Bug 232469 – [GPU Process] [Filters 4/23] Make CSSFilter be a composite of FilterFunctions

• Status: RESOLVED
• Resolution: FIXED
• Component: Layout and Rendering

[Said Abou-Hallawa]

This will make CSSFilter have an item of SVGFilter in this list of FilterFunctions.

[Said Abou-Hallawa]

Created attachment 442784 [details]
Patch

[Said Abou-Hallawa]

Created attachment 442785 [details]
Patch for review

[Simon Fraser (smfr)]

Comment on attachment 442785 [details]
Patch for review

View in context: https://bugs.webkit.org/attachment.cgi?id=442785&action=review

> Source/WebCore/ChangeLog:8
> +        In this patch, the CSS referenced filter is built as an SVGFilter and it

"reference filter"

> Source/WebCore/css/CSSFilterImageValue.cpp:120
> +    auto texture = ImageBuffer::create(size, renderingMode, 1, DestinationColorSpace::SRGB(), PixelFormat::BGRA8);

texture is a weird name for this.

> Source/WebCore/css/CSSFilterImageValue.cpp:130
> +    auto cssFilter = CSSFilter::create(m_filterOperations, renderingMode);
> +    
>      cssFilter->setSourceImageRect(imageRect);
>      if (!cssFilter->build(renderer, m_filterOperations, FilterConsumer::FilterFunction))

Why do we have to pass m_filterOperations both to the constructor, and to the build() function?

> Source/WebCore/rendering/CSSFilter.cpp:223
> +    auto builder = makeUnique<SVGFilterBuilder>();

Can SVGFilterBuilder just be on the stack?

> Source/WebCore/rendering/RenderLayerFilters.h:66
> +    GraphicsContext* beginFilterEffect(RenderElement&, const LayoutRect& filterBoxRect, const LayoutRect& dirtyRect, const LayoutRect& layerRepaintRect, GraphicsContext& destinationContext);

I tend to prefer GraphicsContext& destinationContext as the first argument.

> Source/WebCore/rendering/svg/SVGRenderTreeAsText.cpp:436
> +        auto builder = makeUnique<SVGFilterBuilder>();

Can this live on the stack?

[Said Abou-Hallawa]

Some of the layout tests are failing because of the need to set the sourceImageRect before building the CSSFilter. The sourceImageRect has to include the Outsets of the FilterEffects. So currently to fix these tests we need to set the sourceImageRect twice and to build the CSSFilter twice. The plan to remove this dependency and be able to build the CSSFilter without having to set sourceImageRect and have the FilterEffects be able to calculate the outsets without knowing the sourceImageRect.

The reset of the failures are happing because of the color space which should be fixed once the calculation of the result color space out of the operating color space. So for now, I am going to skip these few tests and remove them by the end.

[Said Abou-Hallawa]

Created attachment 442946 [details]
Patch

[Radar WebKit Bug Importer]

<rdar://problem/85047148>

[Said Abou-Hallawa]

Created attachment 443851 [details]
Patch

[Said Abou-Hallawa]

Created attachment 443864 [details]
Patch

[Said Abou-Hallawa]

Created attachment 443899 [details]
Patch

[EWS]

Committed r285618 (244119@main): <https://commits.webkit.org/244119@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 443899 [details].