Link MediaSessionTabHelper with native MediaSession [CL is going to be split]

content/browser/media/session/media_session.h

Index: content/browser/media/session/media_session.h
diff --git a/content/browser/media/session/media_session.h b/content/browser/media/session/media_session.h
index 9239c7854a18b18326e2c2983451e49490922982..8a99a766cafb6c961ad2d5c4b6c3a5dafd114442 100644
--- a/content/browser/media/session/media_session.h
+++ b/content/browser/media/session/media_session.h
@@ -11,6 +11,7 @@
 #include "base/id_map.h"
 #include "base/macros.h"
 #include "base/optional.h"
+#include "base/observer_list.h"
 #include "content/browser/media/session/audio_focus_manager.h"
 #include "content/browser/media/session/media_session_uma_helper.h"
 #include "content/common/content_export.h"
@@ -29,6 +30,7 @@ namespace content {
 class AudioFocusDelegate;
 class AudioFocusManagerTest;
 class MediaSessionPlayerObserver;
+class MediaSessionObserver;
 class MediaSessionStateObserver;
 class MediaSessionVisibilityBrowserTest;
 
@@ -71,6 +73,11 @@ class MediaSession : public WebContentsObserver,
 
   ~MediaSession() override;
 
+  void AddObserver(MediaSessionObserver* observer);
+  void RemoveObserver(MediaSessionObserver* observer);
+  // Only passes ownership and does not observe.
+  void PassObserverOwnership(std::unique_ptr<MediaSessionObserver> owned_observer);

[whywhat, 2016/10/20 15:38:33]

this seems like an antipattern, along with the map below...

[Zhiqiang Zhang (Slow), 2016/10/20 16:22:32]

It's a problem that I've been struggling with. I discussed with Mounir and he's
fine to have MSO that are owned and not owned by MediaSession.

As you said in the other comment, maybe we could make MSO be owned by
WebContents, but actually it's the same as owned by MediaSession, since MS and
WC have the same lifetime. However there might be some other cases we don't want
it be owned by WC or MS, such as in tests, or the observer (CastMediaBlocker) in
chromecast/

+
   void SetMetadata(const base::Optional<MediaMetadata>& metadata);
   const base::Optional<MediaMetadata>& metadata() const { return metadata_; }
 
@@ -237,6 +244,12 @@ class MediaSession : public WebContentsObserver,
   base::Optional<MediaMetadata> metadata_;
   base::CallbackList<void(State)> media_session_state_listeners_;
 
+  base::ObserverList<MediaSessionObserver> observers_;
+
+  using OwnedObserversType =
+  std::map<MediaSessionObserver*, std::unique_ptr<MediaSessionObserver>>;

[boliu, 2016/10/19 20:39:40]

would scoped_vector be better? is there going to be that many observers in
practice that a hashed data structure is needed?

[Zhiqiang Zhang (Slow), 2016/10/20 16:22:32]

The map is used when removing observers from the session. We don't expect many
observers, so I think using std::map is not very expensive.

[boliu, 2016/10/20 16:28:33]

Hmm? Usually the argument goes that vectors are slower for random access, but is
ok if N is tiny, where vector's memory efficiency wins out. I've never seen
anyone argue hash map is slow before :p

[Zhiqiang Zhang (Slow), 2016/10/20 17:07:03]

You are right, actually base::ObserverList uses std::vector<>.
Done.

+  OwnedObserversType owned_observers_;
+
   DISALLOW_COPY_AND_ASSIGN(MediaSession);
 };