Closed Bug 820471 Opened 12 years ago Closed 11 years ago

Intermittent xpconnect/tests/chrome/test_weakmaps.xul | Garbage gray cycle should be collected.

Categories

(Core :: XPCOM, defect)

Product:
Core
Component:
XPCOM
Platform:
x86
Windows 7
Type:
defect
Priority:
Not set
Severity:
normal

Tracking

()

Status:
RESOLVED WORKSFORME

People

(Reporter: emorley, Assigned: mccr8)

References

Details

(Keywords: intermittent-failure, Whiteboard: [leave open]) 

+++ This bug was initially created as a clone of Bug #749566 +++

Bug 749566 was always on OS X, and hasn't occurred for a while. This is WINNT and seemingly much more frequent, so breaking out.

Rev3 WINNT 6.1 mozilla-inbound pgo test mochitest-other on 2012-12-08 03:03:57 PST for push 33c70f306fd4

slave: talos-r3-w7-075

https://tbpl.mozilla.org/php/getParsedLog.php?id=17742813&tree=Mozilla-Inbound

{
11680 INFO TEST-START | chrome://mochitests/content/chrome/js/xpconnect/tests/chrome/test_weakmaps.xul
11681 INFO TEST-PASS | chrome://mochitests/content/chrome/js/xpconnect/tests/chrome/test_weakmaps.xul | Using window.navigator as a weak map key should produce an exception because it can't be wrapper preserved.
11682 ERROR TEST-UNEXPECTED-FAIL | chrome://mochitests/content/chrome/js/xpconnect/tests/chrome/test_weakmaps.xul | Garbage gray cycle should be collected.
}
Rev3 WINNT 5.1 mozilla-inbound pgo test mochitest-other on 2012-12-08 12:32:13 PST for push 46b7a3a221ad
slave: talos-r3-xp-033
https://tbpl.mozilla.org/php/getParsedLog.php?id=17751154&tree=Mozilla-Inbound

Rev3 WINNT 6.1 mozilla-inbound pgo test mochitest-other on 2012-12-08 12:32:10 PST for push 46b7a3a221ad
slave: talos-r3-w7-028
https://tbpl.mozilla.org/php/getParsedLog.php?id=17751181&tree=Mozilla-Inbound
Hmm. I landed some weak map related changes on 12/03, but that seems a little too old to have caused this. Bill, can you think of any weak map or conservative scanning related changes that might have landed in the last week?
Assignee: nobody → continuation
(bug 812380 landed on Dec 3)
I can't think of anything. However, I always suspected this test had problems. I can't find the bug for it now, but we've always had the problem that nonMarkingTrace in jsweakmap.cpp checks the mark bits before marking the value. This is incorrect for non-marking traces, but ever time we try to fix it, this test fails.
Retriggers on a run that has failed seem to always come back failing too. Compiler bug?

eg this push:
https://tbpl.mozilla.org/?tree=Mozilla-Inbound&rev=25d2aefdca37
(In reply to Ed Morley [UTC+0; email:edmorley@moco] from comment #9)
> Retriggers on a run that has failed seem to always come back failing too.
> Compiler bug?

That's pretty strange...
Yeah, looks like this is PGO only. Awesome. That does suggest some kind of odd compiler bug.
Actually, this could be related to bug 747066, which merged to central on the morning of 12/7.
That sounds plausible...
(In reply to Bill McCloskey (:billm) from comment #18)
> Actually, this could be related to bug 747066, which merged to central on
> the morning of 12/7.

Shall I back that out temporarily to confirm either way?
(In reply to Ed Morley [UTC+0; email:edmorley@moco] from comment #21)
> Shall I back that out temporarily to confirm either way?

I'd rather not. How often does this fail on PGO builds? Could we use a try run and retrigger the test a few times?
It's failed 27 times since 2012-12-08 03:03:57 PST, but many of those were on the same pushes (comment 9).
Too many intermittent failures - test disabled on Windows:
https://hg.mozilla.org/integration/mozilla-inbound/rev/48729e21f1d4
Whiteboard: [test disabled on Windows][leave open]
Depends on: 822297
None of these since it was disabled.
Status: NEW → RESOLVED
Closed: 11 years ago
Resolution: --- → WORKSFORME
I just reenabled this on Windows, so ni me or something if it starts showing up again.  I'm hopeful that it won't recurs, because we're using precise stack rooting now.
Whiteboard: [test disabled on Windows][leave open] → [leave open]
You need to log in before you can comment on or make changes to this bug.