Bug 94990 – Share immutable ElementAttributeData between elements with identical attributes.

• Component: DOM
• Status: RESOLVED FIXED

[Andreas Kling]

<rdar://12141736>

[Andreas Kling]

Created attachment 160542 [details]
Love letter to EWS

[Build Bot]

Comment on attachment 160542 [details]
Love letter to EWS

Attachment 160542 [details] did not pass mac-ews (mac):
Output: http://queues.webkit.org/results/13574935

[Early Warning System Bot]

Comment on attachment 160542 [details]
Love letter to EWS

Attachment 160542 [details] did not pass qt-wk2-ews (qt):
Output: http://queues.webkit.org/results/13591469

[Andreas Kling]

Created attachment 160545 [details]
Love letter to EWS + cheesy GCC buildfix

[Andreas Kling]

Created attachment 161939 [details]
Patch!!11

[Andreas Kling]

I should note that there are a number of follow-up optimizations enabled by this, such as avoiding parsing the same style/class attributes more than once.

[Antti Koivisto]

Comment on attachment 161939 [details]
Patch!!11

View in context: https://bugs.webkit.org/attachment.cgi?id=161939&action=review

Nice. r=me

> Source/WebCore/dom/Document.cpp:6178
> +class ImmutableAttributeDataCacheKey {

Document is super bloaty. Would be nice to factor this a separate file/class.

> Source/WebCore/dom/Document.cpp:6192
> +    bool operator!=(ImmutableAttributeDataCacheKey& other) const

const

> Source/WebCore/dom/Document.cpp:6235
> +    if (!cacheHash || cacheIterator->second)

The logic here is bit strange. I suspect it could be written in a nicer way.

[Andreas Kling]

Committed r127438: <http://trac.webkit.org/changeset/127438>

[Andreas Kling]

Perf-o-matic results are in, looks like a 3.6% improvement on Parser/html5-full-render:malloc

http://webkit-perf.appspot.com/graph.html#tests=[[6268923,2001,963028]]&sel=none&displayrange=7&datatype=running

Booyakasha!

@Antti: I agree that it can be factored better, I'll improve upon this in coming patches.