Make Pepper resize the buffer through the command buffer.

webkit/plugins/ppapi/ppb_surface_3d_impl.cc

 // Copyright (c) 2010 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 "webkit/plugins/ppapi/ppb_surface_3d_impl.h"
 
+#include "gpu/command_buffer/client/gles2_implementation.h"
 #include "gpu/command_buffer/common/command_buffer.h"
 #include "ppapi/c/dev/ppb_graphics_3d_dev.h"
 #include "webkit/plugins/ppapi/common.h"
 #include "webkit/plugins/ppapi/ppapi_plugin_instance.h"
 
 namespace webkit {
 namespace ppapi {
 
 namespace {
 
@@ skipping 81 matching lines @@
   if (context == context_)
     return true;
 
   // Unbind from the current context.
   if (context_) {
     context_->SetSwapBuffersCallback(NULL);
   }
   if (context) {
     // Resize the backing texture to the size of the instance when it is bound.
     // TODO(alokp): This should be the responsibility of plugins.
-    context->ResizeBackingTexture(instance()->position().size());
-
-    // This is a temporary hack. The SwapBuffers is issued to force the resize
-    // to take place before any subsequent rendering. This might lead to a
-    // partially rendered frame being displayed. It is also not thread safe
-    // since the SwapBuffers is written to the command buffer and that command
-    // buffer might be written to by another thread.
-    // TODO(apatrick): Figure out the semantics of binding and resizing.
-    context->SwapBuffers();
+    const gfx::Size& size = instance()->position().size();

[neb, 2011/01/25 19:16:06]

I'm wondering if we can split this right now into a part that sets up callbacks
and the part that touches the command buffer because we'll use a different
command buffer in nacl proxy.

[Antoine Labour, 2011/01/25 19:32:24]

Right, I have a follow up CL where the command buffer implementation is created
explicitly in the ppb_context_3d_impl instead of implicitly by the delegate. In
the case of a proxy'ed plugin, it would not do that, and we would skip the
ResizeChromium (here) and the SwapBuffers (below), giving responsibility of that
to the plugin side (same with NaCl).
Since it's in a follow-up CL, I don't want to do too much refactoring just yet,
keeping this one simple.

+    context->GetGLES2Implementation()->ResizeCHROMIUM(

[apatrick, 2011/01/25 18:22:48]

The issue with calling this concurrently with GL calls on another thread remain.
Not necessarily a problem, maybe just something to be documented.

[Antoine Labour, 2011/01/25 19:32:24]

In truth, none of the pepper calls are thread-safe, aside of CallOnMainThread,
and this is no different.
I could add a note here, but it's the same with SwapBuffers below, and virtually
all the pepper calls. The GL calls themselves aren't thread safe (the context
lookup), so in practice you can't call GL from another thread currently.

+        size.width(), size.height());
 
     context->SetSwapBuffersCallback(
         NewCallback(this, &PPB_Surface3D_Impl::OnSwapBuffers));
   }
   context_ = context;
   return true;
 }
 
 bool PPB_Surface3D_Impl::SwapBuffers(PP_CompletionCallback callback) {
   if (!context_)
@@ skipping 31 matching lines @@
 }
 
 void PPB_Surface3D_Impl::OnSwapBuffers() {
   if (bound_to_instance_)
     instance()->CommitBackingTexture();
 }
 
 }  // namespace ppapi
 }  // namespace webkit