Late initialize MediaController, for less resource i.e. ProcessThread, usage by PeerConnection.

webrtc/api/webrtcsession.h

Index: webrtc/api/webrtcsession.h
diff --git a/webrtc/api/webrtcsession.h b/webrtc/api/webrtcsession.h
index e8b2a05af62cce359093c914e2da8b76557c9465..1e2bd4c3d129ef9a6fbbd06d311a979407f4bb89 100644
--- a/webrtc/api/webrtcsession.h
+++ b/webrtc/api/webrtcsession.h
@@ -130,8 +130,7 @@ class WebRtcSession : public AudioProviderInterface,
     ERROR_TRANSPORT = 2,  // transport error of some kind
   };
 
-  WebRtcSession(webrtc::MediaControllerInterface* media_controller,
-                rtc::Thread* signaling_thread,
+  WebRtcSession(rtc::Thread* signaling_thread,
                 rtc::Thread* worker_thread,
                 cricket::PortAllocator* port_allocator);
   virtual ~WebRtcSession();
@@ -149,6 +148,7 @@ class WebRtcSession : public AudioProviderInterface,
       const MediaConstraintsInterface* constraints,
       rtc::scoped_ptr<DtlsIdentityStoreInterface> dtls_identity_store,
       const PeerConnectionInterface::RTCConfiguration& rtc_configuration);
+  void LateInitialize(webrtc::MediaControllerInterface* media_controller);

[pthatcher1, 2016/02/24 23:10:09]

It's not clear what you can do with a WebRtcSession that has been Initialized,
but not LateInitialized.  Is that a state that even has any value?  If not, why
bother have two Initialize methods?  If so, what things can you do in one place
and not the other.

Honestly, this doesn't smell right at all, to have two initialize methods.

[the sun, 2016/02/25 13:15:39]

Right. When reading the code I was under the impression that StatsCollector
would be fetching info about transports even if there are no active channels,
but that turns out not being correct. I've reworked the code to lazy-init
WebRtcSession, but get stuck on the
PeerConnectionInterfaceTest.CloseAndTestMethods test case. Apparently it is
expected that PC methods continue to return results even after the PC has been
closed.

I'll need to meditate a little on that. If you have any ideas let me know. 
Agree it reeks.

   // Deletes the voice, video and data channel and changes the session state
   // to STATE_CLOSED.
   void Close();
@@ -461,19 +461,21 @@ class WebRtcSession : public AudioProviderInterface,
   bool initial_offerer_ = false;
 
   rtc::scoped_ptr<cricket::TransportController> transport_controller_;
-  MediaControllerInterface* media_controller_;
+  MediaControllerInterface* media_controller_ = nullptr;
   rtc::scoped_ptr<cricket::VoiceChannel> voice_channel_;
   rtc::scoped_ptr<cricket::VideoChannel> video_channel_;
   rtc::scoped_ptr<cricket::DataChannel> data_channel_;
-  cricket::ChannelManager* channel_manager_;
   IceObserver* ice_observer_;
   PeerConnectionInterface::IceConnectionState ice_connection_state_;
   bool ice_connection_receiving_;
   rtc::scoped_ptr<SessionDescriptionInterface> local_desc_;
   rtc::scoped_ptr<SessionDescriptionInterface> remote_desc_;
   // If the remote peer is using a older version of implementation.
-  bool older_version_remote_peer_;
-  bool dtls_enabled_;
+  bool older_version_remote_peer_ = false;
+  bool dtls_enabled_ = false;
+  bool disable_encryption_ = false;
+  rtc::scoped_refptr<rtc::RTCCertificate> certificate_;
+  rtc::scoped_ptr<DtlsIdentityStoreInterface> dtls_identity_store_;
   // Specifies which kind of data channel is allowed. This is controlled
   // by the chrome command-line flag and constraints:
   // 1. If chrome command-line switch 'enable-sctp-data-channels' is enabled,