Make SkImage_Gpu share it's pixelref with the surface to prevent premature return to scratch pool.

Make SkImage_Gpu share it's pixelref with the surface to prevent premature return to scratch pool.

BUG=crbug.com/263329
TEST=Surface unit test, function Test_crbug263329

Committed: http://code.google.com/p/skia/source/detail?r=10378

[Justin Novosad, 2013-07-25 17:18:53]

PTAL

[bsalomon, 2013-07-25 17:26:50]

lgtm

[reed1, 2013-07-25 17:30:17]

hmmm, is this the only way to fix it? I find SkBitmap a bit tricky, and was
hoping to have SkImage be a clean break from that...

[Justin Novosad, 2013-07-25 18:01:51]

On 2013/07/25 17:30:17, reed1 wrote:
> hmmm, is this the only way to fix it? I find SkBitmap a bit tricky, and was
> hoping to have SkImage be a clean break from that...

We could pass a SkPixelRef* instead of a SkBitmap.

Ideally, we should re-architect the scratch texture system to make GrTextures
responsible for their own insertion into the scratch texture pool, as opposed to
SkGrPixelRef that takes car of it now.  But that change is much deeper, and for
now I need a change that is safe and clean so that it can be merged to M29.

[reed1, 2013-07-25 18:14:13]

How about we add a big comment in the public header stating something like:

/**
 * GrTexture is a more logical parameter for this factory, but its interactions
 * with scratch cache still has issues, so for now we take SkBitmap instead.
 * This will be changed in the future.
 */

So we are reminded to fix this, and callers are fore-warned

[Justin Novosad, 2013-07-25 18:30:14]

On 2013/07/25 18:14:13, reed1 wrote:
> How about we add a big comment in the public header stating something like:
> 
> /**
>  * GrTexture is a more logical parameter for this factory, but its
interactions
>  * with scratch cache still has issues, so for now we take SkBitmap instead.
>  * This will be changed in the future.
>  */
> 
> So we are reminded to fix this, and callers are fore-warned

Done.

Also, I think these factory methods do not need to be exposed in the public API.
They should only ever need to be called by SkSurface (and maybe unit tests)

[bsalomon, 2013-07-25 21:25:27]

On 2013/07/25 18:30:14, junov wrote:
> On 2013/07/25 18:14:13, reed1 wrote:
> > How about we add a big comment in the public header stating something like:
> > 
> > /**
> >  * GrTexture is a more logical parameter for this factory, but its
> interactions
> >  * with scratch cache still has issues, so for now we take SkBitmap instead.
> >  * This will be changed in the future.
> >  */
> > 
> > So we are reminded to fix this, and callers are fore-warned
> 
> Done.
> 
> Also, I think these factory methods do not need to be exposed in the public
API.
> They should only ever need to be called by SkSurface (and maybe unit tests)

lgtm

[commit-bot: I haz the power, 2013-07-25 21:26:12]

CQ is trying da patch. Follow status at
https://skia-tree-status.appspot.com/cq/junov@chromium.org/20354003/28001

[commit-bot: I haz the power, 2013-07-25 23:29:42]

Change committed as 10378