Jim Meyering <jim <at> meyering.net> writes: > > [3/5] dirname-lgpl: adjust clients that don't need full dirname > > Fallout from the split. In particular, test-mkdir and test-rename no longer > > need libintl. Meanwhile, openat and friends no longer drag in dirname, but > > still drag in xalloc due to openat-die.
There's still a problem - rename depends on canonicalize-lgpl, but coreutils uses gnulib-tool --avoid=canonicalize-lgpl --import canonicalize. On the surface, this works just fine - canonicalize_file_name is just as easy to write as a call to the more powerful canonicalize_filename_mode, rather than a wrapper around realpath. Unfortunately, canonicalize_filename_mode is GPL, and liberally uses xmalloc. Which means that the GPL version of canonicalize_file_name(), and thus rpl_rename(), can now call exit() (making it unsafe for library use), and that test-rename fails to link for coreutils. But we can't just solve it by adding @LIBINTL@ to the link line for test-rename, since then it would fail to compile when using just canonicalize-lgpl rather than canonicalize. In other words, I don't know of any way to write modules/rename-tests to conditionally include @LIBINTL@ only if the GPL rather than the LGPL canonicalize_file_name is in use. I think the only clean solution is to bite the bullet and decide that canonicalize-lgpl is the only module which can provide canonicalize_file_name, which in turn means that coreutils needs to quit doing --avoid canonicalize- lgpl. -- Eric Blake
