Fixed oversized SkRegion bound problem for complexclip_aa GM

In the case of entirely blank lines count_left_right_zeros was returning all the empty pixels on the left and 0 empty pixels on the right. In the complexclip_aa case there are other filled lines that bump the left zero count down but the right count remains pegged at 0. This resulted in an oversized bounding box.

[bsalomon, 2012-08-01 17:35:34]

On 2012/08/01 15:58:53, robertphillips wrote:

The SkGpuDevice change LGTM :)

[reed1, 2012-08-01 17:50:46]

very good catch

http://codereview.appspot.com/6447076/diff/1/src/core/SkAAClip.cpp
File src/core/SkAAClip.cpp (right):

http://codereview.appspot.com/6447076/diff/1/src/core/SkAAClip.cpp#newcode223
src/core/SkAAClip.cpp:223: 
// Document what we return, and how we handle empty

http://codereview.appspot.com/6447076/diff/1/src/core/SkAAClip.cpp#newcode411
src/core/SkAAClip.cpp:411: count_left_right_zeros(base + yoff->fOffset, width,
&L, &R);
assert(L + R < width || (L == width && R == width))

[robertphillips, 2012-08-01 18:22:25]

http://codereview.appspot.com/6447076/diff/1/src/core/SkAAClip.cpp
File src/core/SkAAClip.cpp (right):

http://codereview.appspot.com/6447076/diff/1/src/core/SkAAClip.cpp#newcode223
src/core/SkAAClip.cpp:223: 
On 2012/08/01 17:50:46, reed1 wrote:
> // Document what we return, and how we handle empty

Done.

http://codereview.appspot.com/6447076/diff/1/src/core/SkAAClip.cpp#newcode411
src/core/SkAAClip.cpp:411: count_left_right_zeros(base + yoff->fOffset, width,
&L, &R);
On 2012/08/01 17:50:46, reed1 wrote:
> assert(L + R < width || (L == width && R == width))

Done.

[robertphillips, 2012-08-02 12:45:41]

committed as r4906