cc: Push valid property values when CalcDrawProps skips layer.

cc/trees/layer_tree_host.cc

 // Copyright 2011 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "cc/trees/layer_tree_host.h"
 
 #include <algorithm>
 #include <stack>
 
 #include "base/bind.h"
@@ skipping 660 matching lines @@
   return !output_surface_lost_;
 }
 
 bool LayerTreeHost::UpdateLayers(ResourceUpdateQueue* queue,
                                  size_t memory_allocation_limit_bytes) {
   DCHECK(!output_surface_lost_);
 
   if (!root_layer())
     return false;
 
+  if (device_viewport_size().IsEmpty())

[enne (OOO), 2013/08/02 17:00:48]

This is wrong for WebView.  You can't just add it back.

[boliu, 2013/08/02 17:44:50]

Not commenting on correctness, just a question:

Is not possible for content area in desktop chrome be resized to nothing, open a
new tab, and it would fall into this case?

[enne (OOO), 2013/08/02 17:47:41]

It is possible, but in that case the compositor thread wouldn't draw anything
either.  This viewport would get passed along and used on the compositor thread.

In the WebView case, you can pass an arbitrary viewport to draw with, so if you
skip on the main thread you may not update layers that need updating and then
you'd attempt to draw them on the compositor with a non-empty viewport.

This patch may give you the correct bounds on the compositor thread, but you
wouldn't get the PictureLayer::Update call so you wouldn't get the right
content.

[danakj, 2013/08/02 18:35:11]

I think checkerboards should be fine in this case. In general, any layer could
be outside the viewport and not get updated, and then we draw with a different
viewport. I'm not sure we need to make this work specially when the first
viewport was empty. It will still clip everything, and we'll activate without
any tiles rastered.

[enne (OOO), 2013/08/02 18:41:49]

Checkerboarding would be wrong for software raster in WebView, as the
PictureLayer needs to be updated so PictureLayerImpl can get its picture.

LayerTreeHostCommon does not clip or skip updating layers that are outside the
viewport.  The only time we drop layers from a RSLL is when they are outside a
clipping render surface (not the root), because we can't change the size of that
clip rect from the compositor thread.

I still think this check is wrong for WebView.

[danakj, 2013/08/02 21:49:13]

Ah, okie.

+    return false;
+
   if (contents_texture_manager_ && memory_allocation_limit_bytes) {
     contents_texture_manager_->SetMaxMemoryLimitBytes(
         memory_allocation_limit_bytes);
   }
 
   return UpdateLayers(root_layer(), queue);
 }
 
 static Layer* FindFirstScrollableLayer(Layer* layer) {
   if (!layer)
@@ skipping 477 matching lines @@
 void LayerTreeHost::DidLoseUIResources() {
   // When output surface is lost, we need to recreate the resource.
   for (UIResourceClientMap::iterator iter = ui_resource_client_map_.begin();
        iter != ui_resource_client_map_.end();
        ++iter) {
     UIResourceLost(iter->first);
   }
 }
 
 }  // namespace cc