fast/dom/gc-8.html crash in alternate-thread marking code in ScriptInterpreter

• Component: New Bugs
• Status: RESOLVED FIXED

[Brady Eidson]

While working on a large patch, I started to see fast/dom/gc-8.html fail with some regularity.  Talking with a few other members of the WebKit team I've learn that others have seen it anywhere from occasionally to regularly, as well.

I'm attaching the stack trace but I can't capture this in the debugger - the test by itself seems to not fail, and adding tests before it to cause its failure is more of an art than a science.

[Brady Eidson]

Created attachment 11813 [details]
Crash log for the layout test gc-8.html

fast/dom/gc-8.html crash log, here you go!  To repro this, I simply ran run-webkit-tests fast/dom - it crashes about 2/3 the way through the top level directory

[Darin Adler]

I've been seeing the same crash.

[Dave Hyatt]

Did anyone see this before I landed the ICU abstraction?  I'm suspicious of that change now even though the crash log looks unrelated.

[Brady Eidson]

While the first time I ever saw this problem was today, I haven't updated since Thursday, so the ICU abstraction is out for my tree...

[Brady Eidson]

lemme amend and clarify my statement - I'm running on 18092 + local mods.  Hyatt landed the ICU abstraction in 18098. 

[Darin Adler]

Created attachment 11814 [details]
patch, including debug harness I used to diagnose and test

[Geoffrey Garen]

Comment on attachment 11814 [details]
patch, including debug harness I used to diagnose and test

r=me

[Darin Adler]

Committed revision 18171.

[Alexey Proskuryakov]

Marking as fixed.