Re: [RFC] Add dummy toplevel configure.ac

[Zack Weinberg <zackw at panix dot com>]

On Wed, Mar 15, 2006 at 10:36:33AM +1100, Ben Elliston wrote:
> > Instead, how about something like this in the existing configure.in?
> > 
> > AC_PREREQ(2.13)
> > dnl ... and _not_ any later version.  It is too much work to test the
> > dnl actual version number, which is hidden in a macro whose name
> > dnl changes repeatedly in the releases between 2.13 and current.
> > dnl However, there's a simpler test: the m4_ macros exist only in
> > dnl 2.50 and above.
> > ifdef([m4_ifdef],
> >   [AC_FATAL([toplevel configure should be created using autoconf 2.13])
> > ])dnl
> > 
> > AC_FATAL is available in autoconf 2.50 (yes, I downloaded the old
> > release tarball and checked) so this should work with any version.
> > [Well, I'm not sure if versions exist that don't provide 'ifdef'
> > at all, only 'm4_ifdef', but there we're just plain screwed - how
> > can you query the existence of the very primitive one uses to query
> > existence?]
> 
> How about using AC_MSG_ERROR, which has been available practically
> forever?

That issues the error when the generated script is run, whereas we want
the error at generation time.  I'm happy with AC_FATAL here, it's available
in all cases where it's needed.  I'm less happy with the test, perhaps you
have a better idea there?

zw