office-gobmx/bin/module-deps.pl
Jan-Marek Glogowski 250e4886d8 Fail module-deps.pl on more errors
* Fail on the error code returned from make
* Collect dependency errors in the tree and fail and print them

The "perldoc -f open" has an example in the "Opening a filehandle
into a command" section, which uses waitpid in the parent. Since
we parse the pipe output, there seems to be no need for waitpid,
because we end when the pipe is closed by the writer.
waitpid always returns -1 at this point and since there aren't
any zombie processes, it seems to be fine to ignore.
Additionally strace shows, clone, execve and waitpid calls, so
the explicit exit in the example seems bogus, if the process
is actually overwritten.
And the pipe is just open in the parent process anyway.

Change-Id: I2e77aa33a0eaa5d6d3e06904bb0af53f4b66ef91
Reviewed-on: https://gerrit.libreoffice.org/c/core/+/126789
Tested-by: Jenkins
Reviewed-by: Jan-Marek Glogowski <glogow@fbihome.de>
2021-12-14 02:33:11 +01:00

584 lines
14 KiB
Perl
Executable file

#!/usr/bin/env perl
use strict;
use warnings;
use Getopt::Long qw(GetOptions VersionMessage);
use Pod::Usage;
my $gnumake;
my $src_root;
my $makefile_build;
my $verbose = 0;
my $no_leaf;
my $from_file;
my $to_file;
my $output_file;
my $preserve_libs = 0;
my $toposort = 0;
my %merged_libs;
sub logit($)
{
print STDERR shift if ($verbose);
}
sub read_deps()
{
my $p;
my $to;
my $invalid_tolerance = 100;
my $line_count = 0;
my %deps;
my $child_pid = 0;
if (defined $to_file)
{
open($to, ">$to_file") or die "can not open file for writing $to_file";
}
if (defined $from_file) {
open ($p, $from_file) || die "can't read deps from cache file: $!";
} else {
$child_pid = open ($p, "-|", "ENABLE_PRINT_DEPS=1 $gnumake -qrf $makefile_build") // die "couldn't launch make: $!";
exit if (!$child_pid);
}
$|=1;
print STDERR "reading deps ";
while (<$p>) {
my $line = $_;
$line_count++;
print STDERR '.' if (!$verbose && $line_count % 10 == 0);
logit($line);
print $to $line if defined $to_file;
chomp ($line);
if ($line =~ m/^MergeLibContents:\s+(\S+.*)\s*$/) {
for my $dep (split / /, $1) {
$merged_libs{$dep} = 1 if $dep ne '';
}
} elsif ($line =~ m/^LibraryDep:\s+(\S+) links against (.*)$/) {
# if ($line =~ m/^LibraryDep:\s+(\S+)\s+links against/) {
$deps{$1} = ' ' if (!defined $deps{$1});
$deps{$1} = $deps{$1} . ' ' . $2;
