Fix hang in Android WebView release build tests

Fix hang in Android WebView release build tests

SurfaceTexture.detachFromGLContext is not present in ICS and the old
workaround of using a NullCompositor lead to the test creating
TextureImageTransportSurface in ImageTransportSurface::CreateSurface.
This path is wrong and not well tested in single-process mode and recently
lead to a deadlock hang in release build only tests (see bug).

Fixed by creating the real compositor but just avoid calling the missing
method in ICS. The Java method is loaded lazily so this does not cause
an error at run time.


BUG=167236, 161864
Android only change. Ran through Android trybots.
NOTRY=true

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=175480

[boliu, 2013-01-07 23:01:34]

Unfortunately there are no release trybots yet, but this works for me locally.

[no sievers, 2013-01-07 23:28:27]

On 2013/01/07 23:01:34, boliu wrote:
> Unfortunately there are no release trybots yet, but this works for me locally.

What about AttachToGLContext()? Isn't that getting triggered from compositing
without the null compositor?

[boliu, 2013-01-07 23:35:44]

On 2013/01/07 23:28:27, Daniel Sievers wrote:
> On 2013/01/07 23:01:34, boliu wrote:
> > Unfortunately there are no release trybots yet, but this works for me
locally.
> 
> What about AttachToGLContext()? Isn't that getting triggered from compositing
> without the null compositor?

Weirdly, AttachToGLContext is not tripped in tests, but added the check anyway.
Any others that you know of?

And I'll wait for my local tests to finish before I spam trybots again.

[boliu, 2013-01-07 23:49:53]

On 2013/01/07 23:35:44, boliu wrote:
> On 2013/01/07 23:28:27, Daniel Sievers wrote:
> > On 2013/01/07 23:01:34, boliu wrote:
> > > Unfortunately there are no release trybots yet, but this works for me
> locally.
> > 
> > What about AttachToGLContext()? Isn't that getting triggered from
compositing
> > without the null compositor?
> 
> Weirdly, AttachToGLContext is not tripped in tests, but added the check
anyway.
> Any others that you know of?
> 
> And I'll wait for my local tests to finish before I spam trybots again.

Actually I'll wait for l-g-t-m's since android trybot is pretty backed up.

[no sievers, 2013-01-08 00:35:12]

On 2013/01/07 23:49:53, boliu wrote:
> On 2013/01/07 23:35:44, boliu wrote:
> > On 2013/01/07 23:28:27, Daniel Sievers wrote:
> > > On 2013/01/07 23:01:34, boliu wrote:
> > > > Unfortunately there are no release trybots yet, but this works for me
> > locally.
> > > 
> > > What about AttachToGLContext()? Isn't that getting triggered from
> compositing
> > > without the null compositor?
> > 
> > Weirdly, AttachToGLContext is not tripped in tests, but added the check
> anyway.
> > Any others that you know of?
> > 
> > And I'll wait for my local tests to finish before I spam trybots again.
> 
> Actually I'll wait for l-g-t-m's since android trybot is pretty backed up.

lgtm. should we print a warning once somewhere that it won't render on older
sdks?

[boliu, 2013-01-08 00:45:35]

On 2013/01/08 00:35:12, Daniel Sievers wrote:
> lgtm. should we print a warning once somewhere that it won't render on older
> sdks?

Good point. Factored out the check into an anonymous function in
surface_texture_bridge.cc, and added the LOG(WARNING) if check fails.

[joth, 2013-01-08 01:54:57]

Thanks for getting to the root of  this Bo.

On 2013/01/07 23:35:44, boliu wrote:
> On 2013/01/07 23:28:27, Daniel Sievers wrote:
> > On 2013/01/07 23:01:34, boliu wrote:
> > > Unfortunately there are no release trybots yet, but this works for me
> locally.
> > 
> > What about AttachToGLContext()? Isn't that getting triggered from
compositing
> > without the null compositor?
> 
> Weirdly, AttachToGLContext is not tripped in tests, but added the check
anyway.

I expect this is because the ICS bots aren't running with the GL functor setup
(we only do that in the downstream tree) and that would be needed to get the
attach call.
(OTOH Detach is probably called in all cases in some destructor chain)


> Any others that you know of?
> 
> And I'll wait for my local tests to finish before I spam trybots again.

[boliu, 2013-01-08 01:56:36]

Where's the rubberstamp?

[joth, 2013-01-08 02:10:43]

lgtm

[commit-bot: I haz the power, 2013-01-08 03:08:27]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/boliu@chromium.org/11787031/9001

[commit-bot: I haz the power, 2013-01-08 03:08:56]

Change committed as 175480