#!/usr/bin/env perl
# vim:ts=4:sw=4:expandtab
-#
-# © 2010-2011 Michael Stapelberg and contributors
-#
-# syntax: ./complete-run.pl --display :1 --display :2
-# to run the test suite on the X11 displays :1 and :2
-# use 'Xdummy :1' and 'Xdummy :2' before to start two
-# headless X11 servers
-#
-
+# © 2010-2012 Michael Stapelberg and contributors
+package complete_run;
use strict;
use warnings;
-use EV;
-use AnyEvent;
-use IO::Scalar; # not in core :\
-use File::Temp qw(tempfile tempdir);
use v5.10;
-use DateTime;
-use Data::Dumper;
+use utf8;
+# the following are modules which ship with Perl (>= 5.10):
+use Pod::Usage;
use Cwd qw(abs_path);
-use Proc::Background;
+use File::Temp qw(tempfile tempdir);
+use Getopt::Long;
+use POSIX ();
use TAP::Harness;
use TAP::Parser;
use TAP::Parser::Aggregator;
-use File::Basename qw(basename);
-use AnyEvent::I3 qw(:all);
-use Try::Tiny;
-use Getopt::Long;
-use Time::HiRes qw(sleep gettimeofday tv_interval);
-use X11::XCB;
-use IO::Socket::UNIX; # core
-use POSIX; # core
+use Time::HiRes qw(time);
+use IO::Handle;
+# these are shipped with the testsuite
+use lib qw(lib);
+use StartXServer;
+use StatusLine;
+use TestWorker;
+# the following modules are not shipped with Perl
+use AnyEvent;
+use AnyEvent::Util;
use AnyEvent::Handle;
-
-# open a file so that we get file descriptor 3. we will later close it in the
-# child and dup() the listening socket file descriptor to 3 to pass it to i3
-open(my $reserved, '<', '/dev/null');
-if (fileno($reserved) != 3) {
- warn "Socket file descriptor is not 3.";
- warn "Please don't start this script within a subshell of vim or something.";
- exit 1;
-}
-
-# install a dummy CHLD handler to overwrite the CHLD handler of AnyEvent / EV
-# XXX: we could maybe also use a different loop than the default loop in EV?
-$SIG{CHLD} = sub {
-};
-
-# reads in a whole file
-sub slurp {
- open my $fh, '<', shift;
- local $/;
- <$fh>;
-}
-
-my $coverage_testing = 0;
+use AnyEvent::I3 qw(:all);
+use X11::XCB::Connection;
+use JSON::XS; # AnyEvent::I3 depends on it, too.
+
+binmode STDOUT, ':utf8';
+binmode STDERR, ':utf8';
+
+# Close superfluous file descriptors which were passed by running in a VIM
+# subshell or situations like that.
+AnyEvent::Util::close_all_fds_except(0, 1, 2);
+
+# convinience wrapper to write to the log file
+my $log;
+sub Log { say $log "@_" }
+
+my %timings;
+my $help = 0;
+# Number of tests to run in parallel. Important to know how many Xephyr
+# instances we need to start (unless @displays are given). Defaults to
+# num_cores * 2.
+my $parallel = undef;
my @displays = ();
+my %options = (
+ valgrind => 0,
+ strace => 0,
+ xtrace => 0,
+ coverage => 0,
+ restart => 0,
+);
+my $keep_xserver_output = 0;
my $result = GetOptions(
- "coverage-testing" => \$coverage_testing,
+ "coverage-testing" => \$options{coverage},
+ "keep-xserver-output" => \$keep_xserver_output,
+ "valgrind" => \$options{valgrind},
+ "strace" => \$options{strace},
+ "xtrace" => \$options{xtrace},
"display=s" => \@displays,
+ "parallel=i" => \$parallel,
+ "help|?" => \$help,
);
+pod2usage(-verbose => 2, -exitcode => 0) if $help;
+
+# Check for missing executables
+my @binaries = qw(
+ ../i3
+ ../i3bar/i3bar
+ ../i3-config-wizard/i3-config-wizard
+ ../i3-dump-log/i3-dump-log
+ ../i3-input/i3-input
+ ../i3-msg/i3-msg
+ ../i3-nagbar/i3-nagbar
+ );
+
+foreach my $binary (@binaries) {
+ die "$binary executable not found, did you run “make”?" unless -e $binary;
+ die "$binary is not an executable" unless -x $binary;
+}
+
+qx(Xephyr -help 2>&1);
+die "Xephyr was not found in your path. Please install Xephyr (xserver-xephyr on Debian)." if $?;
+
@displays = split(/,/, join(',', @displays));
@displays = map { s/ //g; $_ } @displays;
-@displays = qw(:1) if @displays == 0;
+# 2: get a list of all testcases
+my @testfiles = @ARGV;
+
+# if no files were passed on command line, run all tests from t/
+@testfiles = <t/*.t> if @testfiles == 0;
+
+my $numtests = scalar @testfiles;
+
+# No displays specified, let’s start some Xephyr instances.
