Route touch-events for WebViewGuest directly to guest renderer.

content/browser/frame_host/render_widget_host_view_child_frame.cc

Index: content/browser/frame_host/render_widget_host_view_child_frame.cc
diff --git a/content/browser/frame_host/render_widget_host_view_child_frame.cc b/content/browser/frame_host/render_widget_host_view_child_frame.cc
index 1324f77e15c2eef56d5b3e58849ed79b69f6f723..3e80cdabf0383befeaf683bd77600edf74a8a25e 100644
--- a/content/browser/frame_host/render_widget_host_view_child_frame.cc
+++ b/content/browser/frame_host/render_widget_host_view_child_frame.cc
@@ -33,16 +33,12 @@ RenderWidgetHostViewChildFrame::RenderWidgetHostViewChildFrame(
       next_surface_sequence_(1u),
       last_output_surface_id_(0),
       current_surface_scale_factor_(1.f),
+      has_started_rendering_(false),
       ack_pending_count_(0),
       frame_connector_(nullptr),
       weak_factory_(this) {
-  if (use_surfaces_) {
+  if (use_surfaces_)
     id_allocator_ = CreateSurfaceIdAllocator();
-    if (host_->delegate() && host_->delegate()->GetInputEventRouter()) {
-      host_->delegate()->GetInputEventRouter()->AddSurfaceIdNamespaceOwner(
-          GetSurfaceIdNamespace(), this);
-    }
-  }
 
   host_->SetView(this);
 }
@@ -267,6 +263,14 @@ void RenderWidgetHostViewChildFrame::OnSwapCompositorFrame(
     return;
   }
 
+  if (!has_started_rendering_) {
+    if (host_->delegate() && host_->delegate()->GetInputEventRouter()) {
+      host_->delegate()->GetInputEventRouter()->AddSurfaceIdNamespaceOwner(

[tdresser, 2015/11/02 15:38:24]

It isn't clear to me why this was postponed - is it a performance optimization,
or was there a correctness issue?

[kenrb, 2015/11/02 16:04:22]

I don't really like this, but I understand why it is here and don't
immedatiately see a better alternative. I think it is that the inner WebContents
for the guestview isn't attached to the outer WebContents at the point that this
is created, so it is not able to register the surfaceID namespace with the
correct RenderWidgetHostInputEventRouter.

The current code registers it as soon as the surfaceID is created, which is
better, but I don't know how else to do this.

[wjmaclean, 2015/11/02 19:23:10]

I suppose one alternative would be to leave this registration in the
constructor, then have WebViewGuest *de-register* and then re-register at time
of first swap ... I *think* that would work, though we'd have to consider it an
experiment. I'm not sure how it would affect the RWHVA that's attached to the
RWHVG.

[tdresser, 2015/11/02 21:10:19]

This seems reasonable to me.

+          GetSurfaceIdNamespace(), this);
+    }
+    has_started_rendering_ = true;
+  }
+
   cc::RenderPass* root_pass =
       frame->delegated_frame_data->render_pass_list.back();