2 # vim:ts=4:sw=4:expandtab
3 # © 2010-2011 Michael Stapelberg and contributors
8 # the following are modules which ship with Perl (>= 5.10):
11 use File::Temp qw(tempfile tempdir);
16 use TAP::Parser::Aggregator;
17 use Time::HiRes qw(time);
18 # these are shipped with the testsuite
23 # the following modules are not shipped with Perl
27 use AnyEvent::I3 qw(:all);
28 use X11::XCB::Connection;
29 use JSON::XS; # AnyEvent::I3 depends on it, too.
31 # Close superfluous file descriptors which were passed by running in a VIM
32 # subshell or situations like that.
33 AnyEvent::Util::close_all_fds_except(0, 1, 2);
35 # convinience wrapper to write to the log file
37 sub Log { say $log "@_" }
40 my $coverage_testing = 0;
44 # Number of tests to run in parallel. Important to know how many Xdummy
45 # instances we need to start (unless @displays are given). Defaults to
50 my $result = GetOptions(
51 "coverage-testing" => \$coverage_testing,
52 "valgrind" => \$valgrind,
54 "display=s" => \@displays,
55 "parallel=i" => \$parallel,
59 pod2usage(-verbose => 2, -exitcode => 0) if $help;
61 @displays = split(/,/, join(',', @displays));
62 @displays = map { s/ //g; $_ } @displays;
64 # 2: get a list of all testcases
65 my @testfiles = @ARGV;
67 # if no files were passed on command line, run all tests from t/
68 @testfiles = <t/*.t> if @testfiles == 0;
70 my $numtests = scalar @testfiles;
72 # When the user specifies displays, we don’t run multi-monitor tests at all
73 # (because we don’t know which displaynumber is the X-Server with multiple
77 # No displays specified, let’s start some Xdummy instances.
80 ($dpyref, $multidpy) = start_xdummy($parallel, $numtests);
84 # 1: create an output directory for this test-run
85 my $outdir = "testsuite-";
86 $outdir .= POSIX::strftime("%Y-%m-%d-%H-%M-%S-", localtime());
87 $outdir .= `git describe --tags`;
89 mkdir($outdir) or die "Could not create $outdir";
90 unlink("latest") if -e "latest";
91 symlink("$outdir", "latest") or die "Could not symlink latest to $outdir";
94 # connect to all displays for two reasons:
95 # 1: check if the display actually works
96 # 2: keep the connection open so that i3 is not the only client. this prevents
97 # the X server from exiting (Xdummy will restart it, but not quick enough
100 for my $display (@displays) {
102 my $x = X11::XCB::Connection->new(display => $display);
104 die "Could not connect to display $display\n";
106 # start a TestWorker for each display
107 push @single_worker, worker($display, $x, $outdir);
112 if (defined($multidpy)) {
113 my $x = X11::XCB::Connection->new(display => $multidpy);
115 die "Could not connect to multi-monitor display $multidpy\n";
117 push @multi_worker, worker($multidpy, $x, $outdir);
121 # Read previous timing information, if available. We will be able to roughly
122 # predict the test duration and schedule a good order for the tests.
123 my $timingsjson = StartXDummy::slurp('.last_run_timings.json');
124 %timings = %{decode_json($timingsjson)} if length($timingsjson) > 0;
126 # Re-order the files so that those which took the longest time in the previous
127 # run will be started at the beginning to not delay the whole run longer than
129 @testfiles = map { $_->[0] }
130 sort { $b->[1] <=> $a->[1] }
131 map { [$_, $timings{$_} // 999] } @testfiles;
133 printf("\nRough time estimate for this run: %.2f seconds\n\n", $timings{GLOBAL})
134 if exists($timings{GLOBAL});
136 # Forget the old timings, we don’t necessarily run the same set of tests as
137 # before. Otherwise we would end up with left-overs.
138 %timings = (GLOBAL => time());
140 my $logfile = "$outdir/complete-run.log";
141 open $log, '>', $logfile or die "Could not create '$logfile': $!";
142 say "Writing logfile to '$logfile'...";
146 my $num = @testfiles;
147 my $harness = TAP::Harness->new({ });
149 my @single_monitor_tests = grep { m,^t/([0-9]+)-, && $1 < 500 } @testfiles;
150 my @multi_monitor_tests = grep { m,^t/([0-9]+)-, && $1 >= 500 } @testfiles;
152 my $aggregator = TAP::Parser::Aggregator->new();
153 $aggregator->start();
155 status_init(displays => [ @displays, $multidpy ], tests => $num);
157 my $single_cv = AE::cv;
158 my $multi_cv = AE::cv;
160 # We start tests concurrently: For each display, one test gets started. Every
161 # test starts another test after completing.
