Define EncodedVideoSource and RtcCapturedEncodingVideoCapturer.

content/renderer/media/video_capture_impl_manager.h

Index: content/renderer/media/video_capture_impl_manager.h
diff --git a/content/renderer/media/video_capture_impl_manager.h b/content/renderer/media/video_capture_impl_manager.h
index d22b017b08a3400c7314fe0219d30cfc42b363e7..d9f51494d7671f6d8e213bb1e6ca6a1b458d3b78 100644
--- a/content/renderer/media/video_capture_impl_manager.h
+++ b/content/renderer/media/video_capture_impl_manager.h
@@ -18,9 +18,11 @@
 #include "base/synchronization/lock.h"
 #include "content/common/content_export.h"
 #include "media/video/capture/video_capture.h"
+#include "media/video/encoded_video_source.h"
 
 namespace content {
 
+class RtcEncodingVideoCapturerFactory;
 class VideoCaptureImpl;
 class VideoCaptureMessageFilter;
 
@@ -50,6 +52,10 @@ class CONTENT_EXPORT VideoCaptureImplManager
     return filter_.get();
   }
 
+  RtcEncodingVideoCapturerFactory* encoding_capturer_factory() const {

[Ami GONE FROM CHROMIUM, 2013/06/08 00:18:01]

It's almost never a good idea to return a raw pointer to an object whose
lifetime is governed by a scoped_refptr<>.

[hshi1, 2013/06/12 00:50:24]

The problem I'm facing is that the factory pointer is supposed to be owned by
the PeerConnectionFactory (which is the pc_factory_ member of
MediaStreamDependencyFactory). In the current libjingle design, factory lives in
a talk_base::scoped_ptr in pc_factory_. So, it seems wrong to also hold
encoding_capturer_factory_ in a scoped_refptr of vc_manager_. Furthermore, I
believe the vc_manager_ outlives the pc_factory_ (the latter is owned by
MediaStreamDependencyFactory which is outlived by vc_manager_), so we need to
avoid vc_manager_ using the encoder_factory_ after pc_factory_ is freed in
MediaStreamDependencyFactory::CleanupPeerConnectionFactory().

Would the following work?
(1) vc_manager_ ctor constructs and holds a (not scoped) pointer to
encoder_factory_, and pass itself to encoder_factory_;
(2) at CPCF, the encoder_factory_ is passed to the PC factory;
(3) the dtor of encoder_factory_ notifies vc_manager_ that it is destroyed.
After that vc_manager_ will clear the pointer and no longer call
encoder_factory_ when video capturer impls are added/removed.

[hshi1, 2013/06/13 22:03:46]

I've made the following change:
(1) the encoding_capturer_factory is now created by
MediaStreamDependencyFactory, then passed to and owned by its
PeerConnectionFactory;
(2) VCImplManager receives a weak pointer to the factory. VCImplManager will
outlive the PeerConnectionFactory.

+    return encoding_capturer_factory_.get();
+  }
+
  protected:
   virtual ~VideoCaptureImplManager();
 
@@ -74,6 +80,8 @@ class CONTENT_EXPORT VideoCaptureImplManager
   base::Thread thread_;
   scoped_refptr<base::MessageLoopProxy> message_loop_proxy_;
 
+  scoped_refptr<RtcEncodingVideoCapturerFactory> encoding_capturer_factory_;
+
   DISALLOW_COPY_AND_ASSIGN(VideoCaptureImplManager);
 };