From owner-freebsd-hackers  Tue Apr 23 16:50:31 1996
Return-Path: owner-hackers
Received: (from root@localhost)
          by freefall.freebsd.org (8.7.3/8.7.3) id QAA19310
          for hackers-outgoing; Tue, 23 Apr 1996 16:50:31 -0700 (PDT)
Received: from vent.pipex.net (root@vent.pipex.net [158.43.128.5])
          by freefall.freebsd.org (8.7.3/8.7.3) with SMTP id QAA19277
          for <hackers@freebsd.org>; Tue, 23 Apr 1996 16:50:20 -0700 (PDT)
Received:  from dial.pipex.com by vent.pipex.net (8.6.12/PIPEX simple 1.20)
	id AAA16731; Wed, 24 Apr 1996 00:49:38 +0100
Received: (from jraynard@localhost) by dial.pipex.com (8.6.12/8.6.12) id XAA01870; Tue, 23 Apr 1996 23:47:57 GMT
Date: Tue, 23 Apr 1996 23:47:57 GMT
From: James Raynard <jraynard@dial.pipex.com>
Message-Id: <199604232347.XAA01870@dial.pipex.com>
To: julian@ref.tfs.com
CC: hackers@freebsd.org
In-reply-to: <199604222244.PAA20222@ref.tfs.com> (julian@ref.tfs.com)
Subject: Re: Flaws in system() implementation?
Sender: owner-hackers@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

>>>>> "JULIAN Elischer" <julian@ref.tfs.com> writes:
> 
> > Inspired by Rich Stevens' implementation of system() in his book
> > "Advanced Programming in the Unix Environment", I've been looking at
> > the FreeBSD implementation and was struck by the following points:-
> > 
> > 1. Use of a wait union when an int would do.
> This is the correct POSIX thing to do..
> check the man page for wait.

According to the 2.1.0-RELEASE man page, wait and friends use an int*
for the status variable.

> > 3. Returns 0 if fork() fails, when -1 seems more appropriate.
>  POSIX says:

Thanks for posting this, I don't have access to any POSIX specs
(except the odd snippet in books). What's the best way of getting hold
of them?

>  If "command" is a null pointer, the system() function returns non-zero
>  only if a command processor is available.

OK (Out of interest, what kind of circumstances would result in a
command processor not being available?)

>  If "command" is a nin null pointer the system() function returnsa
>  the termination status of the command language interpretter in the
>  format specified by the waitpid() function.

OK

>  [...]
>  If a child process cannot be  created,
>  or if the termination status of hte command interpretter cannot be obtained
>  system() returns -1
>  and sets errno to indicate the error.

Doesn't that imply that system() should return -1 if fork() fails?
(Forgive me if I'm being obtuse, I'm not very experienced at reading
standardese).

> > 4. The SIGINT and SIGQUIT are not ignored until after the fork().
> >    If the child runs first, one of these signals could in theory be
> the child MUST run first in vfork()

I know, but I'm trying not to use vfork(), in view of the recent
discussion (in -current, I think) about replacing it with a call to
fork().

Cheers
James