3 Michael Stapelberg <michael+i3@stapelberg.de>
6 This document explains how the i3 testsuite works, how to use it and extend it.
7 It is targeted at developers who not necessarily have been doing testing before
8 or have not been testing in Perl before. In general, the testsuite is not of
9 interest for end users.
14 The i3 testsuite is a collection of files which contain testcases for various
15 i3 features. Some of them test if a certain workflow works correctly (moving
16 windows, focus behaviour, …). Others are regression tests and contain code
17 which previously made i3 crash or lead to unexpected behaviour. They then check
18 if i3 still runs (meaning it did not crash) and if it handled everything
21 The goal of having these tests is to automatically find problems and to
22 automatically get a feel for whether a change in the source code breaks any
23 existing feature. After every modification of the i3 sourcecode, the developer
24 should run the full testsuite. If one of the tests fails, the corresponding
25 problem should be fixed (or, in some cases, the testcase has to be modified).
26 For every bugreport, a testcase should be written to test the correct
27 behaviour. Initially, it will fail, but after fixing the bug, it will pass.
28 This ensures (or increases the chance) that bugs which have been fixed once
29 will never be found again.
31 Also, when implementing a new feature, a testcase might be a good way to be
32 able to easily test if the feature is working correctly. Many developers will
33 test manually if everything works. Having a testcase not only helps you with
34 that, but it will also be useful for every future change.
38 For several reasons, the i3 testsuite has been implemented in Perl:
40 1. Perl has a long tradition of testing. Every popular/bigger Perl module which
41 you can find on CPAN will not only come with documentation, but also with
42 tests. Therefore, the available infrastructure for tests is comprehensive.
43 See for example the excellent http://search.cpan.org/perldoc?Test::More
44 and the referenced http://search.cpan.org/perldoc?Test::Tutorial.
46 2. Perl is widely available and has a well-working package infrastructure.
47 3. The author is familiar with Perl :).
48 4. It is a good idea to use a different language for the tests than the
49 implementation itself.
51 Please do not start programming language flamewars at this point.
55 ==== Script: complete-run
57 The testcases are run by a script called +complete-run.pl+. It runs all
58 testcases by default, but you can be more specific and let it only run one or
59 more testcases. Also, it takes care of starting up a separate instance of i3
60 with an appropriate configuration file and creates a folder for each run
61 containing the appropriate i3 logfile for each testcase. The latest folder can
62 always be found under the symlink +latest/+. Unless told differently, it will
63 run the tests on a separate X server instance (using the Xdummy script).
65 .Example invocation of complete-run.pl+
66 ---------------------------------------
70 # output omitted because it is very long
72 Files=78, Tests=734, 27 wallclock secs ( 0.38 usr 0.48 sys + 17.65 cusr 3.21 csys = 21.72 CPU)
75 $ ./complete-run.pl t/04-floating.t
76 [:3] i3 startup: took 0.07s, status = 1
77 [:3] Running t/04-floating.t with logfile testsuite-2011-09-24-16-06-04-4.0.2-226-g1eb011a/i3-log-for-04-floating.t
78 [:3] t/04-floating.t finished
80 output for t/04-floating.t:
81 ok 1 - use X11::XCB::Window;
82 ok 2 - The object isa X11::XCB::Window
83 ok 3 - Window is mapped
84 ok 4 - i3 raised the width to 75
85 ok 5 - i3 raised the height to 50
86 ok 6 - i3 did not map it to (0x0)
87 ok 7 - The object isa X11::XCB::Window
88 ok 8 - i3 let the width at 80
89 ok 9 - i3 let the height at 90
90 ok 10 - i3 mapped it to x=1
91 ok 11 - i3 mapped it to y=18
92 ok 12 - The object isa X11::XCB::Window
93 ok 13 - i3 let the width at 80
94 ok 14 - i3 let the height at 90
98 Files=1, Tests=14, 0 wallclock secs ( 0.01 usr 0.00 sys + 0.19 cusr 0.03 csys = 0.23 CPU)
101 $ less latest/i3-log-for-04-floating.t
102 ----------------------------------------
106 The testsuite makes extensive use of the IPC (Inter-Process Communication)
107 interface which i3 provides. It is used for the startup process of i3, for
108 terminating it cleanly and (most importantly) for modifying and getting the
109 current state (layout tree).
