Looking at chromium code it seems we should still try to focus the control if the label is not focusable itself, or such: https://source.chromium.org/chromium/chromium/src/+/master:third_party/blink/renderer/core/html/forms/html_label_element.cc;l=234;drc=ee4cff87f02e46e1fbbdaef0aa123e05761b35e8. WebKit has similar code.

So we should probably keep the override but do something like:

{
  nsIFrame* frame = GetPrimaryFrame(FlushType::Frames));
  if (frame && frame->IsFocusable()) {
    return nsGenericHTMLElement::Focus(aOptions, aCallerType, aError);
  }
}

if (RefPtr<Element> elem = GetLabeledElement()) {
  return elem->Focus(aOptions, aCallerType, aError);
}

Or such.

I pushed this to try on behalf of Mu-an. Hopefully there are some pre-existing tests that start failing with this change. https://treeherder.mozilla.org/#/jobs?repo=try&revision=7513d59996433b09d6a38e60bdcc8fa4b4e0ec3e

Happy to help pushing an updated patch or so on.

@emilio We applied your suggestion and added WPT tests for the expected focus forwarding behavior. The test for an empty tabindex attribute currently fails in Firefox and passes in Chrome.

Can you push this branch again to Try? Thanks for your help!

In D67676#2065527, @dgraham wrote:

Can you push this branch again to Try?

Here is the try push: https://treeherder.mozilla.org/#/jobs?repo=try&revision=559e16cb645bc29be952e67575dab8290411133e

Thanks @edgar!

That looks pretty good. The only thing that's making things orange and are not flaky is this test you're adding (the empty tabindex subtest).

Per the discussion in matrix, it seems this is an expected failure and Firefox shouldn't consider empty tabindex as focusable. If so what needs to get done is to annotate this test as failing, pointing to a bug.

This would be a matter of creating a file called testing/web-platform/meta/html/semantics/forms/the-label-element/forward-focus-to-associated-element.html.ini with something like this as contents:

[forward-focus-to-associated-element.html]
  [focusing a label with empty tabindex should forward focus to the labelable element]
    expected: FAIL
    bug: <link-to-relevant-bug>

That should make the test pass locally when you do ./mach wpt testing/web-platform/tests/html/semantics/forms/the-label-element/forward-focus-to-associated-element.html (and also on automation).

With that I think this should be good to land, though I suggested the implementation so @edgar should probably be the one to stamp it if it looks good to him :)

We added the expected test failure file and the label tests pass locally. According to the HTML spec, invalid tabindex values—non-integer or empty—fall back to platform focus behavior, and labels are not focusable.

The user agent should follow platform conventions to determine if the element should be considered as a focusable area

—https://html.spec.whatwg.org/#the-tabindex-attribute

We found this report describing the tabindex focus issue: https://bugzilla.mozilla.org/show_bug.cgi?id=1128054.

Thanks again for all of your help getting this branch merged.

Thank you! Removing mats from the reviewer list because his feedback was addressed.

Do you have commit access? If not, you can add the "check-in needed" tag, and it'll get pushed for you, or I can push it too if you want.

Also the commit message should be adjusted as we're no longer just removing it.

Check-in Needed handled, landing queued.