Closed Bug 1312110 Opened 8 years ago Closed 8 years ago

nsCSSRuleProcessor::CascadeSheet() allocates a JS regexp object during InterruptCallback

Categories

(Core :: CSS Parsing and Computation, defect)

x86
Windows 8
defect
Not set
critical

Tracking

()

RESOLVED DUPLICATE of bug 1310335

People

(Reporter: jchen, Unassigned)

References

Details

(Keywords: crash)

This bug was filed from the Socorro interface and is 
report bp-0bb229d2-8612-46ce-9100-1233a2161021.
=============================================================

Low volume crash that's been happening in several recent nightlies, but I'm not sure when it started. It does appear across several installations/machines so it's unlikely to be a bogus crash.
This is another regression from bug 1308039 (though we've been marking these blocking bug 1279086 so I'll continue that).

Painting during interrupt callback ends up in nsCSSRuleProcessor::CascadeSheet(), which calls into DocumentRule::UseForPresentation(), which calls nsContentUtils::IsPatternMatching(), which attempts to allocate a new JS engine regular expression object.

It looks like bug 398962 added support for regular expressions to a moz-document CSS rule, or something like that, and it was implemented by calling into the JS engine.

One side note, I'm not sure why these crashes are not annotated with the same release assert as the other crashes. We're crashing inside a JS engine allocation. There may be more release asserts to be added here.
Blocks: 1279086
Component: JavaScript Engine → CSS Parsing and Computation
Summary: Crash in JSObject::create → nsCSSRuleProcessor::CascadeSheet() allocates a JS regexp object during InterruptCallback
Sadly it seems hard to call into irregexp without the SpiderMonkey machinery wrapped around it.  Based on a quick look at that code, it seems fairly intertwined with SpiderMonkey.  For example, NativeRegExpMacroAssembler uses SpiderMonkey's macro assembler machinery...

I wonder if we can bypass resolving such moz-document rules when painting in this mode?  That seems like a questionable thing to do for example for users who are using such rules in their user stylesheet to customize a website...
This is still present on Nightly.
Crash Signature: [@ JSObject::create] → [@ JSObject::create] [@ JSObject* js::Allocate<T> ]
Status: NEW → RESOLVED
Closed: 8 years ago
Resolution: --- → DUPLICATE
Crash Signature: [@ JSObject::create] [@ JSObject* js::Allocate<T> ]
You need to log in before you can comment on or make changes to this bug.