[WIP] Mus: Pass the mojo connector to OzonePlatform::InitializeForGPU

Mus: Pass the mojo connector to OzonePlatform::InitializeForGPU

OzonePlatform::InitializeForGPU is currently called without any
initialization parameter in gl::init::GetAllowedGLImplementations.
This latter function is itself called from gl::init::InitializeGLOneOff
and for mus we actually have the following call stack:

gl::init::InitializeGLOneOff
gpu::GpuInit::InitializeAndStartSandbox
ui::GpuMain::GpuMain
ui::ws::GpuServiceProxy::GpuServiceProxy
ui::ws::WindowServer::WindowServer
ui::Service::OnStart

This CL transmits the mojo connector from ui::Service::OnStart to
gpu::GpuInit::InitializeAndStartSandbox and adds a call to
InitializeForGPU before InitializeGLOneOff, so that the Ozone GPU
component is initialized with the mojo connector.

BUG=620934, 620927
R=rjkroege
TEST=manually tested with "chrome --mash --ozone-platform=x11"
CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.win:win_optional_gpu_tests_rel

[fwang, 2016-09-15 08:45:21]

Description was changed from

==========
Mus: Pass the mojo connector to OzonePlatform::InitializeForGPU

OzonePlatform::InitializeForGPU is currently called without any
initialization parameter in gl::init::GetAllowedGLImplementations.
This latter function is itself called from gl::init::InitializeGLOneOff
and for mus we actually have the following call stack:

gl::init::InitializeGLOneOff
gpu::GpuInit::InitializeAndStartSandbox
ui::GpuMain::GpuMain
ui::ws::GpuServiceProxy::GpuServiceProxy
ui::ws::WindowServer::WindowServer
ui::Service::OnStart

This CL transmits the mojo connector from ui::Service::OnStart to
gpu::GpuInit::InitializeAndStartSandbox and adds a call to
InitializeForGPU before InitializeGLOneOff, so that the Ozone GPU
component is initialized with the mojo connector.

BUG=620934, 620927
R=rjkroege
TEST=manually tested with "chrome --mash --ozone-platform=x11"
==========

to

==========
Mus: Pass the mojo connector to OzonePlatform::InitializeForGPU

OzonePlatform::InitializeForGPU is currently called without any
initialization parameter in gl::init::GetAllowedGLImplementations.
This latter function is itself called from gl::init::InitializeGLOneOff
and for mus we actually have the following call stack:

gl::init::InitializeGLOneOff
gpu::GpuInit::InitializeAndStartSandbox
ui::GpuMain::GpuMain
ui::ws::GpuServiceProxy::GpuServiceProxy
ui::ws::WindowServer::WindowServer
ui::Service::OnStart

This CL transmits the mojo connector from ui::Service::OnStart to
gpu::GpuInit::InitializeAndStartSandbox and adds a call to
InitializeForGPU before InitializeGLOneOff, so that the Ozone GPU
component is initialized with the mojo connector.

BUG=620934, 620927
R=rjkroege
TEST=manually tested with "chrome --mash --ozone-platform=x11"
CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.win:win_optional_gpu_tests_rel
==========

[fwang, 2016-09-15 08:47:27]

Description was changed from

==========
Mus: Pass the mojo connector to OzonePlatform::InitializeForGPU

OzonePlatform::InitializeForGPU is currently called without any
initialization parameter in gl::init::GetAllowedGLImplementations.
This latter function is itself called from gl::init::InitializeGLOneOff
and for mus we actually have the following call stack:

gl::init::InitializeGLOneOff
gpu::GpuInit::InitializeAndStartSandbox
ui::GpuMain::GpuMain
ui::ws::GpuServiceProxy::GpuServiceProxy
ui::ws::WindowServer::WindowServer
ui::Service::OnStart

This CL transmits the mojo connector from ui::Service::OnStart to
gpu::GpuInit::InitializeAndStartSandbox and adds a call to
InitializeForGPU before InitializeGLOneOff, so that the Ozone GPU
component is initialized with the mojo connector.

BUG=620934, 620927
R=rjkroege
TEST=manually tested with "chrome --mash --ozone-platform=x11"
CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.win:win_optional_gpu_tests_rel
==========

to

==========
Mus: Pass the mojo connector to OzonePlatform::InitializeForGPU

OzonePlatform::InitializeForGPU is currently called without any
initialization parameter in gl::init::GetAllowedGLImplementations.
This latter function is itself called from gl::init::InitializeGLOneOff
and for mus we actually have the following call stack:

gl::init::InitializeGLOneOff
gpu::GpuInit::InitializeAndStartSandbox
ui::GpuMain::GpuMain
ui::ws::GpuServiceProxy::GpuServiceProxy
ui::ws::WindowServer::WindowServer
ui::Service::OnStart

This CL transmits the mojo connector from ui::Service::OnStart to
gpu::GpuInit::InitializeAndStartSandbox and adds a call to
InitializeForGPU before InitializeGLOneOff, so that the Ozone GPU
component is initialized with the mojo connector.

