Simon Josefsson wrote: > Below is a strange error that can be seen on: > > http://autobuild.josefsson.org/gnulib/log-201002171652277193000.txt > > There are no make -j switches here, which I suspected could be involved. > > ./init.sh: line 139: `test-acos': not a valid identifier > FAIL: test-pread.sh > ... > ./init.sh: line 139: `test-acos': not a valid identifier > FAIL: test-xalloc-die.sh > ... > ./init.sh: line 139: `test-acos': not a valid identifier > FAIL: test-xstrtoll.sh > > It is just three of the self-tests that fails in this way. > > The relevant init.sh line is: > > if test -n "$base_names_"; then > for base_ in $base_names_; do > # Create a function named $base whose sole job is to invoke > # $base_$EXEEXT, assuming its containing dir is already in PATH. > -> eval "$base_() { $base_$EXEEXT"' "$@"; }' > done > fi > > Could the reason for 'test-acos' showing here be that it is the first > (alphabetically) test?
Right. The *.exe names are sorted, and test-acos is the first one encountered. Is there a more functional shell on that system? If so, is it already detected and thus mentioned somewhere in config.status?
