Re: [Bug-gnulib] proposed gettext.m4 patch for inttypes.h (merge from coreutils)

[Bruno Haible]

Paul Eggert wrote:
> I got this diagnostic:
>
> aclocal: macro `gt_HEADER_INTTYPES_H' required but not defined
>
> Copying gnulib/m4/inttypes.m4 fixed this.  However, that's a minor
> maintenance hassle, as inttypes.m4 is not needed for coreutils.

This problem affects all of
   codeset.m4, glibc21.m4, intdiv0.m4, intmax.m4, inttypes.m4,
   inttypes_h.m4, inttypes-pri.m4, isc-posix.m4, lcmessage.m4, longdouble.m4,
   longlong.m4, nls.m4, po.m4, printf-posix.m4, signed.m4, stdint_h.m4,
   uintmax_t.m4, ulonglong.m4, wchar_t.m4, wint_t.m4
in packages that call AM_GNU_GETTEXT([external]), These .m4 files are
needed for 'aclocal', although they are not needed for the final 'configure'
file.

The problem is that 'aclocal' doesn't know which macros are really needed,
it looks which macros are potentially needed.

Can someone please fix 'aclocal' so that, after determining which macros
are potentially needed, it runs autoconf with tracing enabled, to determine
which macros are really needed?

The alternative is to recommend that people don't define autoconf macros
with arguments, and that only autoconf macros that take no arguments are
created and documented. I think this "solution" would be inferior...

Bruno