Disabled support for running GPU thread in browser process with --single-proc...

Disabled support for running GPU thread in browser process with --single-process.

It isn't possible to have two UI threads with glib because it is not thread safe.

TEST=try
BUG=69674
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=71471

[jonathan.backer, 2011-01-14 19:30:30]

http://codereview.chromium.org/6296006/diff/1/chrome/browser/gpu_process_host.cc
File chrome/browser/gpu_process_host.cc (right):

http://codereview.chromium.org/6296006/diff/1/chrome/browser/gpu_process_host...
chrome/browser/gpu_process_host.cc:578: // the message loop for the CPU child
window. A long-term fix might be to make
I don't think that we need to run a UI thread for message pumping: I ran the gpu
process as a separate process, but changed the type to an IO message pump in
gpu_main.cc, and it seemed to work just fine.

http://codereview.chromium.org/6296006/diff/1/chrome/browser/gpu_process_host...
chrome/browser/gpu_process_host.cc:580: // GPU child window so the GPU thread
can run in the browser process.
The way that the code currently works (it has changed a few times now ;-D --- by
my own hand) is that the GPU child window is "owned" by the browser.
Modifications to this window have to go through the browser (for example window
resizes end up at the GPU through the glResizeCHROMIUM command, which is finally
acted on in the browser via a sync IPC call). This is very different from the
windows code.

My current hypothesis about the problem is that X is not thread safe. This is
consistent with part of the browser architecture: it has two connections to the
X server -- one on the UI thread and one on the BACKGROUND_X11 thread. The
purpose of the second thread is to quickly answer some geometry requests from
the renderer. The fact that we need two connections says to me that a single
connection is not thread safe.

I would rephrase the TODO:
This cannot be a UI message pump on Linux because glib is not thread safe.
Changing this to an IO message pump does not completely resolve the problem,
most likely because we're sharing a connection to the X server with the browser.

[apatrick_chromium, 2011-01-14 19:33:16]

Comment updated.

[jonathan.backer, 2011-01-14 19:38:59]

LGTM