gecko-dev/testing/web-platform
James Graham 983c29daef Bug 1703921 - Update wpt metadata r=twisniewski
Differential Revision: https://phabricator.services.mozilla.com/D111324
2021-04-10 20:04:49 +00:00
..
meta Bug 1703921 - Update wpt metadata r=twisniewski 2021-04-10 20:04:49 +00:00
mozilla Bug 1694238- correct expectation r=aryx DONTBUILD 2021-03-31 15:33:38 +00:00
outbound
products
tests Bug 1703921 - Ensure h2 server terminates threads when connections are closed r=twisniewski 2021-04-10 20:04:49 +00:00
update
README.md
fissionregressions.py Bug 1695972 - Add a mach command for finding fission-specific regressions in wpt, r=kashav,jmaher 2021-03-02 21:19:00 +00:00
mach_commands.py Bug 1695972 - Add a mach command for finding fission-specific regressions in wpt, r=kashav,jmaher 2021-03-02 21:19:00 +00:00
mach_commands_base.py
mach_test_package_commands.py
manifestdownload.py
manifestupdate.py
metamerge.py Bug 1639003 - Convert mach wpt-metadata-merge to Python 3, r=ahal 2021-03-11 15:58:51 +00:00
metasummary.py
moz.build Bug 1701037 - map 'encoding-detection' web-platform-test folder to 'Core :: Internationalization' component in Bugzilla. r=hsivonen DONTBUILD 2021-03-26 09:03:05 +00:00
python.ini
runtests.py
test_metamerge.py
unittestrunner.py
vcs.py
wptrunner.ini

README.md

web-platform-tests

This directory contains the W3C web-platform-tests. They can be run using mach:

mach wpt

To run only certain tests, pass these as additional arguments to the command. For example to include all tests in the dom directory:

mach wpt testing/web-platform/tests/dom

Tests may also be passed by id; this is the path plus any query or fragment part of a url and is suitable for copying directly from logs e.g. on treeherder:

mach wpt /web-nfc/idlharness.https.window.html

A single file can produce multiple tests, so passing test ids rather than paths is sometimes necessary to run exactly one test.

The testsuite contains a mix of various test types including javascript (testharness) tests, reftests and wdspec tests. To limit the type of tests that get run, use --test-type=<type> e.g. --test-type=reftest for reftests.

Note that if only a single testharness test is run the browser will stay open by default (matching the behaviour of mochitest). To prevent this pass --no-pause-after-test to mach wpt.

Tests can be run in headless mode using the --headless command line argument.

Running in Android (GeckoView)

You can run the tests against a Gecko-based browser (GeckoView) on an Android emulator. As shown below, to do so you must start an emulator, build Firefox for Android and then run mach wpt with the org.mozilla.geckoview.test package. The package will be installed interactively by mach and tests will run against TestRunnerActivity.

./mach android-emulator --version x86-7.0
./mach build
./mach wpt --package=org.mozilla.geckoview.test

FAQ

  • I fixed a bug and some tests have started to pass. How do I fix the UNEXPECTED-PASS messages when web-platform-tests is run?

    You need to update the expectation data for those tests. See the section on expectations below.

  • I want to write some new tests for the web-platform-tests testsuite. How do I do that?

    See the section on writing tests below. You can commit the tests directly to the Mozilla repository under testing/web-platform/tests and they will be automatically upstreamed when the patch lands. For this reason please ensure that any tests you write are testing correct-per-spec behaviour even if we don't yet pass, get proper review, and have a commit message that makes sense outside of the Mozilla context. If you are writing tests that should not be upstreamed yet for some reason they must be located under testing/web-platform/mozilla/tests.

  • How do I write a test that requires the use of a Mozilla-specific feature?

    Tests in the mozilla/tests/ directory use the same harness but are not synced with any upstream. Be aware that these tests run on the server with a /_mozilla/ prefix to their URLs.

  • How do I write tests that require user interaction or other features not accessible from content js?

    For testharness tests this is possible using the testdriver API.

    For Gecko-specific testharness tests, the specialPowers extension is available. Note that this should only be used when no other approach works; such tests can't be shared with other browsers. If you're using specialPowers for something that could be tested in other browsers if we extended testdriver or added test-only APIs, please file a bug.

Writing tests

Documentation for writing tests, and everything else that isn't specific to the gecko integration can be found at https://web-platform-tests.org.

Directories

tests/ contains the tests themselves. This is a copy of a certain revision of web-platform-tests. Any patches modifying this directory will be upstreamed next time the tests are imported.

meta/ contains Gecko-specific expectation data. This is explained in the following section.

mozilla/tests contains tests that will not be upstreamed and may make use of Mozilla-specific features.

mozilla/meta contains metadata for the Mozilla-specific tests.

Expectation Data

With the tests coming from upstream, it is not guaranteed that they all pass in Gecko-based browsers. For this reason it is necessary to provide metadata about the expected results of each test. This is provided in a set of manifest files in the meta/ subdirectories.

There is one manifest file per test with "non-default" expectations. By default tests are expected to PASS, and tests with subtests are expected to have an overall status of OK. The manifest file of a test has the same path as the test file but under the meta directory rather than the tests directory and has the suffix .ini.