111 See [http://i3wm.org/docs/ipc.html] for documentation on the IPC interface.
115 In order to open new windows, change attributes, get events, etc., the
116 testsuite uses X11::XCB, a new (and quite specific to i3 at the moment) Perl
117 module which uses the XCB protocol description to generate Perl bindings to
118 X11. They work in a very similar way to libxcb (which i3 uses) and provide
119 relatively high-level interfaces (objects such as +X11::XCB::Window+) aswell as
120 access to the low-level interface, which is very useful when testing a window
123 === Filesystem structure
125 In the git root of i3, the testcases live in the folder +testcases+. This
126 folder contains the +complete-run.pl+ and +Xdummy+ scripts and a base
127 configuration file which will be used for the tests. The different testcases
128 (their file extension is .t, not .pl) themselves can be found in the
129 conventionally named subfolder +t+:
131 .Filesystem structure
132 --------------------------------------------
134 │ ├── complete-run.pl
138 │ │ ├── SocketActivation.pm
139 │ │ └── StartXDummy.pm
143 │ │ ├── 02-fullscreen.t
145 │ │ ├── omitted for brevity
147 │ │ └── 74-regress-focus-toggle.t
149 --------------------------------------------
151 == Anatomy of a testcase
153 Learning by example is definitely a good strategy when you are wondering how to
154 write a testcase. Let's take +t/11-goto.t+ as an easy example and go through it
157 .t/11-goto.t: Boilerplate
158 ----------------------
160 # vim:ts=4:sw=4:expandtab
165 my $x = X11::XCB::Connection->new;
166 -----------------------
168 This is what we call boilerplate. It exists at the top of every test file (to
169 some extent). The first line is the shebang, which specifies that this file is
170 a Perl script. The second line contains VIM specific settings on how to
171 edit/format this file (use spaces instead of tabs, indent using 4 spaces).
172 Afterwards, the +i3test+ module is used. This module contains i3 testsuite
173 specific functions which you are strongly encouraged to use. They make writing
174 testcases a lot easier and will make it easier for other people to read your
177 The next line uses the +File::Temp+ module. This is specific to this testcase,
178 because it needs to generate a temporary name during the test. Many testcases
179 use only the +i3test+ module.
181 The last line opens a connection to X11. You might or might not need this in
182 your testcase, depending on whether you are going to open windows (etc.) or
183 only use i3 commands.
186 ----------------------
187 my $tmp = fresh_workspace;
190 ----------------------
192 The first line calls i3test's +fresh_workspace+ function which looks for a
193 currently unused workspace, switches to it, and returns its name. The variable
194 +$tmp+ will end up having a value such as +"/tmp/87kBVcHbA9"+. Note that this
195 is not (necessarily) a valid path, it's just a random workspace name.
197 So, now that we are on a new workspace, we ensure that the workspace uses
198 horizontal orientation by issuing the +split h+ command (see the i3 User's
199 Guide for a list of commands). This is not strictly necessary, but good style.
200 In general, the +cmd+ function executes the specified i3 command by using the
201 IPC interface and returns once i3 acknowledged the command.
204 ----------------------
205 #####################################################################
206 # Create two windows and make sure focus switching works
207 #####################################################################
209 my $top = open_window($x);
210 my $mid = open_window($x);
211 my $bottom = open_window($x);
212 ----------------------
214 In every major section of a testcase, you should put a comment like the one
215 above. This makes it immediately clear how the file is structured.
217 The +open_window+ function opens a standard window, which will then be put into
218 tiling mode by i3. If you want a floating window, use the
219 +open_floating_window+ function. These functions accept the same parameters as
220 +X11::XCB::Window->new+, see the i3test documentation at TODO.
