Detect change on the intersection of video and viewport.

media/base/media_observer.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 MEDIA_BASE_MEDIA_OBSERVER_H_
 #define MEDIA_BASE_MEDIA_OBSERVER_H_
 
 #include "media/base/cdm_context.h"
 #include "media/base/pipeline_metadata.h"
+#include "ui/gfx/geometry/rect.h"
 
 namespace media {
 
 // This class is an observer of media player events.
 class MEDIA_EXPORT MediaObserver {
  public:
   MediaObserver();
   virtual ~MediaObserver();
 
+  struct ViewportIntersectionInfo {

[szager1, 2016/11/23 17:44:58]

As we mentioned in the meeting, we'd prefer not to ship layout information from
blink out to the browser.  Instead, I suggest you move the logic in
RemotingMediaController that analyzes ViewportIntersectionInfo (including the
de-bounce timer) into a blink class, and instead export something that doesn't
directly reveal layout information.

[ojan, 2016/11/23 18:43:43]

To expand on this a bit, my mental model is that we want followup patches that
change the heuristics to not need extra plumbing through WebMediaPlayer. You
should be able to make a change that just touches a couple files in
third_party/WebKit.

One possible API here is to a callback that requests the video to be remote
playbacked. Actually, now that I think about it, is the existing
WebMediaPlayer::requestRemotePlayback callback sufficient? HTMLMediaElement can
call that when the video is in a state that it should remote playback.

[xjz, 2016/11/23 23:43:25]

Done.

[xjz, 2016/11/23 23:43:25]

As szager1 suggested, moved the logic determining whether media is filling most
of the viewport (with the debouncer timer) into blink. But I don't think we can
use WebMediaPlayer::requestRemotePlayback callback here. IIUC,
WebMediaPlayer::requestRemotePlayback callback is for Remote Playback API, which
is different from our Media Remoting application. Also, starting media remoting
is not completely controlled by filling most of the viewport signal. Other
criteria need to be checked in our control logic. So I added
WebMediaPlayer::stablyFilledMostViewport(), which will be called when media
element starts/stops stably filling most of the viewport.

+    ViewportIntersectionInfo() {}
+    ~ViewportIntersectionInfo() {}
+
+    gfx::Rect root_rect;
+    gfx::Rect intersect_rect;
+  };
+
   // Called when the media element entered/exited fullscreen.
   virtual void OnEnteredFullscreen() = 0;
   virtual void OnExitedFullscreen() = 0;
 
+  // Called when the intersection of the media element and viewport changes.
+  virtual void OnViewportIntersectionChanged(
+      const ViewportIntersectionInfo& info) {}
+
   // Called when CDM is attached to the media element. The |cdm_context| is
   // only guaranteed to be valid in this call.
   virtual void OnSetCdm(CdmContext* cdm_context) = 0;
 
   // Called after demuxer is initialized.
   virtual void OnMetadataChanged(const PipelineMetadata& metadata) = 0;
 };
 
 }  // namespace media
 
 #endif  // MEDIA_BASE_MEDIA_OBSERVER_H_