OSDN Git Service

* dglib.pm: Initial commit.
authormatthewg <matthewg@138bc75d-0d04-0410-961f-82ee72b054a4>
Thu, 28 Sep 2006 01:50:45 +0000 (01:50 +0000)
committermatthewg <matthewg@138bc75d-0d04-0410-961f-82ee72b054a4>
Thu, 28 Sep 2006 01:50:45 +0000 (01:50 +0000)
* compareSumTests3: Initial commit.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@117269 138bc75d-0d04-0410-961f-82ee72b054a4

contrib/ChangeLog
contrib/compareSumTests3 [new file with mode: 0755]
contrib/dglib.pm [new file with mode: 0644]

index c0b3f98..069659d 100644 (file)
@@ -1,3 +1,8 @@
+2006-09-27  Matthew Sachs  <msachs@apple.com>
+
+       * dglib.pm: Initial commit.
+       * compareSumTests3: Initial commit.
+
 2006-09-18  Bernhard Fischer  <aldot@gcc.gnu.org>
 
        * dg-cmp-results.sh (compare-$$.awk): Print name[old]
 2006-09-18  Bernhard Fischer  <aldot@gcc.gnu.org>
 
        * dg-cmp-results.sh (compare-$$.awk): Print name[old]
diff --git a/contrib/compareSumTests3 b/contrib/compareSumTests3
new file mode 100755 (executable)
index 0000000..bcaad20
--- /dev/null
@@ -0,0 +1,252 @@
+#!/usr/bin/perl
+
+# Three-way DejaGNU comparison; uses dglib.pm.  Run perldoc on this file for
+# usage.
+#
+# Author: Matthew Sachs <msachs@apple.com>
+#
+# Copyright (c) 2006 Free Software Foundation.
+#
+# This file is part of GCC.
+#
+# GCC is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2, or (at your option)
+# any later version.
+#
+# GCC is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with GCC; see the file COPYING.  If not, write to
+# the Free Software Foundation, 51 Franklin Street, Fifth Floor,
+# Boston, MA 02110-1301, USA.
+
+=pod
+
+=head1 SYNOPSIS
+
+compareSumTests3 -- Two-way or three-way compare between DejaGNU .sum files
+
+=head1 USAGE
+
+       compareSumTests3 old1.sum [old2.sum] new.sum
+       compareSumTests3 -i 1:2 -x 2:3 old1.sum old2.sum new.sum
+
+=head1 DESCRIPTION
+
+Gives results in terms of 'new' (e.g. things that work in 'new' and don't in
+other compilers are improvements, things that don't in 'new' and do in others
+are regressions, and it tells you which of the two old compilers (or both)
+the test is a regression from.
+
+We treat any DG result other than PASS or XFAIL as a failure, e.g.
+UNRESOLVED, UNTESTED or test was not run.
+
+We merge some tests into 'logical tests' with multiple subphases.
+For instance, some tests will have compile, execute, and link
+subtests.  For these tests, if one of the phases fails, we
+indicate which phase the failure originates in.  For instance,
+in the following test results:
+
+       gcc.c-torture/compile_execute/xxxx.c: [FAIL:C,FAIL:X,PASS]
+
+the "compile_execute" replaces the compile or execute portion of the test name,
+and "FAIL:C" and "FAIL:X" indicates where the combined test failed.
+
+=head1 OPTIONS
+
+=head2 OVERVIEW
+
+=over 4
+
+=item *
+
+C<-i X:Y>: Only display differences between the two indicated runs.
+
+=item *
+
+C<-p>: Give plain output, suitable for piping to another program.
+
+=item *
+
+C<-x X:Y>: Exclude differences between the two indicated runs.
+
+=back
+
+=head2 PLAIN OUTPUT FORMAT
+
+In the plain
+output format, the category headers are not displayed and there are no tabs
+in front of each result line.  Instead, each result line has two characters
+followed by a space in front of it.  The first character will be either an 'I'
+for improvement or 'R' for regression; the second character will be a 1, 2, or 3,
+indicating which run was the odd one out.
