On Thursday 02 September 2010, Ralf Wildenhues wrote: > Well, the only qualm I have with this is that it will in the long > run make it less likely that we are portable to /bin/sh too. Not > a problem per se, but being able to do > cd tests > ./foo.test (or $srcdir/foo.test) > > to quickly run a test interactively may then expose a different > behavior than ordinary 'make check' would. We could adjust the > documentation to state that in order to run tests manually, one > should do > $good_shell ./foo.test > > but if you haven't used CONFIG_SHELL for configure (which is the > case for virtually all users), they might not even know the value > of $good_shell that would be used during 'make check'. Could > cause slightly less reliably reproducible bug reports. True. A not-so-big issue IMHO, but an issue nonetheless. > The patch has obvious advantages though. How to best ameliorate > the disadvantages? Hmm... maybe by making `tests/defs' re-execute the current test with $CONFIG_SHELL unless a proper environment variable (let's say `AM_TESTSUITE_NO_REEXEC') is set? Then we can set AM_TESTSUITE_NO_REEXEC in TEST_ENVIRONMENT (or its equivalent under `parallel-tests' option), to avoid useless re-execs.
I think we shouls apply the current patch as-is, and leave the outlined follow-up patch as material for the tests-init branch. WDYT? Thanks, Stefano
