* Eric Blake wrote on Tue, Nov 03, 2009 at 02:11:07PM CET: > According to Ralf Wildenhues on 10/31/2009 9:40 AM: > > I'm seeing three failures on i386-unknown-freebsd6.4 due to $SHELL being > > zsh. Two are fixed with the patch below, OK? > > Yes, please apply.
Thanks, done. > > stdout: > > SHELL=/usr/local/bin/zsh > > + sed s/parent/child/ > > + set +x > > ../../autoconf/tests/m4sh.at:1290: ./child > > + ./child > > + SHELL=/bogus > > + export SHELL > > + cp stdout expout > > + mv child child.bak > > + set +x > > ../../autoconf/tests/m4sh.at:1295: $CONFIG_SHELL ./script > > + ./script > > So SHELL was zsh, but CONFIG_SHELL was unset; later, when we play with > SHELL, we lose out on the fact that it used to be zsh, and our setting of > /bogus means the parent script is run with the wrong shell. I'm not sure > if that is a bug in the testsuite or in m4sh. Probably a bug in the testsuite. We currently document that $some_whell ./script should only be used in conjunction with adding CONFIG_SHELL=$some_shell to the environment of ./script. Note that adding export CONFIG_SHELL to atlocal is the right solution only iff all m4sh (and thus configure) script invocation inside the testsuite really use $CONFIG_SHELL to start the script. Sigh. Cheers, Ralf
