Implement HTMLMediaElement::srcObject.

Source/core/html/HTMLMediaElement.h

Index: Source/core/html/HTMLMediaElement.h
diff --git a/Source/core/html/HTMLMediaElement.h b/Source/core/html/HTMLMediaElement.h
index e1706610fe53f2519267eaa2daa3379d166dd87a..1f4ee72d6f9dc2d7396a38d3c7488f4e52c3c522 100644
--- a/Source/core/html/HTMLMediaElement.h
+++ b/Source/core/html/HTMLMediaElement.h
@@ -59,6 +59,7 @@ class HTMLTrackElement;
 class KURL;
 class MediaController;
 class MediaControls;
+class MediaProvider;
 class MediaError;
 class HTMLMediaSource;
 class TextTrackList;
@@ -111,6 +112,8 @@ public:
 
     // network state
     void setSrc(const AtomicString&);
+    MediaProvider* srcObject() const;
+    void setSrcObject(MediaProvider*);
     const KURL& currentSrc() const { return m_currentSrc; }
 
     enum NetworkState { NETWORK_EMPTY, NETWORK_IDLE, NETWORK_LOADING, NETWORK_NO_SOURCE };
@@ -384,11 +387,14 @@ private:
     void prepareForLoad();
     void loadInternal();
     void selectMediaResource();
+    void loadResource(const KURL&);
     void loadResource(const KURL&, ContentType&, const String& keySystem);
     void startPlayerLoad();
     void setPlayerPreload();
     blink::WebMediaPlayer::LoadType loadType() const;
     void scheduleNextSourceChild();
+    void loadFromObject();
+    void loadFromAttribute();
     void loadNextSourceChild();
     void userCancelledLoad();
     void clearMediaPlayer(int flags);
@@ -506,8 +512,9 @@ private:
     double m_defaultPlaybackStartPosition;
 
     // Loading state.
-    enum LoadState { WaitingForSource, LoadingFromSrcAttr, LoadingFromSourceElement };
+    enum LoadState { WaitingForSource, LoadingFromSrcObject, LoadingFromSrcAttr, LoadingFromSourceElement };
     LoadState m_loadState;
+    RawPtrWillBeMember<MediaProvider> m_srcObject;

[sof, 2014/09/19 06:26:23]

Could you explain what keeps m_srcObject alive&valid in a non-Oilpan setting? If
it is held onto by someone else & guaranteed to live beyond the media element,
then it would be preferable to make that clear in a comment.

[perkj_chrome, 2014/09/19 10:48:06]

oh, good catch. This is probably not correct 
Is there a generic solution or do I have to make the same solution used by
EventTarget.h ? That defines ref and deref and use
DEFINE_EVENT_TARGET_REFCOUNTING?
Can I disable OILPAN for MediaStreams and test? 

https://code.google.com/p/chromium/codesearch#chromium/src/third_party/WebKit...


I was trying to understand how/if this works without OILPAN today if
createObjectURL is used used to register a MediaStream and the last reference
from JS to the MediaStream is abandonded. It looks like the MediaStream object
itself would not survive? 

[sof, 2014/09/19 13:06:11]

Yes, exactly - ref-forwarding is how this is usually addressed. A better example
might be DOMSettableTokenList here, i.e., you add a ref+deref pure virtual pair
to MediaProvider

#if !ENABLE(OILPAN)
    virtual void ref() = 0;
    virtual void deref() = 0;
#endif

and then provide impls on the objects implementing MediaProvider -- i.e., on
MediaSource:

#if !ENABLE(OILPAN)
    virtual void ref() OVERRIDE {
RefCountedGarbageCollected<MediaStream>::ref(); }
    virtual void deref() OVERRIDE {
RefCountedGarbageCollected<MediaStream>::deref(); }
#endif

As Oilpan is disabled by default, you don't have to do anything to compile&test.
It didn't look as if it would survie. Unless there are cycles to be concerned
with between the MediaProvider implementations and HTMLMediaElement, the latter
should keep a ref ptr to its srcObject, non Oilpan. It's simpler and more
regular with Oilpan, and the handling here looks correct to me.

     RefPtrWillBeMember<HTMLSourceElement> m_currentSourceNode;
     RefPtrWillBeMember<Node> m_nextChildNodeToConsider;