Deliver RenderCallbackErrors even when mixer inputs are paused.

Deliver RenderCallbackErrors even when mixer inputs are paused.

See the bug for more details.  Fixes potential stuck playback.

- Separates error callback registration from mixer input addition.
- Removes |mixer_inputs_| in favor of letting AudioConverter manage.
- Adds a new unittest to verify paused inputs receive errors.

This fixes the problem implicitly because AudioRendererImpl will
deliver a decode error which shuts down the pipeline.

BUG=379374
TEST=new unittest.

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=275585

[DaleCurtis, 2014-05-31 01:06:57]

Alternate solution considered: Tracking an error state inside AudioRendererMixer
and shutting down the sink immediately upon error.  Becomes messy for a few
reason, but mainly since now AudioRendererMixerManager needs to check this error
state and track it separately from "good" mixers.

[scherkus (not reviewing), 2014-06-03 02:01:31]

few questions -- also not super familiar w/ the code so might have to chat
tomorrow

out of curiosity did you actually run into the stuck audio issue, or is this
something you noticed by inspection?

https://codereview.chromium.org/301223012/diff/1/media/base/audio_renderer_mi...
File media/base/audio_renderer_mixer.cc (right):

https://codereview.chromium.org/301223012/diff/1/media/base/audio_renderer_mi...
media/base/audio_renderer_mixer.cc:56: AudioConverter::InputCallback* input,
would using a map/list/vector work? there doesn't seem to be a real need to tie
|input| with the callback

do we expect the size of these data structures to be really large?

[DaleCurtis, 2014-06-03 20:26:19]

cjohn@ reported seeing this with Play music when killing pulseaudio while
paused; see http://b/15346061 --  I verified this was the root cause.

https://codereview.chromium.org/301223012/diff/1/media/base/audio_renderer_mi...
File media/base/audio_renderer_mixer.cc (right):

https://codereview.chromium.org/301223012/diff/1/media/base/audio_renderer_mi...
media/base/audio_renderer_mixer.cc:56: AudioConverter::InputCallback* input,
On 2014/06/03 02:01:31, scherkus wrote:
> would using a map/list/vector work? there doesn't seem to be a real need to
tie
> |input| with the callback
> 
> do we expect the size of these data structures to be really large?

This is using a map so I'm not sure what you're asking.  At the time I wrote
this, I ran into issues putting callbacks in a list/vector and then finding them
again since each placement makes a copy.  It seems to work correctly now though,
so I've switched this to a list and added a note that calls must specify the
same callback to Add/Remove.

[DaleCurtis, 2014-06-06 17:38:02]

scherkus: Ping!

[scherkus (not reviewing), 2014-06-06 21:46:57]

lgtm

[DaleCurtis, 2014-06-06 21:51:18]

The CQ bit was checked by dalecurtis@chromium.org

[commit-bot: I haz the power, 2014-06-06 21:53:08]

CQ is trying da patch. Follow status at
 https://chromium-status.appspot.com/cq/dalecurtis@chromium.org/301223012/40001

[commit-bot: I haz the power, 2014-06-07 01:47:05]

Change committed as 275585