+
+=back
+
+=head2 SELECTING CHANGE SUBSETS
+
+The following options cause only a selected subset of changes to be displayed.
+These options ask for a "run", a number which is used to select
+one of the three runs (C<old1>, C<old2>, or C<new>.)  C<1> and C<2> signify C<old1> and C<old2>
+respectively; 3 signifies C<new>. If multiple options are given, the changes displayed
+will be those which obey all of the given restrictions.
+
+Typical usage of these options is to express something like "give me all changes
+between 2 and 3, except for those where there was the same difference betwen 1 and 2
+(as between 2 and 3.)"  This would be given as:
+
+       -i 2:3 -x 1:2
+
+=over 4
+
+=item *
+
+C<-i X:Y>: Only differences which are present between the two runs given
+are displayed. For instance, if C<-i 1:2> is given and test A passes in
+runs 1 and 2 but fails in run 3, that result will not be displayed.
+
+=item *
+
+C<-x X:Y>: Differences which are identical to a difference between the two runs
+given will B<not> be displayed. For instance, if C<-x 1:2> is given and
+test A passes in run 1 and fails in runs 2 and 3, that result will not be
+displayed (since C<-x> will cause the difference between 1 and 2 to be ignored,
+and the difference in 1 and 3 parallels the difference between 1 and 2.)
+This option may only be used in conjunction with C<-i>.
+
+=back
+
+=cut
+
+use strict;
+use warnings;
+use Getopt::Long;
+
+use FindBin qw($Bin);
+use lib "$Bin";
+use dglib;
+
+my %options;
+my $error = undef;
+
+if(!GetOptions(
+       "p" => \$options{p},
+       "i=s" => \$options{i},
+       "x=s" => \$options{x},
+)) {
+       $error = "";
+} elsif(@ARGV != 2 and @ARGV != 3) {
+       $error = "";
+} elsif($options{x} and !$options{i}) {
+       $error = "-x may only be given in conjunction with -i.";
+} else {
+       foreach my $opt("i", "x") {
+               if($options{$opt} and
+                 ($options{$opt} !~ /^([123]):([123])$/ or
+                  $1 == $2)
+               ) {
+                       $error = "Invalid -$opt argument.";
+               }
+       }
+}
+
+if(defined($error)) {
+       print STDERR "$error\n" if $error;
+       print STDERR "Usage: compareSumTests3 [-p] [-i X:Y [-x X:Y]] old1.sum old2.sum new.sum\n";
+       print STDERR "Try 'perldoc $0' for further information.\n";
+       exit 1;
+} 
+
+my(@sumfiles) = @ARGV;
+-f $_ || die "$_ is not a regular file!\n" foreach @sumfiles;
+my(%results, @inc_changes, @exc_changes, %checksums);
+
+# We decrement the values given so that they correspond
+# to indices into our results array.
