Added proxy / stub pair for GPU surfaces.

Added proxy / stub pair for GPU surfaces.

This doesn't do anything useful yet beyond allowing the renderer process to manage the lifetime of an offscreen GLSurface in the GPU process. More to come.
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=86482

[apatrick_chromium, 2011-05-23 23:43:43]

jam, i need OWNERS review for the gypi files and the stuff in
content/renderer/gpu_*. I guess I'll move those into a gpu/ directory too...

kbr, please look at the rest.

[Ken Russell (switch to Gerrit), 2011-05-24 00:44:32]

Basically LGTM. I don't know what the design constraints are; upon first
impression it seemed to me that having a full-blown route, message handler, etc.
for each surface was heavyweight.

Is the new code robust in the case where the GPU process terminates and
restarts, or otherwise loses the device? If not, it should be.

[apatrick_chromium, 2011-05-24 01:08:10]

On 2011/05/24 00:44:32, kbr wrote:
> Basically LGTM. I don't know what the design constraints are; upon first
> impression it seemed to me that having a full-blown route, message handler,
etc.
> for each surface was heavyweight.
> 
> Is the new code robust in the case where the GPU process terminates and
> restarts, or otherwise loses the device? If not, it should be.

I think it will be convenient for GpuSurfaceStub to field IPCs directly in the
future for synchronizing resizing and swapbuffers and suchlike. Also, it gives
GpuSurfaceStub and GpuSurfaceProxy a way to detect the IPC channel going down
via OnChannelError. Lastly, I want an ID to identify the surface sort of like a
transfer buffer but in this case independent of any particular command buffer.
The surface IDs are in a channel wide namespace. It'll need to be able to handle
the GPU process going down and context lost. There are the beginnings of that in
there already, for example GpuSurfaceProxy::OnChannelError notes that the
channel is down so that future calls (which aren't implemented yet) can fail
fast.

[jam, 2011-05-24 20:37:32]

content lgtm

it might be good to move all the gpu related code in content\renderer to a gpu
subdirectory so that you don't get hassled waiting for owners lgtm