Don't spawn a new engine if the pipe is busy

Don't spawn a new engine if the pipe is busy

[Oleksandr, 2013-06-06 12:21:13]

On 2013/06/06 10:36:39, Felix H. Dahlke wrote:

I don't think this change is enough. We might eliminate some unnecessary process
spawning, but what if, for example 5 processes all call the engine
simultaneously. Won't we end up with 5 calls to WaitNamedPipeW when there is no
pipe available yet? In that case it will immediately fail and we'll be back into
process spawning code 5 times. 
I was thinking maybe we should use a global mutex here, so that we'd be sure we
only start the process once. Also maybe we could do a check in the engine itself
if other instances are running, and kill the current instance, if there are.

[Felix Dahlke, 2013-06-06 12:36:39]

On 2013/06/06 12:21:13, Oleksandr wrote:
> On 2013/06/06 10:36:39, Felix H. Dahlke wrote:
> 
> I don't think this change is enough. We might eliminate some unnecessary
process
> spawning, but what if, for example 5 processes all call the engine
> simultaneously. Won't we end up with 5 calls to WaitNamedPipeW when there is
no
> pipe available yet? In that case it will immediately fail and we'll be back
into
> process spawning code 5 times. 
> I was thinking maybe we should use a global mutex here, so that we'd be sure
we
> only start the process once. Also maybe we could do a check in the engine
itself
> if other instances are running, and kill the current instance, if there are.

Yeah, this is only one side of the table. I'm currently working on making sure
that the engine tries to open the pipe right away, and that it dies right away
if that fails. That should suffice, I think.

[Wladimir Palant, 2013-06-06 13:28:06]

LGTM