[BlackBerry] GC in didAllocate() isn't safe

• Component: WebKit BlackBerry
• Status: RESOLVED LATER

[Yong Li]

Not all calls to Heap::didAllocate() are done in operation Allocation state.

We need to do it in another way

[Yong Li]

Created attachment 148578 [details]
proposed patch

I know this is a bit ugly. Any suggestions about how to hack this neatly?

For example, can I add an argument to GCActivityCallback::didAllocate() indicating it is safe to trigger GC?

[WebKit Review Bot]

Attachment 148578 [details] did not pass style-queue:

Failed to run "['Tools/Scripts/check-webkit-style', '--diff-files', u'Source/JavaScriptCore/ChangeLog', u'Source..." exit_code: 1
Source/JavaScriptCore/runtime/GCActivityCallback.h:53:  Missing space inside { }.  [whitespace/braces] [5]
Total errors found: 1 in 4 files


If any of these errors are false positives, please file a bug against check-webkit-style.

[Yong Li]

Created attachment 150786 [details]
fix style

[Oliver Hunt]

Comment on attachment 150786 [details]
fix style

Why is this necessary?

[Yong Li]

(In reply to comment #4)
> (From update of attachment 150786 [details])
> Why is this necessary?

This is a follow-up of https://bugs.webkit.org/show_bug.cgi?id=89236. 

The initial idea is use small heap size when system is under memory pressure. I ended up with putting platform-specific GC policy in GCActivityCallback. However I found it isn't safe to do GC in didAllocate. The only place it makes sense and is safe to collect garbage is reportExtraMemoryCost.

[Rob Buis]

Any JSC folks want to review this?

[Geoffrey Garen]

This looks like a design flaw in bug 89236 more than anything else. We don't want to modify cross-platform logic to fix this, since we don't want the behavior in bug 89236 in the first place.