Re: backward compatability of tools

[Eric Siegerman]

On Wed, Feb 19, 2003 at 01:51:01PM -0800, Paul Eggert wrote:
> Eric Siegerman <address@hidden> writes:
> > [The "regression" test for shell-function support] first appeared in 2.55, 
> > in mid-November, 2002 (not
> > counting betas).  How long would it be appropriate to keep
> > waiting for complaints?
> 
> I'd say 6 months is enough.

My first thought was to agree.  But then, for a tool like
autoconf, there's a fair lag time between when a new version
comes out and when it starts being heavily exercised by end
users.  First, package maintainers have to upgrade to the new
autoconf release, then they have to cut stable releases of their
own, then end users have to download those and (try to) install
them.

Six months after AC 2.55 came out, it might be just starting to
make it into the real world where the ancient boxes in question
might -- or might not -- still live.  I don't know one way or the
other; does anyone else have a feel for how widely AC 2.55+ has
been disseminated?

> After that, the people who don't want to
> assume shell functions will likely stick with Autoconf 2.13 anyway.

Hmmm, that brings up GCC.  I know they have their own reasons for
sticking with 2.13 (or had, last time I checked), but AC's
dropping old-box support might be one more, given that GCC is
seen as (among other things) a way to bootstrap the rest of GNU
onto weird systems.

If people don't think this is an appropriate line of reasoning,
I'll accept that, but it seemed worth mentioning.

--

|  | /\
|-_|/  >   Eric Siegerman, Toronto, Ont.        address@hidden
|  |  /
A distributed system is one on which I cannot get any work done,
because a machine I have never heard of has crashed.
        - Leslie Lamport