Workaround a skia limitation when drawing repeated patterns

Workaround a skia limitation when drawing repeated patterns, and remove extra transformations applied to pattern.
Windows and Linux rebaselines will follow (see Issue 17632).

Fixes:
LayoutTests/fast/canvas/patternfill-repeat.html
LayoutTests/svg/W3C-SVG-1.1/pservers-grad-06-b.svg

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=8099

[brettw, 2009-01-13 01:46:22]

Is this encountered a lot? A more efficient approach would be to redefine a
PlatformPattern to include a clipping rect, which the caller would set up if it
exists.

[Michael Moss, 2009-01-13 20:04:14]

As per our discussion, I tested this with CSS and SVG patterns (e.g.
LayoutTests/css1/color_and_background/background_repeat.html) and it doesn't
come into play. It seems to only be in the Canvas code path.

[brettw, 2009-01-13 20:23:15]

http://codereview.chromium.org/17365/diff/1/2
File WebCore/platform/graphics/skia/PatternSkia.cpp (right):

http://codereview.chromium.org/17365/diff/1/2#newcode80
Line 80: shader->setLocalMatrix(matrix);
It seems like when we create the fake bitmap that we'll need to translate by
(-1,-1) since there was an extra row of pixels added.

[Michael Moss, 2009-01-13 20:49:16]

http://codereview.chromium.org/17365/diff/1/2
File WebCore/platform/graphics/skia/PatternSkia.cpp (right):

http://codereview.chromium.org/17365/diff/1/2#newcode80
Line 80: shader->setLocalMatrix(matrix);
My (limited) understanding of skia and webkit drawing is that they both use
top-left as origin, but fake bitmap pads the bottom and right sides, so it
shouldn't need to be shifted to realign the original image. Running the layout
test with modified expand/draw/translate values seems to bear this out.

[brettw, 2009-01-13 20:50:33]

Ah, got it. For some reason I thought you were padding all the way around.

LGTM

[Michael Moss, 2009-01-14 00:38:40]

While investigating another layout test, Evan noticed that all the pattern
transformations, not just offsets, were apparently being applied multiple times.
It seems wrong to just ignore the passed in patternTransform here, but as far as
we could tell, it's just not needed, so I took it out.

[Michael Moss, 2009-01-14 23:01:53]

WebKit patch: https://bugs.webkit.org/show_bug.cgi?id=23332