Build WebRTC with data channel only.

webrtc/pc/channel.cc

Index: webrtc/pc/channel.cc
diff --git a/webrtc/pc/channel.cc b/webrtc/pc/channel.cc
index f5428a43effac7a93fce554e11946c220946337c..8b2edfc16f6e78f18d7297e4b3a11f869f87aa17 100644
--- a/webrtc/pc/channel.cc
+++ b/webrtc/pc/channel.cc
@@ -1607,10 +1607,14 @@ bool VoiceChannel::GetStats(VoiceMediaInfo* stats) {
 }
 
 std::vector<webrtc::RtpSource> VoiceChannel::GetSources(uint32_t ssrc) const {
+#ifdef HAVE_MEDIA
   return worker_thread()->Invoke<std::vector<webrtc::RtpSource>>(
       RTC_FROM_HERE,
       Bind(&WebRtcVoiceMediaChannel::GetSources,
            static_cast<WebRtcVoiceMediaChannel*>(media_channel()), ssrc));
+#else
+  return std::vector<webrtc::RtpSource>();
+#endif

[pthatcher1, 2017/05/03 18:05:53]

Would it be easier to move the VoiceChannel and VideoChannel into their own file
and not include those if HAVE_MEDIA is false?

[Zhi Huang, 2017/05/04 01:08:03]

We have a dependency chain:
peerconnection -> audiortpreceiver -> voicechannel:basechannel. We still need
some ifdef to exclude the video channel. Plus the effort of splitting the file,
it feels to me that this way is eaiser. What do you think?

[pthatcher1, 2017/05/05 01:45:12]

It seems like we could do something where we make pure interfaces that
PeerConnection relies on and then we wouldn't need this file at all. We would
just need to change ChannelManager to return nulls (or not use ChannelManager at
all).  

Is it worth it?  To just remove this ifdef, probably not.  But if it saves more
space, maybe it is worth it.  It might be worth trying something really quick to
see if it does save space (change VoiceChannel, VideoChannel, and RtpDataChannel
into interfaces and don't compile in any impls).

 }
 
 void VoiceChannel::StartMediaMonitor(int cms) {