} elsif ($line =~ m/^LibraryDep:\s+links against/) {
# these need fixing, we call gb_LinkTarget__use_$...
# and get less than normal data back to gb_LinkTarget_use_libraries
# print STDERR "ignoring unhelpful external dep\n";
} elsif ($invalid_tolerance < 0) {
# print "read all dependencies to: '$line'\n";
last;
} else {
# print "no match '$line'\n";
$invalid_tolerance--;
}
}
close ($p);
if ($child_pid) {
my $err = $? >> 8;
# make query mode returns 0 or 1, depending on the build status
if ($err != 0 && $err != 1) {
print STDERR " error\n" if (!$verbose);
die("Errorcode $err from make - aborting!");
}
}
print STDERR " done\n";
return \%deps;
}
# graphviz etc. don't like some names
sub clean_name($)
{
my $name = shift;
$name =~ s/[\-\/\.]/_/g;
return $name;
}
# first create nodes for each entry
sub clean_tree($)
{
my $deps = shift;
my %tree;
for my $name (sort keys %{$deps}) {
my $need_str = $deps->{$name};
$need_str =~ s/^\s+//g;
$need_str =~ s/\s+$//g;
my @needs = split /\s+/, $need_str;
$name =~ m/^([^_]+)_(\S+)$/ || die "invalid target name: '$name'";
my $type = $1;
my $target = clean_name ($2);
$type eq 'Executable' || $type eq 'Library' ||
$type eq 'CppunitTest' || die "Unknown type '$type'";
my %result;
$result{type} = $type;
$result{target} = $target;
$result{merged} = 0;
my @clean_needs;
for my $need (@needs) {
push @clean_needs, clean_name($need);
}
$result{deps} = \@clean_needs;
if (defined $tree{$target}) {
logit("warning -duplicate target: '$target'\n");
delete($tree{$target});
}
$tree{$target} = \%result;
logit("$target ($type): " . join (',', @clean_needs) . "\n");
}
return \%tree;
}
sub has_child_dep($$$)
{
my ($tree,$search,$name) = @_;
my $node = $tree->{$name};
return defined $node->{flat_deps}->{$search};
}
# flatten deps recursively into a single hash per module
sub build_flat_dep_hash($$);
sub build_flat_dep_hash($$)
{
my ($tree, $name) = @_;
my %flat_deps;
my $node = $tree->{$name};
return if (defined $node->{flat_deps});
# build flat deps for children
for my $child (@{$node->{deps}}) {
build_flat_dep_hash($tree, $child)
}
for my $child (@{$node->{deps}}) {
$flat_deps{$child} = 1;
for my $dep (@{$tree->{$child}->{deps}}) {
$flat_deps{$dep} = 1;
}
}
$node->{flat_deps} = \%flat_deps;
# useful debugging ...
if (defined $ENV{DEP_CACHE_FILE}) {
logit("node '$name' has flat-deps: '" . join(',', keys %flat_deps) . "' " .
"vs. '" . join(',', @{$node->{deps}}) . "'\n");
}
}
# many modules depend on vcl + sal, but vcl depends on sal
# so we want to strip sal out - and the same for many
# similar instances
sub prune_redundant_deps($)
{
my $tree = shift;
for my $name (sort keys %{$tree}) {
build_flat_dep_hash($tree, $name);
}
}
# glob on libo directory
sub create_lib_module_map()
{
my %l2m;
# hardcode the libs that don't have a directory
$l2m{'merged'} = 'merged';
for (glob($src_root."/*/Library_*.mk"))
{
/.*\/(.*)\/Library_(.*)\.mk/;
# add module -> module
$l2m{$1} = $1;
# add lib -> module
$l2m{$2} = $1;
}
return \%l2m;
}
# call prune redundant_deps
# rewrite the deps array
sub optimize_tree($)
{
my $tree = shift;
prune_redundant_deps($tree);
my @errors;
for my $name (sort keys %{$tree}) {
my $result = $tree->{$name};
if (!defined($result->{target})) {
push @errors, "missing target for dependency '$name'!";
next;
}
logit("minimising deps for $result->{target}\n");
my @newdeps;
for my $dep (@{$result->{deps}}) {
# is this implied by any other child ?
logit("checking if '$dep' is redundant\n");
my $preserve = 1;
for my $other_dep (@{$result->{deps}}) {
next if ($other_dep eq $dep);
if (has_child_dep($tree,$dep,$other_dep)) {
logit("$dep is implied by $other_dep - ignoring\n");
$preserve = 0;
last;
}
}
push @newdeps, $dep if ($preserve);
}
# re-write the shrunk set to accelerate things
$result->{deps} = \@newdeps;
}
if (scalar @errors > 0) {
print STDERR join("\n", @errors) . "\n";
die("Missing targets for dependencies - aborting!");
}
return $tree;
}
# walking through the library based graph and creating a module based graph.
sub collapse_lib_to_module($)
{
my $tree = shift;
my %digraph;
my $l2m = create_lib_module_map();
my %unknown_libs;
for my $lib_name (sort keys %{$tree}) {
my $result = $tree->{$lib_name};
$unknown_libs{$lib_name} = 1 && next if (!grep {/$lib_name/} keys %$l2m);
