Moving RenderSurface creation outside of CalcDrawProps

cc/trees/layer_tree_host_impl.h

Index: cc/trees/layer_tree_host_impl.h
diff --git a/cc/trees/layer_tree_host_impl.h b/cc/trees/layer_tree_host_impl.h
index 878747874c7298b93b554ab645f0c15fe1309dd5..8a16ee3d40d7651fbc29076de8cf565e1854a210 100644
--- a/cc/trees/layer_tree_host_impl.h
+++ b/cc/trees/layer_tree_host_impl.h
@@ -185,6 +185,7 @@ class CC_EXPORT LayerTreeHostImpl
   virtual void CommitComplete();
   virtual void Animate(base::TimeTicks monotonic_time);
   virtual void UpdateAnimationState(bool start_ready_animations);
+  virtual bool IsTesting();

[awoloszyn, 2014/08/14 17:37:27]

This is used to trigger RenderSurface creation on impl-side, because many of our
unit-tests create only the impl-side version of the tree. I tried with templates
to hide some of this, and only made it much worse. The only other thing I can
think of is creating a virtual that we call into before CDP, with a null default
behavior, and the render-surface creation for testing. (If preferred I can
change to that).

[danakj, 2014/08/14 17:44:20]

Oh gosh. Maybe we need to be refactoring our unit tests before we attempt this,
because leaking unit test problems into production code sounds like bad times.

Maybe (some of?) those tests that directly make impl things should be making
main thread things instead.

This is a huge conceptual change in the compositor and a lot of our tests have
assumptions built in which are breaking now, or will break in the future as we
rewrite the world. We need to consider how we can make those tests reflect the
real world not some past one with bandaids.

How can we fix our tests to not need this?

   void ActivateAnimations();
   void MainThreadHasStoppedFlinging();
   void UpdateBackgroundAnimateTicking(bool should_background_tick);