The format of these files is similar to ini files, but with a couple of important differences; sections can be nested using indentation, and only : is permitted as a key-value separator. For example the expectation file for a test with one failing subtest and one erroring subtest might look like:

[filename.html]
    [Subtest name for failing test]
        expected: FAIL

    [Subtest name for erroring test]
        expected: ERROR

Expectations can also be made platform-specific using a simple python-like conditional syntax e.g. for a test that times out on linux but otherwise fails:

[filename.html]
    expected:
        if os == "linux": TIMEOUT
        FAIL

The available variables for the conditions are those provided by mozinfo.

Tests that are intermittent may be marked with multiple statuses using a list of possibilities e.g. for a test that usually passes, but intermittently fails:

[filename.html]
    [Subtest name for intermittent test]
        expected: [PASS, FAIL]

For more information on manifest files, see the wptrunner documentation.

Autogenerating Expectation Data

After changing some code it may be necessary to update the expectation data for the relevant tests. This can of course be done manually, but tools are available to automate much of the process.

First one must run the tests that have changed status, and save the raw log output to a file:

mach wpt /url/of/test.html --log-wptreport new_results.json

Then the wpt-update command may be run using this log data to update the expectation files:

mach wpt-update new_results.json

Disabling Tests

Tests are disabled using the same manifest files used to set expectation values. For example, if a test is unstable on Windows, it can be disabled using an ini file with the contents:

[filename.html]
    disabled:
        if os == "win": https://bugzilla.mozilla.org/show_bug.cgi?id=1234567

For intermittents it's generally preferable to give the test multiple expectations rather than disable it.

Fuzzy Reftests

Reftests where the test doesn't exactly match the reference can be marked as fuzzy. If the difference is inherent to the test, it should be encoded in a meta element, but where it's a Gecko-specific difference it can be added to the metadata file, using the same syntax e.g.

[filename.html]
    fuzzy: maxDifference=10-15;totalPixels=200-300

In this case we specify that we expect between 200 and 300 pixels, inclusive, to be different, and the maximum difference in any colour channel to be between 10 and 15.

Enabling Prefs

Some tests require specific prefs to be enabled before running. These prefs can be set in the expectation data using a prefs key with a comma-seperate list of pref.name:value items to set e.g.

[filename.html]
    prefs: [dom.serviceWorkers.enabled:true,
            dom.serviceWorkers.exemptFromPerDomainMax:true,
            dom.caches.enabled:true]

Disabling Leak Checks

When a test is imported that leaks, it may be necessary to temporarily disable leak checking for that test in order to allow the import to proceed. This works in basically the same way as disabling a test, but with the key 'leaks' e.g.

[filename.html]
    leaks:
        if os == "linux": https://bugzilla.mozilla.org/show_bug.cgi?id=1234567

Setting per-Directory Metadata

Occasionally it is useful to set metadata for an entire directory of tests e.g. to disable then all, or to enable prefs for every test. In that case it is possible to create a __dir__.ini file in the metadata directory corresponding to the tests for which you want to set this metadata e.g. to disable all the tests in tests/feature/unsupported/, one might create meta/feature/unsupported/__dir__.ini with the contents:

disabled: Feature is unsupported

Settings set in this way are inherited into subdirectories. It is possible to unset a value that has been set in a parent using the special token @Reset (usually used with prefs), or to force a value to true or false using @True and @False. For example to enable the tests in meta/feature/unsupported/subfeature-supported one might create an ini file meta/feature/unsupported/subfeature-supported/__dir__.ini like:

disabled: @False

Setting Metadata for Release Branches

Run info properties can be used to set metadata for release branches that differs from nightly (e.g. for when a feature depends on prefs that are only set on nightly), for example:

[filename.html]
  expected:
    if release_or_beta: FAIL

Note that in general the automatic metadata update will work better if the nonstandard configuration is used explicitly in the conditional, and placed at the top of the set of conditions, i.e. the following would cause problems later:

[filename.html]
  expected:
    if nightly_build: PASS
    FAIL

This is because on import the automatic metadata updates are run against the results of nightly builds, and we remove any existing conditions that match that configuration to avoid building up stale configuration options.

Test Manifest

web-platform-tests use a large auto-generated JSON file as their manifest. This stores data about the type of tests, their references, if any, and their timeout, gathered by inspecting the filenames and the contents of the test files. It it not necessary to manually add new tests to the manifest; it is automatically kept up to date when running mach wpt.

Running Tests In Other Browsers

web-platform-tests is cross browser, and the runner is compatible with multiple browsers. Therefore it's possible to check the behaviour of tests in other browsers. By default Chrome, Edge and Servo are supported. In order to run the tests in these browsers use the --product argument to wptrunner:

mach wpt --product chrome dom/historical.html

By default these browsers run without expectation metadata, but it can be added in the testing/web-platform/products/<product> directory. To run with the same metadata as for Firefox (so that differences are reported as unexpected results), pass --meta testing/web-platform/meta to the mach command.

Results from the upstream CI for many browser, including Chrome and Safari, are available on wpt.fyi. There is also a gecko dashboard which by default shows tests that are failing in Gecko but not in Chrome and Safari, organised by component, based on the wpt.fyi data.