Eric Blake wrote:
> According to Jim Meyering on 1/14/2010 1:08 AM:
>> Think of a set-up function that (when $EXEEXT is nonempty)
>> iterates through the *.$EXEEXT executables in a specified directory...
>>
>> create_exe_shim_functions ()
>> {
>> case $EXEEXT in
>> '') return 0 ;;
>> .exe) ;;
>> *) echo "$0: unexpected value of $EXEEXT" 1>&2; return 1 ;;
>> esac
>
> So we still have to pass EXEEXT through TESTS_ENVIRONMENT in
> modules/*-tests for any test that uses a .sh file. But other than that,
> all of our tests-*.sh no longer need to worry about EXEEXT, once they've
> been converted to use init.sh. I like it.
>
> Or go one step further, do the search for *.exe without regards to
> $EXEEXT, skipping this case block, and simplifying modules/*-tests.
I like the idea of not having to modify all modules/*-tests, but have
two reservations:
- I like having no shim-eval overhead when $EXEEXT is empty.
- What if a script were to run something named foo.exe _in unix land_?
Then if there is also a "foo" executable in that directory, it would
mistakenly run "foo" via our shim that creates a foo.exe *function*.
An alternative would be to make gnulib-tool always add EXEEXT.
>> # Remove the .$EXEEXT suffix.
>> base=${file%.$EXEEXT}
>
> This is not portable to Solaris /bin/sh. Unless we take the same measures
> as coreutils to ensure a (mostly) POSIX-compliant shell, I think we are
> better off using only features supported by common /bin/sh.
Good point. Not worth it.
base=$(echo "$file" | sed "s/$EXEEXT//")
