Interoperability between browsers is critical to Chromium's mission of improving the web. We believe that leveraging and contributing to a shared test suite is one of the most important tools in achieving interoperability between browsers. The web-platform-tests repository is the primary shared test suite where all browser engines are collaborating.
Chromium has a 2-way import/export process with the upstream web-platform-tests repository, where tests are imported into web_tests/external/wpt and any changes to the imported tests are also exported to web-platform-tests.
See https://web-platform-tests.org/ for general documentation on web-platform-tests, including tips for writing and reviewing tests.
[TOC]
To contribute changes to web-platform-tests, just commit your changes directly to web_tests/external/wpt and the changes will be automatically upstreamed within 24 hours.
Changes involving adding, removing or modifying tests can all be upstreamed.
Any changes outside of
external/wpt will not be
upstreamed, and any changes *-expected.txt, OWNERS, and MANIFEST.json,
will also not be upstreamed.
Running the web tests will automatically regenerate MANIFEST.json to pick up any local modifications.
Most tests are written using testharness.js, see Writing Web Tests and Web Tests Tips for general guidelines.
Tests in web-platform-tests are expected to match behavior defined by the relevant specification. In other words, all assertions that a test makes should be derived from a specification's normative requirements, and not go beyond them. It is often necessary to change the specification to clarify what is and isn't required.
When implementation experience is needed to inform the specification work, tentative tests can be appropriate. It should be apparent in context why the test is tentative and what needs to be resolved to make it non-tentative.
testdriver.js
provides a means to automate tests that cannot be written purely using web
platform APIs, similar to internals.* and eventSender.* in regular Blink
web tests.
If no testdriver.js API exists, check if it's a known issue and otherwise consider filing a new issue. For instructions on how to add a new testing API, see WPT Test Automation for Chromium
Some specs may define testing APIs (e.g. WebUSB), which may be polyfilled with internal API like MojoJS. MojoJS is only allowed in WPT for this purpose. Please reach out to [email protected] before following the process below for adding a new test-only API:
- Create a full list of
*.mojom.m.jsfiles that you need, including all dependencies. Generated modules load dependencies recursively by default, so you can check the network panel of DevTools to see the full list of dependencies it loads. - Check linux-archive-rel.json and add any
missing
*.mojom.m.jsfiles to themojojs.ziparchive. Globs are supported infilename. Do not copy Mojom bindings into WPT. - Meanwhile in Chromium, you can create a helper for your WPT tests to do
browser-specific setup using
test-only-api.js.
See
webxr_util.js
as an example. You can write tests using this helper right away, but they
will not work upstream (i.e. on https://wpt.fyi ) until your change in step
2 is included in official channels, as
mojojs.zipis built alongside with Chrome.
An alternative to the above options is to write manual tests that are automated
with scripts from
wpt_automation.
Injection of JS in manual tests is determined by loadAutomationScript in
testharnessreport.js.
Such tests still require case-by-case automation to run for other browser engines, but are more valuable than purely manual tests.
Manual tests that have no automation are still imported, but skipped in NeverFixTests; see issue 738489.
Changes made in web_tests/external/wpt are automatically exported to GitHub.
It's still possible to make direct pull requests to web-platform-tests, see https://web-platform-tests.org/writing-tests/github-intro.html.
Entirely new top-level directories should generally be added upstream, since
that's the only way to add an OWNERS file upstream. After adding a new top-level
directory upstream, you should add a line for it in W3CImportExpectations.
Adding the new directory (and W3CImportExpectations entry) in Chromium and
later adding an OWNERS file upstream also works.
It is sometimes desirable to write WPT tests that either test Chromium-specific behaviors, or that cannot yet be upstreamed to WPT (e.g. because the spec is very nascent). For these cases, we maintain a separate directory, wpt_internal that runs under the WPT testing infrastructure (e.g. uses wptserve, etc), but which is not upstreamed to WPT.
Please see the wpt_internal
README for more details.
Note: A significant downside of wpt_internal is that your tests may be
broken by upstream changes to the resources scripts (e.g. testharness.js), as
wpt_internal does not use the forked version of testharness.js used by all
other non-external/wpt tests. Use of new failure
notifications is recommended to ensure you are
notified of breakages.
