Re: Bootstrapping from cvs

[Paul Eggert]

> From: Akim Demaille <address@hidden>
> Date: 16 Sep 2002 11:30:35 +0200
> 
> I suggest installing 2.54 from its tarball.

What I usually do is check out a fresh copy of the Autoconf sources
from CVS.  That way, they'll have the proper time stamps.  This takes
a bit longer over the Internet, but I'm multitasking anyway so it
doesn't take much of my time.

Here's an idea to fix this problem: add an option to 'cvs update' so
that it sets the time stamps of the updated files to equal the time
stamps in the repository.  That way, I could easily maintain a local
copy of the latest CVS version here with "cvs update".  To avoid
messing up my time stamps I wouldn't ever do a "make" directly in my
copy; instead, I would always make a copy of the copy first.

An even more ambitious idea would be to have 'cvs update' to do the
right thing by default: it could make sure that in the case of simple
updates, that if timeof(S)<timeof(T) in CVS, then timeof(S)<timeof(T)
in the working files, even if the working file timestamps disagree
with the CVS timestamps.  This more-ambitious approach would allow
'cvs update' to not screw up the timestamps that 'make' depends on.
But it would be harder to implement and trickier to explain.

> From: Jim Meyering <address@hidden>
> Date: Mon, 16 Sep 2002 15:42:58 +0200
> 
> The CVS folks found a way around this.
> Their current repository has a file called noautomake.sh in it,

I'd rather not head down that path.  I have often been burned by
Makefile.in being inconsistent with Makefile.am, configure being
inconsistent with configure.ac, etc.

I can well sympathize with the problem of bootstrapping Autoconf.  I
have a similar problem with Automake; I still can't run CVS Automake
on Solaris, and some of my problems have been bootstrapping-related.
But I would rather solve the underlying problems than mask them.