+if (@displays == 0) {
+ @displays = start_xserver($parallel, $numtests, $keep_xserver_output);
+}
+
+# 1: create an output directory for this test-run
+my $outdir = "testsuite-";
+$outdir .= POSIX::strftime("%Y-%m-%d-%H-%M-%S-", localtime());
+$outdir .= `git describe --tags`;
+chomp($outdir);
+mkdir($outdir) or die "Could not create $outdir";
+unlink("latest") if -l "latest";
+symlink("$outdir", "latest") or die "Could not symlink latest to $outdir";
+
# connect to all displays for two reasons:
# 1: check if the display actually works
# 2: keep the connection open so that i3 is not the only client. this prevents
-# the X server from exiting (Xdummy will restart it, but not quick enough
-# sometimes)
-my @conns;
-my @wdisplays;
+# the X server from exiting
+my @single_worker;
for my $display (@displays) {
my $screen;
- my $x = X11::XCB->new($display, $screen);
+ my $x = X11::XCB::Connection->new(display => $display);
if ($x->has_error) {
- say STDERR "WARNING: Not using X11 display $display, could not connect";
+ die "Could not connect to display $display\n";
} else {
- push @conns, $x;
- push @wdisplays, $display;
+ # start a TestWorker for each display
+ push @single_worker, worker($display, $x, $outdir, \%options);
}
}
-my $config = slurp('i3-test.config');
+# Read previous timing information, if available. We will be able to roughly
+# predict the test duration and schedule a good order for the tests.
+my $timingsjson = StartXServer::slurp('.last_run_timings.json');
+%timings = %{decode_json($timingsjson)} if length($timingsjson) > 0;
+
+# Re-order the files so that those which took the longest time in the previous
+# run will be started at the beginning to not delay the whole run longer than
+# necessary.
+@testfiles = map { $_->[0] }
+ sort { $b->[1] <=> $a->[1] }
+ map { [$_, $timings{$_} // 999] } @testfiles;
+
+# Run 000-load-deps.t first to bail out early when dependencies are missing.
+my $loadtest = "t/000-load-deps.t";
+if ((scalar grep { $_ eq $loadtest } @testfiles) > 0) {
+ @testfiles = ($loadtest, grep { $_ ne $loadtest } @testfiles);
+}
-# 1: get a list of all testcases
-my @testfiles = @ARGV;
+printf("\nRough time estimate for this run: %.2f seconds\n\n", $timings{GLOBAL})
+ if exists($timings{GLOBAL});
-# if no files were passed on command line, run all tests from t/
-@testfiles = <t/*.t> if @testfiles == 0;
+# Forget the old timings, we don’t necessarily run the same set of tests as
+# before. Otherwise we would end up with left-overs.
