Howdy everybody, we recognized here, that there are two problems with passing all the AC_DEFINE's to (preprocessed) Fortran compilation (through $DEFS or through a header file): 1.) If passed through $(DEFS): Some compilers (at least IBM xlf* on AIX) won't accept '-D' and bail out with an error. See https://lists.gnu.org/archive/html/autoconf-patches/2011-04/msg00004.html also. 2.) Most of the C-defines or aren't relevant for Fortran, and their substitution within preprocessed Fortran might lead to surprising results, take AC_C_INLINE as an example. And for Fortran, it won't matther if we HAVE_UNISTD_H.
A colleague of mine proposed that it would be cleaner to have language specific definers like "AC_DEFINE", "AC_FORTRAN_DEFINE" (and maybe others, but I only care about Fortran). Some clever Autoconf macro could then check how to pass defines to the Fortran compiler and build up FDEFS (which automake would use in turn for building up the compiling command line). On the other hand, if we find an AC_CONFIG_HEADERS ([ ... config.inc ...]) in configure.ac, just leave FDEFS empty. Of course it would be also nice if we had some 'autoheader --fortran' to create config.inc.in from the AC_FORTRAN_DEFINEs it can find. What do you think about all this: In your opinion, is this solution clean? Does it suffice to implement just AC_FORTRAN_DEFINE or would you generalize that to other languages, too? Best, Nicolai
