Delete Tentative "Required Context Role" Tests

[cookiecrook]

Spawning an issue tracker so this action doesn't fall through the cracks.

Consider Deleting Tentative Required Context Role Tests

• Valerie: “We should maybe discuss removing these tests, since we landed the PR to no longer dictate what browsers should do?” The PR: w3c/aria#2284
• Implementation concerns over Orphaned Role requirements in Core-AAM, HTML-AAM… and possibly elsewhere. w3c/aria#2166 (comment)
• Tentative vs Optional test documentation https://web-platform-tests.org/writing-tests/file-names.html\#test-features
• Jocelyn implemented part of this in Chrome in July 2024 (see pass results in History) but rolled the change back when the ARIA group changed the spec with PR w3c/aria#2284
• Conformance checker tests in wpt: https://github.com/web-platform-tests/wpt/tree/master/conformance-checkers
• Decisions:
  • @zcorpan will talk to @jcsteh about rolling back not exposing orphaned roles
  • When rolled back we will remove tests and add comment to closed issue referencing the commit in case anyone wants to revive this someday.

FYI @spectranaut who raised the original comment for yesterday's meeting.

[jcsteh]

@zcorpan will talk to @jcsteh about rolling back not exposing orphaned roles

I'm happy to roll this back, but unfortunately, it's not as trivial as it sounds. That area of the code has evolved significantly since that landed, so we're going to need to manually unpick the threads rather than just doing a straightforward revert. In other words, it may take us some time to get to this.

When rolled back we will remove tests

Do we need to wait to remove the tests until this is rolled back in Gecko? The spec says:

For most roles, user agents can either recover the error by ignoring the role

So we're permitted to ignore the role, even if other browsers don't. We just don't have interop here, but as written, the spec doesn't require interop here.

I'm also okay with changing the spec further to state that user agents (browsers, not AT) should not ignore the role even though it's an author error, in which case we probably should wait for the Gecko revert before removing the tests.

[jcsteh]

https://bugzilla.mozilla.org/show_bug.cgi?id=2003965

[cookiecrook]

Do we need to wait to remove the tests until this is rolled back in Gecko?

Okay, we can either remove them now.

The old language was:

When an element has a role but is not contained in the required context (for example, an orphaned listitem without the required accessible parent of role list), User Agents MUST ignore the role token, and return the computedrole as if the ignored role token had not been included.

The new language is:

When an element has a role but is not contained in the required context (for example, an orphaned listitem without the required accessible parent of role list), this is an authoring error, but the user agent behavior is not specified as a single rule. For most roles, user agents can either recover the error by ignoring the role, or respect the author's intended role in scenarios deemed by the implementation to be harmless. Please note that this permissiveness in how engines treat author role errors might be overridden in a language-specific mapping document such as [[HTML-AAM]].

So I agree with your assessment that we don't need wait on anything to delete these. The behavior is undefined, so Gecko doesn't actually need to change anything. We just need to remove the tests.

[jcsteh]

Personally, I'd prefer that this wasn't undefined behaviour and that we instead specified what user agents (browsers) should do in this case, but still note that it's an authoring error and thus a spec conformance failure. This is one of those cases where I think it's reasonable for AT to behave differently, but not for browsers. But that's not a hill I'm willing to die on.

Regardless, I've filed the Gecko bug to get this reverted when we can get to it, as I still think that's the right thing to do. It just doesn't need to block removal of the tests.

[cookiecrook]

Personally, I'd prefer that this wasn't undefined behaviour and that we instead specified what user agents (browsers) should do in this case, but still note that it's an authoring error and thus a spec conformance failure.

What if the expected UA/browser behavior was, "Dev tools should throw a console log warning."

[jcsteh]

That seems fine, but I still think we should be clear in the spec about whether browsers expose the role or ignore it. We seem to have settled on browsers exposing the role despite the author error; that will be interoperable across all three major browser engines once we revert Gecko's orphaned role implementation.

[cookiecrook]

We seem to have settled on browsers exposing the role despite the author error; that will be interoperable across all three major browser engines once we revert Gecko's orphaned role implementation.

I don't have any confidence that there is interop here, or that all roles are exposed in every situation for any implementation.