Weston test suite

Weston test suite aims to test features of the Weston compositor and libweston. The automatic tests are executed as part of meson test and in the Gitlab CI. In addition to automatic tests, there are few manual tests that have not been automated, but being manual means they are also not routinely (or ever) executed.

Test execution

The test execution hierarchy is:

  • meson test

    • a test program

      • a fixture setup

        • a test

          • a sub-test from a data array

When meson test is executed, it will run all defined test programs potentially in parallel and collect their exit status. Therefore it is important to design each test program to be executable in parallel with every other test program.

A test program is essentially one .c source code file that is built into one executable file (not a library, module, or plugin). Each test program is possible to run manually without Meson straight from the build directory without any environment or command line setup, e.g. with GDB or Valgrind.

A test program may define one fixture setup function. The function may be defined alone or with a data array of an arbitrary data type. If an array is defined, the fixture setup will be called and all the tests in the program executed for each element in the array serially. Fixture setups are used for setting up the Weston compositor for the tests that need it. The array is useful for running the compositor with different settings for the same tests, e.g. with Pixman-renderer and GL-renderer.

A test in a test program is defined with one of the macros TEST(), TEST_P(), or PLUGIN_TEST(). TEST() defines a single test with no sub-tests. TEST_P() defines a data-driven array of tests: a set of sub-tests. PLUGIN_TEST() is used specifically by plugin tests that require access to weston_compositor.

All tests and sub-tests are executed serially in a test program. The test harness does not fork() which means that any test that crashes or hits an assert failure will quit the whole test program on the spot, leaving following tests in that program not executed.

The test suite has no tests that are expected to fail in general. All tests that test for a failure must check the exact error condition expected and succeed if it is met or fail for any other or no error.

Types of tests

Aside from manual vs. automatic, there are three types of tests:

Standalone tests

Standalone tests do not launch the full compositor.

Plugin tests

Plugin tests launch the Weston compositor and execute the list of tests from an idle callback handler in the compositor context, blocking the compositor while they run.

Client tests

Client tests launch the Weston compositor and execute the list of tests in a new thread that is created from an idle callback handler. This means the compositor runs independently from the tests and one can write a test like as a normal Wayland client.

The type of all the tests in a test program is defined by the fixture setup function. A fixture setup function is any defined function with a specific signature and registered with either DECLARE_FIXTURE_SETUP() or DECLARE_FIXTURE_SETUP_WITH_ARG().

Standalone tests

Standalone tests do not have a fixture setup function defined in the test program or the fixture setup function calls weston_test_harness_execute_standalone() explicitly. All test cases must be defined with TEST() or TEST_P().

This is the simplest possible test example:

TEST(always_success)
{
     /* true */
}

Plugin tests

Plugin tests must have a fixture setup function that calls weston_test_harness_execute_as_plugin(). All test cases must be defined with PLUGIN_TEST() which declares an implicit function argument weston_compositor *compositor.

The compositor fixture manufactures the necessary environment variables and the command line argument array to launch Weston, and calls wet_main() directly. An idle task handler is registered, which gets invoked when initialization is done. All tests are executed from that idle handler, and then the compositor exits.

This is an example of a plugin test that just logs a line:

static enum test_result_code
fixture_setup(struct weston_test_harness *harness)
{
     struct compositor_setup setup;

     compositor_setup_defaults(&setup);

     return weston_test_harness_execute_as_plugin(harness, &setup);
}
DECLARE_FIXTURE_SETUP(fixture_setup);

PLUGIN_TEST(plugin_registry_test)
{
     /* struct weston_compositor *compositor; */
     testlog("Got compositor %p\n", compositor);
}

Client tests

Plugin tests must have a fixture setup function that calls weston_test_harness_execute_as_client(). All test cases must be defined with TEST() or TEST_P().

The compositor fixture manufactures the necessary environment variables and the command line argument array to launch Weston, and calls wet_main() directly. An idle task handler is registered, which gets invoked when initialization is done. The idle handler creates a new thread and returns. The new thread will execute all tests and then signal the compositor to exit.

This is an incomplete example of an array of sub-tests and another test as clients:

static enum test_result_code
fixture_setup(struct weston_test_harness *harness)
{
     struct compositor_setup setup;

     compositor_setup_defaults(&setup);

     return weston_test_harness_execute_as_client(harness, &setup);
}
DECLARE_FIXTURE_SETUP(fixture_setup);

struct bad_source_rect_args {
     int x, y, w, h;
};

static const struct bad_source_rect_args bad_source_rect_args[] = {
     { -5,  0,  20,  10 },
     {  0, -5,  20,  10 },
     {  5,  6,   0,  10 },
     {  5,  6,  20,   0 },
     {  5,  6, -20,  10 },
     {  5,  6,  20, -10 },
     { -1, -1,  20,  10 },
     {  5,  6,  -1,  -1 },
};

TEST_P(test_viewporter_bad_source_rect, bad_source_rect_args)
{
     const struct bad_source_rect_args *args = data;
     struct client *client;
     struct wp_viewport *vp;

     client = create_client_and_test_surface(100, 50, 123, 77);

     vp = create_viewport(client);

     testlog("wp_viewport.set_source x=%d, y=%d, w=%d, h=%d\n",
             args->x, args->y, args->w, args->h);
     set_source(vp, args->x, args->y, args->w, args->h);

     expect_protocol_error(client, &wp_viewport_interface,
                           WP_VIEWPORT_ERROR_BAD_VALUE);
}

TEST(test_roundtrip)
{
     struct client *client;

     client = create_client_and_test_surface(100, 50, 123, 77);
     client_roundtrip(client);
}

Writing tests

Test programs do not have a main() of their own. They all share the main() from the test harness and only define test cases and a fixture setup.

It is recommended to have one test program (one .c file) contain only one type of tests to keep the fixture setup simple. See Standalone tests, Plugin tests and Client tests how to set up each type in a test program.

Note

TODO: Currently it is not possible to gracefully skip or fail a test. You can skip with exit(RESULT_SKIP) but that will quit the whole test program and all defined tests that were not ran yet will be counted as failed. You can fail a test by any means, e.g. exit(RESULT_FAIL), but the same caveat applies. Succeeded tests must simply return and not call any exit function.