On Saturday 20 November 2010, Ralf Wildenhues wrote: > * Stefano Lattarini wrote on Mon, Nov 15, 2010 at 06:18:15PM CET: > > Tests defs: don't let useless variables leak in test scripts. > > Tests defs: new subroutine `skip' for test skipping. > > Tests defs: some cleanup and minor fixes. > > > No "ticking clock" for this patches at the moment; the clock will be > > started only after the current testsuite regressions have been solved. > > Patches (1) through (3) will have to wait until somebody has figured out > why the *+#&$)! /bin/sh on Tru64 exits upon > set +e > case `(set -o) 2>/dev/null` in *posix*) set -o posix;; esac > > if 'set -e' has been called at some earlier point (leading to the > instspc*test and probably other spurious test failures). Might be > some internal shell data corruption, not sure. > Wild guess: what about this? case `(set -o) 2>/dev/null` in *posix*) set -o posix;; *) :;; esac
Anyway, in the long run, I think it would be simpler and more reliable to run the tests with configure-detected $SHELL, and add proper configure checks to reject overly buggy shells. The pending patch of mine "Testsuite: use $SHELL to run tests which are shell scripts": <http://lists.gnu.org/archive/html/automake-patches/2010-09/msg00022.html> can help with this (it implements the fist step). Regards, Stefano