Bug 103593 – [Shadow] Internals should have a method to cause distribution.

• Status: RESOLVED
• Resolution: WONTFIX
• Component: DOM

[Shinya Kawanaka]

We would like to have a method to ensure distribution for measuring Shadow DOM performance.
Though content.getDistributeNodes() can cause distribution, however it creates a vector of distributed elements.

We woud like to measure just distribution performance, we have to have it.

[Shinya Kawanaka]

Created attachment 176639 [details]
Patch

[Build Bot]

Comment on attachment 176639 [details]
Patch

Attachment 176639 [details] did not pass mac-ews (mac):
Output: http://queues.webkit.org/results/15031507

[Shinya Kawanaka]

Created attachment 176648 [details]
WIP

[Hajime Morrita]

Comment on attachment 176639 [details]
Patch

Is there any other way to make this happen without internals?
Measuring performance through internals API sounds a bit too artificial.

[Shinya Kawanaka]

(In reply to comment #4)
> (From update of attachment 176639 [details])
> Is there any other way to make this happen without internals?
> Measuring performance through internals API sounds a bit too artificial.

We have:
 - offsetLeft
 - getDistributedNodes()
But this causes layout (getDistributedNodes() causes layout now).

I have no idea we have another way.

[Dimitri Glazkov (Google)]

(In reply to comment #5)
> (In reply to comment #4)
> > (From update of attachment 176639 [details] [details])
> > Is there any other way to make this happen without internals?
> > Measuring performance through internals API sounds a bit too artificial.
> 
> We have:
>  - offsetLeft
>  - getDistributedNodes()
> But this causes layout (getDistributedNodes() causes layout now).
> 
> I have no idea we have another way.

As long as this produces meaningful measurement, I am okay with using internals in perf tests.

[Build Bot]

Comment on attachment 176648 [details]
WIP

Attachment 176648 [details] did not pass win-ews (win):
Output: http://queues.webkit.org/results/15027929

[Hajime Morrita]

(In reply to comment #6)
> > 
> > I have no idea we have another way.
> 
> As long as this produces meaningful measurement, I am okay with using internals in perf tests.

Talked to Shinya offline and found that we can omit layout time 
by setting display:none to the host.
If even null-layout prevents meaningful measurement, 
the speed of content distribution must not matter.
Unfortunately as expected, it matters at least for now ;-)

[Shinya Kawanaka]

(In reply to comment #8)
> (In reply to comment #6)
> > > 
> > > I have no idea we have another way.
> > 
> > As long as this produces meaningful measurement, I am okay with using internals in perf tests.
> 
> Talked to Shinya offline and found that we can omit layout time 
> by setting display:none to the host.
> If even null-layout prevents meaningful measurement, 
> the speed of content distribution must not matter.
> Unfortunately as expected, it matters at least for now ;-)

Let close this for now. When the time when we really need this comes, let's reopen this.