222 .t/11-goto.t: Helper function
223 ----------------------
225 # Returns the input focus after sending the given command to i3 via IPC
226 # and syncing with i3
233 return $x->input_focus;
235 ----------------------
237 This section defines a helper function which will be used over and over in this
238 testcase. If you have code which gets executed more than once or twice
239 (depending on the length of your test, use your best judgement), please put it
240 in a function. Tests should be short, concise and clear.
242 The +focus_after+ function executes a command and returns the X11 focus after
243 the command was executed. The +sync_with_i3+ command makes sure that i3 could
244 push its state to X11. See <<i3_sync>> to learn how this works exactly.
246 .t/11-goto.t: Test assumptions
247 ----------------------
248 $focus = $x->input_focus;
249 is($focus, $bottom->id, "Latest window focused");
251 $focus = focus_after('focus left');
252 is($focus, $mid->id, "Middle window focused");
253 ----------------------
255 Now, we run the first two real tests. They use +Test::More+'s +is+ function,
256 which compares two values and prints the differences if they are not the same.
257 After the arguments, we supply a short comment to indicate what we are testing
258 here. This makes it vastly more easy for the developer to spot which testcase
259 is the problem in case one fails.
261 The first test checks that the most recently opened window is focused.
262 Afterwards, the command +focus left+ is issued and it is verified that the
263 middle window now has focus.
265 Note that this is not a comprehensive test of the +focus+ command -- we would
266 have to test wrapping, focus when using a more complex layout, focusing the
267 parent/child containers, etc. But that is not the point of this testcase.
268 Instead, we just want to know if +$x->input_focus+ corresponds with what we are
269 expecting. If not, something is completely wrong with the test environment and
270 this trivial test will fail.
272 .t/11-goto.t: Test that the feature does not work (yet)
273 ----------------------
274 #####################################################################
275 # Now goto a mark which does not exist
276 #####################################################################
278 my $random_mark = mktemp('mark.XXXXXX');
280 $focus = focus_after(qq|[con_mark="$random_mark"] focus|);
281 is($focus, $mid->id, "focus unchanged");
282 ----------------------
284 Syntax hint: The qq keyword is the interpolating quote operator. It lets you
285 chose a quote character (in this case the +|+ character, a pipe). This makes
286 having double quotes in our string easy.
288 In this new major section, a random mark (mark is an identifier for a window,
289 see "VIM-like marks" in the i3 User’s Guide) will be generated. Afterwards, we
290 test that trying to focus that mark will not do anything. This is important: Do
291 not only test that using a feature has the expected outcome, but also test that
292 using it without properly initializing it does no harm. This command could for
293 example have changed focus anyways (a bug) or crash i3 (obviously a bug).
295 .t/11-goto.t: Test that the feature does work
296 ----------------------
297 cmd "mark $random_mark";
299 $focus = focus_after('focus left');
300 is($focus, $top->id, "Top window focused");
302 $focus = focus_after(qq|[con_mark="$random_mark"] focus|);
303 is($focus, $mid->id, "goto worked");
304 ----------------------
306 Remember: Focus was on the middle window (we verified that earlier in "Test
307 assumptions"). We now mark the middle window with our randomly generated mark.
308 Afterwards, we switch focus away from the middle window to be able to tell if
309 focusing it via its mark will work. If the test works, the goto command seems
312 .t/11-goto.t: Test corner case
313 ----------------------
314 # check that we can specify multiple criteria
316 $focus = focus_after('focus left');
317 is($focus, $top->id, "Top window focused");
319 $focus = focus_after(qq|[con_mark="$random_mark" con_mark="$random_mark"] focus|);
320 is($focus, $mid->id, "goto worked");