BUG=620934, 620927
R=rjkroege
TEST=manually tested with "chrome --mash --ozone-platform=x11"
CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.win:win_optional_gpu_tests_rel
==========

[fwang, 2016-09-15 09:00:36]

@Robert: So following-up on the discussion for
https://codereview.chromium.org/2327073002/ this is commit is about the
remaining important part for mus (mojo connector passed to Ozone GPU component).
Currently, I'm only able to run Mus/Ozone with X11 so that's what I tested
manually to verify that the mojo connector is now properly passed to
InitializeForUI & InitializeForGPU. I don't know if there is a simpler way to
pass that connector.

[tonikitoo, 2016-09-15 14:20:10]

On 2016/09/15 09:00:36, fwang wrote:
> @Robert: So following-up on the discussion for
> https://codereview.chromium.org/2327073002/ this is commit is about the
> remaining important part for mus (mojo connector passed to Ozone GPU
component).
> Currently, I'm only able to run Mus/Ozone with X11 so that's what I tested
> manually to verify that the mojo connector is now properly passed to
> InitializeForUI & InitializeForGPU. I don't know if there is a simpler way to
> pass that connector.

I believe now ui::OzonePlatform::InitializeForGPU(..) would be called twice: (1)
from gpu_main.cc - the call you are adding; (2) from the existing call in
GetAllowedGLImplementations call.
In order to avoid this, it might make sense to not call the later if on a Mus
code path. For example, I would add 

diff --git a/ui/gl/init/gl_factory_ozone.cc b/ui/gl/init/gl_factory_ozone.cc
index af7a40dc..5d05661 100644
--- a/ui/gl/init/gl_factory_ozone.cc
+++ b/ui/gl/init/gl_factory_ozone.cc
@@ -60,7 +60,9 @@ scoped_refptr<GLSurface> CreateDefaultOffscreenGLSurface(
 }  // namespace
 
 std::vector<GLImplementation> GetAllowedGLImplementations() {
-  ui::OzonePlatform::InitializeForGPU();
+  // comment goes here
+  if (!usingMus())
+    ui::OzonePlatform::InitializeForGPU();
   return GetSurfaceFactoryOzone()->GetAllowedGLImplementations();
 }

[fwang, 2016-09-15 15:05:03]

> I believe now ui::OzonePlatform::InitializeForGPU(..) would be called twice:
(1)
> from gpu_main.cc - the call you are adding; (2) from the existing call in
> GetAllowedGLImplementations call.
> In order to avoid this, it might make sense to not call the later if on a Mus
> code path.

Actually, it will always be called twice for Ozone build. And for non-mus in
--in-process-gpu it is also even called before as I said in
https://codereview.chromium.org/2327073002/patch/1/10002.

InitializeForGPU is implemented in a way that it becomes a non-op after the
second call, so repeated calls should not be an issue. However I agree that it
might probably be better to move the call from GetAllowedGLImplementations to
InitializeAndStartSandbox *only* when using mus+ozone. However, it seems that
this require to make ui/gl/init/ and gpu/ipc/service/ depends on
services/shell/runner/common/ (in order to add shell::ShellIsRemote if
conditionals in both files), which we may not want to.

[fwang, 2016-09-16 09:33:19]

I actually just realized that the ui/gpu components are in the same process for
Mus/X11. This means that the patch is not really useful yet since the connector
is already available at the UI init.

[rjkroege, 2016-09-21 13:27:21]

On 2016/09/16 09:33:19, fwang wrote:
> I actually just realized that the ui/gpu components are in the same process
for
> Mus/X11. This means that the patch is not really useful yet since the
connector
> is already available at the UI init.

So the whole mojo-i-fication of ozone is wip and sort of half-done in tree while
I go fix something else. In the meantime, ozone will keep on using the param
traits ipc.

I think something like this patch will be necessary but it's not time yet.

[fwang, 2016-09-21 14:38:24]

For the record, the "right" approach is given in
https://groups.google.com/a/chromium.org/d/msg/ozone-dev/x998tKrMBbo/ED329B04...
where we will use a connector for a mojo:gpu service (instead of passing the one
for mojo:ui). I guess this discussion is moved on ozone-dev for now so I'm
removing the review request.

[fwang, 2016-09-21 14:47:36]

fwang@igalia.com changed reviewers:
- rjkroege@chromium.org

[fwang, 2017-02-23 12:02:24]

I'm closing this CL.