Make context initialization order deterministic in WebFrame::swap().

third_party/WebKit/Source/bindings/core/v8/WindowProxyManager.h

Index: third_party/WebKit/Source/bindings/core/v8/WindowProxyManager.h
diff --git a/third_party/WebKit/Source/bindings/core/v8/WindowProxyManager.h b/third_party/WebKit/Source/bindings/core/v8/WindowProxyManager.h
index f05f501ed29d34552f860647dd7446eedfb9b473..b6d71559a1f1f6be986a18066cc0e3a5d64f4eb4 100644
--- a/third_party/WebKit/Source/bindings/core/v8/WindowProxyManager.h
+++ b/third_party/WebKit/Source/bindings/core/v8/WindowProxyManager.h
@@ -30,10 +30,10 @@ class WindowProxyManagerBase : public GarbageCollected<WindowProxyManagerBase> {
   void clearForClose();
   void CORE_EXPORT clearForNavigation();
 
-  void CORE_EXPORT
-  releaseGlobals(HashMap<DOMWrapperWorld*, v8::Local<v8::Object>>&);
-  void CORE_EXPORT
-  setGlobals(const HashMap<DOMWrapperWorld*, v8::Local<v8::Object>>&);
+  using GlobalsVector =
+      Vector<std::pair<DOMWrapperWorld*, v8::Local<v8::Object>>>;

[haraken, 2017/03/09 22:04:46]

Would you add a comment and explain why this needs to be a vector?

[dcheng, 2017/03/09 22:10:20]

I thought about asking for a comment, but it felt like a very fragile comment:
if something in the embedder changes and it starts getting an isolated world
context *while* initializing the main context, we'll have similar issues. So
this is kind of a bandaid fix while the extensions code is fixed (but is still
useful to have, because deterministic initialization seems like a good thing).

[Alexander Semashko, 2017/03/09 22:26:10]

Added a comment. Is it ok?

+  void CORE_EXPORT releaseGlobals(GlobalsVector&);
+  void CORE_EXPORT setGlobals(const GlobalsVector&);
 
  protected:
   using IsolatedWorldMap = HeapHashMap<int, Member<WindowProxy>>;