[webkit-dev] WPT first test policy proposal

Tim Nguyen tim_nguyen2 at apple.com
Fri Nov 19 13:06:05 PST 2021


Hello everyone,

I would like to start a discussion on a policy to enforce WPT usage as a first choice, that would be enforced via check-webkit-style on Changelog files.

Why use WPT?

Contributing to WPT has many benefits:
interoperability/compatibility issues with WPT we write may be detected by other browser vendors and we would get faster feedback and turnaround to fix them
creates/encourages discussion in case of disagreement with other browser vendors
WebKit greatly benefits from WPT coverage, it is time to provide our own coverage to other browsers
Improves WebKit’s score generally for WPT (which has tests mostly contributed by Chromium, Firefox at the moment)

Are there reasons to not use WPT?

Common reasons for not writing WPT that have been mentioned are:
"WPTs are less pleasant to write.”
This is not true imo, the WPT harness is documented (https://web-platform-tests.org/writing-tests/index.html <https://web-platform-tests.org/writing-tests/index.html>), unlike WebKit internals, making it easier for new contributors to figure out things.

“When you actually move a regression test to WPT proper, commit history is lost, and you don't know what kind of user facing problem it's preventing any more.”
Use `<link rel=“help” href=“…”>` with a reference to the webkit bug. Ensure you actually export the WPT and merge the WPT PR as well!

There are things WPT can’t test
True (e.g. tree dumps). My proposal below covers that.

Here is my current proposal:

Every LayoutTests/ changelog adding new test files would contain:

WPT-exemption-reason:  NONE (OOPS!)

If that line is removed or no valid reason is provided, webkit-style would throw out an error. Valid reasons could be:

WPT-exemption-reason: urgency (webkit.org/b/XXX <http://webkit.org/b/XXX>)
I’m personally not in favour of this reason existing, but some have expressed interest in it, this may be used in case of urgency (security fixes etc.), where a bug must be filed to port it to WPT (alternatively the test path could be added to a new allowlist file). Eventually this reason would be removed if two way sync with the WPT repo is available.

WPT-exemption-reason: test-harness-deficiency (webkit.org/b/XXX <http://webkit.org/b/XXX>)
If the WebKit implementation of the WPT harness lacks a certain feature that should be implemented in the WPT harness, a new bug may be filed, or an existing bug may be referenced.
Maintainers of the WPT harness may watch this reason to prioritize work needed on the harness.

WPT-exemption-reason: requires-internals: NONE (OOPS!)
requires-internals should have a valid explanation of which internals are needed. (Tree dumps would be a valid reason, although usually reftests can cover those)

In addition: 

WPT-PR: https://github.com/web-platform-tests/wpt/pulls/XX <https://github.com/web-platform-tests/wpt/pulls/XX> must be present in LayoutTests/imported/w3c/Changelog, this is to ensure tests are not lost at the next import.

This would also be enforced in webkit-style, although eventually this might be replaced by a fully automated process in the future similar to Chrome/Firefox.

Any feedback is absolutely welcome!

Cheers,
Tim
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-dev/attachments/20211119/cf1eceaf/attachment.htm>


More information about the webkit-dev mailing list