[Mojo Video Capture] Stop service when last client disconnects.

services/video_capture/public/cpp/device_to_feedback_observer_adapter.h

 // Copyright 2016 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #ifndef SERVICES_VIDEO_CAPTURE_PUBLIC_CPP_DEVICE_TO_FEEDBACK_OBSERVER_ADAPTER_H_
 #define SERVICES_VIDEO_CAPTURE_PUBLIC_CPP_DEVICE_TO_FEEDBACK_OBSERVER_ADAPTER_H_
 
 #include "media/capture/video/video_capture_device.h"
 #include "services/video_capture/public/interfaces/device.mojom.h"
 
 namespace video_capture {
 
 // Adapter that allows a video_capture::mojom::Device to be used in place of
 // a media::VideoFrameConsumerFeedbackObserver
 class DeviceToFeedbackObserverAdapter
     : public media::VideoFrameConsumerFeedbackObserver {
  public:
-  DeviceToFeedbackObserverAdapter(mojom::DevicePtr device);
+  DeviceToFeedbackObserverAdapter(mojom::DevicePtr* device);
   ~DeviceToFeedbackObserverAdapter() override;
 
   // media::VideoFrameConsumerFeedbackObserver:
   void OnUtilizationReport(int frame_feedback_id, double utilization) override;
 
  private:
-  mojom::DevicePtr device_;
+  mojom::DevicePtr* device_;

[mcasas, 2017/04/27 00:22:15]

I'm surprised by this change, it seems unrelated to other
changes in this CL and it's in itself peculiar: mojom Ptrs
are move-only types, so holding a weak reference to one
such is non-intuitive.  The previous behaviour looks more
correct to me.

[chfremer, 2017/04/27 19:52:49]

Thanks for catching this. You are right, this change should not actually belong
to this CL. I believe it must somehow have propagated itself here through the
countless merges and reshuffles that the CLs have gone through on their way to
review. The change may make sense at a later CL. I guess if it is still needed,
we will see it when the time comes.

Reverted.

 };
 
 }  // namespace video_capture
 
 #endif  // SERVICES_VIDEO_CAPTURE_PUBLIC_CPP_DEVICE_TO_FEEDBACK_OBSERVER_ADAPTER_H_