Re: Broken makefile given Autoconf version mismatch

[Noah Misch]

On Thu, Apr 13, 2006 at 09:54:31AM +0200, Alexandre Duret-Lutz wrote:
> Noah:
> | `automake' can generate a Makefile.in that does not fit
> | `configure' if the version of Autoconf it uses to trace
> | `configure.ac' differs from the version used to generate
> | `configure'.
> 
> The Autoconf version used by Automake should also be the same
> used by aclocal.  Maybe aclocal could output some code in
> aclocal.m4 to ensure that the autoconf version that reads
> aclocal.m4 is the same as the version that was traced to
> generate it.  
> 
> The consequences of such a test are not entirely clear to me,
> but it looks like it would prevent the version mismatch we are
> discussing:

This solution is elegant.  It would have prevented the problem I described,
along with a whole family of potential, similar failures.  There could be subtle
consequences that I am not seeing, either, but your description makes sense, and
no obvious snags come to mind.

This will also affect `autoheader'; we'll probably want to make it return $?=63
on version mismatch, like the Automake-supplied tools.

> In case this check doesn't sound appropriate for
> non-automake-based package, we could also arrange that it be
> only performed inside AM_INIT_AUTOMAKE.

Yes.  The check could be in a macro that AM_INIT_AUTOMAKE calls, and that
non-Automake packages could call directly, should they want the protection.


Many thanks!  This is the sort of solution I had hoped for.  It's good that
Autoconf 2.60 will explicitly clean up $datarootdir breakage, but this will
reduce the need for such workarounds in the future.