+if($options{i}) {
+       $options{i} =~ /(\d+):(\d+)/;
+       @inc_changes = ($1 - 1, $2 - 1);
+}
+if($options{x}) {
+       $options{x} =~ /(\d+):(\d+)/;
+       @exc_changes = ($1 - 1, $2 - 1);
+}
+
+
+my %analyzed_results = compareSumFiles(\@sumfiles);
+
+foreach my $cat (qw(improvements regressions miscellaneous)) {
+       if(@sumfiles == 3) {
+               my @subcounts;
+               if(!$options{p}) {
+                       $subcounts[$_] = @{$analyzed_results{$cat}->[$_] || []} for(0..2);
+                       print "\u$cat: ", ($subcounts[0]+$subcounts[1]+$subcounts[2]), "\n";
+               }
+
+               for(my $i = 0; $i < 3; $i++) {
+                       if(!$options{p} and $cat ne "miscellaneous") {
+                               if($i == 0) {
+                                       if($cat eq "regressions") {
+                                               print "\tSuccess in old1 only: $subcounts[$i]\n";
+                                       } else {
+                                               print "\tFailure in old1 only: $subcounts[$i]\n";
+                                       }
+                               } elsif($i == 1) {
+                                       if($cat eq "regressions") {
+                                               print "\tSuccess in old2 only: $subcounts[$i]\n";
+                                       } else {
+                                               print "\tFailure in old2 only: $subcounts[$i]\n";
+                                       }
+                               } else {
+                                       if($cat eq "regressions") {
+                                               print "\tFailure in new only: $subcounts[$i]\n";
+                                       } else {
+                                               print "\tSuccess in new only: $subcounts[$i]\n";
+                                       }
+                               }
+                       }
+
+                       foreach my $test (sort {$a->{name} cmp $b->{name}} @{$analyzed_results{$cat}->[$i] || []}) {
+                               if(!$options{p}) {
+                                       if($cat eq "miscellaneous") {
+                                               print "\t";
+                                       } else {
+                                               print "\t\t";
+                                       }
+                               } else {
+                                       if($cat eq "regressions") {
+                                               print "R";
+                                       } else {
+                                               print "I";
+                                       }
+
+                                       print $i+1, " ";
+                               }
+                               printf "%s [%s,%s,%s]\n", $test->{name}, $test->{data}->[0], $test->{data}->[1], $test->{data}->[2];
+                       }
+               }
+       } else {
+               if(!$options{p}) {
+                       my $subcount = @{$analyzed_results{$cat}};
+                       print "\u$cat: $subcount\n";
+               }
+
+               foreach my $test (sort {$a->{name} cmp $b->{name}} @{$analyzed_results{$cat}}) {
+                       if(!$options{p}) {
+                               print "\t";
+                       } else {
+                               if($cat eq "regressions") {
+                                       print "R";                              } else {
+                                       print "I";
+                               }
+
+                               print "  ";
+                       }
+                       printf "%s [%s,%s]\n", $test->{name}, $test->{data}->[0], $test->{data}->[1], $test->{data}->[2];
+               }
+       }
+}
diff --git a/contrib/dglib.pm b/contrib/dglib.pm
new file mode 100644 (file)
index 0000000..2d2016e
--- /dev/null
@@ -0,0 +1,424 @@
+# Library of functions for dealing with DejaGNU, or which are otherwise
+# generally useful for the DejaGNU tool stack.
+#
+# Author: Matthew Sachs <msachs@apple.com>
+#
+# Functions:
+#      parseLogFile: See "sub parseLogFile" below for details.  This function
+#              returns a detailed parse of a DejaGNU log or sum file.
+#      ispass: Takes a DejaGNU result (e.g. "PASS", "XPASS") and returns
+#              true if and only if it is a passing result (PASS, XFAIL, or
+#              KFAIL.)
+#
+# Copyright (c) 2006 Free Software Foundation.
+#
+# This file is part of GCC.
+#
+# GCC is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2, or (at your option)
+# any later version.
+#
+# GCC is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with GCC; see the file COPYING.  If not, write to
+# the Free Software Foundation, 51 Franklin Street, Fifth Floor,
+# Boston, MA 02110-1301, USA.
+
+package dglib;
+
+use strict;
+use warnings;
+use Exporter;
+
+our @ISA = qw(Exporter);
+our @EXPORT = qw(ispass parseLogFile);
+
+use File::Basename;
+use POSIX qw(mktime);
+
+
+# Create a group hierarchy, returning the leaf node
+sub mkGroupPath {
+       my($root, $groups, @newgroups) = @_;
+
+       my $parent = $root;
+       my $fullname = "";
+       foreach my $group(@newgroups) {
+               $fullname .= "/" if $fullname;
+               $fullname .= $group;
+               if(exists($groups->{$fullname})) {
+                       $parent = $groups->{$fullname};
+               } else {
+                       my $newgroup = {name => $group, parent => $parent};
+                       $groups->{$fullname} = $newgroup;
+                       $parent->{testgroup} ||= [];
+                       push @{$parent->{testgroup}}, $newgroup;
+                       $parent = $newgroup;
+               }
+       }
+
+       return $parent;
+}
+
+# Extract information from DejaGNU log or sum files.
+# Options, if provided, should be a hashref with zero or more of the following keys:
+#      gccdir:
+#              Passing in the full path to the root of the gcc/testsuite directory
+#              will help in the parsing, but if it isn't provided, it will be guessed.
+#      diagnostics:
+#              If set to 0, diagnostics will not be returned.  This can save a lot
+#              of memory if you are not using this information.
+#      fullname:
+#              If set to 0, the fullname key will not be included in tests.
+# Returns a hash with the following keys:
+#      incomplete: 1 if the summary file appears truncated, otherwise 0
+#      diagnostics: List of (type, value) for any errors detected.  Type can be ERROR, WARNING, or NOTE.
+#      test: Array of root-level tests, with keys:
+#              name: Name of the test, relative to the enclosing test group.
+#              fullname: Fully-qualified name of the test.
+#              result: DejaGNU result (PASS, FAIL, XPASS, &c)
+#              detail: For multi-phase (e.g. compile/link/execute), this will be
+#                      the furthest phase which the test was able to attempt,
+#                      so if the result is FAIL and this is "link phase", the test
+#                      compiled but failed to link.  This key may contain other
+#                      auxiliary data.
+#              pseudotest: If 1, this test may not really exist; see "pseudotest" below.
+#      testgroup: Array of root-level testgroups, with keys:
+#              name: Name of the group.
+#              parent: Parent test group.
+#              test: As per above.
+#              testgroup: Child test groups.
+#      compiler: Version string from compiler used to run the tests (if detected)
+sub parseLogFile($;$) {
+       my($logfile, $options) = @_;
+       $options ||= {};
+       my $gccdir = $options->{gccdir} || "";
+       my $return_diags = exists($options->{diagnostics}) ? $options->{diagnostics} : 1;
+       my $emit_fullname = exists($options->{fullname}) ? $options->{fullname} : 1;
+       my $is_gdb = 0;
+       my $gdbhack = "";
+
+       my %ret = (incomplete => 1, diagnostics => [], testgroup => []);
+       my(%testindex, %groupindex);
+
+       open(LOGFILE, $logfile) or die "Couldn't open log file $logfile: $!\n";
+
+       my($currgroup, $currtest, $lastrun);
+       $currgroup = \%ret;
+
+       my %monmap = (Jan => 0, Feb => 1, Mar => 2, Apr => 3, May => 4, Jun => 5, Jul => 6, Aug => 7, Sep => 8, Oct => 9, Nov => 10, Dec => 11);
+
+       # We don't want gccdir matching on an empty string.
+       $gccdir ||= "this will never match, or my name isn't Reginald St. Croix";
+
+       my $line = 1;
+       while(<LOGFILE>) {
+               chomp;
+               s/\x{d}$//; #^M
+               next if $_ eq "";
+
+               if(/^gcc version/) {
+                       $ret{compiler} = $_;
+               } elsif(/^got a .* signal, interrupted by user /) {
+                       $ret{incomplete} = 2;
+               } elsif(/^\s*=== gdb/) {
+                       $is_gdb = 1;
+                       # The log file from the GDB test suite is prone to have random crap
+                       # in front of test result lines, so we need to be looser about how
+                       # we parse those for GDB.
