Remove incorrect GRAssert calls in GrResourceCache::removeAll

Remove incorrect GrAssert calls in GrResourceCache::removeAll

The asserts in GrResourceCache are too aggressive, and assume that all resources are always purged.  Locked resources will not be purged.

BUG=114861
Committed: https://code.google.com/p/skia/source/detail?r=3226

[twiz, 2012-02-18 01:59:35]

Brian, PTAL.

I think that the assumptions in GrResouceCache::removeAll are not correct, as
I've observed failures where the cache is non-empty after a purge state:
 fUnlockedEntryCount = 0
 fEntryCount = 1
 fHead, fTail = ...<some pointer>
 fClientDetachedCount = 0
 fClientDetachedBytes = 0

Is this a valid configuration?  Can a resource be locked, yet not detached?

[bsalomon, 2012-02-21 18:36:58]

On 2012/02/18 01:59:35, twiz wrote:
> Brian, PTAL.
> 
> I think that the assumptions in GrResouceCache::removeAll are not correct, as
> I've observed failures where the cache is non-empty after a purge state:
>  fUnlockedEntryCount = 0
>  fEntryCount = 1
>  fHead, fTail = ...<some pointer>
>  fClientDetachedCount = 0
>  fClientDetachedBytes = 0
> 
> Is this a valid configuration?  Can a resource be locked, yet not detached?

You're right, the assumption is wrong; resources that are locked and undetached
will still be present in the cache.

LGTM, but it'd also be nice to do some verification in that case (e.g. all the
resources remaining are locke, the bytes add up, ...). You can file a Skia Issue
against me if you like for that.

[twiz, 2012-02-21 19:34:15]

On 2012/02/21 18:36:58, bsalomon wrote:
> On 2012/02/18 01:59:35, twiz wrote:
> > Brian, PTAL.
> > 
> > I think that the assumptions in GrResouceCache::removeAll are not correct,
as
> > I've observed failures where the cache is non-empty after a purge state:
> >  fUnlockedEntryCount = 0
> >  fEntryCount = 1
> >  fHead, fTail = ...<some pointer>
> >  fClientDetachedCount = 0
> >  fClientDetachedBytes = 0
> > 
> > Is this a valid configuration?  Can a resource be locked, yet not detached?
> 
> You're right, the assumption is wrong; resources that are locked and
undetached
> will still be present in the cache.
> 
> LGTM, but it'd also be nice to do some verification in that case (e.g. all the
> resources remaining are locke, the bytes add up, ...). You can file a Skia
Issue
> against me if you like for that.

Issue added:  http://code.google.com/p/skia/issues/detail?id=504