Refactoring RTree integration to support SkBBoxHierarchy polymorphism in SkPicture.

Refactoring RTree integration to support SkBBoxHierarchy polymorphism in SkPicture.
This moves the rtree creation into a virtual method.
Committed: https://code.google.com/p/skia/source/detail?r=6242

[junov1, 2012-10-30 19:12:23]

PTAL

https://codereview.appspot.com/6811057/diff/1/gyp/utils.gyp
File gyp/utils.gyp (right):

https://codereview.appspot.com/6811057/diff/1/gyp/utils.gyp#newcode18
gyp/utils.gyp:18: '../src/core',
I admit this is yucky. Any better ideas for accessing SkTSort.h from utils? If
we really don't like this, I can cancel the moving of SkRTree to utils.

[reed1, 2012-10-30 19:19:33]

It doesn't seem wrong to move SkRTree into utils, though lets not make its
header public if we can help it (everything in src/utils?).

I think we should keep the high-level hint (flag bit) in the public API for
picture recording for now. I understand the notion of wanting a plug-n-play
structure so we can try out different implementations, but lets find a way to
keep that under the hood if we can.

https://codereview.appspot.com/6811057/diff/1/include/utils/SkRTree.h
File include/utils/SkRTree.h (right):

https://codereview.appspot.com/6811057/diff/1/include/utils/SkRTree.h#newcode62
include/utils/SkRTree.h:62: */
Esp. if this is now a utility, and not specific to SkPicture, why are we
introducing defaults for min ad max children? The fewer default arguments the
better it seems.

[junov1, 2012-10-31 14:59:50]

New patch.

To keep the SkRTree header private, I used an abstract factory pattern. This way
only the creation interface is public.

[junov1, 2012-10-31 15:19:23]

Patch 3: Changed factory interface

[reed1, 2012-10-31 20:35:44]

Do we really need to make this switchable in the public now? Perhaps this is
correct, but I had not envisioned that we exposed so much publicly. I was hoping
to just have the hint to "optimize" or not (if we even need that), and reserve
this factoring for our own experiments. What do you think? Can we not do all of
our experiments/timings w/ just .skp files and a private api?

[junov1, 2012-11-01 13:58:54]

Ok, then I have to put SkRTree back in core since core can't depend on utils.
Perhaps we could use an SkConfig macro to chose a BBoxHierarchy implementation,
rather than an API method?

[junov1, 2012-11-01 15:20:22]

Patch

[reed1, 2012-11-01 16:53:50]

lgtm