+%timings = (GLOBAL => time());
-# 2: create an output directory for this test-run
-my $outdir = "testsuite-";
-$outdir .= DateTime->now->strftime("%Y-%m-%d-%H-%M-%S-");
-$outdir .= `git describe --tags`;
-chomp($outdir);
-mkdir($outdir) or die "Could not create $outdir";
-unlink("latest") if -e "latest";
-symlink("$outdir", "latest") or die "Could not symlink latest to $outdir";
+my $logfile = "$outdir/complete-run.log";
+open $log, '>', $logfile or die "Could not create '$logfile': $!";
+$log->autoflush(1);
+say "Writing logfile to '$logfile'...";
# 3: run all tests
my @done;
my $aggregator = TAP::Parser::Aggregator->new();
$aggregator->start();
-my $cv = AnyEvent->condvar;
+status_init(displays => \@displays, tests => $num);
+
+my $single_cv = AE::cv;
# We start tests concurrently: For each display, one test gets started. Every
# test starts another test after completing.
-take_job($_) for @wdisplays;
+for (@single_worker) {
+ $single_cv->begin;
+ take_job($_, $single_cv, \@testfiles);
+}
+
+$single_cv->recv;
+
+$aggregator->stop();
+
+# print empty lines to seperate failed tests from statuslines
+print "\n\n";
+
+for (@done) {
+ my ($test, $output) = @$_;
+ say "no output for $test" unless $output;
+ Log "output for $test:";
+ Log $output;
+ # print error messages of failed tests
+ say for $output =~ /^not ok.+\n+((?:^#.+\n)+)/mg
+}
+
+# 4: print summary
+$harness->summary($aggregator);
+
+close $log;
+
+# 5: Save the timings for better scheduling/prediction next run.
+$timings{GLOBAL} = time() - $timings{GLOBAL};
+open(my $fh, '>', '.last_run_timings.json');
+print $fh encode_json(\%timings);
+close($fh);
+
+# 6: Print the slowest test files.
+my @slowest = map { $_->[0] }
+ sort { $b->[1] <=> $a->[1] }
+ map { [$_, $timings{$_}] }
+ grep { !/^GLOBAL$/ } keys %timings;
+say '';
+say 'The slowest tests are:';
+printf("\t%s with %.2f seconds\n", $_, $timings{$_})
+ for @slowest[0..($#slowest > 4 ? 4 : $#slowest)];
+
+# When we are running precisely one test, print the output. Makes developing
+# with a single testcase easier.
+if ($numtests == 1) {
+ say '';
+ say 'Test output:';
+ say StartXServer::slurp($logfile);
+}
+
+END { cleanup() }
+
+exit 0;
#
# Takes a test from the beginning of @testfiles and runs it.
# triggered to finish testing.
#
sub take_job {
- my ($display) = @_;
-
- my $test = shift @testfiles;
- return unless $test;
- my $dont_start = (slurp($test) =~ /# !NO_I3_INSTANCE!/);
- my $logpath = "$outdir/i3-log-for-" . basename($test);
-
- my ($fh, $tmpfile) = tempfile('i3-run-cfg.XXXXXX', UNLINK => 1);
- say $fh $config;
- say $fh "ipc-socket /tmp/nested-$display";
- close($fh);
-
- my $activate_cv = AnyEvent->condvar;
- my $time_before_start = [gettimeofday];
- my $start_i3 = sub {
- # remove the old unix socket
- unlink("/tmp/nested-$display-activation");
-
- # pass all file descriptors up to three to the children.
- # we need to set this flag before opening the socket.
- open(my $fdtest, '<', '/dev/null');
- $^F = fileno($fdtest);
- close($fdtest);
- my $socket = IO::Socket::UNIX->new(
- Listen => 1,
- Local => "/tmp/nested-$display-activation",
- );
-
- my $pid = fork;
- if (!defined($pid)) {
- die "could not fork()";
- }
- if ($pid == 0) {
- $ENV{LISTEN_PID} = $$;
- $ENV{LISTEN_FDS} = 1;
- $ENV{DISPLAY} = $display;
- $^F = 3;
-
- close($reserved);
- POSIX::dup2(fileno($socket), 3);
-
- # now execute i3
- my $i3cmd = abs_path("../i3") . " -V -d all --disable-signalhandler";
- my $cmd = "exec $i3cmd -c $tmpfile >$logpath 2>&1";
- exec "/bin/sh", '-c', $cmd;
-
- # if we are still here, i3 could not be found or exec failed. bail out.
