Delay construction of StyleResolverState until after style sharing.

Delay construction of StyleResolverState until after style sharing.

In some (synthetic) cases, style sharing is so efficient that the
construction of StyleResolverState negatively impacts performance.
Since we only need an ElementResolveContext (plus some parent information)
to determine style sharing, we can delay the StyleResolverState
construction until after that.

R=rune@opera.com
BUG=437281
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=186346

[andersr, 2014-12-02 13:02:12]

https://codereview.chromium.org/772803002/diff/1/Source/core/css/resolver/Ele...
File Source/core/css/resolver/ElementResolveContext.h (right):

https://codereview.chromium.org/772803002/diff/1/Source/core/css/resolver/Ele...
Source/core/css/resolver/ElementResolveContext.h:39: explicit
ElementResolveContext(const Document&);
I wonder if this is even reachable.

[rune, 2014-12-02 13:16:28]

On 2014/12/02 at 13:02:12, andersr wrote:
>
https://codereview.chromium.org/772803002/diff/1/Source/core/css/resolver/Ele...
> File Source/core/css/resolver/ElementResolveContext.h (right):
> 
>
https://codereview.chromium.org/772803002/diff/1/Source/core/css/resolver/Ele...
> Source/core/css/resolver/ElementResolveContext.h:39: explicit
ElementResolveContext(const Document&);
> I wonder if this is even reachable.

You're saying Element is always non-null?

[andersr, 2014-12-02 13:30:18]

On 2014/12/02 13:16:28, rune wrote:
> On 2014/12/02 at 13:02:12, andersr wrote:
> >
>
https://codereview.chromium.org/772803002/diff/1/Source/core/css/resolver/Ele...
> > File Source/core/css/resolver/ElementResolveContext.h (right):
> > 
> >
>
https://codereview.chromium.org/772803002/diff/1/Source/core/css/resolver/Ele...
> > Source/core/css/resolver/ElementResolveContext.h:39: explicit
> ElementResolveContext(const Document&);
> > I wonder if this is even reachable.
> 
> You're saying Element is always non-null?

Yes, it's at least non-null in *almost* all cases. Try-runs here:
https://codereview.chromium.org/776543002/ , although I'm not sure if
null-element cases are covered.

Not that it matters a lot for this patch.

[andersr, 2014-12-02 14:08:17]

https://codereview.chromium.org/772803002/diff/1/Source/core/css/resolver/Sty...
File Source/core/css/resolver/StyleResolver.cpp (right):

https://codereview.chromium.org/772803002/diff/1/Source/core/css/resolver/Sty...
Source/core/css/resolver/StyleResolver.cpp:572: StyleResolverState
state(document(), elementContext, defaultParent);
I ran blink_perf.css locally on this patch, and I didn't find any performance
issues related to the copying of ElementResolveContext.

If it turns out that I'm wrong, it's pretty easy to mitigate anyway by always
creating the ElementResolveContext separately, and just keeping a reference in
StyleResolverState.

[rune, 2014-12-02 20:04:32]

lgtm

[andersr, 2014-12-02 21:06:44]

The CQ bit was checked by andersr@opera.com

[commit-bot: I haz the power, 2014-12-02 21:07:22]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/772803002/1

[commit-bot: I haz the power, 2014-12-02 23:10:57]

Committed patchset #1 (id:1) as
https://src.chromium.org/viewvc/blink?view=rev&revision=186346