Rewrite VideoCaptureManager to streamline the lifetimes of

VideoCaptureController vs. VideoCaptureDevice.

The goal of this change is to eliminate code and statefulness from
VCM, VCC and VCD and to move towards more consistent threading
behavior.

Move most of the state of the VCM to the IO thread only. Track both
the VCC and the VCD objects in a single collection that lives on the
IO thread. Move the allocation of the Controller to the IO thread.
Always allocate a Controller before a Device, and have the Controller
outlive its Device. Device creation and destruction is farmed out to
the device thread, but the decision to create or destroy occurs on the
IO thread.

VCM::Open() no longer creates the device but instead just records the
ID of the device to be created. De-duplication of active devices will
occur when a client actually Starts() a session.

From the Controller, the changes to the VCM enable some
simplifications here. Remove the notion of pending_clients_ and device
restart. This is possible to do because the decision of which devices
are open lives on the IO thread, so we don't have to worry about
inconsistencies.

Grant the Controller sole responsibility for tracking its clients,
rather than having the Manager maintain a parallel list of handlers.
In turn, the Manager asks the Controller how many clients remain, and
takes on responsibility for actually starting and stopping the Device.
Removing the Controller's calls into the manager enables some new unit
tests on the public API of the Controller.

Two small, subtle bugfixes are included: 

[1] In VCC, we were sometimes still delivering events to clients after
|session_close|. This was uncovered by unit tests.

[2] In VCM, the "in_use" state of devices (meaning whether any client
had opened the device) was being returned to the MediaStreamManager in
the devices-enumerated event. This change sets it to always be false,
as it would appear that the MSM intends |in_use| to indicate per-
session, not a systemwide per- device, bit.

Several new unit tests are included.

TEST=Manual tests of windows webcam creation, basic tests of chromecast mirroring
BUG=284829,289731,289684

Review URL: https://chromiumcodereview.appspot.com/22866015

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@222883 0039d316-1c4b-4281-b951-d872f2087c98

1 files changed, 1 insertions, 1 deletions

diff --git a/media/video/capture/video_capture_device.h b/media/video/capture/video_capture_device.h
index 8ef43e7..0eaf39e 100644
--- a/media/video/capture/video_capture_device.h
+++ b/media/video/capture/video_capture_device.h
@@ -90,7 +90,7 @@ class MEDIA_EXPORT VideoCaptureDevice {
 #if defined(OS_WIN)
     // This class wraps the CaptureApiType, so it has a by default value if not
     // inititalized, and I (mcasas) do a DCHECK on reading its value.
-    class CaptureApiClass{
+    class CaptureApiClass {
      public:
       CaptureApiClass():  capture_api_type_(API_TYPE_UNKNOWN) {}
       CaptureApiClass(const CaptureApiType api_type)