- exit 1;
- }
+ my ($worker, $cv, $tests) = @_;
- my $child_watcher;
- $child_watcher = AnyEvent->child(pid => $pid, cb => sub {
- say "child died. pid = $pid";
- undef $child_watcher;
- });
-
- # close the socket, the child process should be the only one which keeps a file
- # descriptor on the listening socket.
- $socket->close;
-
- # We now connect (will succeed immediately) and send a request afterwards.
- # As soon as the reply is there, i3 is considered ready.
- my $cl = IO::Socket::UNIX->new(Peer => "/tmp/nested-$display-activation");
- my $hdl;
- $hdl = AnyEvent::Handle->new(fh => $cl, on_error => sub { $activate_cv->send(0) });
-
- # send a get_tree message without payload
- $hdl->push_write('i3-ipc' . pack("LL", 0, 4));
-
- # wait for the reply
- $hdl->push_read(chunk => 1, => sub {
- my ($h, $line) = @_;
- $activate_cv->send(1);
- undef $hdl;
- });
-
- return $pid;
- };
-
- my $pid;
- $pid = $start_i3->() unless $dont_start;
-
- my $kill_i3 = sub {
- # Don’t bother killing i3 when we haven’t started it
- return if $dont_start;
-
- # When measuring code coverage, try to exit i3 cleanly (otherwise, .gcda
- # files are not written) and fallback to killing it
- if ($coverage_testing) {
- my $exited = 0;
- try {
- say "Exiting i3 cleanly...";
- i3("/tmp/nested-$display")->command('exit')->recv;
- $exited = 1;
- };
- return if $exited;
- }
+ my $test = shift @$tests
+ or return $cv->end;
- say "[$display] killing i3";
- kill(9, $pid) or die "could not kill i3";
- };
-
- # This will be called as soon as i3 is running and answered to our
- # IPC request
- $activate_cv->cb(sub {
- my $time_activating = [gettimeofday];
- my $start_duration = tv_interval($time_before_start, $time_activating);
- my ($status) = $activate_cv->recv;
- if ($dont_start) {
- say "[$display] Not starting i3, testcase does that";
- } else {
- say "[$display] i3 startup: took " . sprintf("%.2f", $start_duration) . "s, status = $status";
- }
+ my $display = $worker->{display};
+
+ Log status($display, "$test: starting");
+ $timings{$test} = time();
+ worker_next($worker, $test);
- say "[$display] Running $test with logfile $logpath";
-
- my $output;
- my $parser = TAP::Parser->new({
- exec => [ 'sh', '-c', qq|DISPLAY=$display LOGPATH="$logpath" /usr/bin/perl -It/lib $test| ],
- spool => IO::Scalar->new(\$output),
- merge => 1,
- });
-
- my @watchers;
- my ($stdout, $stderr) = $parser->get_select_handles;
- for my $handle ($parser->get_select_handles) {
- my $w;
- $w = AnyEvent->io(
- fh => $handle,
- poll => 'r',
- cb => sub {
- # Ignore activity on stderr (unnecessary with merge => 1,
- # but let’s keep it in here if we want to use merge => 0
- # for some reason in the future).