321 ----------------------
323 Now we test the same feature, but specifying the mark twice in the command.
324 This should have no effect, but let’s be sure: test it and see if things go
327 .t/11-goto.t: Test second code path
328 ----------------------
329 #####################################################################
330 # Check whether the focus command will switch to a different
331 # workspace if necessary
332 #####################################################################
334 my $tmp2 = fresh_workspace;
336 is(focused_ws(), $tmp2, 'tmp2 now focused');
338 cmd qq|[con_mark="$random_mark"] focus|;
340 is(focused_ws(), $tmp, 'tmp now focused');
341 ----------------------
343 This part of the test checks that focusing windows by mark works across
344 workspaces. It uses i3test's +focused_ws+ function to get the current
347 .t/11-goto.t: Test second code path
348 ----------------------
350 ----------------------
352 The end of every testcase has to contain the +done_testing+ line. This tells
353 +complete-run.pl+ that the test was finished successfully. If it does not
354 occur, the test might have crashed during execution -- some of the reasons why
355 that could happen are bugs in the used modules, bugs in the testcase itself or
356 an i3 crash resulting in the testcase being unable to communicate with i3 via
360 == Appendix A: The i3 sync protocol
362 Consider the following situation: You open two windows in your testcase, then
363 you use +focus left+ and want to verify that the X11 focus has been updated
364 properly. Sounds simple, right? Let’s assume you use this straight-forward
367 .Racey focus testcase
369 my $left = open_window($x);
370 my $right = open_window($x);
372 is($x->input_focus, $left->id, 'left window focused');
375 However, the test fails. Sometimes. Apparantly, there is a race condition in
376 your test. If you think about it, this is because you are using two different
377 pieces of software: You tell i3 to update focus, i3 confirms that, and then you
378 ask X11 to give you the current focus. There is a certain time i3 needs to
379 update the X11 state. If the testcase gets CPU time before X11 processed i3's
380 requests, the test will fail.
382 image::i3-sync.png["Diagram of the race condition", title="Diagram of the race condition"]
384 One way to "solve" this would be to add +sleep 0.5;+ after the +cmd+ call.
385 After 0.5 seconds it should be safe to assume that focus has been updated,
388 In practice, this usually works. However, it has several problems:
390 1. This is obviously not a clean solution, but a workaround. Ugly.
391 2. On very slow machines, this might not work. Unlikely, but in different
392 situations (a delay to wait for i3 to startup) the necessary time is much
393 harder to guess, even for fast machines.
394 3. This *wastes a lot of time*. Usually, your computer is much faster than 0.5s
395 to update the status. However, sometimes, it might take 0.4s, so we can’t
398 To illustrate how grave the problem with wasting time actually is: Before
399 removing all sleeps from the testsuite, a typical run using 4 separate X
400 servers took around 50 seconds on my machine. After removing all the sleeps,
401 we achieved times of about 25 seconds. This is very significant and influences
402 the way you think about tests -- the faster they are, the more likely you are
403 to check whether everything still works quite often (which you should).
405 What I am trying to say is: Delays adds up quickly and make the test suite
408 The real solution for this problem is a mechanism which I call "the i3 sync
409 protocol". The idea is to send a request (which does not modify state) via X11
410 to i3 which will then be answered. Due to the request's position in the event
411 queue (*after* all previous events), you can be sure that by the time you
412 receive the reply, all other events have been dealt with by i3 (and, more
415 image::i3-sync-working.png["Diagram of the i3 sync solution", title="Diagram of the i3 sync solution"]
417 === Implementation details
419 The client which wants to sync with i3 initiates the protocol by sending a
420 ClientMessage to the X11 root window:
424 # Generate a ClientMessage, see xcb_client_message_t
425 my $msg = pack "CCSLLLLLLL",
426 CLIENT_MESSAGE, # response_type
429 $root, # destination window
430 $x->atom(name => 'I3_SYNC')->id,
432 $_sync_window->id, # data[0]: our own window id
433 $myrnd, # data[1]: a random value to identify the request
438 # Send it to the root window -- since i3 uses the SubstructureRedirect
439 # event mask, it will get the ClientMessage.
440 $x->send_event(0, $root, EVENT_MASK_SUBSTRUCTURE_REDIRECT, $msg);
443 i3 will then reply with the same ClientMessage, sent to the window specified in
444 +data[0]+. In the reply, +data[0]+ and +data[1]+ are exactly the same as in the
445 request. You should use a random value in +data[1]+ and check that you received
446 the same one when getting the reply.
