Streamline frame detach

Source/core/frame/Frame.cpp

Index: Source/core/frame/Frame.cpp
diff --git a/Source/core/frame/Frame.cpp b/Source/core/frame/Frame.cpp
index 302a3e1e1328c2dd483c8800fa04416b5323c0b4..7cfa6122bd1c4e176c0b212155e58524e60655ca 100644
--- a/Source/core/frame/Frame.cpp
+++ b/Source/core/frame/Frame.cpp
@@ -70,8 +70,9 @@ Frame::Frame(FrameClient* client, FrameHost* host, FrameOwner* owner)
     frameCounter.increment();
 #endif
 
+    m_host->incrementFrameCount();
+
     if (m_owner) {
-        page()->incrementSubframeCount();
         if (m_owner->isLocal())
             toHTMLFrameOwnerElement(m_owner)->setContentFrame(*this);
     } else {
@@ -102,6 +103,22 @@ void Frame::trace(Visitor* visitor)
     visitor->trace(m_domWindow);
 }
 
+void Frame::detach()
+{
+    // client() should never be null because that means we somehow re-entered

[dcheng, 2014/09/25 23:35:43]

This was a little hard for me to follow. We needed this comment originally
because this happened after all the other stuff in
FrameLoader::detachFromParent. This still happens 'last' so to speak, but it's
not so obvious. I don't have any good suggestions though...

[Nate Chapin, 2014/09/25 23:57:50]

It may be that this comment should be on the client() null-check in
LocalFrame::detach().

[dcheng, 2014/09/26 00:40:25]

I'm not sure it really makes sense there either. I guess let's just leave it
here =/

+    // the frame detach code... but it is sometimes.
+    // FIXME: Understand why this is happening so we can document this insanity.
+    // http://crbug.com/371084 is a probable explanation.
+    if (!client())
+        return;
+    m_host->decrementFrameCount();
+    m_host = nullptr;

[dcheng, 2014/09/25 23:35:43]

I don't think we need to do it in this patch, but I wonder if it's worth
enforcing the invariant that host(), client() and page() must all go to null at
the same time?

[Nate Chapin, 2014/09/25 23:57:50]

That seems like a good idea for the future.

[dcheng, 2014/09/26 00:40:25]

Mind adding a FIXME?

[Nate Chapin, 2014/09/26 20:25:51]

Now that I look at it, because page() is directly derived from host(), I think
this invariant is already true.

+    // After this, we must no longer talk to the client since this clears
+    // its owning reference back to our owning LocalFrame.
+    m_client->detached();
+    m_client = nullptr;
+}
+
 void Frame::detachChildren()
 {
     typedef WillBeHeapVector<RefPtrWillBeMember<Frame> > FrameVector;
@@ -206,8 +223,6 @@ void Frame::disconnectOwnerElement()
     if (m_owner) {
         if (m_owner->isLocal())
             toHTMLFrameOwnerElement(m_owner)->clearContentFrame();
-        if (page())
-            page()->decrementSubframeCount();
     }
     m_owner = nullptr;
 }