Mac h264 Flash: Work around scoped binder incompatibility.

Mac h264 Flash: Work around scoped binder incompatibility.

gfx::ScopedTextureBinder has an optimization where it restores the state
from the ContextState, instead of querying the GL driver.

This backfires when GL_ACTIVE_TEXTURE has changed, because
gfx::ScopedTextureBinder will re-bind whatever texture the ContextState
thinks was active, which will not be accurate, because it will not be
updated gfx::ScopedActiveTexture.

Fixing this will require that some of the scoped binders be
fundamentally re-evaluated. In the short-term, do a manual re-bind of
the relevant context state.

BUG=599314

[ccameron, 2016-04-08 09:01:57]

ccameron@chromium.org changed reviewers:
+ dcastagna@chromium.org, reveman@chromium.org

[ccameron, 2016-04-08 09:01:57]

The alternatives that I see are as follows:
1. a "scoped don't don't restore from ContextState" object
2. an implicit "don't restore from ContextState" when setting active texture
3. having active texture binders modify ContextState

I'm quite happy to put together any of those 3, if they are preferred.

[reveman, 2016-04-08 12:43:35]

Is GLImageIOSurface the only code that is affected by this problem?

dcastagna@, I would prefer if we fixed this in whatever way we think is the best
long term solution without temporary workarounds but I'll let you make the call
on what's the best solution as you're the author of these scoped classes.

[Daniele Castagna, 2016-04-08 22:26:30]

On 2016/04/08 at 12:43:35, reveman wrote:
> Is GLImageIOSurface the only code that is affected by this problem?
> 
Likely, those two binders are not used in many places.

> dcastagna@, I would prefer if we fixed this in whatever way we think is the
best long term solution without temporary workarounds but I'll let you make the
call on what's the best solution as you're the author of these scoped classes.

Should we try to just avoid using the GLRestorer? crrev.com/1870273002

[ccameron, 2016-04-09 00:30:28]

On 2016/04/08 22:26:30, Daniele Castagna wrote:
> On 2016/04/08 at 12:43:35, reveman wrote:
> > Is GLImageIOSurface the only code that is affected by this problem?
> > 
> Likely, those two binders are not used in many places.
> 
> > dcastagna@, I would prefer if we fixed this in whatever way we think is the
> best long term solution without temporary workarounds but I'll let you make
the
> call on what's the best solution as you're the author of these scoped classes.
> 
> Should we try to just avoid using the GLRestorer? crrev.com/1870273002

Closing this in favor of https://codereview.chromium.org/1870323002. That patch
fixes the scoped binding issue and the memory leak issues.