# new collapsed name.
my $name = $l2m->{$lib_name};
# sal has no dependencies, take care of it
# otherwise it doesn't have target key
if (!@{$result->{deps}}) {
if (!exists($digraph{$name})) {
my @empty;
$digraph{$name}{deps} = \@empty;
$digraph{$name}{target} = $result->{target};
$digraph{$name}{merged} = $result->{merged};
}
}
for my $dep (@{$result->{deps}}) {
my $newdep;
$newdep = $l2m->{$dep};
die "Mis-named */Library_*.mk file - should match rules: '$dep'" if (!defined $newdep);
$dep = $newdep;
# ignore: two libraries from the same module depend on each other
next if ($name eq $dep);
if (exists($digraph{$name}))
{
my @deps = @{$digraph{$name}{deps}};
# only add if we haven't seen already that edge?
if (!grep {/$dep/} @deps)
{
push @deps, $dep;
$digraph{$name}{deps} = \@deps;
}
}
else
{
my @deps;
push @deps, $dep;
$digraph{$name}{deps} = \@deps;
$digraph{$name}{target} = $result->{target};
$digraph{$name}{merged} = $result->{merged};
}
}
}
logit("warn: no module for libs were found and dropped: [" .
join(",", (sort (keys(%unknown_libs)))) . "]\n");
return optimize_tree(\%digraph);
}
sub prune_leaves($)
{
my $tree = shift;
my %newtree;
my %name_has_deps;
# we like a few leaves around:
for my $nonleaf ('desktop', 'sw', 'sc', 'sd', 'starmath') {
$name_has_deps{$nonleaf} = 1;
}
# find which modules are depended on by others
for my $name (keys %{$tree}) {
for my $dep (@{$tree->{$name}->{deps}}) {
$name_has_deps{$dep} = 1;
}
}
# prune modules with no deps
for my $name (keys %{$tree}) {
delete $tree->{$name} if (!defined $name_has_deps{$name});
}
return optimize_tree($tree);
}
sub annotate_mergelibs($)
{
my $tree = shift;
print STDERR "annotating mergelibs\n";
for my $name (keys %{$tree}) {
if (defined $merged_libs{$name}) {
$tree->{$name}->{merged} = 1;
# print STDERR "mark $name as merged\n";
}
}
}
sub dump_graphviz($)
{
my $tree = shift;
my $to = \*STDOUT;
open($to, ">$output_file") if defined($output_file);
print $to <<END;
digraph LibreOffice {
edge [color="#31CEF0", len=0.4]
edge [fontname=Arial, fontsize=10, fontcolor="#31CEF0"]
END
;
my @merged_names;
my @normal_names;
for my $name (sort keys %{$tree}) {
if ($tree->{$name}->{merged}) {
push @merged_names, $name;
} else {
push @normal_names, $name;
}
}
print $to "node [fontname=Verdana, fontsize=10, height=0.02, width=0.02,".
'shape=Mrecord,color="#BBBBBB"' .
"];" . join(';', @normal_names) . "\n";
print $to "node [fontname=Verdana, fontsize=10, height=0.02, width=0.02,".
'shape=box,style=filled,color="#CCCCCC"' .
