Bug 51374 – Cairo's ContextShadow may mis-render some box shadows

• Status: RESOLVED
• Resolution: FIXED
• Component: Platform

[Simon Fraser (smfr)]

After porting ContextShadow to Mac, I sees some combinations of border-radius and shadow blur that it mis-renders. Example coming.

[Simon Fraser (smfr)]

Created attachment 77081 [details]
Testcase

[Simon Fraser (smfr)]

I think the top of ContextShadow::drawRectShadow() should read:


    float radiusTwice = m_blurDistance * 2;

    // Size of the tiling side.
    int sideTileWidth = 1;

    // Find the extra space needed from the curve of the corners.
    int extraWidthFromCornerRadii = radiusTwice + max(topLeftRadius.width(), bottomLeftRadius.width()) +
        radiusTwice + max(topRightRadius.width(), bottomRightRadius.width());
    int extraHeightFromCornerRadii = radiusTwice + max(topLeftRadius.height(), topRightRadius.height()) +
        radiusTwice + max(bottomLeftRadius.height(), bottomRightRadius.height());

    // The length of a side of the buffer is the enough space for four blur radii,
    // the radii of the corners, and then 1 pixel to draw the side tiles.
    IntSize shadowTemplateSize = IntSize(sideTileWidth + extraWidthFromCornerRadii,
                                         sideTileWidth + extraHeightFromCornerRadii);

    // drawShadowedRect still does not work with rotations.
    // https://bugs.webkit.org/show_bug.cgi?id=45042
    CGContextRef cgContext = context->platformContext();
    if ((!context->getCTM().isIdentityOrTranslationOrFlipped()) || (shadowTemplateSize.width() > rect.width())
        || (shadowTemplateSize.height() > rect.height()) || (m_type != BlurShadow)) {
        drawRectShadowWithoutTiling(cgContext, rect, topLeftRadius, topRightRadius, bottomLeftRadius, bottomRightRadius, 1);
        return;
    }

[Alejandro G. Castro]

(In reply to comment #1)
> Created an attachment (id=77081) [details]
> Testcase

Yep, it fails for me also in webkitgtk, we had solved this issues I guess in some of the refactorings before applying the patch we did not check it correctly.

[Alejandro G. Castro]

Created attachment 77099 [details]
Proposed patch

[Alejandro G. Castro]

(In reply to comment #2)
> I think the top of ContextShadow::drawRectShadow() should read:
> 
> [...]
>
>     if ((!context->getCTM().isIdentityOrTranslationOrFlipped()) || (shadowTemplateSize.width() > rect.width())
>         || (shadowTemplateSize.height() > rect.height()) || (m_type != BlurShadow)) {
>         drawRectShadowWithoutTiling(cgContext, rect, topLeftRadius, topRightRadius, bottomLeftRadius, bottomRightRadius, 1);
>         return;
>     }

You are right, the problem is the calculation of the internal space required for the corner shadows, I've refactored the code with your proposal in the patch. Thanks.

[Martin Robinson]

Comment on attachment 77099 [details]
Proposed patch

Great, though we should definitely include the test case with pixel results to avoid this in the future.

[Alejandro G. Castro]

Created attachment 77327 [details]
Patch

[Martin Robinson]

Comment on attachment 77327 [details]
Patch

Thank you!

[Alejandro G. Castro]

Landed http://trac.webkit.org/changeset/75237