Several minor improvements of DirectX capturer

webrtc/modules/desktop_capture/win/dxgi_output_duplicator.h

Index: webrtc/modules/desktop_capture/win/dxgi_output_duplicator.h
diff --git a/webrtc/modules/desktop_capture/win/dxgi_output_duplicator.h b/webrtc/modules/desktop_capture/win/dxgi_output_duplicator.h
index 020ca640edba19769610ce5dae4beea487e32110..0d5b13479609b20e60222b4074bc60ba887b6fbb 100644
--- a/webrtc/modules/desktop_capture/win/dxgi_output_duplicator.h
+++ b/webrtc/modules/desktop_capture/win/dxgi_output_duplicator.h
@@ -111,7 +111,7 @@ class DxgiOutputDuplicator {
   // Returns a DesktopRect in the coordinate of |offset|.
   DesktopRect TargetRect(DesktopRect rect, DesktopVector offset);
 
-  const D3dDevice& device_;
+  const D3dDevice device_;

[Jamie, 2016/09/17 00:49:30]

No need for const if it's not a reference.

Please update the CL description to specify that this is a collection of
*reference-counted* pointers.

[Hzj_jie, 2016/09/19 19:09:57]

Updated the description.
I remember we have discussed const decorator before, as in Chromium const class
fields are not so common. But I do think it's a good idea for both maintainness
and correctness.
For the maintainness, readers can simply assume these fields won't be changed
during the lifetime of the instance.
For the correctness, since the logic is somehow complex, unexpected changing to
these variables may trigger undebuggable situation, as we are using lots of
Windows APIs, and their callstacks are not always available.

[Jamie, 2016/09/19 19:16:07]

Thanks for the detailed rationale. If the style guide doesn't prohibit this,
then I'm happy to keep them const.

   const Microsoft::WRL::ComPtr<IDXGIOutput1> output_;
   const DesktopRect desktop_rect_;
   Microsoft::WRL::ComPtr<IDXGIOutputDuplication> duplication_;