Same as Blink web tests, you can use
run_web_tests.py to run any WPT test.
One thing to note is that glob patterns for WPT tests are not yet supported.
See Running WPT tests in Content Shell for debugging etc.
Anyone who can review code and tests in Chromium can also review changes in external/wpt that will be automatically upstreamed. There will be no additional review in web-platform-tests as part of the export process.
If upstream reviewers have feedback on the changes, discuss on the pull request created during export, and if necessary work on a new pull request to iterate until everyone is satisfied.
When reviewing tests, check that they match the relevant specification, which may not fully match the implementation. See also Write tests against specifications.
Chromium has a mirror of the GitHub repo and periodically imports a subset of the tests to run as part of the regular Blink web test testing process.
The goals of this process are to be able to run web-platform-tests unmodified locally just as easily as we can run the Blink tests, and ensure that we are tracking tip-of-tree in the web-platform-tests repository as closely as possible, and running as many of the tests as possible.
There is an automatic process for updating the Chromium copy of web-platform-tests. The import is done by the builder wpt-importer builder.
The easiest way to check the status of recent imports is to look at:
- Recent logs on LUCI for wpt-importer builder
- Recent CLs created by WPT Autoroller.
The import jobs will generally be green if either there was nothing to do, or a CL was successfully submitted.
For maintainers:
- The source lives in third_party/blink/tools/wpt_import.py.
- If the importer starts misbehaving, it can be disabled by landing a CL to skip the update step.
Test owners can elect to have the importer automatically file bugs against a
component when imported changes introduce failures. This includes new tests that
fail in Chromium, as well as new failures introduced to an existing test. To
opt-in to this functionality, create an DIR_METADATA file in the appropriate
external/wpt/ subdirectory that contains at least wpt.notify and
monorail.component fields. For example, external/wpt/css/css-grid/DIR_METADATA
looks like:
monorail {
component: "Blink>Layout>Grid"
}
team_email: "[email protected]"
wpt {
notify: YES
}
When a test under external/wpt/css/css-grid/ newly fails in a WPT import, the
importer will automatically file a bug against the Blink>Layout>Grid component
in crbug.com, with details of which test failed and the
output.
Note that we are considering making the notifications opt-out instead of opt-in: see https://crbug.com/845232
We control which tests are imported via a file called W3CImportExpectations, which has a list of directories to skip while importing.
In addition to the directories and tests explicitly skipped there, tests may also be skipped for a couple other reasons, e.g. if the file path is too long for Windows. To check what files are skipped in import, check the recent logs for wpt-importer builder.
If you wish to un-skip some of the directories currently skipped in
W3CImportExpectations, you can modify that file locally and commit it, and on
the next auto-import, the new tests should be imported.
If you want to import immediately (in order to try the tests out locally, etc)
you can also run wpt-import, but this is not required.
Remember your import might fail due to GitHub's limit for unauthenticated requests, so consider passing your GitHub credentials to the script.
If there are new test failures that start after an auto-import, there are several possible causes, including:
- New baselines for flaky tests were added (https://crbug.com/701234).
- Modified tests should have new results for non-Release builds but they weren't added (https://crbug.com/725160).
- New baselines were added for tests with non-deterministic test results (https://crbug.com/705125).
Because these tests are imported from the Web Platform tests, it is better to have them in the repository (and marked failing) than not, so prefer to add test expectations rather than reverting. However, if a huge number of tests are failing, please revert the CL so we can fix it manually.
If you upload a CL with any changes in third_party/blink/web_tests/external/wpt, once you add reviewers the exporter will create a provisional pull request with those changes in the upstream WPT GitHub repository. The exporter runs on wpt-exporter builder.
Once you're ready to land your CL, please follow the link posted by the bot and check the status of the required checks of the GitHub PR. If it's green, go ahead landing your CL and the exporter will automatically merge the PR.
If GitHub status is red on the PR, please try to resolve the failures before merging. If you run into any issues, or if you have a CL with WPT changes that the exporter did not pick up, please reach out to [email protected].
Additional things to note:
- CLs that change over 1000 files will not be exported.
- All PRs use the
chromium-exportlabel. - All PRs for CLs that haven't yet been landed in Chromium also use the
do not merge yetlabel. - The exporter cannot create upstream PRs for in-flight CLs with binary files (e.g. webm files). An export PR will still be made after the CL lands.
