Fix for nested rect drawing bug

This CL fixes crbug.com 273121 (regression in canvas 2d fill with rect and inverted winding rect) and adds a GM that would've caught the problem.

[bsalomon, 2013-08-18 20:32:49]

The logic for determining whether it the path is nested rects is feeling like it
has really detailed knowledge of the particular implementation of the rect
renderer. Should the renderer itself be making these decisions?

Broadening that a bit, what do you think about wrapping the various special
cases (ovals, rects, nested rects...) into a GrPathRenderer subclass?

Part of the reason I ask is that I believe in certain configurations, like when
we're relying on MSAA or NVPR we may not want the special cases at all and then
disabling them would just mean not installing that subclass.

[robertphillips, 2013-08-19 13:06:15]

I think that's a good idea (having a PathRenderer subclass that collects the
special cases). This is a Priority-2 and I would be more comfortable cherry
picking this change rather than a path renderer change back to M30.

[bsalomon, 2013-08-19 13:19:45]

On 2013/08/19 13:06:15, robertphillips wrote:
> I think that's a good idea (having a PathRenderer subclass that collects the
> special cases). This is a Priority-2 and I would be more comfortable cherry
> picking this change rather than a path renderer change back to M30.

gotcha, lgtm

[robertphillips, 2013-08-20 12:11:51]

committed as r10803

[robertphillips, 2013-08-28 13:45:13]

Actually committed as r10804

[robertphillips, 2013-08-28 13:50:51]

Merged into Chromium M30 branch at r10962
(https://code.google.com/p/skia/source/detail?r=10962)