3 # Program: find-cycles.pl
5 # Synopsis: Given a list of possibly cyclic dependencies, merge all the
6 # cycles. This makes it possible to topologically sort the
7 # dependencies between different parts of LLVM.
9 # Syntax: find-cycles.pl < LibDeps.txt > FinalLibDeps.txt
11 # Input: cycmem1: cycmem2 dep1 dep2
12 # cycmem2: cycmem1 dep3 dep4
15 # Output: cycmem1 cycmem2: dep1 dep2 dep3 dep4
18 # This file was written by Eric Kidd, and is placed into the public domain.
29 # Read our dependency information.
32 my ($module, $dependency_str) = /^([^:]*): ?(.*)$/;
33 die "Malformed data: $_" unless defined $dependency_str;
34 my @dependencies = split(/ /, $dependency_str);
35 $DEPS{$module} = \@dependencies;
38 # Partition our raw dependencies into sets of cyclically-connected nodes.
41 # Print out the finished cycles, with their dependencies.
43 foreach my $cycle (@CYCLES) {
44 my @modules = sort keys %{$cycle};
46 # Merge the dependencies of all modules in this cycle.
48 foreach my $module (@modules) {
49 @dependencies{@{$DEPS{$module}}} = 1;
52 # Prune the known cyclic dependencies.
53 foreach my $module (@modules) {
54 delete $dependencies{$module};
57 # Warn about possible linker problems.
58 my @archives = grep(/\.a$/, @modules);
60 print STDERR "find-cycles.pl: Circular dependency between *.a files:\n";
61 print STDERR "find-cycles.pl: ", join(' ', @archives), "\n";
62 print STDERR "find-cycles.pl: Some linkers may have problems.\n";
63 push @modules, @archives; # WORKAROUND: Duplicate *.a files. Ick.
66 # Add to our output. (@modules is already as sorted as we need it to be.)
67 push @output, (join(' ', @modules) . ': ' .
68 join(' ', sort keys %dependencies) . "\n");
73 #==========================================================================
74 # Depedency Cycle Support
75 #==========================================================================
76 # For now, we have cycles in our dependency graph. Ideally, each cycle
77 # would be collapsed down to a single *.a file, saving us all this work.
79 # To understand this code, you'll need a working knowledge of Perl 5,
80 # and possibly some quality time with 'man perlref'.
85 sub found_cycles ($@);
88 # Find all multi-item cycles.
89 my @modules = sort keys %DEPS;
90 foreach my $module (@modules) { find_cycles($module); }
92 # Build fake one-item "cycles" for the remaining modules, so we can
93 # treat them uniformly.
94 foreach my $module (@modules) {
95 unless (defined $CYCLES{$module}) {
96 my %cycle = ($module, 1);
97 $CYCLES{$module} = \%cycle;
101 # Find all our unique cycles. We have to do this the hard way because
102 # we apparently can't store hash references as hash keys without making
105 foreach my $cycle (values %CYCLES) {
106 unless ($seen{$cycle}) {
108 push @CYCLES, $cycle;
113 # Walk through our graph depth-first (keeping a trail in @path), and report
114 # any cycles we find.
115 sub find_cycles ($@) {
116 my ($module, @path) = @_;
117 if (str_in_list($module, @path)) {
118 found_cycle($module, @path);
120 return if defined $SEEN{$module};
122 foreach my $dep (@{$DEPS{$module}}) {
123 find_cycles($dep, @path, $module);
128 # Give a cycle, attempt to merge it with pre-existing cycle data.
129 sub found_cycle ($@) {
130 my ($module, @path) = @_;
132 # Pop any modules which aren't part of our cycle.
133 while ($path[0] ne $module) { shift @path; }
134 #print join("->", @path, $module) . "\n";
136 # Collect the modules in our cycle into a hash.
138 foreach my $item (@path) {
140 if (defined $CYCLES{$item}) {
141 # Looks like we intersect with an existing cycle, so merge
143 foreach my $old_item (keys %{$CYCLES{$item}}) {
144 $cycle{$old_item} = 1;
149 # Update our global cycle table.
150 my $cycle_ref = \%cycle;
151 foreach my $item (keys %cycle) {
152 $CYCLES{$item} = $cycle_ref;
154 #print join(":", sort keys %cycle) . "\n";
157 sub str_in_list ($@) {
158 my ($str, @list) = @_;
159 foreach my $item (@list) {
160 return 1 if ($item eq $str);