review unlabeled section vs explicit region role

[cookiecrook]

@jcsteh wrote:

<nav role="region" data-testname="region without label" data-expectedrole="navigation" class="ex">x</nav>

This is an interesting case. In Gecko, computedRole currently returns "region" here. While it's clear from the HTML AAM spec that <section> without a label shouldn't return "region", I'm not sure where it's specified that role="region" without a label shouldn't map to "region". Core AAM says that region without a label shouldn't be exposed as a landmark to individual platforms, but it says nothing about the fact that it shouldn't be exposed as a "region" as far as ARIA (not the platform APIs) are concerned. You can intuit the latter from the former, but unless I'm missing something, there's no spec concept which allows a role to be completely ignored/invalidated based on a missing name. There's "Name required true", but that's specified for button as well, and I'm not sure we'd want to completely ignore the role for an unlabelled button.

Originally posted by @jcsteh in web-platform-tests/wpt#38925 (comment)

[cookiecrook]

Source: wai-aria/role/region-roles.html

Current results: https://wpt.fyi/results/wai-aria/role/region-roles.html?label=experimental&label=master&aligned

WebKit and Chromium pass both tests as written.

[cookiecrook]

@scottaohara Can you help us determine if either:

• there's a relevant portion of a spec indicating Gecko has a bug, or
• the test is incorrect and both WebKit and Chromium have the same bug.

[cookiecrook]

The ARIA spec says about regions:

User agents SHOULD treat elements with role region and have an accessible name as navigational landmarks. (Typo addressed in ARIA #1919)

But to @jcsteh's point, that's not the same as "User Agents MUST NOT expose the region role on elements without a label." So where is the disconnect?

[cookiecrook]

Ah, here it is, @jcsteh:

Certain landmark roles require names from authors. In situations where an author has not specified names for these landmarks, it is considered an authoring error. The user agent MUST treat such elements as if no role had been provided. If a valid fallback role had been specified, or if the element had an implicit ARIA role, then user agents would continue to expose that role, instead. Instances of such roles are as follows:

• form
• region

Citation: https://w3c.github.io/aria/#document-handling_author-errors_roles

I'd like to close this issue, but I will wait til you confirm and file the related Gecko bug.

[jcsteh]

Aha. Thank you.

I already assumed this was something we wanted in the spec, even if it wasn't clear yet, so I already have patches up for review to fix this in Gecko. Those patches are in https://bugzilla.mozilla.org/show_bug.cgi?id=1822112 and friends.

Please feel free to close this.