Use separate rtp module lists for send and receive in PacketRouter.

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 25f44c1c66fcbaccc7543121971f0c9cda6dc4e1..4e7cc130781565f1a36e7a847bed0083734c11ac 100644
--- a/webrtc/voice_engine/channel.cc
+++ b/webrtc/voice_engine/channel.cc
@@ -125,6 +125,12 @@ class RtpPacketSenderProxy : public RtpPacketSender {
     rtp_packet_sender_ = rtp_packet_sender;
   }
 
+  bool HasPacketSender() const {
+    RTC_DCHECK(thread_checker_.CalledOnValidThread());
+    rtc::CritScope lock(&crit_);
+    return rtp_packet_sender_ != nullptr;
+  }
+
   // Implements RtpPacketSender.
   void InsertPacket(Priority priority,
                     uint32_t ssrc,
@@ -2922,6 +2928,9 @@ void Channel::SetCongestionControlObjects(
     RtpPacketSender* rtp_packet_sender,
     TransportFeedbackObserver* transport_feedback_observer,
     PacketRouter* packet_router) {
+  bool sender = rtp_packet_sender != nullptr ||
+                (rtp_packet_sender_proxy_.get() &&
+                 rtp_packet_sender_proxy_->HasPacketSender());

[the sun, 2016/01/25 15:42:01]

Can this property change as a consequence of this call to
SetCongestionControlObjects? Does it need to be evaluated twice - once before
SetPacketSender and once after?

[stefan-webrtc, 2016/01/25 16:00:26]

It can change to null as the stream is being teared down, but a channel can only
be a sender or a receiver and this expression makes sure we can figure out which
it is. 

If rtp_packet_sender isn't null, we know that the channel (stream) is either
being teared down or that it is a receiving channel. If it's true we know that
it must be a sending channel since no receivers have an rtp_packet_sender.

To check the former case, we check if the rtp_packet_sender_proxy_ has already
been configured with an rtp_packet_sender, since in that case we know that it
already is a sending channel (and that it is actually being teared down).

Because of this I don't think we need to evaluate it twice.

[the sun, 2016/01/26 09:17:03]

Very well, but I think this code is becoming hard to follow. How about we split
it into three functions instead?

SetSenderCongestionControlObjects(x, y, z)
SetReceiverCongestionControlObjects(z)
ResetCongestionControlObjects()

I think the logic would be simpler.

[stefan-webrtc, 2016/01/26 10:50:38]

Agree, done.

   RTC_DCHECK(packet_router != nullptr || packet_router_ != nullptr);
   if (transport_feedback_observer) {
     RTC_DCHECK(feedback_observer_proxy_.get());
@@ -2937,9 +2946,9 @@ void Channel::SetCongestionControlObjects(
   }
   _rtpRtcpModule->SetStorePacketsStatus(rtp_packet_sender != nullptr, 600);
   if (packet_router != nullptr) {
-    packet_router->AddRtpModule(_rtpRtcpModule.get());
+    packet_router->AddRtpModule(_rtpRtcpModule.get(), sender);
   } else {
-    packet_router_->RemoveRtpModule(_rtpRtcpModule.get());
+    packet_router_->RemoveRtpModule(_rtpRtcpModule.get(), sender);
   }
   packet_router_ = packet_router;
 }