Added isIntersectionOfRects to SkClipStack

This replicates a small chunk of GrClip's functionality in SkClipStack

[reed1, 2012-07-24 20:30:10]

lgtm

1. are you saving the return value of getBounds for some awesome other value?
2. could you extend the enum to include game-of-thrones/intersection-of-rects
facet, rather than a var parameter?

[robertphillips, 2012-07-24 20:50:07]

1) Nope - it has just evolved that way. If we adopt my proposal for #2,
returning the 'BoundBits' could make sense. 

2) If we wanted to move in this direction I think the clearest thing to do would
be to make the return be something like 'BoundBits' where we could just OR
together everything we know about the bound.

[bsalomon, 2012-07-24 20:50:23]

On 2012/07/24 20:30:10, reed1 wrote:
> lgtm
> 
> 1. are you saving the return value of getBounds for some awesome 
other value?

I had earlier objected to returning a bool from this function that doesn't mean
success/failure. I thought it would be too easy to confuse it with a
SkMatrix::invert() pattern where false means failure and the result param wasn't
updated. This seemed especially since the clipstack can be unbounded (in which
case a reasonable expectation could be failure.) I thought callers should be
really clear that the function doesn't fail on infinite clips but rather they
get the bounds of the unwritable area.





> 2. could you extend the enum to include game-of-thrones/intersection-of-rects
> facet, rather than a var parameter?

[robertphillips, 2012-07-24 22:08:05]

committed as r4745

[robertphillips, 2012-07-26 15:38:43]

Any thoughts on returning bit flags from getBound?

[bsalomon, 2012-07-26 16:24:56]

On 2012/07/26 15:38:43, robertphillips wrote:
> Any thoughts on returning bit flags from getBound?

Seems like an OK idea to me. Another approach would be to add bool isRect() as a
separate function.

[reed1, 2012-07-26 17:57:23]

does isIntersectionOfRects mean that the finite-bounds is the exact answer, or
just that there were no paths involved, or something else?

Can we store and return a bitfield to encompass bounds-type and the above
boolean?

Rec is getting very big. We should be sure that the extra fields that are
"caches" are worth the extra storage/copying costs.

[robertphillips, 2012-07-26 18:46:46]

isIntersectionOfRects means that the finite-bounds is the exact answer.
Currently it is only set if no paths were involved and only Replace & Intersect
ops were used.

I would hope the bitfield would include both the bound type enum and some extra
flags about the bound (e.g., isIntersectionOfRects) both in Rec and as a return
value.