Make DecodeMemoryToTarget handle more configs.

Make DecodeMemoryToTarget handle more configs.

In SkImageDecoder::DecodeMemoryToTarget, if SkBitmapToImageInfo fails due to an unsupported config, copy to 8888.

Needed for https://codereview.chromium.org/15145004/ and
https://codereview.chromium.org/12851012/ for all bitmaps to be
decoded properly in render_pictures and bench_pictures.

R=reed@google.com

Committed: https://code.google.com/p/skia/source/detail?r=9196

[scroggo, 2013-05-14 21:38:00]

New approach.

[reed1, 2013-05-15 20:12:23]

wrap the call(s) to SkImageDecoder to never return kIndex8 (but return 8888
instead)? That way the main logic can just proceed, blissfully unaware that
sometimes there is a copyTo happening behind the scenes...

[scroggo, 2013-05-17 13:40:09]

On 2013/05/15 20:12:23, reed1 wrote:
> wrap the call(s) to SkImageDecoder to never return kIndex8 (but return 8888
> instead)? That way the main logic can just proceed, blissfully unaware that
> sometimes there is a copyTo happening behind the scenes...

New patch up. Not quite as simple as we might have hoped, but the decode calls
are wrapped (by separate functions).

[reed1, 2013-05-20 14:03:24]

lgtm, with suggestion for generalizing your allocator for future use (could be
separate CL or ignored)

https://codereview.chromium.org/14772035/diff/12001/src/images/SkImageDecoder...
File src/images/SkImageDecoder.cpp (right):

https://codereview.chromium.org/14772035/diff/12001/src/images/SkImageDecoder...
src/images/SkImageDecoder.cpp:292: TargetAllocator(void* target)
For more generality (not trying to derail the CL...)

- pass in a length and config
- then can use those to decide when to dynamically allocate, rather than
hard-coding 8888

[scroggo, 2013-05-20 16:58:21]

Committed patchset #5 manually as r9196 (presubmit successful).