Add new ANA stats to GetStats() to count the number of actions taken by each controller.

webrtc/modules/audio_coding/include/audio_coding_module.h

Index: webrtc/modules/audio_coding/include/audio_coding_module.h
diff --git a/webrtc/modules/audio_coding/include/audio_coding_module.h b/webrtc/modules/audio_coding/include/audio_coding_module.h
index 44eec756666ef9432d2dc274f417ce7a038ea7c0..404cba9d52a8836327e982be39a4105d581b912f 100644
--- a/webrtc/modules/audio_coding/include/audio_coding_module.h
+++ b/webrtc/modules/audio_coding/include/audio_coding_module.h
@@ -823,6 +823,8 @@ class AudioCodingModule {
 
   virtual void GetDecodingCallStatistics(
       AudioDecodingCallStats* call_stats) const = 0;
+
+  virtual AudioEncoder::AudioEncoderStats GetAudioEncoderStatistics() const = 0;

[minyue-webrtc, 2017/08/31 07:24:21]

I think it is better to do similar to GetDecodingCallStatistics()

[ossu, 2017/08/31 13:13:10]

Why do you think so? Functions with output parameters are much more unwieldy to
call and only has a significant benefit in a few special cases (i.e. having it
fill in a (large) caller-managed buffer).

For this function, the copy could (and very probably will) be elided thanks to
Named Return Value Optimization.

[minyue-webrtc, 2017/08/31 13:20:17]

Copying is fine with me, and I understand the merits of not passing pointers.
What I would prefer, though is to let stats defined from top, since stats are
often defined in specs.

[ivoc, 2017/08/31 14:57:50]

I agree with ossu on return values vs passing pointers, but I'm not sure I
understand what minyue is arguing for. Minyue: do you think this function should
pass in the stats struct by pointer or that the stats struct should be defined
in common_types.h like AudioDecodingCallStats, or both?

[ivoc, 2017/09/01 15:27:16]

I moved the stats struct to common_types.h.

 };
 
 }  // namespace webrtc