cc: Allow the main thread to cancel commits

cc/trees/layer_tree_host_unittest.cc

Index: cc/trees/layer_tree_host_unittest.cc
diff --git a/cc/trees/layer_tree_host_unittest.cc b/cc/trees/layer_tree_host_unittest.cc
index 3dbce1b32c0d3dc1c0f5fb679828d4b75c1805c4..e176ac2dec7da10f58e24a927d951968dfae9cee 100644
--- a/cc/trees/layer_tree_host_unittest.cc
+++ b/cc/trees/layer_tree_host_unittest.cc
@@ -1812,7 +1812,7 @@ class LayerTreeHostTestContinuousCommit : public LayerTreeHostTest {
   virtual void DidCommit() OVERRIDE {
     if (num_draw_layers_ == 2)
       return;
-    layer_tree_host()->root_layer()->SetNeedsDisplay();
+    layer_tree_host()->SetNeedsCommit();
   }
 
   virtual void CommitCompleteOnThread(LayerTreeHostImpl* impl) OVERRIDE {
@@ -3028,6 +3028,9 @@ class LayerTreeHostTestDeferredInitialize : public LayerTreeHostTest {
 
   virtual void SetupTree() OVERRIDE {
     layer_ = FakePictureLayer::Create(&client_);
+    // Force commits to not be aborted so new frames get drawn, otherwise
+    // the renderer gets deferred initialized but nothing new needs drawing.
+    layer_->set_always_update_resources(true);

[danakj, 2013/07/23 15:35:27]

Is this saying that when we switch to GL mode, the TileManager isn't creating GL
resources which would cause us to draw again? I thought it was and that was why
we were flakily getting 2 draws in this test.

Either way if we're supposed to draw, we need to produce a frame a different
type of frame. Sounds like switching modes needs to cause a full root layer
damage?

[enne (OOO), 2013/07/23 16:58:21]

I am not convinced of that.  If you switch modes but there's been no damage,
then the current frame is still good enough and you shouldn't need to draw
again.

[danakj, 2013/07/23 17:00:11]

I guess it depends what it means to switch modes, like if the embedder switched
to software mode then it would want a software frame, etc.

But.. then it could be the one responsible for causing that damage, it wouldn't
need to be implicit.

OK This LGTM as is, thanks!

[boliu, 2013/07/23 17:19:44]

Right. The reason there's a SetNeedsCommit is mailbox resources could be dropped
and needs to be committed again to the impl side. It is not for triggering draws
(although this test did depend on that behavior)

     layer_tree_host()->SetRootLayer(layer_);
     LayerTreeHostTest::SetupTree();
   }
@@ -3311,9 +3314,11 @@ class LayerTreeHostTestLayersPushProperties : public LayerTreeHostTest {
         ++expected_push_properties_grandchild_;
         break;
       case 16:
+        // SetNeedsDisplay does not always set needs commit (so call it
+        // explicitly), but is a property change.
         child_->SetNeedsDisplay();
-        // The modified layer needs commit
         ++expected_push_properties_child_;
+        layer_tree_host()->SetNeedsCommit();
         break;
       case 17:
         EndTest();