Allow child processes to abort quitting.

Allow child processes to abort quitting.
This is useful if a renderer process starts to quit but is then given more work to do from the browser (e.g., a new RenderView), such that it should no longer quit.

BUG=65953
TEST=Call AddRefProcess after ReleaseProcess, before queue is emptied.

[Charlie Reis, 2011-05-03 00:41:33]

This is useful for the window.opener CL I'm working on at
http://codereview.chromium.org/6319001/.

In that CL, we make it possible for RenderViews to be swapped out and later
swapped in again, and we allow their process to exit if its ref count reaches 0.
 However, it's possible for the user to go away and come back quickly, such that
we start processing a new navigation while a quit is pending on the MessageLoop.
 That's messy, since the process then exits as soon as the MessageLoop becomes
idle.

By adding CancelQuit, we can make it possible to put the process back in service
if the ref count increases again.

Does this sound like a sane change?

Thanks,
Charlie

[jam, 2011-05-03 00:53:29]

see all the ChildProcess messages regarding shutdown.  This is to avoid the race
condition that you're seeing.  The child process only quits if the browser
process tells is it's ok.

The browser side is currently only implemented in ChildProcessHost, which is
used by all processes except the renderer.  So if you implement this in
BrowserRenderProcessHost, you can reuse the same logic.

[Charlie Reis, 2011-05-04 16:49:07]

On 2011/05/03 00:53:29, John Abd-El-Malek wrote:
> see all the ChildProcess messages regarding shutdown.  This is to avoid the
race
> condition that you're seeing.  The child process only quits if the browser
> process tells is it's ok.
> 
> The browser side is currently only implemented in ChildProcessHost, which is
> used by all processes except the renderer.  So if you implement this in
> BrowserRenderProcessHost, you can reuse the same logic.

Thanks.  I've taken that approach in http://codereview.chromium.org/6927014/.