[256] in NetBSD-Development
Re: Lola & project/netbsd changes
daemon@ATHENA.MIT.EDU (Greg Hudson)
Thu Dec 8 09:21:45 1994
To: jhawk@MIT.EDU
Cc: netbsd-dev@MIT.EDU
In-Reply-To: Your message of "Thu, 08 Dec 1994 04:28:46 EST."
<199412080928.EAA13102@lola-granola.MIT.EDU>
Date: Thu, 08 Dec 1994 09:21:27 EST
From: Greg Hudson <ghudson@MIT.EDU>
> Why is it that /bin/athena is a symlink to /usr/athena/bin? Under
> the standard Athena paradigm, /bin/athena is supposed to include
> everything you need to get to /usr/athena/bin (i.e. attach, kinit,
> aklog, etc.). Unfortunately, this is not the case. As a result,
> /usr/athena is merely a symlink to the syspacks instead of an
> "attach" symlink. Can someone explain to me the rationale for
> breaking the paradigm?
You do not need "kinit, aklog, etc." to get to /usr/athena, even if
/usr/athena is mounted with "attach" and not "ln -s". Although using
attach to mount /usr/athena might be nice, it seems ridiculous to have
a separate /bin/athena for the sake of exactly one binary.
Using "attach" for /usr/athena is not particularly beneficial; people
are only going to be mounting /usr/athena once, and that will be done
for them by an install script once I have it ready. Matt Braun has
argued that hesiod provides for a useful abstraction for version
control, but this isn't really true; symlinks work much better, since
they take effect immediately and not when someone re-attaches the
locker. (My SunOS machine in the zone was broken for some time
because /usr/athena moved after the packs build and I hadn't
re-attached the locker.)
There is actually another reason to keep /bin/athena and
/usr/athena/bin separate, and that's that config_afs doesn't function
correctly if AFS can't be mounted, since the "fs" binary isn't local.
This isn't particularly catastrophic; you just don't get the right
suidcells. I don't know if this is a compelling reason, though.
