[Android, RemotePlayback] Implement HTMLMediaElement.remote.connect().

third_party/WebKit/Source/modules/remoteplayback/RemotePlayback.h

Index: third_party/WebKit/Source/modules/remoteplayback/RemotePlayback.h
diff --git a/third_party/WebKit/Source/modules/remoteplayback/RemotePlayback.h b/third_party/WebKit/Source/modules/remoteplayback/RemotePlayback.h
index fa24f558fc5ac27779afcac8b4549c1b52f4e452..3407e6ad6eb02e0d0d530aa49b2372337864570f 100644
--- a/third_party/WebKit/Source/modules/remoteplayback/RemotePlayback.h
+++ b/third_party/WebKit/Source/modules/remoteplayback/RemotePlayback.h
@@ -20,6 +20,7 @@ class ExecutionContext;
 class HTMLMediaElement;
 class LocalFrame;
 class RemotePlaybackAvailability;
+class ScriptPromiseResolver;
 
 class RemotePlayback final
     : public RefCountedGarbageCollectedEventTargetWithInlineData<RemotePlayback>
@@ -38,6 +39,7 @@ public:
     ExecutionContext* getExecutionContext() const override;
 
     ScriptPromise getAvailability(ScriptState*);
+    ScriptPromise connect(ScriptState*);
 
     String state() const;
 
@@ -46,7 +48,7 @@ public:
     DECLARE_VIRTUAL_TRACE();
 
 private:
-    RemotePlayback(LocalFrame*, WebRemotePlaybackState, bool availability);
+    explicit RemotePlayback(HTMLMediaElement&);
 
     void stateChanged(WebRemotePlaybackState) override;
     void availabilityChanged(bool available) override;
@@ -54,6 +56,8 @@ private:
     WebRemotePlaybackState m_state;
     bool m_availability;
     HeapVector<Member<RemotePlaybackAvailability>> m_availabilityObjects;
+    RawPtrWillBeWeakMember<HTMLMediaElement> m_mediaElement;

[philipj_slow, 2016/03/24 07:31:57]

Hmm, why weak? I guess that RawPtr is needed pre-Oilpan to avoid a reference
cycle, but it would be nice to just have a strong reference in the Oilpan world
unless that has some unintended side effect?

Also, m_mediaElement is never cleared in the non-Oilpan case so I guess the
pointer can become stale?

[mlamouri (slow - plz ping), 2016/03/24 10:47:06]

+1. I believe oilpan will deal with the cycle:
https://docs.google.com/presentation/d/1XPu03ymz8W295mCftEC9KshH9Icxfq81YwIJQ...

And that would allow someone to do this:
```
var remote = media.remote;
[.. stuff happens ..]
[.. media is GC'd ..]
remote.connect();
```

If `media` is a weak member, it would not be possible to connect at that point
but if it's a member it wouldn't be GC'd because `remote` has a reference to it.
Should ~HTMLMediaElement have something to prevent this or is there a strong
binding that would prevent this to happen?

[philipj_slow, 2016/03/24 11:06:23]

~HTMLMediaElement would be one way, or perhaps WeakPtrWillBeMember, in which
case HTMLMediaElement::createWeakPtr is needed. So much fun given that
non-Oilpan is going away, but it can't be left as a potential use-after-free :/

[whywhat, 2016/03/24 16:08:40]

Sounds like if it's a Member in Oilpan world, it should be a RefPtr in
non-Oilpan world so the same trick with holding a reference to remote works?

[philipj_slow, 2016/03/29 05:51:45]

The trouble is that one can't really get the same behavior with and without
Oilpan in a situation like this. I think the choice is between making GC
observable in both cases, and making it observable only pre-Oilpan.

+    HeapVector<Member<ScriptPromiseResolver>> m_connectPromiseResolvers;
 };
 
 } // namespace blink