Menu
▾
▴
[Valgrind-developers] valgrind: r12231 - trunk/auxprogs
|
From: <sv...@va...> - 2011-10-25 20:06:26
|
Author: florian Date: 2011-10-25 21:01:41 +0100 (Tue, 25 Oct 2011) New Revision: 12231 Log: Add script to summarize the outcome of nightly builds. Added: trunk/auxprogs/nightly-build-summary Modified: trunk/auxprogs/Makefile.am Modified: trunk/auxprogs/Makefile.am =================================================================== --- trunk/auxprogs/Makefile.am 2011-10-25 09:20:05 UTC (rev 12230) +++ trunk/auxprogs/Makefile.am 2011-10-25 20:01:41 UTC (rev 12231) @@ -6,6 +6,7 @@ gen-mdg \ gsl16test \ gsl19test \ + nightly-build-summary \ posixtestsuite-1.5.1-diff-results EXTRA_DIST = \ Added: trunk/auxprogs/nightly-build-summary =================================================================== --- trunk/auxprogs/nightly-build-summary (rev 0) +++ trunk/auxprogs/nightly-build-summary 2011-10-25 20:01:41 UTC (rev 12231) @@ -0,0 +1,476 @@ +#!/usr/bin/env perl + +#----------------------------------------------------------------- +# Quick and dirty script to summarize build information for a +# set of nightly runs. +# +# The results of the nighly regression runs are extracted from +# the GMANE mail archive. The URL for a given mail sent to the +# valgrind-developers mailing list is +# +# http://article.gmane.org/gmane.comp.debugging.valgrind.devel/<integer> +# +# The script extracts information about the regression run from a +# block of information at the beginning of the mail. That information +# was added beginning October 4, 2011. Therefore, only regression runs +# from that date or later can be analyzed. +# +# There is unfortunately no good way of figuring out the interval +# of integers in the above URL that include all nightly regression +# runs. +# +# The function get_regtest_data does all the work. It returns a hash +# whose keys are the dates at which nightly runs took place. The value +# is in turn a hash. +# +# Each such hash has the following keys: +# "builds" array of hashes +# "num_builds" int +# "num_failing_builds" int +# "num_passing_builds" int +# "num_testcase_failures" int +# "num_failing_testcases" int +# "failure_frequency" hash indexed by testcase name; value = int +# +# "builds" is an array of hashes with the following keys +# "arch" string (architecture) +# "distro" string (distribution, e.g. Fedora-15) +# "failures" array of strings (failing testcases) +# "valgrind revision" integer +# "VEX revision" integer +# "GCC version" string +# "C library" string +# "uname -mrs" string +# "Vendor version" string +# +#----------------------------------------------------------------- +use strict; +use warnings; + +use LWP::Simple; +use Getopt::Long; + +my $prog_name = "nightly-build-summary"; + +my $debug = 0; +my $keep = 0; + +my $usage=<<EOF; +USAGE + + $prog_name + + --from=INTEGER beginning of mail interval; > 14800 + + [--to=INTEGER] end of mail interval; default = from + 100 + + [--debug] verbose mode (debugging) + + [--keep] write individual emails to files (debugging) + + [--dump] write results suitable for post-processing + + [--readable] write results in human readable form (default) + +EOF + + +#----------------------------------------------------------------- +# Search for a line indicating that this is an email containing +# the results of a valgrind regression run. +# Return 1, if found and 0 oherwise. +#----------------------------------------------------------------- +sub is_regtest_result { + my (@lines) = @_; + + foreach my $line (@lines) { + return 1 if ($line =~ "^valgrind revision:"); + } + + return 0; +} + + +#----------------------------------------------------------------- +# Extract information from the run. Don't prep the data here. This +# is done later on. +#----------------------------------------------------------------- +sub get_raw_data { + my (@lines, $msgno) = @_; + my ($i, $n, $line, $date); + + $n = scalar @lines; + + my %hash = (); + +# 1) Get the date from the mail header. This comes first + for ($i = 0; $i < $n; ++$i) { + $line = $lines[$i]; + + if ($line =~ /^Date:/) { + $date = (split(/ /, $line))[1]; + last; + } + } + die "no date found in message $msgno" if ($i == $n); + +# 2) Locate the section with the info about the environment of this nightly run + for ($i = $i + 1; $i < $n; ++$i) { + last if ($lines[$i] =~ /^valgrind revision:/); + } + die "no info block in message $msgno" if ($i == $n); + +# 3) Read the info about the build: compiler, valgrind revision etc. +# and put it into a hash. + for ( ; $i < $n; ++$i) { + $line = $lines[$i]; + last if ($line =~ /^$/); # empty line indicates end of section + my ($key, $value) = split(/:/, $line); + $value =~ s/^[ ]*//; # removing leading blanks + $hash{$key} = $value; + } + + if ($debug) { + foreach my $key (keys %hash) { + my ($val) = $hash{$key}; + print "regtest env: KEY = |$key| VAL = |$val|\n"; + } + } + +# 4) Find out if the regression run failed or passed + $hash{"failures"} = []; + for ($i = $i + 1; $i < $n; ++$i) { + $line = $lines[$i]; + if ($line =~ /Running regression tests/) { + return %hash if ($line =~ /done$/); # regtest succeeded; no failures + die "cannot determine regtest outcome for message $msgno" + if (! ($line =~ /failed$/)); + last; + } + } + +# 5) Regtest failed; locate the section with the list of failing testcases + for ($i = $i + 1; $i < $n; ++$i) { + $line = $lines[$i]; +# Match for end-of-line == because line might be split. + last if ($line =~ /==$/); + } + die "cannot locate failing testcases in message $msgno" if ($i == $n); + +# 6) Get list of failing testcases + for ($i = $i + 1; $i < $n; ++$i) { + $line = $lines[$i]; + + last if ($line =~ /^$/); + + my ($testcase) = (split(/\s+/, $line))[0]; + print "ADD failing testcase $testcase\n" if ($debug); + push @{$hash{"failures"}}, $testcase; + } + + return ($date, %hash); +} + + +#----------------------------------------------------------------- +# Extract architecture; get a pretty name for the distro +#----------------------------------------------------------------- +sub prep_regtest_data { + my (%hash) = @_; + my ($val, $arch, $distro); + + $val = $hash{"uname -mrs"}; + die "uname -mrs info is missing" if (! defined $val); + $arch = (split(/ /, $val))[2]; + + $val = $hash{"Vendor version"}; + die "Vendor version info is missing" if (! defined $val); + + if ($val =~ /Fedora release ([0-9]+)/) { + $distro = "Fedora-$1"; + } elsif ($val =~ /openSUSE ([0-9]+)\.([0-9]+)/) { + $distro = "openSUSE-$1.$2"; + } elsif ($val =~ /SUSE Linux Enterprise Server 11 SP1/) { + $distro = "SLES-11-SP1"; + } elsif ($val =~ /Red Hat Enterprise Linux AS release 4/) { + $distro = "RHEL-4"; + } else { + $distro = "UNKNOWN"; + } + +# Add architecture and distribution to hash + $hash{"arch"} = $arch; + $hash{"distro"} = $distro; + + return %hash; +} + + +#----------------------------------------------------------------- +# Precompute some summary information and record it +#----------------------------------------------------------------- +sub precompute_summary_info +{ + my (%dates) = @_; + + foreach my $date (sort keys %dates) { + my %failure_frequency = (); + + my %nightly = %{ $dates{$date} }; + my @builds = @{ $nightly{"builds"} }; + + $nightly{"num_builds"} = scalar (@builds); + $nightly{"num_failing_builds"} = 0; + $nightly{"num_testcase_failures"} = 0; + + foreach my $build (@builds) { + my %regtest_data = %{ $build }; + + my @failures = @{ $regtest_data{"failures"} }; + my $num_fail = scalar (@failures); + + ++$nightly{"num_failing_builds"} if ($num_fail != 0); + $nightly{"num_testcase_failures"} += $num_fail; + +# Compute how often a testcase failed + foreach my $test ( @failures ) { + if (defined $failure_frequency{$test}) { + ++$failure_frequency{$test}; + } else { + $failure_frequency{$test} = 1; + } + } + } + + $nightly{"num_passing_builds"} = + $nightly{"num_builds"} - $nightly{"num_failing_builds"}; + + $nightly{"num_failing_testcases"} = scalar (keys %failure_frequency); + + $nightly{"failure_frequency"} = { %failure_frequency }; + + $dates{$date} = { %nightly }; + } + + return %dates; +} + + +#----------------------------------------------------------------- +# Get messages from GMANE, and build up a database of results. +#----------------------------------------------------------------- +sub get_regtest_data { + my ($from, $to) = @_; + + my $url_base = "http://article.gmane.org/gmane.comp.debugging.valgrind.devel/"; + + my %dates = (); + + my $old_date = "-1"; + my @builds = (); + + for (my $i = $from; $i <= $to; ++$i) { + my $url = "$url_base" . "$i"; + + my $page = get("$url"); + + if ($keep) { + open (EMAIL, ">$i"); + print EMAIL $page; + close(EMAIL); + } + +# Detect if the article does not exist. Happens for too large --to= values + last if ($page eq "No such file.