Add a conservativelyContainsRect function to SkPath.

Add a conservativelyContainsRect function to SkPath.
Committed: https://code.google.com/p/skia/source/detail?r=6411

[reed1, 2012-11-13 16:43:59]

The unit test seems very daunting to review for correctness. Can it be

1. more table driven (at least collect all the rect data in one or more arrays)?
2. is there some more formulaic way to generate paths and rects and test for
containment, possibly using regions?
3. what about explicit tests w/ round-rects?

https://codereview.appspot.com/6852044/diff/1004/src/core/SkPath.cpp
File src/core/SkPath.cpp (right):

https://codereview.appspot.com/6852044/diff/1004/src/core/SkPath.cpp#newcode313
src/core/SkPath.cpp:313: 
const SkPoint&

[reed1, 2012-11-13 16:45:31]

1. make a utility for now, so as not to expose it to chrome/android yet?
2. any notion how much of the perf is in the switch/bookkeeping -vs- the
cross-products?
3. do we want a bench?

[reed1, 2012-11-13 16:46:03]

the edge_cross-rect code is very clever and fast looking. nice job.

[reed1, 2012-11-13 16:48:20]

https://codereview.appspot.com/6852044/diff/1004/include/core/SkPath.h
File include/core/SkPath.h (right):

https://codereview.appspot.com/6852044/diff/1004/include/core/SkPath.h#newcod...
include/core/SkPath.h:306: * will ever return true for single convex contour
paths.
How does it treat edge sharing? Does a rect contains itself?

[bsalomon, 2012-11-13 17:40:24]

On 2012/11/13 16:43:59, reed1 wrote:
> The unit test seems very daunting to review for correctness. Can it be
> 
> 1. more table driven (at least collect all the rect data in one or more
arrays)?

Will do.

> 2. is there some more formulaic way to generate paths and rects and test for
> containment, possibly using regions?
??

> 3. what about explicit tests w/ round-rects?

There are round rect tests in the current test.

> 
> https://codereview.appspot.com/6852044/diff/1004/src/core/SkPath.cpp
> File src/core/SkPath.cpp (right):
> 
>
https://codereview.appspot.com/6852044/diff/1004/src/core/SkPath.cpp#newcode313
> src/core/SkPath.cpp:313: 
> const SkPoint&

[bsalomon, 2012-11-13 17:42:02]

On 2012/11/13 16:48:20, reed1 wrote:
> https://codereview.appspot.com/6852044/diff/1004/include/core/SkPath.h
> File include/core/SkPath.h (right):
> 
>
https://codereview.appspot.com/6852044/diff/1004/include/core/SkPath.h#newcod...
> include/core/SkPath.h:306: * will ever return true for single convex contour
> paths.
> How does it treat edge sharing? Does a rect contains itself?

Yes, sharing is allowed (a rect contains itself). Empty rects are allowed. Will
update the dox.

[bsalomon, 2012-11-13 21:22:57]

Added benches, rewrote unit test, updated comments.

In the oval bench roughly 15% of the new func's is spent in RawIter::next() and
the almost all the rest is the edge/rect test.

https://codereview.appspot.com/6852044/diff/1004/src/core/SkPath.cpp
File src/core/SkPath.cpp (right):

https://codereview.appspot.com/6852044/diff/1004/src/core/SkPath.cpp#newcode313
src/core/SkPath.cpp:313: 
On 2012/11/13 16:44:00, reed1 wrote:
> const SkPoint&

Done.

[reed1, 2012-11-13 21:47:19]

lgtm