Re: [Bug-gnulib] addition: wait-process.h, wait-process.c, 2nd round

[Paul Eggert]

Bruno Haible <address@hidden> writes:

> I think this scenario is
>   - extremely rare, due to the time it requires until a PID is reused,
>   - due to an inherent limitation of waitpid in POSIX: there is no way to
>     ask waitpid: "Tell me when the process terminates but keep the process
>     in zombie state afterwards."

If you can assume the XSI standard extension to POSIX, it sounds like
there is a way to solve the problem: you can pass WNOWAIT to waitpid.
WNOWAIT keeps the process whose status is returned in a waitable
state. The process may be waited for again with identical results, and
this sounds like it's what you want.  See
<http://www.opengroup.org/onlinepubs/007904975/basedefs/sys/wait.h.html>.

So you could use WNOWAIT if available, and fall back on the current
code if it's not available.


> > I'm a bit dubious about the semantics, portability, and
> > understandability of structs whose members are volatile.
> 
> It's the same as with const:

Yes, but the same objections apply to const.  Anyway, your other
arguments were good ones, though I admit that I still have a few
qualms, as it's unlikely that compiler bugs in this area will be found
and fixed.