Call AppendGpuCommandLine for sync compositor

content/browser/android/in_process/synchronous_compositor_factory_impl.cc

 // Copyright 2014 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 "content/browser/android/in_process/synchronous_compositor_factory_impl.h"
 
 #include "base/observer_list.h"
 #include "content/browser/android/in_process/synchronous_compositor_external_begin_frame_source.h"
 #include "content/browser/android/in_process/synchronous_compositor_impl.h"
 #include "content/browser/android/in_process/synchronous_compositor_output_surface.h"
+#include "content/browser/gpu/gpu_data_manager_impl_private.h"
 #include "content/public/browser/browser_thread.h"
 #include "content/renderer/gpu/frame_swap_message_queue.h"
 #include "gpu/blink/webgraphicscontext3d_in_process_command_buffer_impl.h"
 #include "gpu/command_buffer/client/gl_in_process_context.h"
 #include "gpu/command_buffer/common/gles2_cmd_utils.h"
 #include "ui/gl/android/surface_texture.h"
 #include "ui/gl/gl_surface.h"
 #include "ui/gl/gl_surface_stub.h"
 #include "webkit/common/gpu/context_provider_in_process.h"
 
@@ skipping 136 matching lines @@
   ObserverList<StreamTextureFactoryContextObserver> observer_list_;
 
   DISALLOW_COPY_AND_ASSIGN(VideoContextProvider);
 };
 
 
 SynchronousCompositorFactoryImpl::SynchronousCompositorFactoryImpl()
     : record_full_layer_(true),
       num_hardware_compositors_(0) {
   SynchronousCompositorFactory::SetInstance(this);
+
+  GpuDataManagerImpl* gpu_data_manager = GpuDataManagerImpl::GetInstance();
+  DCHECK(gpu_data_manager);
+  gpu_data_manager->AppendGpuCommandLine(
+      base::CommandLine::ForCurrentProcess());

[no sievers, 2015/01/23 20:21:50]

We shouldn't append the command line twice. 
Can we have GpuDataManagerImplPrivate::AppendGpuCommandLine()
DCHECK(!cmdline->HasSwitch(kGpuDriverBugWorkarounds)) before we try to add it?

We don't ever try to spin up the GPU channel with WebView, right?
You should also add something in browser_main_loop.cc so that it doesn't call 
BrowserGpuChannelHostFactory::Initialize(established_gpu_channel) if running
in-process, even after the TODO there gets addressed.

Can you then file another bug for how to sort this out when we use mixed
contexts?
It can probably be done by making context creation here atomic or so.

[boliu, 2015/01/23 20:30:29]

I did think about that.

One is the factory here is a singleton already, so this should only be called
once.

And even if some other code calls AppendGpuCommandLine, command line args is
actually a hash map, so technically "doesn't hurt", except run time cost.
Doesn't matter if we do or not. We used to just do it in webview (before this
broke) and it didn't hurt.

So I'm inclined to leave browser_main_loop alone in this CL. Make it easier to
merge etc.

WYDT?
I don't quite understand what you mean by "atomic context creation".

[boliu, 2015/01/23 20:36:24]

Actually DCHECK isn't that bad. At least will notify us if something else ends
up calling AppendGpuCommandLine on initialization.

Added that in PS2

 }
 
 SynchronousCompositorFactoryImpl::~SynchronousCompositorFactoryImpl() {}
 
 scoped_refptr<base::MessageLoopProxy>
 SynchronousCompositorFactoryImpl::GetCompositorMessageLoop() {
   return BrowserThread::GetMessageLoopProxyForThread(BrowserThread::UI);
 }
 
 bool
@@ skipping 123 matching lines @@
   DCHECK(!service_.get());
   service_ = service;
 }
 
 void SynchronousCompositorFactoryImpl::SetRecordFullDocument(
     bool record_full_document) {
   record_full_layer_ = record_full_document;
 }
 
 }  // namespace content