CC: Add a scheduled action for ManageTiles

cc/trees/layer_tree_host_impl.cc

Index: cc/trees/layer_tree_host_impl.cc
diff --git a/cc/trees/layer_tree_host_impl.cc b/cc/trees/layer_tree_host_impl.cc
index 816383a6eb136e180b8874cce0576b24b1852120..f86009503d2b2180ceebf83fed3f49e94d1f4ede 100644
--- a/cc/trees/layer_tree_host_impl.cc
+++ b/cc/trees/layer_tree_host_impl.cc
@@ -2508,9 +2508,18 @@ void LayerTreeHostImpl::SetTreePriority(TreePriority priority) {
   if (new_state.tree_priority == priority)
     return;
 
+  TRACE_EVENT0("cc", "LayerTreeHostImpl::SetTreePriority");
+
   new_state.tree_priority = priority;
   tile_manager_->SetGlobalState(new_state);
   manage_tiles_needed_ = true;
+
+  // In smoothness-takes-priority, we may finish rastering without
+  // activating. We need to guarantee ManageTiles is called after this
+  // case, or we will block activation until after the next compositor
+  // frame (webkit is live-locked in the mean time).
+  if (priority != SMOOTHNESS_TAKES_PRIORITY)

[reveman, 2013/09/10 19:21:20]

why not when entering SMOOTHNESS_TAKES_PRIORITY or NEW_CONTENTS_TAKES_PRIORITY?
are we more likely to call ManageTiles() at a later time in that case?

[epennerAtGoogle, 2013/09/10 19:36:55]

The reason we are calling it here is to not starve activation, which is only
possible when exiting smoothness mode. It's primarily called on every frame, so
I think that is sufficient for the rest of the cases.

I don't mind calling it more often if that's desired, but I don't see the
necessity for doing so.

[reveman, 2013/09/10 19:57:54]

How are we guaranteeing that a new frame is produced immediately after enter
smoothness mode? If we're not always calling ManageTiles here can we add a unit
test to ensure that a new frame is produced and ManageTiles is called as
expected?

+    ManageTiles();
 }
 
 void LayerTreeHostImpl::ResetCurrentFrameTimeForNextFrame() {