Discuss tentative orphaned role test that are no longer required in the specs (and have conflicting historical WPT results for Firefox/Gecko)

[cookiecrook]

Discuss tentative orphaned role test that are no longer required in the specs (and have conflicting historical WPT results for Firefox/Gecko)

The following appear to fail for all implementations....

• https://wpt.fyi/results/wai-aria/role/grid-roles.tentative.html
• https://wpt.fyi/results/wai-aria/role/list-roles.tentative.html
• https://wpt.fyi/results/wai-aria/role/listbox-roles.tentative.html
• https://wpt.fyi/results/wai-aria/role/menu-roles.tentative.html
• https://wpt.fyi/results/wai-aria/role/tab-roles.tentative.html

History is that the specs used to require the orphaned role recovery, but no longer does (implementor cost didsn't outweigh the end user gain), so I was considering that we should remove all these test files. However, if you expand the test run historical data, it looks like Firefox/Gecko has been passing these tests, even though the current run is a failure?

Safari/WebKit passes the ~"orphaned table tree role" tests, and both Firefox/Gecko and Safari/WebKit have historical passes logged there.

• https://wpt.fyi/results/wai-aria/role/tree-roles.tentative.html

Adding Agenda+ to discuss with the Mozilla team, WebKit contributors, and others whether we should keep or remove these. I think WPT policy would be to remove them, but there may be a benefit to marking them optional or otherwise keeping them around for historical reasons.

[cookiecrook]

@jcsteh I know the new meeting time didn't end up any better for you in Australia, but if you have thoughts on these before the June 3rd meeting, that'd be helpful.

[jcsteh]

I would vote for removing them. The quick summary:

1. Originally, these were part of Interop 2025.
2. As such, we (reluctantly) implemented the orphaned roles part of the spec in Firefox.
3. Other implementer concerns were raised and so these became tentative. I think there was also a spec tweak.
4. Firefox kept the orphaned roles implementation until very recently, since the spec "allows" a UA to do it, even if it doesn't "require" it.
5. In March, I finally ripped it out. Aside from cleanup, the more pressing reason was that some major sites such as Google Docs rely on this and the orphaned role behaviour breaks them badly. Their reliance on this is authoring error, but pragmatically, we had to prioritise users over correctness and we were going to remove this anyway, so that just sealed the deal.

[cookiecrook]

Okay good. I agree with removing the ones where the engines are aligned. @twilco, do you have an opinion on the tentative tree role file?