Preserve FrameSinkSourceMapping nodes that have path to root

cc/surfaces/surface_manager.cc

Index: cc/surfaces/surface_manager.cc
diff --git a/cc/surfaces/surface_manager.cc b/cc/surfaces/surface_manager.cc
index 753191d7460be547e751b117afe584ad3734c89b..d5638be959559b1ecf4f43b9233183a678f63310 100644
--- a/cc/surfaces/surface_manager.cc
+++ b/cc/surfaces/surface_manager.cc
@@ -582,7 +582,8 @@ void SurfaceManager::UnregisterFrameSinkHierarchy(
   // The SurfaceFactoryClient and hierarchy can be registered/unregistered
   // in either order, so empty frame_sink_source_map entries need to be
   // checked when removing either clients or relationships.
-  if (!iter->second.has_children() && !clients_.count(parent_frame_sink_id)) {
+  if (!iter->second.has_children() && !clients_.count(parent_frame_sink_id) &&

[enne (OOO), 2017/03/17 17:00:38]

I'm not sure that this is enough.  The additional check is that it has a source
here, but what if you didn't register the begin frame source in your unit test
until after you attached C? It seems like C would not get a source at that
point.

If so would it make more sense for UnregisterFrameSinkHierarchy to remove the
*child* if it had no children or clients instead of the parent?  Or, maybe
there's some other solution here.

[enne (OOO), 2017/03/17 17:00:38]

I'm not sure that this is enough.  The additional check is that it has a source
here, but what if you didn't register the begin frame source in your unit test
until after you attached C? It seems like C would not get a source at that
point.

If so would it make more sense for UnregisterFrameSinkHierarchy to remove the
*child* if it had no children or clients instead of the parent?  Or, maybe
there's some other solution here.

[Fady Samuel, 2017/03/17 18:42:34]

As discussed offline, this isn't an issue. I've added an additional unit test to
verify this.

+      !iter->second.source) {
     frame_sink_source_map_.erase(iter);
     return;
   }