Ojan, note as well that JS wrapper objects survive GC in "groups": roughly all the objects from the same DOM tree (including document itself) either survive GC altogether or not (and add Mads' comment about necessity of two major GCs to collect wrappers as they are usually weak references).
I am not 100% sure, but apparently that implies that as long as you have window object live, there is not that much you can collect in JS wrappers. yours, anton On Thu, Aug 26, 2010 at 12:14 PM, Mads Ager <[email protected]> wrote: > Hi Ojan, > >> I'm trying to write a perf testing framework for webkit tests. I suspect >> that GC is responsible for some of the flakiness. I've added a call to >> GCController.collect() before running the test, but as best I can tell, that >> doesn't do a full GC. Is there a reason for that? Could we change it to do >> so? > > calling CollectAllGarbage does perform a full GC, so I don't think > that is the issue. > >> It looks like this is what gets hit ultimately: >> Heap::CollectAllGarbage(false); >> http://code.google.com/p/v8/source/browse/trunk/src/execution.cc#691 >> >> Should that false be changed to true? Also, should it clear the compilation >> cache first? > > You could change the flag to true, but I don't think it will make a > difference. If the flag is true it forces a compacting collection in > V8. With a false flag you still get a full GC and potentially a > compaction if there is fragmentation. I think the main issue here is > that in order to get rid of DOM wrappers in the heap you need more > than one GC (at least two - one to notice that it is not referenced > from JS and perform a callback and another one to actually get rid of > the object). Does it help if you do a couple of calls to collect()? > >> Also, I noticed that we don't implement GCController.getJSObjectCount. Why >> is there? Is there some meaningful value we could return there? > > I have no idea why it is there. Not sure what we would want to use it > for. I don't think we should rely on it in testing. > > Cheers, -- Mads > >> Thanks, >> Ojan >> >> -- >> v8-dev mailing list >> [email protected] >> http://groups.google.com/group/v8-dev >> > > -- > v8-dev mailing list > [email protected] > http://groups.google.com/group/v8-dev > -- v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev
