, eg id= attr
#die Dwarn $x;
my @link_attr = ("id=$x->{id}");
(my $link_name = $x->{name}) =~ s/->$//;
$link_name .= " #$x->{id}" if $opt_showid;
push @link_attr, (sprintf "label=%s", $dotnode->($link_name));
printf $dot_fh qq{n%d -> n%d [%s];\n},
$x->{parent_id}, $child_id, join(",", @link_attr);
}
}
}
if ($dbh) {
my $attr_json = $j->encode($x->{attr});
my $leaves_json = $j->encode($x->{leaves});
$node_ins_sth->execute(
$x->{id}, $x->{name}, $x->{title}, $x->{type}, $x->{depth}, $x->{parent_id},
$x->{self_size}, $x->{kids_size}, $x->{kids_node_count},
$x->{child_id} ? join(",", @{$x->{child_id}}) : undef,
$attr_json, $leaves_json,
);
# XXX attribs
}
return $x;
}
my $indent = ": ";
while (<>) {
warn "\t\t\t\t== $_" if $opt_debug;
chomp;
my ($type, $id, $val, $name, $extra) = split / /, $_, 5;
if ($type =~ s/^-//) { # Node type ($val is depth)
printf "%s%s%s %s [#%d @%d]\n", $indent x $val, $name,
($type == NPtype_LINK) ? "->" : "",
$extra||'', $id, $val
if $opt_text;
# this is the core driving logic
while ($val < @stack) {
my $x = leave_node(pop @stack);
warn "N $id d$val ends $x->{id} d$x->{depth}: size $x->{self_size}+$x->{kids_size}\n"
if $opt_verbose;
}
die "panic: stack already has item at depth $val"
if $stack[$val];
die "Depth out of sync\n" if $val != @stack;
my $node = enter_node({
id => $id, type => $type, name => $name, extra => $extra,
attr => { }, leaves => {}, depth => $val, self_size=>0, kids_size=>0
});
$stack[$val] = $node;
$seqn2node{$id} = $node;
}
# --- Leaf name and memory size
elsif ($type eq "L") {
my $node = $seqn2node{$id} || die;
$node->{leaves}{$name} += $val;
$run_size += $val;
printf "%s+%d=%d %s\n", $indent x ($node->{depth}+1), $val, $run_size, $name
if $opt_text;
}
# --- Attribute type, name and value (all rather hackish)
elsif (looks_like_number($type)) {
my $node = $seqn2node{$id} || die;
my $attr = $node->{attr} || die;
# attributes where the string is a key (or always empty and the type is the key)
if ($type == NPattr_NAME or $type == NPattr_NOTE) {
printf "%s~%s(%s) %d [t%d]\n", $indent x ($node->{depth}+1), $attr_type_name[$type], $name, $val, $type
if $opt_text;
warn "Node $id already has attribute $type:$name (value $attr->{$type}{$name})\n"
if exists $attr->{$type}{$name};
$attr->{$type}{$name} = $val;
#Dwarn $attr;
$node->{title} = $name if $type == NPattr_NAME and !$val; # XXX hack
}
# attributes where the number is a key (or always zero)
elsif (NPattr_PADFAKE==$type or NPattr_PADTMP==$type or NPattr_PADNAME==$type) {
printf "%s~%s('%s') %d [t%d]\n", $indent x ($node->{depth}+1), $attr_type_name[$type], $name, $val, $type
if $opt_text;
warn "Node $id already has attribute $type:$name (value $attr->{$type}[$val])\n"
if defined $attr->{$type}[$val];
$attr->{+NPattr_PADNAME}[$val] = $name; # store all as NPattr_PADNAME
}
else {
printf "%s~%s %d [t%d]\n", $indent x ($node->{depth}+1), $name, $val, $type
if $opt_text;
warn "Invalid attribute type '$type' on line $. ($_)";
}
}
elsif ($type eq 'S') { # start of a run
die "Unexpected start token" if @stack;
if ($opt_dot) {
open $dot_fh, ">$opt_dot";
print $dot_fh "digraph {\n"; # }
print $dot_fh "graph [overlap=false]\n"; # target="???", URL="???"
}
if ($dbh) {
# XXX add a size_run table records each run
# XXX pick a table name to store the run nodes in
#$run_ins_sth->execute(
my $table = "node";
$dbh->do("DROP TABLE IF EXISTS $table");
$dbh->do(qq{
CREATE TABLE $table (
id integer primary key,
name text,
title text,
type integer,
depth integer,
parent_id integer,
self_size integer,
kids_size integer,
kids_node_count integer,
child_ids text,
attr_json text,
leaves_json text
)
});
$node_ins_sth = $dbh->prepare(qq{
INSERT INTO $table VALUES (?,?,?,?,?,?, ?,?,?,?,?,?)
});
}
}
elsif ($type eq 'E') { # end of a run
my $top = $stack[0]; # grab top node before we pop all the nodes
leave_node(pop @stack) while @stack;
# if nothing output (ie size(undef))
$top ||= { self_size=>0, kids_size=>0, kids_node_count=>0 };
my $top_size = $top->{self_size}+$top->{kids_size};
printf "Stored %d nodes totalling %s [lines=%d size=%d write=%.2fs]\n",
1+$top->{kids_node_count}, fmt_size($top_size),
$., $top_size, $val;
# the duration here ($val) is from Devel::SizeMe perspective
# ie doesn't include time to read file/pipe and commit to database.
if ($opt_verbose or $run_size != $top_size) {
warn "EOF ends $top->{id} d$top->{depth}: size $top->{self_size}+$top->{kids_size}\n";
warn Dumper($top);
}
die "panic: seqn2node should be empty ". Dumper(\%seqn2node)
if %seqn2node;
if ($dot_fh) {
print $dot_fh "}\n";
close $dot_fh;
system("open -a Graphviz $opt_dot") if $^O eq 'darwin'; # OSX
}
$dbh->commit if $dbh;
}
else {
warn "Invalid type '$type' on line $. ($_)";
next;
}
$dbh->commit if $dbh and $id % 10_000 == 0;
}
die "EOF without end token" if @stack;
=for This is out of date but gives you an idea of the data and stream
SV(PVAV) fill=1/1 [#1 @0]
: +64 sv =64
: +16 av_max =80
: AVelem-> [#2 @1]
: : SV(RV) [#3 @2]
: : : +24 sv =104
: : : RV-> [#4 @3]
: : : : SV(PVAV) fill=-1/-1 [#5 @4]
: : : : : +64 sv =168
: AVelem-> [#6 @1]
: : SV(IV) [#7 @2]
: : : +24 sv =192
192 at -e line 1.
=cut
__DATA__
N 1 0 SV(PVAV) fill=1/1
L 1 64 sv
L 1 16 av_max
N 2 1 AVelem->
N 3 2 SV(RV)
L 3 24 sv
N 4 3 RV->
N 5 4 SV(PVAV) fill=-1/-1
L 5 64 sv
N 6 1 AVelem->
N 7 2 SV(IV)
L 7 24 sv