The email you commit with in Chromium will be the author of the commit on GitHub. You can add it as a secondary address on your GitHub account to link your exported commits to your GitHub profile.
If you are a Googler, you can also register your GitHub account at go/github, making it easier for other Googlers to find you.
This cannot be avoided entirely as the two repositories are independent, but should be rare with frequent imports and exports. When it does happen, manual intervention will be needed and in non-trivial cases you may be asked to help resolve the conflict.
To allow the importer to land CLs without human intervention, it utilizes the Rubber-Stamper bot to approve import CLs.
Adding the Rubber-Stamper as a reviewer is one of the last steps the importer takes, once tests have been rebaselined and the CQ passes. If the Rubber-Stamper cannot approve a CL, it will leave a comment on the CL explaining why - this will also cause the importer to go red.
There are two possibilities when the Rubber-Stamper rejects an import: either it
is a valid rejection, because the import changes code files (.py, .bat,
.sh), or it is invalid and we're missing an allowlist rule for a file the
importer is allowed to modify.
For valid rejections, it is the job of the rotation sheriff to land the CL
manually. You need to un-abandon the import, CR+1 it yourself, and CQ+2 it.
If you don't have permission to do that (e.g. are not a committer), contact
[email protected].
For invalid rejections, message [email protected] or add an exception rule yourself. This is an example CL that adds an exception rule. (Note that you need internal access to access this repository).
To pull the latest versions of the tests that are currently being imported, you can also directly invoke the wpt-import script.
That script will pull the latest version of the tests from our mirrors of the upstream repositories. If any new versions of tests are found, they will be committed locally to your local repository. You may then upload the changes.
Remember your import might fail due to GitHub's limit for unauthenticated requests, so consider passing your GitHub credentials to the script.
- The source lives in third_party/blink/tools/wpt_export.py.
- If the exporter starts misbehaving (for example, creating the same PR over and over again), put it in "dry run" mode by landing this CL.
When manually running the wpt-import and wpt-export scripts, several
requests are made to GitHub to query the status of pull requests, look for
existing exported commits etc. GitHub has a fairly
low request limit for
unauthenticated requests, so it is recommended that you let wpt-export and
wpt-import use your GitHub credentials when sending requests:
- Generate a new personal access token
- Set up your credentials by either:
- Setting the
GH_USERenvironment variable to your GitHub user name and theGH_TOKENenvironment variable to the access token you have just created or - Creating a JSON file with two keys:
GH_USER, your GitHub user name, andGH_TOKEN, the access token you have just generated. After that, pass--credentials-json <path-to-json>towpt-exportandwpt-import.
- Setting the
This section explains the way to debug web platform tests.
Please build blink_tests before running commands below.
It is explained in Running Web Tests.
The way to run web tests is explained in Running the Tests.
Assume that you are writing the test named wpt_internal/fake/foobar.html.
You may want to run only the tests and you do not want to run all tests under
wpt_internal/fake. The following command narrows down the test to only
wpt_internal/fake/foobar.html.
third_party/blink/tools/run_web_tests.py -t Default \
third_party/blink/web_tests/wpt_internal/fake/foobar.htmlDuring the debug, you may want to log what happens during the test.
You can use console.log in JavaScript to log arbitrary strings.
e.g.
console.log('fake has been executed.');
console.log('foo=' + foo);
Logs are written under $root_build_dir/layout-test-results.
If you have tested wpt_internal/fake/foobar.html, the log will be stored in
$root_build_dir/layout-test-results/wpt_internal/fake/foobar-stderr.txt.
You can change output directory with --results-directory=<output directory>.
For some test cases, you may use .headers file to set arbitrary HTTP headers. To verify what is set to headers, you can run an HTTP server used for WPT by yourself. The following command starts the HTTP server for you:
third_party/blink/tools/run_blink_wptserve.pyTo see headers returned by the server, you can use curl -v.
curl will show headers in stderr. You may want to use |& less to
see output if it is too long.
curl -v http://localhost:8081/wpt_internal/fake/foobar.html |& lessYou are able to debug the inside of Chromium with a debugger for particular WPT tests. Refer to Running web tests using the content shell for details.