This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: flag day for Solaris portions of config.{guess,sub}
- From: "Zack Weinberg" <zack at codesourcery dot com>
- To: Paul Eggert <eggert at CS dot UCLA dot EDU>
- Cc: Ben Elliston <bje at wasabisystems dot com>, gcc at gcc dot gnu dot org, binutils at sources dot redhat dot com, gdb at sources dot redhat dot com, rms at gnu dot org
- Date: Mon, 01 Dec 2003 13:29:28 -0800
- Subject: Re: flag day for Solaris portions of config.{guess,sub}
- References: <8765hf4c8z.fsf@wasabisystems.com><87wu9mt79r.fsf@egil.codesourcery.com><871xrs5b9j.fsf@penguin.cs.ucla.edu> <87znegqb31.fsf@codesourcery.com><87brqsw9d9.fsf@penguin.cs.ucla.edu>
Paul Eggert <eggert@CS.UCLA.EDU> writes:
> "Zack Weinberg" <zack@codesourcery.com> writes:
>
>> Once a pattern of canonical names has been chosen for a given family
>> of operating systems, that pattern must not ever change.
>
> That's still too strong. Changing canonical names is not something
> one wants to do lightly of course, but it's not unprecedented. We
> have changed the output of config.guess in the past, notably for
> GNU/Linux.
I would argue that every last one of those changes was a mistake, but
a mistake that cannot now be rectified.
> That being said, I'm sympathetic to the design principle you're
> advocating. Ironically, this whole problem occurred because we didn't
> follow that principle: we changed the pattern of canonical names for
> part of the SunOS family of operating systems from -sunos* to
> -solaris*.
Again, this was a mistake, which *cannot now be rectified*.
Changing it again would be *worse* than the original change was - the
original change happened when solaris2 was still a new thing, not
widely used, and (critically) CPU-sun-solaris2.x / CPU-sun-sunos5.x
patterns did not appear in a large number of autoconf scripts.
My point is really that you and others advocating the change seem to
underestimate the disruption involved by orders of magnitude. I think
it's roughly comparable to the disruption involved in the switch from
autoconf 2.13 to autoconf 2.5x -- every last configure script on the
planet is going to have to be audited for problems, and possibly
modified.
And configure scripts aren't the only things that use config.guess/
config.sub. Consider FTP archives and automatic programs that
retrieve files from those archives. Consider system administrators
using cfengine to manage large heterogeneous networks. Consider old
backup tapes labelled and formatted according to canonical system name.
Is smoothing out a minor irregularity of naming convention really
worth all this disruption?
zw