childprocess slower than background_process

[aslakhellesoy]

After the switch to childprocess things are a lot slower. Compare cucumber's own features running with aruba 0.2.7 and 0.2.8:

aruba 0.2.7:

$ time cucumber features

26 scenarios (26 passed)
184 steps (184 passed)
0m19.330s

real    0m19.889s
user    0m15.434s
sys 0m3.922s

aruba 0.2.8:

$ time cucumber features

26 scenarios (26 passed)
184 steps (184 passed)
0m31.335s

real    0m31.895s
user    0m15.438s
sys 0m4.017s

Also see comment in #41 - the slowness is observable in aruba too. Since I'm running on OS X I'm not sure why things are slower - I suppose both background_process and childprocess use the same fork etc. methods? Or not?

Please close this issue if this can be resolved with a patch to childprocess.

[aslakhellesoy]

Ouch. childprocess has a poll_for_exit method that sleeps. Not sure if that's it...

[msassak]

poll_for_exit was my first suspicion. I did add a sleep when reading output (patch to modify that coming soon) but that shouldn't add up to this much of a difference. I'll look into it and let you know.

[msassak]

I have a fix in my wait-for-io branch here: https://github.com/msassak/aruba/tree/wait-for-io

The problem was because I was stupidly waiting for IO whether the process was still running or not. I think I'll be able to improve this even more, but this helps considerably when running Aruba's own features. On my laptop it reduces runtime from 27 to 6 seconds.

[aslakhellesoy]

Attribution. Closed by 23cd24e.