448 == Appendix B: Socket activation
450 Socket activation is a mechanism which was made popular by systemd, an init
451 replacement. It basically describes creating a listening socket before starting
452 a program. systemd will invoke the program only when an actual connection to
453 the socket is made, hence the term socket activation.
455 The interesting part of this (in the i3 context) is that you can very precisely
456 detect when the program is ready (finished its initialization).
458 === Preparing the listening socket
460 +complete-run.pl+ will create a listening UNIX socket which it will then pass
461 to i3. This socket will be used by i3 as an additional IPC socket, just like
462 the one it will create on its own. Passing the socket happens implicitly
463 because children will inherit the parent’s sockets when fork()ing and sockets
464 will continue to exist after an exec() call (unless CLOEXEC is set of course).
466 The only explicit things +complete-run.pl+ has to do is setting the +LISTEN_FDS+
467 environment variable to the number of sockets which exist (1 in our case) and
468 setting the +LISTEN_PID+ environment variable to the current process ID. Both
469 variables are necessary so that the program (i3) knows how many sockets it
470 should use and if the environment variable is actually intended for it. i3 will
471 then start looking for sockets at file descriptor 3 (since 0, 1 and 2 are used
472 for stdin, stdout and stderr, respectively).
474 The actual Perl code which sets up the socket, fork()s, makes sure the socket
475 has file descriptor 3 and sets up the environment variables follows (shortened
479 .Setup socket and environment
480 -----------------------------
481 my $socket = IO::Socket::UNIX->new(
483 Local => $args{unix_socket_path},
488 $ENV{LISTEN_PID} = $$;
489 $ENV{LISTEN_FDS} = 1;
491 # Only pass file descriptors 0 (stdin), 1 (stdout),
492 # 2 (stderr) and 3 (socket) to the child.
495 # If the socket does not use file descriptor 3 by chance
496 # already, we close fd 3 and dup2() the socket to 3.
497 if (fileno($socket) != 3) {
499 POSIX::dup2(fileno($socket), 3);
504 -----------------------------
506 === Waiting for a reply
508 In the parent process, we want to know when i3 is ready to answer our IPC
509 requests and handle our windows. Therefore, after forking, we immediately close
510 the listening socket (i3 will handle this side of the socket) and connect to it
511 (remember, we are talking about a named UNIX socket) as a client. This connect
512 call will immediately succeed because the kernel buffers it. Then, we send a
513 request (of type GET_TREE, but that is not really relevant). Writing data to
514 the socket will also succeed immediately because, again, the kernel buffers it
515 (only up to a certain amount of data of course).
517 Afterwards, we just blockingly wait until we get an answer. In the child
518 process, i3 will setup the listening socket in its event loop. Immediately
519 after actually starting the event loop, it will notice a new client connecting
520 (the parent process) and handle its request. Since all initialization has been
521 completed successfully by the time the event loop is entered, we can now assume
524 === Timing and conclusion
526 A beautiful feature of this mechanism is that it does not depend on timing. It
527 does not matter when the child process gets CPU time or when the parent process
528 gets CPU time. On heavily loaded machines (or machines with multiple CPUs,
529 cores or unreliable schedulers), this makes waiting for i3 much more robust.
531 Before using socket activation, we typically used a +sleep(1)+ and hoped that
532 i3 was initialized by that time. Of course, this breaks on some (slow)
533 computers and wastes a lot of time on faster computers. By using socket
534 activation, we decreased the total amount of time necessary to run all tests
535 (72 files at the time of writing) from > 100 seconds to 16 seconds. This makes
536 it significantly more attractive to run the test suite more often (or at all)
539 An alternative approach to using socket activation is polling for the existance
540 of the IPC socket and connecting to it. While this might be slightly easier to
541 implement, it wastes CPU time and is considerably uglier than this solution
542 :). After all, +lib/SocketActivation.pm+ contains only 54 SLOC.