Oilpan: extend tracing over WebFrame trees.

Source/web/WebRemoteFrameImpl.cpp

Index: Source/web/WebRemoteFrameImpl.cpp
diff --git a/Source/web/WebRemoteFrameImpl.cpp b/Source/web/WebRemoteFrameImpl.cpp
index ddd1552ed65134b927d39c4eb5d2d695ba96b064..eaa4bcf2ab4f4abf6a73a078805c831d8bc57cc0 100644
--- a/Source/web/WebRemoteFrameImpl.cpp
+++ b/Source/web/WebRemoteFrameImpl.cpp
@@ -131,15 +131,20 @@ WebRemoteFrameImpl::~WebRemoteFrameImpl()
 {
 }
 
+#if ENABLE(OILPAN)
 void WebRemoteFrameImpl::trace(Visitor* visitor)
 {
-#if ENABLE(OILPAN)
     visitor->trace(m_frame);
     visitor->trace(m_ownersForChildren);
+    visitor->registerWeakMembers<WebRemoteFrameImpl, &WebRemoteFrameImpl::clearWeakMembers>(this);

[dcheng, 2014/09/29 04:11:45]

Why do we need to forward this through WebRemoteFrameImpl/WebLocalFrameImpl?
Can't WebFrame::traceFrames() register this itself?

[sof, 2014/09/29 06:41:25]

You could do that, but you have to be careful not to re-register the weak
callback each time you call traceFrame(). Having it here provides a clearer
separation & avoidance of that detail.

I didn't like that quirkiness, which is why it is like in the above, but I've
put up a version that moves weak pointer processing to the WebFrame side
instead. Do we want to do this?

+    WebFrame::traceFrames(visitor, this);
+}
 
-    WebFrame::traceChildren(visitor, this);
-#endif
+void WebRemoteFrameImpl::clearWeakMembers(Visitor* visitor)
+{
+    WebFrame::clearWeakFrames(visitor, this);
 }
+#endif
 
 bool WebRemoteFrameImpl::isWebLocalFrame() const
 {