Bug 56237 – [chromium] backface visibility doesn't work correctly on perspective-projected layers

• Component: Layout and Rendering
• Status: RESOLVED FIXED

[Vangelis Kokkevis]

The math used to compute the direction of the Z axis doesn't produce the right answer for complex transforms (ones with perspective projection, non-uniform scaling, etc)

To see the problem go to:
http://www.webkit.org/blog-files/3d-transforms/morphing-cubes.html

and uncheck "backfaces visible".  You'll notice faces popping in and out.

[Vangelis Kokkevis]

Created attachment 85563 [details]
Patch

[Kenneth Russell]

Comment on attachment 85563 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=85563&action=review

Looks good to me. One comment.

> Source/WebCore/platform/graphics/chromium/LayerRendererChromium.cpp:815
> +        FloatQuad mappedLayer = combinedDrawMatrix.mapQuad(FloatQuad(layerRect));
> +        FloatSize horizontalDir = mappedLayer.p2() - mappedLayer.p1();
> +        FloatSize verticalDir = mappedLayer.p4() - mappedLayer.p1();
> +        FloatPoint3D xAxis(horizontalDir.width(), horizontalDir.height(), 0);
> +        FloatPoint3D yAxis(verticalDir.width(), verticalDir.height(), 0);
> +        FloatPoint3D zAxis = xAxis.cross(yAxis);

It seems odd to do these operations in 2D. Intuitively I would transform the four points of the quad in 3D (probably without the perspective portion of the matrix) and then do the appropriate vector differences and cross products. However I gather that TransformationMatrix doesn't provide enough control to do this. As long as you've tested this math to your satisfaction I'm sure it's OK.

[Vangelis Kokkevis]

Committed r81288: <http://trac.webkit.org/changeset/81288>