Bruno Haible wrote: > Yes, that's my point. Instead of using unnormalized file names nearly > everywhere and normalized file names only in a few places, it is better > to use normalized file names nearly everywhere and unnormalized file names > only in a few places. It causes fewer bugs and less head-aches. And it > is what the previous 'joinpath' function did achieve.
Agreed, I reverted that patch and pushed it before it causes any trouble [1]. I still think it would be nice to remove the os.path.join() vs. joinpath() thought. And double check the paths are normalized everywhere that it matters. Do you still expect GNULIB_TOOL_IMPL=sh+py to pass by the way? I could just do path normalization everywhere (as long as there are no Make variables) which would make everything easier. But it would likely cause small diffs where gnulib-tool.sh doesn't do the same. Collin [1] https://git.savannah.gnu.org/cgit/gnulib.git/commit/?id=7a99ec181aca5cf5839b62e632f7ec4c82b9a845
