597923 - A bunch of time spent in JSCompartment::purge during tab switch.

Status: RESOLVED WORKSFORME

(Core :: JavaScript Engine, defect)

Description

[Jeff Muizelaar [:jrmuizel]]

I don't have a callstacks but these functions include the majority of cpu time.

	62.6%	62.6%	XUL	112 B	JSCompartment::purge(JSContext*)	
	11.1%	11.1%	XUL	752 B	js::mjit::ic::PurgePICs(JSContext*, JSScript*)	

I got profiles using Shark's Unresponsive Application profiling. All of the profiles it produced basically looked the same. I don't have any stacks because of -fomit-frame-pointer. I have about 30-40 tabs open and haven't tried to reproduce the problem yet. This was using a build from 09-19-2010

Comment 1

[Brendan Eich [:brendan]]

Why are we purging PICs on tab switch?

/be

Comment 2

[David Anderson [:dvander] - inactive, e-mail if emergency]

This is probably from a GC.

Comment 3

[David Mandelin [:dmandelin]]

I suspect this is compartments-related. Without full compartments support, the script list could get corrupted.

Without STR, I don't really see how to fix this, but there is a bug on file for cleaning up the scripts list (bug 586181). Once compartments are done, we should audit the script-list code and that bug and fix any problems.

Comment 4

[David Mandelin [:dmandelin]]

Nominating for reconsideration (was beta8+) because:

 - This seems to have happened only once, with no STR and no further reports

 - It was suspected to be caused by having JM with compartments, and now we have
   compartments.

Comment 5

[David Mandelin [:dmandelin]]

It is impossible to actually work on this bug, so closing. Also, my original guess was that it related to not having scripts properly restricted to compartments, which has been fixed.