Decode entire in-memory file for WebAudio

media/filters/audio_file_reader.h

Index: media/filters/audio_file_reader.h
diff --git a/media/filters/audio_file_reader.h b/media/filters/audio_file_reader.h
index 1e42abd7b018fde25b2d6fce976fa26807de0b63..126a70bc64e6b1c95332f5008effa358d55d827b 100644
--- a/media/filters/audio_file_reader.h
+++ b/media/filters/audio_file_reader.h
@@ -6,6 +6,7 @@
 #define MEDIA_FILTERS_AUDIO_FILE_READER_H_
 
 #include <memory>
+#include <vector>
 
 #include "base/macros.h"
 #include "media/base/audio_codecs.h"
@@ -38,14 +39,19 @@ class MEDIA_EXPORT AudioFileReader {
   bool Open();
   void Close();
 
-  // After a call to Open(), attempts to fully fill |audio_bus| with decoded
-  // audio data.  Any unfilled frames will be zeroed out.
-  // |audio_data| must be of the same size as channels().
-  // The audio data will be decoded as floating-point linear PCM with
-  // a nominal range of -1.0 -> +1.0.
-  // Returns the number of sample-frames actually read which will always be
-  // <= audio_bus->frames()
-  int Read(AudioBus* audio_bus);
+  // Returns true if (an estimated) duration of the audio data is
+  // known.  Must be called after Open();
+  bool HasKnownDuration() const;
+
+  // After a call to Open(), attempts to decode the entire data,
+  // updating |decodedAudioPackets| with each decoded packet in order.
+  // The caller must convert these packets into one complete set of
+  // decoded audio data.  The audio data will be decoded as
+  // floating-point linear PCM with a nominal range of -1.0 -> +1.0.
+  // Returns the number of sample-frames actually read which will
+  // always be the total size of all the frames in
+  // |decodedAudioPackets|.
+  int Read(std::vector<std::unique_ptr<AudioBus>>& decodedAudioPackets);

[DaleCurtis, 2017/02/09 18:47:36]

Pass by pointer instead of ref. decoded_audio_packets. I think using a list here
might be better to avoid doubling behavior that comes with the vector. I don't
think there's any benefit for the colocation of elements that the vector
provides in this case.

Don't have any data to back that last statement up though, so up to you.

[Raymond Toy, 2017/02/09 18:56:12]

Good point. I expect the size of this vector to be quite small compared to the
actual decoded audio data.

[Raymond Toy, 2017/02/09 21:45:06]

Passing by pointer.

Decided to keep a vector because it's pretty simple.  I think a vector of 1000
elements would hold about a million audio frames, so it's pretty small in
comparison.

 
   // These methods can be called once Open() has been called.
   int channels() const { return channels_; }