From: Johannes Sixt <j.sixt@viscovery.net>
Date: Thu, 17 Jul 2008 07:22:05 GMT
Subject: Re: [PATCH 1/2] t/test-lib.sh: Let test_must_fail fail on signals only
Message-ID: <487EF31D.8090007@viscovery.net>
In-Reply-To: <20080717063856.GA10450@sigill.intra.peff.net>

Jeff King schrieb:
> On Wed, Jul 16, 2008 at 11:31:41PM -0700, Junio C Hamano wrote:
>> Is it that somebody do not want 255 exit value, or anything that has 7th
>> bit set?  2488df8 (builtin run_command: do not exit with -1., 2007-11-13)
>> suggests otherwise at least for Windows runtime, so what we currently have
>> that does extra truncation ourselves might be sufficient.
> 
> Johannes will have to answer that; however, the truncation there does
> leave the extra 7th bit. Maybe & 0x7f would be more appropriate?

I never found out the real reason why -1 would not be recognized as
"failure"; the conclusion of my debugging session was that MSYS bash has
an issue, and I chose to append '& 0xff' because the documentation of
WEXITSTATUS() says that it can receive only 8 bits of the exit() code. The
intention of 2488df8 was to keep as much information as possible. But if
that extra information hurts, we should better truncate to 7 bits.

The source code of Windows's C runtime suggests that any value that fits
in 4 bytes can be supplied to exit() and can be received by cwait()
(Windows's version of waitpid()); but I haven't looked at how MSYS
implements waitpit() and whether it can receive that much information.

-- Hannes