"];" . join(';', @merged_names) . "\n";
my @errors;
for my $name (sort keys %{$tree}) {
my $result = $tree->{$name};
if (!defined($result->{target})) {
push @errors, "Missing target for dependency '$name'!";
next;
}
logit("minimising deps for $result->{target}\n");
for my $dep (@{$result->{deps}}) {
print $to "$name -> $dep;\n" ;
}
}
if (scalar @errors > 0) {
print STDERR join("\n", @errors) . "\n";
die("Missing targets for dependencies - aborting!");
}
print $to "}\n";
}
sub toposort_visit($$$$);
sub toposort_visit($$$$)
{
my $tree = shift;
my $list = shift;
my $tags = shift;
my $name = shift;
die "dependencies don't form a DAG"
if (defined($tags->{$name}) && $tags->{$name} == 1);
if (!$tags->{$name}) {
$tags->{$name} = 1;
my $result = $tree->{$name};
for my $dep (@{$result->{deps}}) {
toposort_visit($tree, $list, $tags, $dep);
}
$tags->{$name} = 2;
push @{$list}, $name;
}
}
sub dump_toposort($)
{
my $tree = shift;
my @list;
my %tags;
for my $name (sort keys %{$tree}) {
toposort_visit($tree, \@list, \%tags, $name);
}
my $to = \*STDOUT;
open($to, ">$output_file") if defined($output_file);
for (my $i = 0; $i <= $#list; ++$i) {
print $to "$list[$i]\n";
}
}
sub filter_targets($)
{
my $tree = shift;
for my $name (sort keys %{$tree})
{
my $result = $tree->{$name};
if ($result->{type} eq 'CppunitTest' ||
($result->{type} eq 'Executable' &&
$result->{target} ne 'soffice_bin'))
{
delete($tree->{$name});
}
}
}
sub parse_options()
{
my %h = (
'verbose|v' => \$verbose,
'help|h' => \my $help,
'man|m' => \my $man,
'version|r' => sub {
VersionMessage(-msg => "You are using: 1.0 of ");
},
'preserve-libs|p' => \$preserve_libs,
'toposort|t' => \$toposort,
'write-dep-file|w=s' => \$to_file,
'read-dep-file|f=s' => \$from_file,
'no-leaf|l' => \$no_leaf,
'output-file|o=s' => \$output_file);
GetOptions(%h) or pod2usage(2);
pod2usage(1) if $help;
pod2usage(-exitstatus => 0, -verbose => 2) if $man;
($gnumake, $makefile_build) = @ARGV if $#ARGV == 1;
$gnumake = 'make' if (!defined $gnumake);
$makefile_build = 'Makefile.gbuild' if (!defined $makefile_build);
$src_root = defined $ENV{SRC_ROOT} ? $ENV{SRC_ROOT} : ".";
}
sub main()
{
parse_options();
my $deps = read_deps();
my $tree = clean_tree($deps);
filter_targets($tree);
optimize_tree($tree);
annotate_mergelibs($tree);
if (!$preserve_libs && !defined($ENV{PRESERVE_LIBS})) {
$tree = collapse_lib_to_module($tree);
}
if ($no_leaf) {
$tree = prune_leaves($tree);
}
if ($toposort) {
dump_toposort($tree);
} else {
dump_graphviz($tree);
}
}
main()
__END__
=head1 NAME
module-deps - Generate module dependencies for LibreOffice build system
=head1 SYNOPSIS
module_deps [options] [gnumake] [makefile]
=head1 OPTIONS
=over 8
=item B<--help>
=item B<-h>
Print a brief help message and exits.
=item B<--man>
=item B<-m>
Prints the manual page and exits.
=item B<--version>
=item B<-v>
Prints the version and exits.
=item B<--preserve-libs>
=item B<-p>
Don't collapse libs to modules
=item B<--toposort>
=item B<-t>
Output a topological sorting instead of a graph
=item B<--read-dep-file file>
=item B<-f>
Read dependency from file.
=item B<--write-dep-file file>
=item B<-w>
Write dependency to file.
=item B<--output-file file>
=item B<-o>
Write graph or sort output to file
=back
=head1 DESCRIPTION
B<This program> parses the output of LibreOffice make process
(or cached input file) and generates the digraph build dependency,
that must be piped to B<graphviz> program (typically B<dot>).
B<Hacking on it>:
The typical (optimized) B<workflow> includes 3 steps:
=over 3
=item 1
Create cache dependency file: module_deps --write-dep-file lo.dep
=item 2
Use cache dependency file: module_deps --read-dep-file lo.dep -o lo.graphviz
=item 3
Pipe the output to graphviz: cat lo.graphviz | dot -Tpng -o lo.png
=back
=head1 TODO
=over 2
=item 1
Add soft (include only) dependency
=item 2
Add dependency on external modules
=back
=head1 AUTHOR
=over 2
=item Michael Meeks
=item David Ostrovsky
=back
=cut