Add an ALSA output to the motown audio server.

services/media/audio/platform/linux/alsa_output.h

Index: services/media/audio/platform/linux/alsa_output.h
diff --git a/services/media/audio/platform/linux/alsa_output.h b/services/media/audio/platform/linux/alsa_output.h
new file mode 100644
index 0000000000000000000000000000000000000000..d559498afb6272bc1c6ea5c7589aba269adcbae9
--- /dev/null
+++ b/services/media/audio/platform/linux/alsa_output.h
@@ -0,0 +1,65 @@
+// Copyright 2015 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef SERVICES_MEDIA_AUDIO_PLATFORM_LINUX_ALSA_OUTPUT_H_
+#define SERVICES_MEDIA_AUDIO_PLATFORM_LINUX_ALSA_OUTPUT_H_
+
+#include <alsa/asoundlib.h>
+#include <memory>
+
+#include "mojo/services/media/common/cpp/linear_transform.h"
+#include "mojo/services/media/common/interfaces/media_types.mojom.h"
+#include "services/media/audio/audio_output.h"
+#include "services/media/audio/platform/generic/standard_output_base.h"
+
+namespace mojo {
+namespace media {
+namespace audio {
+
+class AlsaOutput : public StandardOutputBase {
+ public:
+  static AudioOutputPtr New(AudioOutputManager* manager);
+  ~AlsaOutput() override;
+
+  MediaResult Configure(LpcmMediaTypeDetailsPtr config);

[jeffbrown, 2015/11/04 20:33:43]

I wonder whether these Mojo objects should be flowing down to this level of
abstraction.  Does it make sense to keep them at the interface boundary alone
and unpack the relevant bits into other structures you use internally?  (I was
debating this earlier with the View Manager too and ended up doing essentially
what you did here and not caring about the boundary.  Still not sure what's
best.)

[johngro, 2015/11/06 20:22:07]

I think that this is proper.  LpcmMediaTypeDetails is currently a structure
which is defined in mojom and meant to be part of the formal interface
definition for the media system.  It seems natural to use the same
structures/constants which you defined for your external interfaces for your
internal interfaces.  If you cannot, then it seems to imply that something is
either wrong with your interface, or the bindings generated by your IDL.

This said, these structures are a PITA to use (because our C++ structure
bindings are generally a PITA to use).  I *really* didn't want to use them at
all because they annoy me so much.  In the end, I did, because I decided that
while these are a pain; duplicating the structure contents and maintaining both
sides as the structure evolves was actually a sin.

Note: I have committed this sin when describing format capabilities in
audio_track_impl.cc.  My confession is in a comment at the top and my penance is
to replace all of that with something better (around the topic of format
negotiation) someday.

+
+ protected:
+  explicit AlsaOutput(AudioOutputManager* manager);
+
+  // AudioOutput implementation
+  MediaResult Init() override;
+  void Cleanup() override;
+
+  // StandardOutputBase implementation
+  bool StartMixJob(MixJob* job, const LocalTime& process_start) override;
+  bool FinishMixJob(const MixJob& job) override;
+
+ private:
+  void HandleAlsaError(snd_pcm_sframes_t code);
+  void HandleAsError(snd_pcm_sframes_t code);
+  void HandleAsUnderflow();
+
+  snd_pcm_t* alsa_device_ = nullptr;
+  snd_pcm_format_t alsa_format_;
+
+  LinearTransform::Ratio frames_per_tick_;
+  uint8_t silence_byte_;
+
+  std::unique_ptr<uint8_t> mix_buf_;
+  uint32_t mix_buf_bytes_;
+  uint32_t mix_buf_frames_;
+
+  bool primed_ = false;
+  bool local_to_output_known_ = false;
+  LinearTransform local_to_output_;
+  uint32_t local_to_output_gen_ = MixJob::INVALID_GENERATION + 1;
+  int64_t frames_sent_;
+};
+
+}  // namespace audio
+}  // namespace media
+}  // namespace mojo
+
+#endif  // SERVICES_MEDIA_AUDIO_PLATFORM_LINUX_ALSA_OUTPUT_H_
+