\n"); + +# Split the page into lines + my @lines = split(/\n/, $page); + +# Check whether it contains a regression test result + next if (! is_regtest_result(@lines)); + print "message $i is a regression test result\n" if ($debug); + +# Get the raw data + my ($date, %regtest_data) = get_raw_data(@lines); + + %regtest_data = prep_regtest_data(%regtest_data); + + if ($date ne $old_date) { + my %nightly = (); + $nightly{"builds"} = [ @builds ]; + $dates{$old_date} = { %nightly } if ($old_date ne "-1"); + + $old_date = $date; + @builds = (); + } + + push @builds, { %regtest_data }; + } + my %nightly = (); + $nightly{"builds"} = [ @builds ]; + $dates{$old_date} = { %nightly } if ($old_date ne "-1"); + +# Convenience: precompute some info we'll be interested in + %dates = precompute_summary_info( %dates ); + + return %dates; +} + + +#----------------------------------------------------------------- +# Write out the results in a form suitable for automatic post-processing +#----------------------------------------------------------------- +sub dump_results { + my (%dates) = @_; + + foreach my $date (sort keys %dates) { + + my %nightly = %{ $dates{$date} }; + my @builds = @{ $nightly{"builds"} }; + + foreach my $build (@builds) { + my %regtest_data = %{ $build }; + + my $arch = $regtest_data{"arch"}; + my $distro = $regtest_data{"distro"}; + my @failures = @{ $regtest_data{"failures"} }; + my $num_fail = scalar (@failures); + my $fails = join(":", sort @failures); + + printf("Regrun: %s %3d %-10s %-20s %s\n", + $date, $num_fail, $arch, $distro, $fails); + } + + my %failure_frequency = %{ $nightly{"failure_frequency"} }; + + foreach my $test (keys %failure_frequency) { + printf("Test: %s %3d %s\n", + $date, $failure_frequency{$test}, $test); + } + + printf("Total: %s builds: %d %d fail %d pass tests: %d fail %d unique\n", + $date, $nightly{"num_builds"}, $nightly{"num_failing_builds"}, + $nightly{"num_passing_builds"}, $nightly{"num_testcase_failures"}, + $nightly{"num_failing_testcases"}); + } +} + + +sub write_readable_results { + my (%dates) = @_; + + foreach my $date (sort keys %dates) { + my %nightly = %{ $dates{$date} }; + + print "$date\n----------\n"; + + printf("%3d builds\n", $nightly{"num_builds"}); + printf("%3d builds fail\n", $nightly{"num_failing_builds"}); + printf("%3d builds pass\n", $nightly{"num_passing_builds"}); + print "\n"; + printf("%3d testcase failures (across all runs)\n", + $nightly{"num_testcase_failures"}); + printf("%3d failing testcases (unique)\n", + $nightly{"num_failing_testcases"}); + print "\n"; + + my @builds = @{ $nightly{"builds"} }; + + if ($nightly{"num_passing_builds"} != 0) { + print "Passing builds\n"; + print "--------------\n"; + foreach my $build (@builds) { + my %regtest_data = %{ $build }; + my @failures = @{ $regtest_data{"failures"} }; + my $num_fail = scalar (@failures); + + if ($num_fail == 0) { + my $arch = $regtest_data{"arch"}; + my $distro = $regtest_data{"distro"}; + + printf("%-8s %-15s\n", $arch, $distro); + } + print "\n"; + } + print "\n"; + } + + if ($nightly{"num_failing_builds"} != 0) { + print "Failing builds\n"; + print "--------------\n"; + foreach my $build (@builds) { + my %regtest_data = %{ $build }; + my @failures = @{ $regtest_data{"failures"} }; + my $num_fail = scalar (@failures); + + if ($num_fail != 0) { + my $arch = $regtest_data{"arch"}; + my $distro = $regtest_data{"distro"}; + + printf("%-8s %-15s %d failures\n", $arch, $distro, $num_fail); + foreach my $test (@failures) { + print " $test\n"; + } + print "\n"; + } + } + print "\n"; + } + + print "Failing testcases and their frequency\n"; + print "-------------------------------------\n"; + my %failure_frequency = %{ $nightly{"failure_frequency"} }; + +# Sorted in decreasing frequency + foreach my $test (sort {$failure_frequency{$b} cmp $failure_frequency{$a} } + keys %failure_frequency) { + printf("%3d %s\n", $failure_frequency{$test}, $test); + } + print "\n"; + } +} + + +sub main +{ + my ($from, $to, $dump, $readable); + + $from = $to = 0; + $dump = $readable = 0; + + GetOptions( "from=i" => \$from, + "to=i" => \$to, + "debug" => \$debug, + "dump" => \$dump, + "keep" => \$keep, + "readable" => \$readable + ) || die $usage; + +# 14800 is about Oct 4, 2011 which is when we began including information +# about the environment + + die $usage if ($from < 14800); + + $to = $from + 100 if ($to == 0); + + if ($from > $to) { + print STDERR "*** invalid [from,to] interval. Try again\n"; + die $usage; + } + + $readable = 1 if ($dump == 0 && $readable == 0); + + print "check message interval [$from...$to]\n" if ($debug); + +# Get mails from GMANE mail archive + + my %dates = get_regtest_data($from, $to); + + dump_results(%dates) if ($dump); + + write_readable_results(%dates) if ($readable); +} + +main(); + +exit 0; Property changes on: trunk/auxprogs/nightly-build-summary ___________________________________________________________________ Name: svn:executable + * |