Adding the ability to create an RtpSender without a track.

talk/app/webrtc/peerconnectioninterface_unittest.cc

Index: talk/app/webrtc/peerconnectioninterface_unittest.cc
diff --git a/talk/app/webrtc/peerconnectioninterface_unittest.cc b/talk/app/webrtc/peerconnectioninterface_unittest.cc
index bee6380bf46a17275bf2bd3fb1c3e2e7ae42497c..3a9ba0c241d6b4a86c64c195c2ebc2e03b6911d9 100644
--- a/talk/app/webrtc/peerconnectioninterface_unittest.cc
+++ b/talk/app/webrtc/peerconnectioninterface_unittest.cc
@@ -1948,8 +1948,10 @@ TEST_F(PeerConnectionInterfaceTest, LocalDescriptionChanged) {
   EXPECT_TRUE(ContainsSender(senders, kVideoTracks[1]));
 
   // Remove an audio and video track.
+  pc_->RemoveStream(reference_collection_->at(0));
   rtc::scoped_ptr<SessionDescriptionInterface> desc_2;
   CreateSessionDescriptionAndReference(1, 1, desc_2.accept());
+  pc_->AddStream(reference_collection_->at(0));

[pthatcher1, 2015/10/20 17:42:49]

Was this test wrong before?

[Taylor Brandstetter, 2015/10/21 00:22:08]

No, there's an intentional change in behavior. Previously a sender only existed
if AddStream AND SetLocalDescription were called. But now, createSender must
return a sender even before SetLocalDescription is called. So the bounds of a
sender's lifetime had to change a bit.

   EXPECT_TRUE(DoSetLocalDescription(desc_2.release()));
   senders = pc_->GetSenders();
   EXPECT_EQ(2u, senders.size());