From: Jeff King <peff@peff.net>
Date: Thu, 17 Jul 2008 06:01:43 GMT
Subject: Re: [PATCH 1/2] t/test-lib.sh: Let test_must_fail fail on signals only
Message-ID: <20080717060143.GA3338@sigill.intra.peff.net>
In-Reply-To: <7v3am9m5ne.fsf@gitster.siamese.dyndns.org>

On Wed, Jul 16, 2008 at 10:38:45PM -0700, Junio C Hamano wrote:

> "Only the least significant 8 bits (that is, status & 0377) shall be
> available to a waiting parent process".  So it is not just "at least on
> Linux" but is a well defined behaviour.
> 
> http://www.opengroup.org/onlinepubs/000095399/functions/exit.html

Ah, thanks. I read that same text in the Linux manpage but didn't think
to check that it was POSIX. However, some of our systems aren't quite
POSIX...check out 2488df84 (builtin run_command: do not exit with -1) by
Johannes Sixt [who is now cc'd]. I assume that was a Windows fallout.

> I would however agree that when we do mean 255 we should probably write
> 255, not (-1).  It is easier to document things that way.

I started to fix the callsites that Stephan mentioned, but it really is
convenient to be able to 'return error("foo")' (or even return
func_that_calls_error(), and tracking down deep calls is time consuming
and error prone). So maybe we should just enhance the change from
2488df84 and special case "-1" into "1"?

diff --git a/git.c b/git.c
index 6b600b5..4f28e8c 100644
--- a/git.c
+++ b/git.c
@@ -240,7 +240,7 @@ static int run_command(struct cmd_struct *p, int argc, const char **argv)
 
 	status = p->fn(argc, argv, prefix);
 	if (status)
-		return status & 0xff;
+		return status == -1 ? 1 : status & 0xff;
 
 	/* Somebody closed stdout? */
 	if (fstat(fileno(stdout), &st))

-Peff