Mus: Implement GpuMain mojo interface

services/ui/ws/window_server.cc

Index: services/ui/ws/window_server.cc
diff --git a/services/ui/ws/window_server.cc b/services/ui/ws/window_server.cc
index f4d960a65c3c858f7f588d525afbef07bb726961..8d986d295898af001a85cd12808b536ad675df97 100644
--- a/services/ui/ws/window_server.cc
+++ b/services/ui/ws/window_server.cc
@@ -59,7 +59,14 @@ WindowServer::WindowServer(WindowServerDelegate* delegate)
       display_compositor_client_binding_(this) {
   user_id_tracker_.AddObserver(this);
   OnUserIdAdded(user_id_tracker_.active_id());
-  display_compositor_request_ = mojo::GetProxy(&display_compositor_);
+  // The display compositor gets its own thread in mus-gpu. The gpu service,
+  // where GL commands are processed resides on its own thread. Various
+  // components of the display compositor such as Display, ResourceProvider,
+  // and GLRenderer block on sync tokens from other command buffers. Thus,
+  // the gpu service must live on a separate thread.

[sadrul, 2016/11/30 21:05:55]

This comment should in gpu_main.cc, right? It seems out of place here in
WindowServer (which should not need to know/care about that detail).

[Fady Samuel, 2016/11/30 22:53:20]

Done.

+  gpu_proxy_->CreateDisplayCompositor(
+      mojo::GetProxy(&display_compositor_),
+      display_compositor_client_binding_.CreateInterfacePtrAndBind());
 }
 
 WindowServer::~WindowServer() {
@@ -743,21 +750,6 @@ void WindowServer::OnTransientWindowRemoved(ServerWindow* window,
 }
 
 void WindowServer::OnGpuServiceInitialized() {
-  // TODO(fsamuel): Currently we: 1. create the gpu service, 2. we initialize
-  // GL and then 3. we request a display compositor interface after
-  // initialization. A display compositor interface can be created at any time,
-  // even prior to completing initialization. If initialization crashes the GPU
-  // process, then we will still get a connection error on the
-  // |display_compositor_client_binding_| binding. We should investigate doing
-  // this in parallel with initialization in the future.
-  // The display compositor gets its own thread in mus-gpu. The gpu service,
-  // where GL commands are processed resides on its own thread. Various
-  // components of the display compositor such as Display, ResourceProvider,
-  // and GLRenderer block on sync tokens from other command buffers. Thus,
-  // the gpu service must live on a separate thread.
-  gpu_proxy_->CreateDisplayCompositor(
-      std::move(display_compositor_request_),
-      display_compositor_client_binding_.CreateInterfacePtrAndBind());
   // TODO(kylechar): When gpu channel is removed, this can instead happen
   // earlier, after GpuServiceProxy::OnInitialized().
   delegate_->StartDisplayInit();