Bug 114440 – Dynamically triggered subframe loads are causing tile churn

• Component: Layout and Rendering
• Status: RESOLVED FIXED

[Antti Koivisto]

We use minimal tile coverage while page is loading. Subframe loads that start after the main load has finishes are causing us to re-enter minimal tiling mode, causing unnecessary churn.

[Antti Koivisto]

This shows up in the tile map with speculative tiles disappearing and reappearing after end of the load.

[Antti Koivisto]

Created attachment 198114 [details]
patch

[Andreas Kling]

Comment on attachment 198114 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=198114&action=review

r=me, good catch. I wonder if there are cases where the window resize optimization is too clever and just causing churn.

> Source/WebCore/ChangeLog:8
> +        In some cases we switch in and out of minimal times multiple times during page loading. This

minimal times -> minimal tiles

> Source/WebCore/ChangeLog:9
> +        creates tile churn where we create speculatively tiles and then throw them out. This can be

speculatively -> speculative

> Source/WebCore/ChangeLog:16
> +            tiles while waiting response for the initial requests.

waiting -> awaiting

> Source/WebCore/ChangeLog:18
> +        (WebCore):

<darin>Please remove this useless line. Better yet, please fix the script to stop generating them!</darin>

> Source/WebCore/ChangeLog:28
> +                Only switch into minimal coverage mode once per page load.

Indentation is different here.

> Source/WebCore/rendering/RenderLayerBacking.cpp:226
> +        useMinimalTilesDuringLoading = !frameView->isVisuallyNonEmpty() || (frame->page()->progress()->isLoadProgressing() && !frameView->wasScrolledByUser());

"!frameView->isVisuallyNonEmpty()" :|

[Antti Koivisto]

http://trac.webkit.org/changeset/148432

[Simon Fraser (smfr)]

Comment on attachment 198114 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=198114&action=review

> Source/WebCore/rendering/RenderLayerBacking.h:312
> +    bool m_didSwitchToFullTileCoverageDuringLoading;

Shame to use a bool on every RLB when only the root needs one.