CycleCollector/CCStatus-2007-03-30: Difference between revisions

No edit summary
(prev/next)
 
(4 intermediate revisions by 2 users not shown)
Line 1: Line 1:
<small>[[CycleCollector/CCStatus-2007-03-26|&laquo; previous meeting]] | [[CycleCollector/Status_Meetings|index]]</small>
<small>[[CycleCollector/CCStatus-2007-03-23|&laquo; previous meeting]] | [[CycleCollector/Status_Meetings|index]] | [[CycleCollector/CCStatus-2007-04-20|next meeting &raquo;]]</small>


=== Agenda ===
=== Agenda ===
Line 13: Line 13:
Graydon:   
Graydon:   


I'm seeing all windows and documents not leaking.
Sees no documents or windows leaked.


David:
David:
Line 23: Line 23:
Igor landed a patch that regressed the leaks a bit.  See {{bug|375808}}
Igor landed a patch that regressed the leaks a bit.  See {{bug|375808}}


We're still missing javascript edges and that might cause leaks of windows.  Igor has been working on a patch that makes the GC code and traversal code share the same code.  So we automatically find all the javascript edges.
We're still missing javascript edges and that might cause leaks of windows.  Igor has been working on a patch that makes the GC code and traversal code share the same code ({{bug|375270}}).  So we automatically find all the javascript edges.


 
* {{bug|375063}} (Leaking nsJSContexts and nsXBLDocGlobalObjects at shutdown):  This is the bug that caused Igor to file {{bug|375270}}. Will probably want to fix {{bug|375270}} instead of landing the fix for {{bug|375063}}.
--
* {{bug|375063}} (Leaking nsJSContexts and nsXBLDocGlobalObjects at shutdown):  This is the bug that caused Igor to write the {{bug|375808}}.


* {{bug|375075}} (Stop leaking the cycle collector on shutdown)
* {{bug|375075}} (Stop leaking the cycle collector on shutdown)
Tried to reland patch for {{bug|372960}} (Make XPConnect traverse more JS edges), but Tinderbox is still going orange. Will probably back out again, seems related to CC performance issue. Tried to turn on jst's timing code, but that made Tinderbox go green so can't debug the issue then.
Made new patch for {{bug|372713}} (Add cycle collection to RDF datasources) to address r- comments.


These are a bunch of safety checks that need to be removed in order to improve the performance:  {{bug|373693}} (Cycle collector makes AddRef about 6x slower) and {{bug|373694}} (Cycle collector makes Release about 11x slower), not done yet.
These are a bunch of safety checks that need to be removed in order to improve the performance:  {{bug|373693}} (Cycle collector makes AddRef about 6x slower) and {{bug|373694}} (Cycle collector makes Release about 11x slower), not done yet.
Line 39: Line 33:
=== Status of Pending CC Code in Bugzilla ===
=== Status of Pending CC Code in Bugzilla ===


* [https://bugzilla.mozilla.org/show_bug.cgi?id=372713 Add cycle collection to RDF datasources]
* [https://bugzilla.mozilla.org/show_bug.cgi?id=372713 Add cycle collection to RDF datasources]: awaiting sr
* [https://bugzilla.mozilla.org/show_bug.cgi?id=372960 Make XPConnect traverse more JS edges]
* [https://bugzilla.mozilla.org/show_bug.cgi?id=368774 Support cycle collection of refcounted non-xpcom objects (aka, the XBL patch)]
* [https://bugzilla.mozilla.org/show_bug.cgi?id=368774 Support cycle collection of refcounted non-xpcom objects (aka, the XBL patch)]
* [https://bugzilla.mozilla.org/show_bug.cgi?id=368869 Make cycle collection suspect all native wrapper roots]
* [https://bugzilla.mozilla.org/show_bug.cgi?id=368869 Make cycle collection suspect all native wrapper roots]: backed out.
* [https://bugzilla.mozilla.org/show_bug.cgi?id=372766 cycle collector shutdown dump based on known should-have-been-freed objects]
 


=== Next Steps ===
=== Next Steps ===

Latest revision as of 06:25, 27 April 2007

« previous meeting | index | next meeting »

Agenda

  • Discuss performance: Boris has filed two bugs (bug 373693 (addref) and bug 373694 (release). Status?
  • dbaron on slow down he was encountering.
  • Performance in general.
  • Outcome of leaks tests?
  • jst on status of landing the CC timing code (bug 374239).
  • Discuss next steps.

Leak Numbers

Graydon:

Sees no documents or windows leaked.

David:

I'm still seeing significant leaks. I generally do run with leak gauge, but I only quit the browser once a week and thats by crashing.

Peterv:

Igor landed a patch that regressed the leaks a bit. See bug 375808

We're still missing javascript edges and that might cause leaks of windows. Igor has been working on a patch that makes the GC code and traversal code share the same code (bug 375270). So we automatically find all the javascript edges.

  • bug 375063 (Leaking nsJSContexts and nsXBLDocGlobalObjects at shutdown): This is the bug that caused Igor to file bug 375270. Will probably want to fix bug 375270 instead of landing the fix for bug 375063.
  • bug 375075 (Stop leaking the cycle collector on shutdown)

These are a bunch of safety checks that need to be removed in order to improve the performance: bug 373693 (Cycle collector makes AddRef about 6x slower) and bug 373694 (Cycle collector makes Release about 11x slower), not done yet.

Status of Pending CC Code in Bugzilla


Next Steps

  • Graydon is working on ES4.
  • Peter will take care of all the ifdefs and will measure before and after timings.
  • Dbaron will continue to monitor performance.

Audio of This Meeting