Fix Omnibox.CharTypedToRepaintLatency regression with D3DVsync experiment

cc/scheduler/begin_frame_source.h

Index: cc/scheduler/begin_frame_source.h
diff --git a/cc/scheduler/begin_frame_source.h b/cc/scheduler/begin_frame_source.h
index 98ea3771b397907cd7da351af4a4fd660d945134..ff4b44c69f0e44e61f1614ec77125cfa5e31c06c 100644
--- a/cc/scheduler/begin_frame_source.h
+++ b/cc/scheduler/begin_frame_source.h
@@ -275,6 +275,12 @@ class CC_EXPORT ExternalBeginFrameSource : public BeginFrameSource {
   void OnBeginFrame(const BeginFrameArgs& args);
 
  protected:
+  // Called on AddObserver and gets missed BeginFrameArgs for the given
+  // observer; returns true if the output |args| argument is valid and
+  // OnBeginFrame should be called.
+  virtual bool GetMissedBeginFrameArgs(BeginFrameObserver* obs,

[sunnyps, 2017/05/30 22:13:17]

nit: Can you make this return BeginFrameArgs? The BFS can look at
BeginFrameArgs::IsValid and decide if needs to be forwarded to the observer as a
missed frame.

[stanisc, 2017/05/31 01:17:52]

Returning BeginFrameArgs would be slightly less optimal because that would
require an extra struct copy in some cases. But that probably doesn't matter
much considering that this is called infrequently. Done.

+                                       BeginFrameArgs* missed_args);
+
   BeginFrameArgs last_begin_frame_args_;

[sunnyps, 2017/05/30 22:13:17]

nit: Can you add a method to get last_begin_frame_args_ instead of accessing it
directly in the subclass?

[stanisc, 2017/05/31 01:17:52]

I am reluctant to do this because this member is already protected which means
it is intended to be accessed directly in derived classes. And the pattern so
far has been to access these protected members directly from derived classes.

   std::unordered_set<BeginFrameObserver*> observers_;
   ExternalBeginFrameSourceClient* client_;