- return if defined($stderr) and $handle == $stderr;
-
- my $result = $parser->next;
- if (defined($result)) {
- # TODO: check if we should bail out
- return;
- }
-
- # $result is not defined, we are done parsing
- say "[$display] $test finished";
- close($parser->delete_spool);
- $aggregator->add($test, $parser);
- push @done, [ $test, $output ];
-
- $kill_i3->();
-
- undef $_ for @watchers;
- if (@done == $num) {
- $cv->send;
- } else {
- take_job($display);
- }
+ # create a TAP::Parser with an in-memory fh
+ my $output;
+ my $parser = TAP::Parser->new({
+ source => do { open(my $fh, '<', \$output); $fh },
+ });
+
+ my $ipc = $worker->{ipc};
+
+ my $w;
+ $w = AnyEvent->io(
+ fh => $ipc,
+ poll => 'r',
+ cb => sub {
+ state $tests_completed = 0;
+ state $partial = '';
+
+ sysread($ipc, my $buf, 4096) or die "sysread: $!";
+
+ if ($partial) {
+ $buf = $partial . $buf;
+ $partial = '';
+ }
+
+ # make sure we feed TAP::Parser complete lines so it doesn't blow up
+ if (substr($buf, -1, 1) ne "\n") {
+ my $nl = rindex($buf, "\n");
+ if ($nl == -1) {
+ $partial = $buf;
+ return;
+ }
+
+ # strip partial from buffer
+ $partial = substr($buf, $nl + 1, '');
+ }
+
+ # count lines before stripping eof-marker otherwise we might
+ # end up with for (1 .. 0) { } which would effectivly skip the loop
+ my $lines = $buf =~ tr/\n//;
+ my $t_eof = $buf =~ s/^$TestWorker::EOF$//m;
+
+ $output .= $buf;
+
+ for (1 .. $lines) {
+ my $result = $parser->next;
+ next unless defined($result);
+ if ($result->is_test) {
+ $tests_completed++;
+ status($display, "$test: [$tests_completed/??] ");
+ } elsif ($result->is_bailout) {
+ Log status($display, "$test: BAILOUT");
+ status_completed(scalar @done);
+ say "";
+ say "test $test bailed out: " . $result->explanation;
+ exit 1;
}
- );
- push @watchers, $w;
+ }
+
+ return unless $t_eof;
+
+ Log status($display, "$test: finished");
+ $timings{$test} = time() - $timings{$test};
+ status_completed(scalar @done);
+
+ $aggregator->add($test, $parser);
+ push @done, [ $test, $output ];
+
+ undef $w;
+ take_job($worker, $cv, $tests);
}
- });
+ );
+}
- $activate_cv->send(1) if $dont_start;
+sub cleanup {
+ $_->() for our @CLEANUP;
+ exit;
}
-$cv->recv;
+# must be in a begin block because we C<exit 0> above
+BEGIN {
+ $SIG{$_} = sub {
+ require Carp; Carp::cluck("Caught SIG$_[0]\n");
+ cleanup();
+ } for qw(INT TERM QUIT KILL PIPE)
+}
-$aggregator->stop();
+__END__
-for (@done) {
- my ($test, $output) = @$_;
- say "output for $test:";
- say $output;
-}
+=head1 NAME
-# 4: print summary
-$harness->summary($aggregator);
+complete-run.pl - Run the i3 testsuite
+
+=head1 SYNOPSIS
+
+complete-run.pl [files...]
+
+=head1 EXAMPLE
+
+To run the whole testsuite on a reasonable number of Xephyr instances (your
+running X11 will not be touched), run:
+ ./complete-run.pl
+
+To run only a specific test (useful when developing a new feature), run:
+ ./complete-run t/100-fullscreen.t
+
+=head1 OPTIONS
+
+=over 8
+
+=item B<--display>
+
+Specifies which X11 display should be used. Can be specified multiple times and
+will parallelize the tests:
+
+ # Run tests on the second X server
+ ./complete-run.pl -d :1
+
+ # Run four tests in parallel on some Xephyr servers
+ ./complete-run.pl -d :1,:2,:3,:4
+
+Note that it is not necessary to specify this anymore. If omitted,
+complete-run.pl will start (num_cores * 2) Xephyr instances.
+
+=item B<--valgrind>
+
+Runs i3 under valgrind to find memory problems. The output will be available in
+C<latest/valgrind-for-$test.log>.
+
+=item B<--strace>
+
+Runs i3 under strace to trace system calls. The output will be available in
+C<latest/strace-for-$test.log>.
+
+=item B<--xtrace>
+
+Runs i3 under xtrace to trace X11 requests/replies. The output will be
+available in C<latest/xtrace-for-$test.log>.
+
+=item B<--coverage-testing>
+
+Exits i3 cleanly (instead of kill -9) to make coverage testing work properly.
+
+=item B<--parallel>
+
+Number of Xephyr instances to start (if you don’t want to start num_cores * 2
+instances for some reason).
+
+ # Run all tests on a single Xephyr instance
+ ./complete-run.pl -p 1