Re: tests: work around Tru64 sh -e issues for instspc*.test.

[Eric Blake]

On 01/20/2011 02:04 PM, Ralf Wildenhues wrote:
> Hi Eric,
> 
> * Eric Blake wrote on Thu, Jan 20, 2011 at 10:01:15PM CET:
>> On 01/20/2011 01:50 PM, Ralf Wildenhues wrote:
>>> Ahh, but it would have needed to be in that very patch, unfortunately.
>>> As it is, my last patch fixed Tru64, but regressed NetBSD by roughly the
>>> same number of failures, because `unset NOT_SET' fails there.
>>
>> But bash 2.05a would dump core on 'unset NOT_SET', as documented in the
>> autoconf manual.
> 
> Erm, old bashs set an error status, but don't dump core.  The core
> dumping was, if I understand correctly, only with MAIL (and only that
> variable) and only one specific bash version.  But I haven't tested
> that now.  If things are different, I'd appreciate an update to
> autoconf.texi.

Yep, I stand corrected.  bash 2.05 doesn't dump core on unset, but
suffered from the same problem still in NetBSD /bin/sh in setting $?
incorrectly (which in turn affects set -e).  The core dump was bash
2.01, and affected MAILCHECK as well as MAIL.

http://groups.google.com/group/gnu.bash.bug/msg/5a41781550c61d62

But since bash 2.05a is pretty much dead these days, while NetBSD
/bin/sh is still alive and kicking, the autoconf section should be
updated to use a modern example in its claim that unset exit status is
still a portability problem to be aware of.

-- 
Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature