Properly reject clipped out draws in BBox pictures.

Properly reject clipped out draws in BBox pictures.

[bungeman, 2013-01-09 23:37:05]

This fixes Issue 1046 [ http://code.google.com/p/skia/issues/detail?id=1046
](rtree and tilegrid GM tests are failing on Win7 DirectWrite). quickReject
actually end up calling getClipBounds to do most of it's work, but also does
caching and handles perspective transformations correctly.

[reed1, 2013-01-10 14:43:19]

lgtm

1. did you locally run gm w/ this change?
2. any idea if this change affects the perf-time for recording rtrees?

[junov, 2013-01-10 14:59:33]

On 2013/01/09 23:37:05, bungeman wrote:
> quickReject
> actually end up calling getClipBounds to do most of it's work, but also does
> caching and handles perspective transformations correctly.

Good find!

So the fact that this only failed with DirectWrite was a fluke?

[bungeman, 2013-01-10 16:35:53]

On 2013/01/10 14:43:19, reed1 wrote:
> lgtm
> 
> 1. did you locally run gm w/ this change?

Yes, this fixes the problem observed in Issue 1046 and everything else still
looks good (unchanged).

> 2. any idea if this change affects the perf-time for recording rtrees?

I've run

out\Release\bench_pictures.exe out\gmskp --repeat 200 --mode record --bbh rtree
--timers w

both with and without this change. 'out\gmskp' contained one .skp files from
each gm. Any change in time is less than the noise level.

[bungeman, 2013-01-10 16:40:38]

On 2013/01/10 14:59:33, junov wrote:
> On 2013/01/09 23:37:05, bungeman wrote:
> > quickReject
> > actually end up calling getClipBounds to do most of it's work, but also does
> > caching and handles perspective transformations correctly.
> 
> Good find!
> 
> So the fact that this only failed with DirectWrite was a fluke?

Looks like it. DirectWrite probably gives the tightest bounds, so the size and
position of the text and being in perspective just happened to give a set of
numbers which caused getClipBounds to end up confused. getClipBounds doesn't
appear to handle perspective matrices well.

[bungeman, 2013-01-10 17:16:11]

Committed revision 7117.

Additional change needed to avoid assertion that empty rects are not passed
(though not sure assertion is needed) committed revision 7118.