RenderWidgetHostViewChildFrame's called a virtual in its ctor.

content/browser/renderer_host/render_widget_host_view_aura.h

Index: content/browser/renderer_host/render_widget_host_view_aura.h
diff --git a/content/browser/renderer_host/render_widget_host_view_aura.h b/content/browser/renderer_host/render_widget_host_view_aura.h
index 53431c5dbaea63be11803ee3a5d5af662eed8bd8..0c10ef0b8304b5a3b159ca8013bd2f445e678a7b 100644
--- a/content/browser/renderer_host/render_widget_host_view_aura.h
+++ b/content/browser/renderer_host/render_widget_host_view_aura.h
@@ -532,6 +532,8 @@ class CONTENT_EXPORT RenderWidgetHostViewAura
   // Forwards a mouse event to this view's parent window delegate.
   void ForwardMouseEventToParent(ui::MouseEvent* event);
 
+  void UpdateNeedsBeginFramesInternal();

[clamy, 2016/10/05 12:35:53]

How is this change (and the similar ones about needs_begin_frames_internal)
related to the issue of the virtual being called in the ctor?

[dtapuska, 2016/10/05 13:21:29]

So I tried to land a change that did vsync aligned synthetic gestures (see
https://codereview.chromium.org/2372873003/) which had part of this CL. 

And this caused a regression because when SetView is called (in the ctor)
SetNeedsBeginFrame method is called
https://cs.chromium.org/chromium/src/content/browser/renderer_host/render_wid...

but its state is completely wrong.

If I don't carry forward some of that original change (as here); it fails an
assert because the frame observer is added twice. And there is a DCHECK that the
frame observer only be added once in the listener table.

[clamy, 2016/10/05 13:47:01]

Ok so let me see if I get this right (sorry not super familiar with this area of
the code): the virtual method that ends up getting called in the
RenderWidgetHostViewChildFrame ctor is SetNeedsBeginFrame? And because it was
called before RenderWidgetHostViewGuest vtable had been populated it didn't
actually do what it was supposed to do, but now that it does it triggers an
assert, hence the change to UpdateNeedsBeginFramesInternal?

If so, could you please update the CL description by adding the name of the
problematic method & explaining that this also required some modification to the
implementation of the method in some inheriting classes now that the proper
version is being called in the ctor? That wasn't immediately clear when reading
it.

+
   // Returns the RenderViewHostDelegateView instance for this view. Returns
   // NULL on failure.
   RenderViewHostDelegateView* GetRenderViewHostDelegateView();
@@ -592,8 +594,13 @@ class CONTENT_EXPORT RenderWidgetHostViewAura
   // The begin frame source being observed.  Null if none.
   cc::BeginFrameSource* begin_frame_source_;
   cc::BeginFrameArgs last_begin_frame_args_;
+
+  // Whether a request for begin frames has been issued.
   bool needs_begin_frames_;
 
+  // Whether or not a frame observer has been added.
+  bool added_frame_observer_;
+
   // Used to record the last position of the mouse.
   // While the mouse is locked, they store the last known position just as mouse
   // lock was entered.