Re: may Automake rely on _AC_SUBST_VARS?

[Ralf Wildenhues]

Hi Eric,

* Eric Blake wrote on Wed, Apr 08, 2009 at 05:53:00AM CEST:
> According to Ralf Wildenhues on 4/7/2009 12:41 PM:
> >> m4_define([_AC_COMPILER_EXEEXT],
> >> m4_defn([_AC_COMPILER_EXEEXT])[m4_provide([_AM_COMPILER_EXEEXT])])
> >>
> >> m4_provide_if([_AM_COMPILER_EXEEXT],
> >>   [AM_CONDITIONAL([am__EXEEXT], [test -n "$EXEEXT"])])
> > 
> > I assume that you mean that the m4_define part should happen before
> > AC_PROG_CC and the m4_provide_if part afterwards?  How would you ensure
> > the former?
> 
> The m4_define part would occur in isolation at the top level, as something
> that gets folded into aclocal.m4 alongside all your other
> m4_define/AC_DEFUN setups.  aclocal.m4 is sourced after autoconf.m4 (so
> you are guaranteed that the original _AC_COMPILER_EXEEXT that you are
> trying to hook has already been defined), and before configure.ac (and
> thus before any invocation of AC_INIT, let alone AC_PROG_CC).

Cool.  I tend to forget about this possibility.

> The m4_provide_if part would be in the same place that you were originally
> thinking of investigating _AC_SUBST_VARS.  And regardless of whether
> AC_PROG_CC or AM_INIT_AUTOMAKE was expanded first, the fact that you
> hooked _AC_COMPILER_EXEEXT should be visible.

Alright.  I think we have it sorted out now.  BTW, Automake really needs
to do the m4_provide_if dance, rather than just tacking code onto the
end of _AC_COMPILER_EXEEXT, because Autoconf may redefine
_AC_COMPILER_EXEEXT and expand its earlier definition inside a shell
conditional statement, see AC_NO_EXECUTABLES.

> > If it works then yes, that sounds easier.  Can Automake rely on the
> > existence of the _AC_COMPILER_EXEEXT macro then, and that it's this
> > macro which is called whenever EXEEXT may be needed?
> 
> Whether it is _AC_COMPILER_EXEEXT or _AC_SUBST_VARS, I think that once we
> have a working use in automake that we should definitely add the
> documenting comment into autoconf.

Yes.  I have pushed the inline documentation patch to Autoconf now,
see this thread for the patch I've pushed to Automake:
<http://thread.gmane.org/gmane.comp.sysutils.automake.patches/3528>.

Thank you for your help on this!

Cheers,
Ralf