Re: Migrating to autoconf 2.52g

[Tim Van Holder]

On Mon, 2002-02-04 at 12:28, John Poltorak wrote:
>
> > autoconf 2.5x provides 'autoupdate', which will try to do just that.
> 
> Does autoupdate create a new configure.in?

It does an in-place update of configure.in, making a backup of the
original one as configure.in~.

> When running autoupdate I get:-
> 
> + autoupdate
> autom4te.: no such file or directory: /dev/null
> autoupdate.: cannot close autom4te ...

I would guess that whatever OS/2 environment you use (EMX I guess) does
not provide emulation for /dev/null (or at least bash and/or perl
don't).  Perhaps autoconf needs to detect this:

DEVNULL=/dev/null
test -f /dev/null || { test -f NUL && DEVNULL=NUL }

or something similar.

> Also I notice a directory called:-
> 
> autom4te..cache
> 
> in the working directory. Is this name correct? - It looks wrong...#

Looks like a bug in perl or bash, also apparent in the error messages
above; it seems that $0 has a trailing '.', leading to incorrect set up
of the directory (though it should not pose a real problem, as autoconf
will always refer to that 'unusual' path; it's only a problem if the
name is ever hard-coded).

> > Be warned though that many applications used configure.ins that relied
> > on undocumented and unsupported autoconf internals; those will often
> > break with autoconf 2.5x, and may require non-trivial rewriting.
> > 
> > Basically, when confronted with an outdated configure.in, back up the
> > configure script and run a recent autoconf/autoheader (possibly in
> > tandem with a recent aclocal & automake).
> > If autoconf does not complain, things will probably be ok.
> > If it complains, try running autoupdate, and then autoconf again.
> 
> Can a test to see if these functions need to be run be incorporated into
> autoconf itself so that they are run automatically when necessary?

I don't know, but I doubt it.  If that were possible, I don't think we
would have needed autoupdate in the first place, as autoconf could have
handled everything directly.