Bug 127323 - platform/mac/accessibility/iframe-aria-hidden.html is flaky
Summary: platform/mac/accessibility/iframe-aria-hidden.html is flaky
Alias: None
Product: WebKit
Classification: Unclassified
Component: Accessibility (show other bugs)
Version: 528+ (Nightly build)
Hardware: Unspecified Unspecified
: P2 Normal
Assignee: Alexey Proskuryakov
Keywords: InRadar
Depends on:
Reported: 2014-01-20 16:58 PST by Alexey Proskuryakov
Modified: 2014-01-21 14:10 PST (History)
12 users (show)

See Also:

proposed patch (3.04 KB, patch)
2014-01-21 11:23 PST, Alexey Proskuryakov
no flags Details | Formatted Diff | Diff

Note You need to log in before you can comment on or make changes to this bug.
Description Alexey Proskuryakov 2014-01-20 16:58:53 PST
platform/mac/accessibility/iframe-aria-hidden.html always fails on the first attempt on WebKit2 bots, and only passes on retry:


PASS body.childrenCount is 3
PASS body.childAtIndex(0).childrenCount is 0
PASS body.childrenCount is 4
FAIL body.childAtIndex(0).childAtIndex(0).childAtIndex(0).role should be AXRole: AXWebArea. Threw exception TypeError: null is not an object (evaluating 'body.childAtIndex(0).childAtIndex(0).childAtIndex(0).role')

When running locally, I see that it always fails on second try. The second attempt will fail:

run-webkit-tests -2 platform/mac/accessibility/iframe-aria-hidden.html --repeat 2 --no-retry-failures

Looking at the test, I think that it needs to be run in window.onload, because otherwise, there is a race with loading the iframe. But when I change it to run in onload, it starts to fail every time. For some reason, AXScrollArea doesn't have any child elements.

So it seems like there are at least two issues - one that the test should be made more robust by switching to onload, and another that it shouldn't fail.
Comment 1 Alexey Proskuryakov 2014-01-20 17:01:59 PST
Marked as flaky in <http://trac.webkit.org/r162392>.

Unfortunately, this reduces test coverage, as we won't confirm that the test passes on retry - we'll just accept failure as an expected outcome. Hopefully this bug can be fixed for good, making this moot.
Comment 2 Alexey Proskuryakov 2014-01-20 17:13:05 PST
Hmm, this is a very recent regression! These changes are ones in blame list:

http://trac.webkit.org/changeset/162354 Update overlay scrollbars in single pass
http://trac.webkit.org/changeset/162355 Remove an unused member variable
http://trac.webkit.org/changeset/162356 Let RenderImage construct its RenderImageResource

I suspect that this is the scroll bars change. Antti, will you have an opportunity to look into this soon? Should we roll out for now to confirm that this change is to blame?
Comment 3 Radar WebKit Bug Importer 2014-01-20 17:28:06 PST
Comment 4 Alexey Proskuryakov 2014-01-21 10:38:10 PST
Rolling out r162354 appears to have helped. I'll update the test and un-mark it.
Comment 5 Alexey Proskuryakov 2014-01-21 11:23:38 PST
Created attachment 221765 [details]
proposed patch
Comment 6 WebKit Commit Bot 2014-01-21 14:10:16 PST
Comment on attachment 221765 [details]
proposed patch

Clearing flags on attachment: 221765

Committed r162474: <http://trac.webkit.org/changeset/162474>
Comment 7 WebKit Commit Bot 2014-01-21 14:10:20 PST
All reviewed patches have been landed.  Closing bug.