Re: New packaging manual draft
On 30 Sep 2000, Manoj Srivastava wrote:
> Yet another version is up at
> http://master.debian.org/%7Esrivasta/new-packaging.txt
> Ummm, could you propose corrected wording, then? Are you
> saying that pre depends and conflicts can now prevent a package being
> on the system in an unconfigured state? Or are all bets off when a
> new package is unpacked?
Reading again I don't see what the problem was - what you have seems fine.
(Reading the dpkg bug lists.. though perhaps it is busted - are we going
to document bugs in dpkg in this manual? #2041)
> Jason> dependencies aren't satisfied. If it is asked to
> Jason> make an installation which would cause an installed
> Jason> package's dependencies to become unsatisfied it will
> Jason> complain [1], unless `--auto-deconfigure' is
> Jason> IIRC this is just completely false. Dpkg breaks reverse dependencies
> Jason> without any error or warning - that is how it must operate.
>
> Hmm. Reverse dependencies are never mentioned here, so I think
> completely false is a trifle strong.
Unless I am mis-reading it that is exactly what it is talking about.
Reverse dependencies being the dependency on the package in question from
other packages in the system. dpkg does not check those.
APT does this all the time to sequence complex situations, consider a
modern upgrade from potato -> woody, APT will sometimes do the following:
Inst libc6
Conf libc6
Inst locales
And you can see that the locales I have installed now:
Package: locales
Version: 2.1.3-10
Depends: libc6 (= 2.1.3-10)
Thus when libc6 is configured the locales has a broken 'reverse depends'
and dpkg does not complain. This is exactly the opposite of what the
quoted paragraph says - right?
> Are you sure that is policy? Seems to me that belongs in a
> devel guide, since it is the only correct way way to create the
> scripts.
Probably then.
Jason
Reply to: