[Oilpan]: Move CSSFontSelectorClient to the oilpan heap using transition types.

Source/core/css/resolver/ViewportStyleResolver.h

 /*
  * Copyright (C) 2012 Intel Corporation. All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
  *
  * 1. Redistributions of source code must retain the above
  *    copyright notice, this list of conditions and the following
  *    disclaimer.
@@ skipping 24 matching lines @@
 #include "platform/Length.h"
 #include "wtf/RefCounted.h"
 #include "wtf/RefPtr.h"
 
 namespace WebCore {
 
 class Document;
 class MutableStylePropertySet;
 class StyleRuleViewport;
 
-class ViewportStyleResolver : public RefCountedWillBeGarbageCollected<ViewportStyleResolver> {
+class ViewportStyleResolver : public NoBaseWillBeGarbageCollected<ViewportStyleResolver> {
     DECLARE_EMPTY_DESTRUCTOR_WILL_BE_REMOVED(ViewportStyleResolver);
 public:
-    static PassRefPtrWillBeRawPtr<ViewportStyleResolver> create(Document* document)
+    static PassOwnPtrWillBeRawPtr<ViewportStyleResolver> create(Document* document)
     {
-        return adoptRefWillBeNoop(new ViewportStyleResolver(document));
+        return adoptPtrWillBeNoop(new ViewportStyleResolver(document));
     }
 
     enum Origin { UserAgentOrigin, AuthorOrigin };
 
     void collectViewportRules(RuleSet*, Origin);
 
-    void clearDocument();

[haraken, 2014/03/25 13:19:30]

Removing clearDocument() looks wrong to me. When a document is cleared,
ViewportStyleResolver::m_document should be cleared.

[wibling-chromium, 2014/03/25 13:27:17]

It wasn't cleared previously. The only caller of clearDocument was the
StyleResolver's destructor and since I changed the StyleResolver's
RefPtr<ViewportStyleResolver> to an OwnPtr<ViewportStyleResolver> it should not
be necessary to clear the document pointer since it is dead in the same cycle.

[haraken, 2014/03/25 13:50:21]

Correct me if I'm wrong:
StyleResolver and ViewportStyleResolver are on the heap and die together.
Document is not on the heap. Thus it's possible that the ViewportStyleResolver
outlives the Document. Then it's possible that ViewportStyleResolver::m_document
becomes a stale pointer, which we should avoid.

[wibling-chromium, 2014/03/25 13:59:25]

My point is that before my change what you said was also true since there was
nothing that prevented the Document to die before the ViewportStyleResolver
before my change. I was merely assuming that the code before was correct and the
person who wrote it knew the Document would live at least as long as the
ViewportStyleResolver since there was no call to clearDocument from Document
code.
Looking at the Document code a bit closer it seems there is an
OwnPtr<StyleEngine> which in turn has an OwnPtr<StyleResolver> which now has an
OwnPtr<ViewportStyleResolver> so from what I can tell the code is correct and
Document will not die before the ViewportStyleResolver.

     void resolve();
 
     void trace(Visitor*);
 
 private:
     explicit ViewportStyleResolver(Document*);
 
     void addViewportRule(StyleRuleViewport*, Origin);
 
     float viewportArgumentValue(CSSPropertyID) const;
     Length viewportLengthValue(CSSPropertyID) const;
 
     Document* m_document;
     RefPtrWillBeMember<MutableStylePropertySet> m_propertySet;
     bool m_hasAuthorStyle;
 };
 
 } // namespace WebCore
 
 #endif // ViewportStyleResolver_h