Make AudioSinkInterface::Data hold a const pointer to the audio buffer.

webrtc/api/call/audio_sink.h

Index: webrtc/api/call/audio_sink.h
diff --git a/webrtc/api/call/audio_sink.h b/webrtc/api/call/audio_sink.h
index e865ead365cdc68e2ec50a2b4a2a0c607a26990d..8d38763f9402e6e6a351c3a051ad5cd72c210e76 100644
--- a/webrtc/api/call/audio_sink.h
+++ b/webrtc/api/call/audio_sink.h
@@ -27,7 +27,7 @@ class AudioSinkInterface {
   virtual ~AudioSinkInterface() {}
 
   struct Data {
-    Data(int16_t* data,
+    Data(const int16_t* data,
          size_t samples_per_channel,
          int sample_rate,
          size_t channels,
@@ -38,7 +38,7 @@ class AudioSinkInterface {
           channels(channels),
           timestamp(timestamp) {}
 
-    int16_t* data;               // The actual 16bit audio data.
+    const int16_t* data;         // The actual 16bit audio data.

[kwiberg-webrtc, 2017/05/10 08:19:51]

Seems like a good change to me---it can't break callers, and it'll only break
AudioSinkInterface implementations that mutate the input buffer, which seems
like a questionable practice...

(Two points that are out f scope for this CL: First, |data| should probably be
an ArrayView instead of a raw pointer; the array length is already implicitly
specified by all the other Data members, but an ArrayView would make things
easier to reason about. Second, what's the point of passing all the arguments in
a struct with named fields instead of as five separate arguments, and provide a
Data constructor that means callers won't use all those field names anyway?)

[the sun, 2017/05/11 07:56:15]

Agree with Karl on all points (let's not address the out-of-scope issue here and
now).

     size_t samples_per_channel;  // Number of frames in the buffer.
     int sample_rate;             // Sample rate in Hz.
     size_t channels;             // Number of channels in the audio data.