In simple cases individual tests can be run by simply loading the page in a browser window. For running larger groups of tests, or running tests frequently, this is not a practical approach and several better options exist.

From the Command Line

The simplest way to run tests is to use the wpt run command from the root of the repository. This will automatically load the tests in the chosen browser, and extract the test results. For example to run the dom/historical.html tests in a local copy of Chrome:

./wpt run chrome dom/historical.html

Or to run in a specified copy of Firefox:

./wpt run --binary ~/local/firefox/firefox firefox dom/historical.html

On Windows edge is also supported, and if you have a sauce labs account, any browser can be run using product sauce. For details on how to pass information to sauce, including credentials see ./wpt run --help. That also details a large number of other options for customising the test run.

From Inside a Browser

Tests that have been merged on GitHub are mirrored at [http://w3c-test.org/].

For running multiple tests inside a browser, there is a test runner located at /tools/runner/index.html.

This allows all the tests, or those matching a specific prefix (e.g. all tests under /dom/) to be run. For testharness.js tests, the results will be automatically collected, while the runner provides a simple UI for manually comparing reftest rendering and running manual tests.

Note, however, it does not currently handle more complex reftests with more than one reference involved.

Because it runs entirely in-browser, this runner cannot deal with edge-cases like tests that cause the browser to crash or hang.

Writing Your Own Runner

Most test runners have two stages: finding all tests, followed by executing them (or a subset thereof).

To find all tests in the repository, it is strongly recommended to use the included wpt manifest tool: the required behaviors are more complex than what are documented (especially when it comes to precedence of the various possibilities and some undocumented legacy ways to define test types), and hence its behavior should be considered the canonical definition of how to enumerate tests and find their type in the repository.

For test execution, please read the documentation for the various test types very carefully and then check your understanding on the mailing list or IRC (webclient, join channel #testing). It’s possible edge-case behavior isn’t properly documented!