[Webkit-unassigned] [Bug 33957] [v8] Implement Node map in intrusive way for better speed

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=33957





--- Comment #6 from anton muhin <antonm at chromium.org>  2010-01-22 06:30:42 PST ---
(In reply to comment #4)
> (From update of attachment 47126 [details])
> + if (!wrapper) return v8::Persistent<v8::Object>();
> 
> Should be two lines.  (This nit repeated many times.)

Hopefully fixed all those, sorry.  Just curious, how difficult it'd be to add
this check into style-checking script?

> + TableChunk
> 
> Why not just use Vector?

I am not sure I fully understand you.  You later corrected it to HashMap. 
Anyway, one thing I'd like to escape is x2 allocation policy used by wtf's
HashTable and Vector for two reasons: less memory and speed.

I know one deficiency of my current implementation: it thrashes when on adds
and removes the element which could lead to allocation and deallocation of next
chunk.  It's quite easy to fix, but would complicate the code slightly.  What
do you think, should I fix it immediately?

> + virtual v8::Persistent<ValueType> get(KeyType* obj)
> 
> Are these methods already virtual?  It seems like we shouldn't need that.  We
> should be able to use templates to solve the dispatch problem at compile time.

I think they already are---WeakReferenceMap declares all those methods as
virtual.

I wholeheartedly agree with you that we should at least try to de-virtualize
them though, but could we save it for another CL?

Running through tests on my box and uploading new patch.  Thanks a lot for
comments, Adam.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.