Delete media type check in Call::NotifyBweOfReceivedPacket.

webrtc/voice_engine/channel.cc

Index: webrtc/voice_engine/channel.cc
diff --git a/webrtc/voice_engine/channel.cc b/webrtc/voice_engine/channel.cc
index 8c9b00179f37c705a4356475d3baaa9aef52da7b..0a5ecb96c3fda5af2b6fdc5b49f1ba2274dcbfbf 100644
--- a/webrtc/voice_engine/channel.cc
+++ b/webrtc/voice_engine/channel.cc
@@ -2711,7 +2711,13 @@ int Channel::ResendPackets(const uint16_t* sequence_numbers, int length) {
 }
 
 void Channel::ProcessAndEncodeAudio(const AudioFrame& audio_input) {
-  RTC_DCHECK(channel_state_.Get().sending);
+  // TODO(nisse): This is not quite enough to avoid race on
+  // destruction. In case we are preempted after this check, and some
+  // other thread calls ChannelManager::DestroyChannel whch calls

[henrika_webrtc, 2017/04/26 14:16:01]

nit, which

Also, in your failing test, why did you not hit the previous DCHECK?

+  // StopSend, StopSend's flush task gets posted *before* we post our
+  // task, our task can get run at the same time as the destructor.

[the sun, 2017/04/26 20:29:03]

Can we use ChannelOwner to allow the task to keep a reference to the Channel?
Seems hard to avoid ref counting in this situation.

+  if (!channel_state_.Get().sending)
+    return;
   std::unique_ptr<AudioFrame> audio_frame(new AudioFrame());
   // TODO(henrika): try to avoid copying by moving ownership of audio frame
   // either into pool of frames or into the task itself.
@@ -2725,7 +2731,8 @@ void Channel::ProcessAndEncodeAudio(const int16_t* audio_data,
                                     int sample_rate,
                                     size_t number_of_frames,
                                     size_t number_of_channels) {
-  RTC_DCHECK(channel_state_.Get().sending);
+  if (!channel_state_.Get().sending)
+    return;
   CodecInst codec;
   GetSendCodec(codec);
   std::unique_ptr<AudioFrame> audio_frame(new AudioFrame());