* Stefano Lattarini wrote on Tue, Jan 18, 2011 at 11:57:24PM CET:
> On Tuesday 18 January 2011, Ralf Wildenhues wrote:
> > This patch fixes all but 9 of the 76 or so instspc*.test failures on
> > Tru64. It is for the tests-init branch, or a tests-init-fixes on top of
> > v1.11-395-ge118126 I guess. I'll wait a couple of days before pushing.
> >
> The patch is ok with me (but see below). BTW, the idea of having an
> 'errexit' clean `tests/defs' sounds good to me, indipendently from the
> issue at hand. But I won't go as far as asking for a separate patch ;-)
That shouldn't be hard though. It can be done after we have
framework_failure_. One needs to remember that `unset NOT_SET' can
fail, but maybe it is sufficient to
{
unset FOO BAR BAZ
unset BLA ...
} || :
(haven't checked Sven's table yet).
> > I haven't analyzed the failure sufficiently to be able to write a patch
> > for autoconf.texi yet. But the very last hunk below makes me suspicious
> > of more removals of '|| Exit 1' "just because".
> >
> IMVHO, we should just reject a shell with such an untrustworthy `set -e'
> behaviour.
I guess one of my points is that there exist systems out there (that we
do not regularly test) on which we may not find better shells, but that
still have a nonzero user base. In that light, I regard your pending
patch as "can have sweeping under the rug effects". ;->
> My pending patch "Testsuite: use $SHELL to run tests which
> are shell scripts":
> <http://lists.gnu.org/archive/html/automake-patches/2011-01/msg00033.html>
> might be a first step in ensuring that the automake tesuite is run with
> a "sane" shell (if that's available).
Yes I will eventually approve (a revised version of) that patch, but I
really would like us to be old-shell clean as far as is possible without
too much jumping through hoops.
> That said, ATM I'd just apply your patch, it's definitely worth to have
> if it removes 60-70 spurious test failures!
Agreed, and pushed.
Thanks!
Ralf
