dglib.pm: Initial commit.

* dglib.pm: Initial commit.
* compareSumTests3: Initial commit.

From-SVN: r117269
This commit is contained in:
Matthew Sachs 2006-09-28 01:50:45 +00:00 committed by Matthew Sachs
parent 3441f1062c
commit ed5ce4e014
3 changed files with 681 additions and 0 deletions

View 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]

252
contrib/compareSumTests3 Executable file
View File

@ -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];
}
}
}

424
contrib/dglib.pm Normal file
View File

@ -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;