Add audio_device_id in mojo media ServiceFactory CreateRenderer api.

media/mojo/interfaces/renderer.mojom

Index: media/mojo/interfaces/renderer.mojom
diff --git a/media/mojo/interfaces/renderer.mojom b/media/mojo/interfaces/renderer.mojom
index 1318bad6d0c03eb41f6e23b6de453388ee02de16..f3ef189cf28fbf2fa00667597edef449ced63fc5 100644
--- a/media/mojo/interfaces/renderer.mojom
+++ b/media/mojo/interfaces/renderer.mojom
@@ -11,9 +11,12 @@ import "ui/gfx/geometry/mojo/geometry.mojom";
 interface Renderer {
   // Initializes the Renderer with one or both of an audio and video stream,
   // executing the callback with whether the initialization succeeded.
+  // Audio stream will be played on device_id, which is defined in
+  // media/audio/audio_device_description.h.
   Initialize(RendererClient client,
              DemuxerStream? audio,
-             DemuxerStream? video) => (bool success);
+             DemuxerStream? video,
+             string device_id) => (bool success);

[alokp, 2016/05/27 04:43:08]

It is possible to target a particular device when using
AudioManager::MakeAudioOutputStream*. In that sense I think it is reasonable to
specify the audio device. The only weird thing is that unlike AudioManager, this
interface does not allow querying for available device IDs.

Perhaps it may not be that weird once mojo:audio service is available? I
understand WebRTC intern is working on bringing it up. Tommi?

[xhwang, 2016/05/27 06:06:35]

+dalecurtis

I am not super familiar with audio, but it seems a bit odd for this interface.

Let me ask this question, if everything lives in the same process and we don't
need mojo. How will we choose which audio stream will be played on which
device_id. This will help me better understand how things should work. My gut
feeling is that we should have some separate channel to pass this information,
but I could be wrong.

[alokp, 2016/05/27 18:31:01]

In that case you would provide an AudioRendererSink for the AudioRenderer to
render to. The default subclass of AudioRendererSink is AudioOutputDevice, which
lets you specify audio device id.

Since the mojo interface does not deal with AudioRenderer or AudioRendererSink,
it should let the clients specify the audio-device id somehow. This is not an
issue for the renderer process because it always renders to the default device.
But it is important for chirp - they want to render TTS sounds to communications
device so that other sounds are attenuated.

 
   // Discards any buffered data, executing callback when completed.
   // NOTE: If an error occurs, RendererClient::OnError() can be called