Hi, On Sat, Jun 25, 2005 at 04:15:14PM +0200, Laurent Fousse wrote: > * Joost van Baal [2005-06-24]: > > My suggestion is: in postinst, change > > > > SYSTRAQUSER=systraq > > > > in > > > > SYSTRAQUSER=Debian-systraq > > > > and in postrm, change > > > > rm -rf /var/lib/systraq > > > > in > > > > if getent passwd Debian-systraq > /dev/null 2>&1; then > > deluser --remove-home --quiet Debian-systraq > /dev/null > > fi > > > > .
Oops, I forgot about /etc/cron.d/systraq ...
> > It'd be cool if the Debian-foo convention could be codified in the
> > Debian policy, and if the adduser tool could be changed to refuse
> > (unless given a --i-am-package-and-therefore-i-know-what-i-am-doing
> > flag) to create users named Debian-foo.
>
> A convention about special usernames for users created by debian
> packages would be nice, yes. But we're far from a consensus on this
> issue, and the exim4 model is, err, somewhat disputed (see bugs
> #223831, #223963, #225031, #233803, #255493, and the discussion that
> followed on debian-devel). I'm more inclined to use debian-systraq
> then, with a lowercase as in debian-{mixmaster, mixminion, tor}.
Thanks for this informative reply. (Yeah, I had this feeling the issue
would be discussed more; I failed to find _all_ bugreports you mention.)
This final summary, I guess, is in #248809 against debian-policy. The
README.Debian-accountname document in the exim4 package is informative
too, it's quoted in #223831 .
Anyway, I guess using the name "debian-systraq" would be better than
what the systraq package currenlty does. Otoh, changing the username is
a hassle. Hrm...
postinst:
if there is a user "systraq"
warn "i will not reuse the user systraq"
warn "make sure you have a /etc/cron.d/systraq file from systraq >= version"
fi
if there is no user "debian-systraq"
create user "debian-systraq"
fi
postrm:
if there is a user "systraq"
warn "i will not remove user systraq"
fi
if there is a user "debian-systraq"
delete user "debian-systraq"
fi
Dealing with /etc/cron.d/systraq , which uses the systraq account name,
could also be handled by verifying whether the systrem is being upgraded
from a "systraq"-account-name systraq package.
Would that be feasable? Or too error-prone?
Bye,
Joost
signature.asc
Description: Digital signature
