Bug 131881

Summary: REGRESSION (r167530): fast/dom/gc-attribute-node.html is failing on Mac
Product: WebKit Reporter: Timothy Hatcher <timothy>
Component: DOMAssignee: Nobody <webkit-unassigned>
Severity: Normal CC: ap, fpizlo, ggaren, mhahnenberg, oliver
Priority: P2 Keywords: LayoutTestFailure, PlatformOnly
Version: 528+ (Nightly build)   
Hardware: Unspecified   
OS: Unspecified   

Description Timothy Hatcher 2014-04-18 21:57:06 PDT
It has been failing on Windows (bug 127480).
Comment 1 Timothy Hatcher 2014-04-18 23:06:05 PDT
Marked as flaky in https://trac.webkit.org/r167531.
Comment 2 Alexey Proskuryakov 2014-04-18 23:12:35 PDT
Could you please post a diff? Looking at flakiness dashboard, I don't see this test having ever failed.
Comment 3 Timothy Hatcher 2014-04-18 23:14:33 PDT
It started flaking with my patch in https://trac.webkit.org/r167530. Not really sure why. My change is unrelated to the tested code.
Comment 4 Alexey Proskuryakov 2014-04-20 15:41:06 PDT
Could be memory corruption from
Comment 5 Timothy Hatcher 2014-04-20 21:05:51 PDT
Comment 6 Alexey Proskuryakov 2014-04-20 22:59:01 PDT
I wanted to conjecture memory corruption from a previous test, but that's not accurate.

Turns out that custom attributes on attribute nodes are broken in JSC. We used to know that and to expect a failure, until Filip's patch <http://trac.webkit.org/changeset/159705> slightly changed memory layout, so the test started to pass accidentally. He then removed the failing expectation.

But there was no real fix, and now your patch slightly changed memory layout again, making the test fail again.

*** This bug has been marked as a duplicate of bug 88045 ***