getUserMeida: report device starting states

content/browser/renderer_host/media/video_capture_host.cc

Index: content/browser/renderer_host/media/video_capture_host.cc
diff --git a/content/browser/renderer_host/media/video_capture_host.cc b/content/browser/renderer_host/media/video_capture_host.cc
index 4c458f5d0da72417407e3eefd565bb9d37e7cc44..16e96bc03849cb6e907dc8ef9e1163c55249ac8f 100644
--- a/content/browser/renderer_host/media/video_capture_host.cc
+++ b/content/browser/renderer_host/media/video_capture_host.cc
@@ -118,6 +118,15 @@ void VideoCaptureHost::OnEnded(VideoCaptureControllerID controller_id) {
                  controller_id));
 }
 
+void VideoCaptureHost::OnStarted(VideoCaptureControllerID controller_id) {
+  DVLOG(1) << __func__;
+  DCHECK_CURRENTLY_ON(BrowserThread::IO);
+  BrowserThread::PostTask(
+      BrowserThread::IO, FROM_HERE,

[chfremer, 2017/02/09 23:34:52]

Why is it necessary to schedule this to the end of the current message queue
instead of invoking device_id_to_observer_map_[controller_id]->OnStateChanged()
directly?

I see that this is also done for OnError() and OnEnded(), but not for the other
methods.
But I couldn't find any documentation explaining why this is done.

[braveyao, 2017/02/14 01:05:49]

The EventHandler methods used to call "PostTask" before. And since cl,
https://codereview.chromium.org/1017503002, some of them are switched to IPC
message methods and then Mojo(I suppose).
I guess they will all be unified eventually in Mojo? 

[chfremer, 2017/02/14 19:21:22]

Thanks for the reference to that CL. It looks like mcasas@ removed the 
PostTask for some of the methods and left it in place for others.
The question is why is the PostTask needed for some but not all of the 
methods?

I feel that if we are adding a new method OnStarted() we need to be 
able to answer this in order to decide whether or not PostTask needs 
to be used in OnStarted().

[braveyao, 2017/02/15 21:47:15]

Done.

Thanks so much for pushing on this!
In the beginning I didn't notice that not all methods need PostTask. Now I found
out that the OnError and OnEnd events still need PostTask to finish some other
clearing jobs before they are handled. (DesktopCaptureApiTest and
VideoCaptureTest may fail if we remove PostTask to these two.) Only OnStart can
be processed directly.

Changes to DoError/DoEnd is too sophisticated, which deserves a separate study
and cl.

+      base::Bind(&VideoCaptureHost::DoStarted, weak_factory_.GetWeakPtr(),
+                 controller_id));
+}
+
 void VideoCaptureHost::Start(int32_t device_id,
                              int32_t session_id,
                              const media::VideoCaptureParams& params,
@@ -283,6 +292,18 @@ void VideoCaptureHost::DoEnded(VideoCaptureControllerID controller_id) {
   DeleteVideoCaptureController(controller_id, false);
 }
 
+void VideoCaptureHost::DoStarted(VideoCaptureControllerID controller_id) {
+  DVLOG(1) << __func__;
+  DCHECK_CURRENTLY_ON(BrowserThread::IO);
+  if (controllers_.find(controller_id) == controllers_.end())
+    return;
+
+  if (base::ContainsKey(device_id_to_observer_map_, controller_id)) {
+    device_id_to_observer_map_[controller_id]->OnStateChanged(
+        mojom::VideoCaptureState::STARTED);
+  }
+}
+
 void VideoCaptureHost::OnControllerAdded(
     int device_id,
     const base::WeakPtr<VideoCaptureController>& controller) {
@@ -306,11 +327,6 @@ void VideoCaptureHost::OnControllerAdded(
     return;
   }
 
-  if (base::ContainsKey(device_id_to_observer_map_, controller_id)) {
-    device_id_to_observer_map_[device_id]->OnStateChanged(
-        mojom::VideoCaptureState::STARTED);
-  }
-
   DCHECK(!it->second);
   it->second = controller;
 }