CycleCollector/CCStatus-2007-05-25: Difference between revisions
Jump to navigation
Jump to search
No edit summary |
No edit summary |
||
| Line 8: | Line 8: | ||
* Jonas will write a test case to generate cycles via document. STATUS: ? | * Jonas will write a test case to generate cycles via document. STATUS: ? | ||
* [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)]: Done. | ||
* [https://bugzilla.mozilla.org/show_bug.cgi?id=377884 Switch nsXPConnect::Traverse to use tracing]: | * [https://bugzilla.mozilla.org/show_bug.cgi?id=377884 Switch nsXPConnect::Traverse to use tracing]: Done. | ||
* [https://bugzilla.mozilla.org/show_bug.cgi?id=379718 using trace API for reference counts]: | * [https://bugzilla.mozilla.org/show_bug.cgi?id=379718 using trace API for reference counts]: Nice to have, but not urgent. Need to do some perf analysis on this one. dbaron will look into this over the next couple of weeks. Peterv has a patch that he's kept up to date and will attach that to the bug. It's not clear that the patch will change anything (leak wise), but it might have help performance and make the code more maintainable, GC stuff gets hidden behind the API etc. | ||
Later: | Later: | ||
* {{bug|381199}}: revisit cycle collector aging strategy | * {{bug|381199}}: revisit cycle collector aging strategy | ||
Dbaron will look at this over the next couple of weeks as well. | |||
== Performance requirements == | == Performance requirements == | ||
| Line 23: | Line 25: | ||
== Next steps on leaks == | == Next steps on leaks == | ||
We're done fixing new leaks AFAWK, now we can concentrate on finding old leaks etc. Leaks are down to 3k or so, so not too much to go after in the normal case. | |||
== Next steps on performance (pauses) == | == Next steps on performance (pauses) == | ||
Revision as of 18:47, 25 May 2007
« previous meeting | index | next meeting »
Agenda
- status of patches in progress: All we've done in the past week is review peterv's and Igor's patches. We still need to land some of these (basically we are almost in the same spot as last week).
Status of Pending CC Code in Bugzilla
- Jonas will write a test case to generate cycles via document. STATUS: ?
- Support cycle collection of refcounted non-xpcom objects (aka, the XBL patch): Done.
- Switch nsXPConnect::Traverse to use tracing: Done.
- using trace API for reference counts: Nice to have, but not urgent. Need to do some perf analysis on this one. dbaron will look into this over the next couple of weeks. Peterv has a patch that he's kept up to date and will attach that to the bug. It's not clear that the patch will change anything (leak wise), but it might have help performance and make the code more maintainable, GC stuff gets hidden behind the API etc.
Later:
- bug 381199: revisit cycle collector aging strategy
Dbaron will look at this over the next couple of weeks as well.
Performance requirements
Criteria for Performance Satisfactory:
- Double the current GC time.
Next steps on leaks
We're done fixing new leaks AFAWK, now we can concentrate on finding old leaks etc. Leaks are down to 3k or so, so not too much to go after in the normal case.