Add an ALSA output to the motown audio server.

services/media/audio/audio_output_manager.cc

Index: services/media/audio/audio_output_manager.cc
diff --git a/services/media/audio/audio_output_manager.cc b/services/media/audio/audio_output_manager.cc
index 59e4ff38a2052c2a748e307355851b6f87412eb3..adbd744fdcbdffae5d114e331418788e4ce63b07 100644
--- a/services/media/audio/audio_output_manager.cc
+++ b/services/media/audio/audio_output_manager.cc
@@ -9,6 +9,9 @@
 #include "services/media/audio/audio_server_impl.h"
 #include "services/media/audio/audio_track_to_output_link.h"
 #include "services/media/audio/platform/generic/throttle_output.h"
+#if defined(OS_LINUX)
+#include "services/media/audio/platform/linux/alsa_output.h"
+#endif
 
 namespace mojo {
 namespace media {
@@ -27,6 +30,36 @@ AudioOutputManager::~AudioOutputManager() {
   DCHECK(!thread_pool_);
 }
 
+void AudioOutputManager::CreateAlsaOutputs() {
+#if defined(OS_LINUX)

[jeffbrown, 2015/11/04 20:33:42]

This to me looks like an opportunity to define a new class to abstract platform
audio sink concerns.

[johngro, 2015/11/06 20:22:06]

Acknowledged.

We should talk sometime.  I'd like to understand more about your approach to
doing this.  

+  // TODO(johngro): Do better than this.  If we really want to support
+  // Linux/ALSA as a platform, we should be creating one output for each
+  // physical output in the system, matching our configuration to the physical
+  // output's configuration, and disabling resampling at the ALSA level.
+  //
+  // If we could own the output entirely and bypass the mixer to achieve lower
+  // latency, that would be even better.
+  AudioOutputPtr audio_out(audio::AlsaOutput::New(this));
+  if (!audio_out) { return; }
+
+  audio::AlsaOutput* alsa_out =
+    static_cast<audio::AlsaOutput*>(audio_out.get());

[jeffbrown, 2015/11/04 20:33:42]

Did you "git cl format"?  I think this is supposed to be 4 spaces or reflow into
something equally hideous.

[johngro, 2015/11/06 20:22:06]

Done.

Solved this a different way.  A long time ago, I used to have the output manager
in just mojo::media, and the audio:: scope was needed, but it no longer is. 
Removing that makes stuff fit on one line.

Oddly enough, I cannot find clear guidance in the public style guide about what
to do if an assignment (or any other operation) exceeds the 80 col length.  I'm
not sure what git CL format chooses to do, but AFAICT there are not a lot of
strict rules here besides use your best judgement and be consistent.

+  DCHECK(alsa_out);
+
+  LpcmMediaTypeDetailsPtr config(LpcmMediaTypeDetails::New());
+  DCHECK(config);

[jeffbrown, 2015/11/04 20:33:42]

Might be able to omit this DCHECK.  Seems like the sort of thing the Mojo
bindings should do internally whenever New() fails.  (I really hate checking the
result of new...)

[johngro, 2015/11/06 20:22:06]

The problem here is that this is not a call to operator new().  It's a call to a
function someone wrote called New, which (in theory) could have extra side
effects.  IOW - its great that the mojo/chrome global new has been overloaded to
DCHECK if allocation fails, but this function could succeed to allocate and
still return NULL.

In this case, however, I know that this is machine generated code which should
Never Do This, so I'll remove the check.

+  config->frames_per_second = 48000;
+  config->samples_per_frame = 2;
+  config->sample_format = LpcmSampleFormat::SIGNED_16;
+
+  if (alsa_out->Configure(config.Pass()) != MediaResult::OK) {
+    return;
+  }
+
+  outputs_.emplace(audio_out);
+#endif
+}
+
 MediaResult AudioOutputManager::Init() {
   // Step #1: Initialize the mixing thread pool.
   //
@@ -43,6 +76,7 @@ MediaResult AudioOutputManager::Init() {
   // platform.  Right now, we just create some hardcoded default outputs and
   // leave it at that.
   outputs_.emplace(audio::ThrottleOutput::New(this));
+  CreateAlsaOutputs();
 
   // Step #3: Being monitoring for plug/unplug events for pluggable audio
   // output devices.