Chromecast: mitigate threading issue between AvPipeline/MediaKeys.

Chromecast: mitigate threading issue between AvPipeline/MediaKeys.

R=lcwu@chromium.org,erickung@chromium.org
BUG=internal b/19550765

Committed: https://crrev.com/36fdd401d567e5c75180610dd6df564fe00c9ffc
Cr-Commit-Position: refs/heads/master@{#318973}

[lcwu1, 2015-03-03 18:52:35]

https://codereview.chromium.org/974623002/diff/1/chromecast/media/cma/pipelin...
File chromecast/media/cma/pipeline/av_pipeline_impl.cc (right):

https://codereview.chromium.org/974623002/diff/1/chromecast/media/cma/pipelin...
chromecast/media/cma/pipeline/av_pipeline_impl.cc:198: // CDM destruction
requires immediate update; don't change threads.
A more detailed explanation of why the unset_cb should be called synchronously
instead of posting it to the media thread is probably warranted. (Especially the
problem that the original asynchronous call caused where the media_keys object
was destroyed before the unset_cb_ got the chance to run to set media_keys_ to
null.)

https://codereview.chromium.org/974623002/diff/1/chromecast/media/cma/pipelin...
chromecast/media/cma/pipeline/av_pipeline_impl.cc:340:
DCHECK(thread_checker_.CalledOnValidThread());
Now that this function can be called from the main thread, won't you hit a
dcheck failure if you still keep the thread checker here?

[gunsch, 2015-03-03 23:04:19]

Testing this actually uncovered another bug: weak-ptr references being used
across threads.

I made AvPipelineImpl ref-counted to avoid this, instead of having it use a
weak-ptr in callbacks.

PTAL.

https://codereview.chromium.org/974623002/diff/1/chromecast/media/cma/pipelin...
File chromecast/media/cma/pipeline/av_pipeline_impl.cc (right):

https://codereview.chromium.org/974623002/diff/1/chromecast/media/cma/pipelin...
chromecast/media/cma/pipeline/av_pipeline_impl.cc:198: // CDM destruction
requires immediate update; don't change threads.
On 2015/03/03 18:52:35, lcwu1 wrote:
> A more detailed explanation of why the unset_cb should be called synchronously
> instead of posting it to the media thread is probably warranted. (Especially
the
> problem that the original asynchronous call caused where the media_keys object
> was destroyed before the unset_cb_ got the chance to run to set media_keys_ to
> null.)

Done.

https://codereview.chromium.org/974623002/diff/1/chromecast/media/cma/pipelin...
chromecast/media/cma/pipeline/av_pipeline_impl.cc:340:
DCHECK(thread_checker_.CalledOnValidThread());
On 2015/03/03 18:52:35, lcwu1 wrote:
> Now that this function can be called from the main thread, won't you hit a
> dcheck failure if you still keep the thread checker here?

Done.

[lcwu1, 2015-03-04 00:06:23]

lgtm

[gunsch, 2015-03-04 00:10:57]

The CQ bit was checked by gunsch@chromium.org

[commit-bot: I haz the power, 2015-03-04 00:11:58]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/974623002/20001

[commit-bot: I haz the power, 2015-03-04 00:15:45]

Committed patchset #2 (id:20001)

[commit-bot: I haz the power, 2015-03-04 00:16:28]

Patchset 2 (id:??) landed as
https://crrev.com/36fdd401d567e5c75180610dd6df564fe00c9ffc
Cr-Commit-Position: refs/heads/master@{#318973}