Re: lock-up on SIGSTOP/CONT

[Sven Mascheck]

Paul Jarc <prj@po.cwru.edu> wrote:

> So do I understand correctly that any of these will avoid the problem:
> - reprogram the child to send itself SIGTSTP instead of SIGSTOP

Adams patch.

> - reprogram the parent to not catch SIGCHLD

Nah, usually it should catch it, but it can ask the kernel to not
deliver it for jobcontrol reasons: sigaction/SA_NOCLDSTOP,
it worked very well.

> - reprogram the parent to wait in the SIGCHLD handler with either
>   WNOHANG or WUNTRACED

I am not sure about WNOHANG (and that's why i was careful with my arguments
in the posting), WUNTRACED seemed to have sufficed for me.  The BSD variant
of SA_NOCLDSTOP.

I haven't even read Stevnes APUE yet, and i am "walking on thin wood"
with these things, not programming too much.

> The kernel isn't supposed to get rid of zombies, [...]
> the kernel should assign its children as children of process 1,

Oh yes you're so certainly right!
BTW, init(1/8) actually tends to fail about this on said
AIX, Free- and OpenBSD versions.

And: "Rex A. Roof" at arbornet.org reminded me that this is a reason,
why you have to insist in suspending a _login_ shell.  With xterm(1)
it's certainly a different issue, in turn.