Omar Polo:

> Not sure if it would help, but since we use that idiom to either start a
> perl script that behaves like a http server or nc(1) to mimick a smtpd
> server, we could just move the timeout handling to perl using, for e.g.,
> alarm().

Maybe.

I'm not enthusiastic about relying on timeouts.  They're a notorious
source of unreliability in regression suites if a machine is too
fast or too slow.

I need to check how the various make(1)s and shells interact with
the process group handling.  I'm probably overthinking this.

-- 
Christian "naddy" Weisgerber                          naddy@mips.inka.de