+                       $gdbhack = ".*";
+               } elsif(/^(Test Run By \S+ on|runtest completed at) ... (.{3}) (\d{1,2}) (\d{1,2}):(\d{1,2}):(\d{1,2}) (\d{4})/) {
+                       my $time = mktime($6, $5, $4, $3, $monmap{$2}, $7 - 1900);
+                       if($1 eq "runtest completed at") {
+                               $ret{end_time} = $time;
+                       } else {
+                               $ret{start_time} = $time;
+                       }
+               } elsif(m<^Running (?!target )\Q$gccdir\E/?(\S+)> or m<^Running (?!target )\S*?((?:gcc|gdb|libstdc\+\+-v3)/testsuite/\S+)>) {
+                       # We keep track of the last "Running foo/bar/baz.exp" line because
+                       # some tests don't bother printing out the full paths of their files,
+                       # and this gives us the directory information.
+
+                       $lastrun = $1;
+                       $lastrun =~ s!/[^/]*/\.\.!!; # foo/bar/../baz -> foo/baz
+                       $currgroup = mkGroupPath(\%ret, \%groupindex, split(m!/!, $lastrun));
+                       #$currgroup->{testfile} = $lastrun;
+               } elsif(/^Executing on (.*?):(.*)/) {
+                       # Okay, if it's "Executing on host", it's a new
+                       # file.  If it's "Executing on unix", it's probably
+                       # a test within the file -- an execution test, specifically --
+                       # (makes sense, no?)  But not always, sometimes we
+                       # see "Executing on unix" outside the context of a
+                       # file.
+
+                       # Try to pick out the gccdir-relative filename.
+                       # If we can't find it, it isn't really a new testfile,
+                       # but a derived file.
+                       my($exectype, $execwhat) = ($1, $2);
+                       next if $execwhat =~ /^dsymutil/;
+                       $execwhat =~
+                               s!.*?\s\Q$gccdir\E/?(\S+).*!$1! or
+                               s!.*?/((?:gcc|gdb|libstdc\+\+-v3)/testsuite/\S+).*!$1! or
+                               $exectype = "unix";
+
+                       if($exectype eq "host" or !$currgroup) {
+                               # New file
+
+                               my $nogroup = 0;
+                               if($execwhat =~ / /) {
+                                       # We probably haven't parsed the file correctly.
+                                       # Try getting it from $lastrun.
+
+                                       $execwhat = dirname($lastrun) . "/" . basename($execwhat) if $lastrun and $execwhat;
+                                       $execwhat =~ s/\s.*//;
+
+                                       # At the end of each tool, it invokes "gcc -v" or "c++ -v"
+                                       # as a test.  We don't really want to treat this as a test.
+                                       if($execwhat =~ m!/(gcc|c\+\+)$!) {
+                                               undef $currtest;
+                                               undef $currgroup;
+                                               $nogroup = 1;
+                                       }
+                               }
+
+                               if(!$nogroup) {
+                                       undef $currtest;
+                                       $execwhat =~ s!/[^/]*/\.\.!!; # foo/bar/../baz -> foo/baz
+
+                                       if($lastrun) {
+                                               my $lastbase = dirname($lastrun);
+                                               my $basegroup = $execwhat;
+                                               $basegroup =~ s!^\Q$lastbase\E/!!;
+                                               $execwhat = "$lastrun/$basegroup";
+                                       }
+
+                                       $currgroup = mkGroupPath(\%ret, \%groupindex, split(m!/!, $execwhat));
+                                       #$currgroup->{testfile} = $execwhat;
+                               }
+                       } else {
+                               # New test within current file
+
+                               $currtest = {};
+                       }
+               } elsif(/^# of/) {
+                       # This line appears should appear near the end of summary files.
+                       # If it doesn't, something went wrong.
+
+                       if($ret{incomplete} == 2) {
+                               #Ah, but we previously saw indication that we were killed via a signal.
