Re: Bug#137019: debconf: apt can't dist-upgrade properly from potato with debconf-tiny installed
On Wed, 6 Mar 2002, Joey Hess wrote:
> > As far as I can make out, the reason why debconf's C/P/R doesn't work is
> > that debconf-tiny in potato provided debconf, and apt appears to assign
> > a very high priority to packages that provide important packages (even
> > higher than the important package itself in the case where it's a mixed
> > virtual package, which seems odd).
>
> Screwy.
Just to clarify..
It was never a design goal to allow important packages to cleanly use any
alternation mechanism without the user having to explicitly choose to
switch to an alternative (via command line or dselect). We have never
really needed this before now.
The virtual debconf has become important by virtue of it's wide use, if
debconf-tiny is installed then it is considered to be the users prefered
choice and APT will try very hard not to switch it automatically. That is
why it gets a very high score. It does not view debconf as superceeding
debconf-tiny. C/P/R is the normal declaration for an alternative choice.
The only way to detect this sort of case is to use the information in the
Release files to realize that debconf-tiny is obsolete and then select
some random package that replaces it. This is why the Release files were
originally created, but the need to implement the feature never came up in
the releases that had Release files, so I never did it.
Anyhow, I think there is a fairly understandable fix which will solve
this, but I don't think it is worth the risk to try and put it in woody at
this late date. It is possible it could disrupt something else, it would
certainly invalidate all the upgrade testing that has already been done.
So, I think using a dummy debconf-tiny package is a fine way to ensure a
safe upgrade..
Jason
--
To UNSUBSCRIBE, email to debian-devel-request@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Reply to: