content: implement unittests backend for Ozone GBM

ui/ozone/public/ozone_gpu_test_helper.h

 // 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.
 
 #ifndef UI_OZONE_PUBLIC_OZONE_GPU_THREAD_HELPER_H_
 #define UI_OZONE_PUBLIC_OZONE_GPU_THREAD_HELPER_H_
 
 #include "base/macros.h"
 #include "base/memory/ref_counted.h"
 #include "base/memory/scoped_ptr.h"
@@ skipping 26 matching lines @@
       const scoped_refptr<base::SingleThreadTaskRunner>& gpu_task_runner);
 
  private:
   scoped_ptr<FakeGpuProcess> fake_gpu_process_;
   scoped_ptr<FakeGpuProcessHost> fake_gpu_process_host_;
   scoped_ptr<base::Thread> io_helper_thread_;
 
   DISALLOW_COPY_AND_ASSIGN(OzoneGpuTestHelper);
 };
 
+// Helper class for tests to initialize both UI and GPU Ozone platform.
+class OZONE_EXPORT OzoneInitializerForTest {

[reveman, 2016/01/11 19:13:51]

Do we really need this class? I feel like the code would be easier to understand
if we just called OzonePlatform::InitializeForUI/GPU() directly from
ContentTestSuite::Initialize().

[dshwang, 2016/01/11 20:17:45]

The first patch set was similar to what you say now. However, in
https://codereview.chromium.org/1208603002/#msg3, you said
"I don't think we want our unit tests to require some form of ipc. Please try to
design ozone code so we can run the same set of unit tests as on other platforms
without using threads and ipc."

It's how OzoneInitializerForTest was introduced.

In addition, gl_unittests needs similar code now;
https://codereview.chromium.org/1484473003/diff/60001/ui/gl/test/run_all_unit...

So this test helper should stay in somewhere ozone/.

[reveman, 2016/01/13 15:05:17]

I didn't mean that we should hide the need for ipc and threads in an ozone class
but rather that we should removed it. The gl_image unit tests are not supposed
to exercise any IPC code. I don't know how hard it would be to fix this but that
the current code rely on threads and ipc makes me concerned as it's not clear
that we can reuse any of this code in a MUS world with Mojo IPC.

[dshwang, 2016/01/13 15:37:32]

I see that you want light-version ozone backend for test which don't bother ipc
or thread.

Current Ozone can not work like it. To test GpuMemoryBuffer, it needs both
browser and gpu ozone backend.
browser backend grabs gpu device and send it to gpu, which require ipc. browser
backend is also needed to create NativePixmap which belong to GpuMemoryBuffer in
the test.
gpu backend creates actual drm buffer and gl image.

In the same sense, gl_unittests needs both browser and gpu ozone backend because
gl_unittests needs to create NativePixmap.

To not use ipc and thread, we need to create OzonePlatform::InitializeForTest(),
which each ozone backend (e.g. gbm, wayland) supports by itself.
It might be too big for only test.

reveman, spang, what do you think?

+ public:
+  static scoped_ptr<OzoneInitializerForTest> Create();
+  ~OzoneInitializerForTest();
+
+ private:
+  OzoneInitializerForTest();
+  bool Initialize();
+
+  scoped_ptr<OzoneGpuTestHelper> gpu_helper_;

[reveman, 2016/01/11 19:13:51]

What is this used for?

[dshwang, 2016/01/11 20:17:45]

This sets up message forwarding between the "gpu" and "ui" threads.
Like ozone_demo, OzoneInitializerForTest initializes both OzonePlatform UI and
OzonePlatform GPU in the same thread. OzoneGpuTestHelper helps to establish ipc
channel for both UI and GPU.

[reveman, 2016/01/13 15:05:17]

I guess this is a more general Ozone testing problem but I think it's a mistake
that we need to spin up threads and use ipc to exercise this code. 

[dshwang, 2016/01/13 15:37:32]

As I explain above, it's accidental problem to reuse current ozone code. It we
want to avoid ipc and thread, ozone should provide something like
ContextProviderInProcess. Current Ozone provide only something like
ContextProviderCommandBuffer.

+
+  DISALLOW_COPY_AND_ASSIGN(OzoneInitializerForTest);
+};
+
 }  // namespace ui
 
 #endif  // UI_OZONE_PUBLIC_OZONE_GPU_THREAD_HELPER_H_