Keep track of the user-facing number of channels in a ChannelBuffer

webrtc/common_audio/channel_buffer.cc

Index: webrtc/common_audio/channel_buffer.cc
diff --git a/webrtc/common_audio/channel_buffer.cc b/webrtc/common_audio/channel_buffer.cc
index 349d4245ccdb6e472db3cb60ee3a951c1e333b3b..3c43b3bf3e923a1a711d2c012b7a18bd717b73c7 100644
--- a/webrtc/common_audio/channel_buffer.cc
+++ b/webrtc/common_audio/channel_buffer.cc
@@ -54,6 +54,7 @@ void IFChannelBuffer::RefreshF() const {
         float_channels[i][j] = int_channels[i][j];
       }
     }
+    fbuf_.set_num_channels(ibuf_.num_channels());

[peah-webrtc, 2016/06/10 11:39:34]

This seems wrong, similarly to what is described below.

[aluebs-webrtc, 2016/06/14 01:40:28]

This is not similar to below.

     fvalid_ = true;
   }
 }
@@ -68,6 +69,7 @@ void IFChannelBuffer::RefreshI() const {
                     ibuf_.num_frames(),
                     int_channels[i]);
     }
+    ibuf_.set_num_channels(fbuf_.num_channels());

[peah-webrtc, 2016/06/10 11:39:34]

This seems wrong:
Afaics:
1) First ibuf_.num_channels() channels are converted from float to int.
2) Then the number of channels are set to fbuf_.num_channels() in
ibuf_.set_num_channels(fbuf_.num_channels());

I guess the above works, because by the usage of ibuf_ it is guaranteed that the
preallocated number of channels is sufficiently large for both ibuf and fbuf.
But this needs to be corrected regardless of that.

Furthermore, if fbuf_ has 1 channel (but 2 allocated internally), and ibuf_ has
2 channels, this is twice as complex as required, right?

[aluebs-webrtc, 2016/06/14 01:40:28]

Good catch. This was not an issue before since they were fixed and the same, but
now it is. I changed the for-loop to only go over fbuf_.num_channels().

     ivalid_ = true;
   }
 }