162 for (@single_worker) {
164 take_job($_, $single_cv, \@single_monitor_tests);
166 for (@multi_worker) {
168 take_job($_, $multi_cv, \@multi_monitor_tests);
176 # print empty lines to seperate failed tests from statuslines
180 my ($test, $output) = @$_;
181 say "no output for $test" unless $output;
182 Log "output for $test:";
184 # print error messages of failed tests
185 say for $output =~ /^not ok.+\n+((?:^#.+\n)+)/mg
189 $harness->summary($aggregator);
193 # 5: Save the timings for better scheduling/prediction next run.
194 $timings{GLOBAL} = time() - $timings{GLOBAL};
195 open(my $fh, '>', '.last_run_timings.json');
196 print $fh encode_json(\%timings);
199 # 6: Print the slowest test files.
200 my @slowest = map { $_->[0] }
201 sort { $b->[1] <=> $a->[1] }
202 map { [$_, $timings{$_}] }
203 grep { !/^GLOBAL$/ } keys %timings;
205 say 'The slowest tests are:';
206 printf("\t%s with %.2f seconds\n", $_, $timings{$_})
207 for @slowest[0..($#slowest > 4 ? 4 : $#slowest)];
209 # When we are running precisely one test, print the output. Makes developing
210 # with a single testcase easier.
211 if ($numtests == 1) {
214 say StartXDummy::slurp($logfile);
222 # Takes a test from the beginning of @testfiles and runs it.
224 # The TAP::Parser (which reads the test output) will get called as soon as
225 # there is some activity on the stdout file descriptor of the test process
226 # (using an AnyEvent->io watcher).
228 # When a test completes and @done contains $num entries, the $cv condvar gets
229 # triggered to finish testing.
232 my ($worker, $cv, $tests) = @_;
234 my $test = shift @$tests
237 my $display = $worker->{display};
239 Log status($display, "$test: starting");
240 $timings{$test} = time();
241 worker_next($worker, $test);
243 # create a TAP::Parser with an in-memory fh
245 my $parser = TAP::Parser->new({
246 source => do { open(my $fh, '<', \$output); $fh },
249 my $ipc = $worker->{ipc};
256 state $tests_completed = 0;
259 sysread($ipc, my $buf, 4096) or die "sysread: $!";
262 $buf = $partial . $buf;
266 # make sure we feed TAP::Parser complete lines so it doesn't blow up
267 if (substr($buf, -1, 1) ne "\n") {
268 my $nl = rindex($buf, "\n");
274 # strip partial from buffer
275 $partial = substr($buf, $nl + 1, '');
278 # count lines before stripping eof-marker otherwise we might
279 # end up with for (1 .. 0) { } which would effectivly skip the loop
280 my $lines = $buf =~ tr/\n//;
281 my $t_eof = $buf =~ s/^$TestWorker::EOF$//m;
286 my $result = $parser->next;
287 if (defined($result) and $result->is_test) {
289 status($display, "$test: [$tests_completed/??] ");
293 return unless $t_eof;
295 Log status($display, "$test: finished");
296 $timings{$test} = time() - $timings{$test};
297 status_completed(scalar @done);
299 $aggregator->add($test, $parser);
300 push @done, [ $test, $output ];
303 take_job($worker, $cv, $tests);
309 $_->() for our @CLEANUP;
313 # must be in a begin block because we C<exit 0> above
316 require Carp; Carp::cluck("Caught SIG$_[0]\n");
318 } for qw(INT TERM QUIT KILL PIPE)
325 complete-run.pl - Run the i3 testsuite
329 complete-run.pl [files...]
333 To run the whole testsuite on a reasonable number of Xdummy instances (your
334 running X11 will not be touched), run:
337 To run only a specific test (useful when developing a new feature), run:
338 ./complete-run t/100-fullscreen.t
346 Specifies which X11 display should be used. Can be specified multiple times and
347 will parallelize the tests:
349 # Run tests on the second X server
350 ./complete-run.pl -d :1
352 # Run four tests in parallel on some Xdummy servers
353 ./complete-run.pl -d :1,:2,:3,:4
355 Note that it is not necessary to specify this anymore. If omitted,
356 complete-run.pl will start (num_cores * 2) Xdummy instances.
360 Runs i3 under valgrind to find memory problems. The output will be available in
361 C<latest/valgrind-for-$test.log>.
365 Runs i3 under strace to trace system calls. The output will be available in
366 C<latest/strace-for-$test.log>.
368 =item B<--coverage-testing>
370 Exits i3 cleanly (instead of kill -9) to make coverage testing work properly.
374 Number of Xdummy instances to start (if you don’t want to start num_cores * 2
375 instances for some reason).
377 # Run all tests on a single Xdummy instance
378 ./complete-run.pl -p 1