Bug 216355 – Don't create event regions when the page has no subscrollers

• Status: RESOLVED
• Resolution: FIXED
• Component: Scrolling

[Antti Koivisto]

Unless the page uses features like touch-action we don't need event regions for plain main frame scrolling.

[Antti Koivisto]

<rdar://problem/67900642>

[Antti Koivisto]

Created attachment 408431 [details]
patch

[Antti Koivisto]

Created attachment 408432 [details]
patch

[Simon Fraser (smfr)]

Comment on attachment 408432 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=408432&action=review

> Source/WebCore/rendering/RenderLayerCompositor.cpp:880
> +        if (scrollingCoordinator && scrollingCoordinator->hasSubscrollers() != hadSubscrollers)
> +            invalidateEventRegionForAllFrames();

I don't like that this triggers a second compositing update just after we've finished one, and in all frames too!

This means that every page with one or more subscrollers now suffers from an extra full layer tree walk in all frames and a second compositing update.

[Antti Koivisto]

> I don't like that this triggers a second compositing update just after we've
> finished one, and in all frames too!

No it doesn't. It just sets m_needsEventRegionUpdate bit in RenderLayerBacking.

[Simon Fraser (smfr)]

Comment on attachment 408432 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=408432&action=review

> Source/WebCore/page/scrolling/ScrollingStateScrollingNode.cpp:39
> +    scrollingStateTree().scrollingNodeAdded();

I wonder if ScrollingStateTree::addNode() is a better place to call this.

> Source/WebCore/page/scrolling/ScrollingStateScrollingNode.cpp:80
> +    scrollingStateTree().scrollingNodeRemoved();

I wonder if ScrollingStateTree::willRemoveNode() is a better place to call this.

>> Source/WebCore/rendering/RenderLayerCompositor.cpp:880
>> +            invalidateEventRegionForAllFrames();
> 
> I don't like that this triggers a second compositing update just after we've finished one, and in all frames too!
> 
> This means that every page with one or more subscrollers now suffers from an extra full layer tree walk in all frames and a second compositing update.

I see now that we'll update regions at the end of the rendering update.

[Antti Koivisto]

> I wonder if ScrollingStateTree::addNode() is a better place to call this.
> I wonder if ScrollingStateTree::willRemoveNode() is a better place to call
> this.

I considered it but this seemed like the most robust solution.

[EWS]

commit-queue failed to commit attachment 408432 [details] to WebKit repository.

[Antti Koivisto]

Created attachment 408461 [details]
patch

[EWS]

Committed r266846: <https://trac.webkit.org/changeset/266846>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 408461 [details].