Bug 128321 – Check selectors exactly when invalidating style

• Status: RESOLVED
• Resolution: FIXED
• Component: DOM

[Antti Koivisto]

instead of heuristics

[Antti Koivisto]

Created attachment 223362 [details]
patch

[Andreas Kling]

Comment on attachment 223362 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=223362&action=review

r=me with some tweaks.

Q: Can we piggyback on the already massaged data in the Document's StyleResolver here?

> Source/WebCore/ChangeLog:9
> +        With selectors are really fast to match with the JIT. Take advantage of it by invalidating
> +        document style exactly when new stylesheets arrive (instead of heuristics).

I don't think you need the "With" here. We should remove it to achieve 60fps on mobile in 2014.

> Source/WebCore/css/StyleInvalidationAnalysis.cpp:40
> +static bool shouldDirtyAllStyle(const Vector<RefPtr<StyleRuleBase>> rules)

We should pass the rules vector by (const) reference here.

> Source/WebCore/css/StyleInvalidationAnalysis.cpp:42
> +    for (auto rule : rules) {

Ref churn here. auto& would avoid it.

> Source/WebCore/css/StyleInvalidationAnalysis.cpp:57
> +    for (auto import : sheet.importRules()) {

Same here.

> Source/WebCore/css/StyleInvalidationAnalysis.cpp:70
> +    for (auto sheet : sheets) {

Sam has convinced me to start writing "auto*" to clarify that there's no by-value copy.

> Source/WebCore/css/StyleInvalidationAnalysis.cpp:84
> +    for (auto sheet : sheets)

Same comment here.

[Antti Koivisto]

(In reply to comment #2)
> (From update of attachment 223362 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=223362&action=review
> 
> r=me with some tweaks.
> 
> Q: Can we piggyback on the already massaged data in the Document's StyleResolver here?

Not really. The new sheet stuff is not in StyleResolver yet. Optimally we would transfer it over instead of reprocessing but everything here is pretty cheap in any case.

[Antti Koivisto]

https://trac.webkit.org/r163592

[Antti Koivisto]

https://trac.webkit.org/r163592
https://trac.webkit.org/r163594