+                               $ret{incomplete} = 1;
+                       } else {
+                               $ret{incomplete} = 0;
+                       }
+               } elsif(/^testcase .* completed/) {
+                       # End of a .exp file
+                       undef $currtest;
+                       undef $currgroup;
+               } elsif(/^$gdbhack(FAIL|PASS|UNRESOLVED|UNSUPPORTED|UNTESTED|XFAIL|XPASS|KFAIL|KPASS): (.*)/) {
+                       # If the currtest already has a name, that means we've already seen
+                       # its results, so what we have now is a new test.  However, if we
+                       # haven't seen results for currtest yet, that means currtest just
+                       # has some diagnostics associated with it but no actual results,
+                       # so just use that one.
+                       undef $currtest if $currtest->{name};
+
+                       my $phase = ""; # compile/link/execute
+                       my($test, $result) = ($2, $1);
+
+                       # Compile/(link/)execute combining
+                       if($test =~ /^(.*) compile\s*$/) {
+                               $test = "$1 compile,link,execute";
+                               $phase = "compile";
+                       } elsif($test =~ /^(.*)-(.*) (link|execute)\s*$/) {
+                               $test = "$1 compile,link,execute";
+                               if($3 eq "link") {
+                                       $phase = "link";
+                               } else {
+                                       $phase = "execute";
+                               }
+                       } elsif($test =~ /(compile|compilation|execute|execution)/) {
+                               my $phasematch = $1;
+                               if($test =~ /^com/) {
+                                       $phase = "compile";
+                               } else {
+                                       $phase = "execute";
+                               }
+                               $test =~ s!\Q$phasematch\E!compile,execute!;
+                       }
+
+                       # gcov tests behave in non-standard fashion.
+                       my $failwhy = "";
+                       $test =~ s/ gcov failed: (.*)// and $failwhy = $1;
+
+                       # And some other tests have random information after a colon :(
+                       # But for scan-assembler, this really is part of the name.
+                       if(!$is_gdb and $test !~ /scan-assembler/ and $test =~ s/:\s*(.+)//) {
+                               $failwhy = $1;
+                       }
+
+                       $test =~ s/\s*$//;
+                       $test =~ s/^\s*$//;
+
+                       # Sometimes there's a test which shows up as:
+                       #       foo (test for excess errors)
+                       #       foo (something else)
+                       #       foo: error executing dg-final
+                       # if it runs, but just:
+                       #       foo
+                       # if it doesn't.  When we see the top form, we create a
+                       # "pseudotest" in the bottom form, so that comparisons
+                       # can be made.
+                       my $basetest = $test;
+                       $basetest =~ s/:? .*//;
+
+                       if(exists($testindex{$test}) and !$testindex{$test}->{pseudotest}) {
+                               $currtest = $testindex{$test};
+                               if(ispass($currtest->{result})) {
+                                       $currtest->{result} = $result;
+                                       $currtest->{detail} = "$phase phase";
+                                       $currtest->{detail} .= "; $failwhy" if $failwhy;
+                               }
+                       } else {
+                               # This might have been created earlier as a pseudotest.
+                               # If so, overwrite it.
+                               $currtest ||= $testindex{$test} || {};
+
+                               $currtest->{name} = basename($test);
+                               if($emit_fullname) {
+                                       $currtest->{fullname} = ($currgroup->{name} || dirname($test)) . "/$currtest->{name}";
+                               }
+                               my $grpname = $currgroup->{name} || "";
+                               $currtest->{name} =~ s/^\s*\Q$grpname\E\s*//;
+                               $currtest->{name} =~ s/^: // if $is_gdb;
+                               # Sometimes there's a test at the root of the group.
+                               # For instance, you'll have:
+                               #       FAIL: foo/bar.c (test for excess errors)
+                               #       UNRESOLVED: foo/bar.c: couldn't open "bar.s": no such file or directory
+                               # In this case, groupname *is* the entire name, so the regex above will make the test name empty.
+                               # In this case, we actually want to use the parent group and make this a test within that group.
+                               my $orig_currgroup = $currgroup;
+                               if(!$currtest->{name}) {
+                                       $currtest->{name} = $grpname;
+                                       $currgroup = $currgroup->{parent};
+                                       $grpname = $currgroup->{name} || "";
+                               }
+
+                               $currtest->{result} = $result;
+                               if($phase and $failwhy) {
+                                       $currtest->{detail} = "$phase phase; $failwhy" if $phase;
+                               } elsif($phase) {
+                                       $currtest->{detail} = "$phase phase";
+                               } elsif($failwhy) {
+                                       $currtest->{detail} = $failwhy;
+                               }
+
+                               $currgroup->{test} ||= [];
+                               push @{$currgroup->{test}}, $currtest;
+                               $testindex{$test} = $currtest;
+                               $currgroup = $orig_currgroup;
+
+                               if($basetest ne $test) {
+                                       if(!exists($testindex{$basetest}) ) {
+                                               my $btbase = basename($basetest);
+                                               $testindex{$basetest} = {
+                                                       name => $btbase,
+                                                       result => $result,
+                                                       pseudotest => 1,
+                                                       fullname => $btbase
+                                               };
+                                               if($emit_fullname) {
+                                                       $testindex{basetest}->{fullname} = ($currgroup->{name} || dirname($basetest)) . "/$btbase";
+                                               }
+                                               push @{$currgroup->{parent}->{test}}, $testindex{$basetest};
+                                       } else {
+                                               # Only let the base test pass if all the sub-tests pass
+                                               $testindex{$basetest}->{result} = $result if !ispass($result);
+                                       }
+                               }
+
+                       }
+               } elsif(/^\s+=== .* Summary ===\s*$/) {
+                       undef $currgroup;
+                       undef $currtest;
+               }
+
+               my $severity;
+               if(/^(ERROR|WARNING|NOTE): (.*)/) {
+                       $severity = $1;
+                       my $message = $2;
+
+                       if($message eq "program timed out.") {
+                               $currtest->{result} = "TIMEDOUT";
+                       } elsif(
+                               $message =~ /can't read "(HOSTCC|libiconv)": no such variable/ or
+                               $message =~ /no files matched glob pattern/ or
+                               $message =~ /error executing dg-final: .*: no such file/
+                       ) {
+                               $severity = "NOTE";
+                       }
+               } else {
+                       $severity = "logline";
+               }
+
+               if($return_diags) {
+                       my $dobj;
+                       if($currtest) {
+                       $currtest->{diagnostics} ||= [];
+                               $dobj = $currtest->{diagnostics};
+                       } elsif($currgroup) {
+                               $currgroup->{diagnostics} ||= [];
+                               $dobj = $currgroup->{diagnostics};
+                       } else {
+                               $dobj = $ret{diagnostics};
+                       }
+
+                       push @$dobj, {message => $_, severity => $severity, line => $line};
+               }
+       } continue {
+               $line++;
+       }
+       close LOGFILE;
+
+       return %ret;
+}
+
+# Split a test into testdivs
+sub splitTest($$) {
+       my($root, $test) = @_;
+
+       $test->{fullname} =~ /^(\S+)\s*(.*)/;
+       my($path, $descriptor) = ($1, $2);
+       my @nodes = split(m!/!, $path);
+       push @nodes, $descriptor if $descriptor;
+       my $lastnode = pop @nodes;
+
+       my $hash = $root;
+       foreach (@nodes) {
+               $hash->{testdiv} ||= {};
+               $hash = $hash->{testdiv}->{$_} ||= {};
+       }
+
+
+       $hash->{test} ||= {};
+       $hash->{test}->{$lastnode} = $test;
+}
+
+
+# ==== Comparison ====
+
+sub ispass($) {
+       my $result = shift;
+
+       if($result eq "PASS" or $result eq "XFAIL" or $result eq "KFAIL") {
+               return 1;
+       } else {
+               return 0;
+       }
+}
+
+1;