On Thu, Jun 13, 2024 at 10:25:28PM +0200, Patrice Dumas wrote: > On Mon, Jun 10, 2024 at 07:11:53PM +0200, Bruno Haible wrote: > > It sounds like a bug in /bin/sh. And indeed, when I use bash to execute this > > script, no problem any more. It's not the first /bin/sh bug on this system > > that we encounter: > > https://lists.gnu.org/archive/html/bug-gnulib/2016-01/msg00030.html > > > > My workaround for this kind of bugs is to define the environment variable > > CONFIG_SHELL=/usr/bin/bash > > in my environment. > > But all invocations of the tp/tests/run_parser_all.sh script should then > > run it through > > ${CONFIG_SHELL-/bin/sh} tp/tests/run_parser_all.sh > > instead of > > tp/tests/run_parser_all.sh > > $(SHELL) is already used in some cases and for rules generated by > autoconf. It seems to abide to CONFIG_SHELL setup, but only when passed > on ./configure command line, not at 'runtime'. I would propose to use > $(SHELL) in every make rule that call a shell script, and also possibly in > shell scripts that call other shell scripts. > > What do you think?
If we could find out why this extra file "0" is being reported as existing, we could try to fix that problem only, rather than requiring bash for all tests or shell scripts.
