Hi Eric, > Typo in the ChangeLog; s/AM_DEFUN_ONCE/AC_DEFUN_ONCE/
Oops. Fixed now. Thanks. > > +m4_version_prereq([2.64],[AC_DEFUN_ONCE],[AC_DEFUN])([AM_ICONV], > > ... rather than have a version check, I'd rather see: > > m4_ifdef([AC_DEFUN_ONCE],[AC_DEFUN_ONCE],[AC_DEFUN])([AM_ICONV], Remember that AC_DEFUN_ONCE existed all the time in autoconf, since 2.52 or earlier, but it was not really usable before 2.64. (See autoconf/NEWS.) With this m4_ifdef, equivalent to an unconditional use of AC_DEFUN_ONCE, users of autoconf < 2.64 without gnulib would encounter warnings that they have not seen so far. > That's great for gettext, where you don't require autoconf 2.64 but also > don't guarantee AC_DEFUN_ONCE. But what about for gnulib, where we > guarantee that even with autoconf 2.59, we have a working AC_DEFUN_ONCE? I don't see how to test for a new vs. an old definition of AC_DEFUN_ONCE. Therefore these users will have a warning occasionally. But users of gnulib tend to live on the "bleeding edge" and can afford to install autoconf 2.64. Bruno
