Re: Problems found by piuparts

[Lars Wirzenius <liw@liw.iki.fi>]

I added a Cc to Manoj since I would like to hear his comment. Whoever
responds may want to remove the Cc to avoid stuffing his inbox
unnecessarily.

su, 2006-02-19 kello 23:42 -0800, Steve Langasek kirjoitti:
> On Mon, Feb 20, 2006 at 08:24:53AM +0200, Lars Wirzenius wrote:
> >         * Use of ucf in postrm during a purge, without checking that ucf
> >         is installed. Since ucf is not an essential package, postrm
> >         during purge cannot rely on it. As it happens, I think it might
> >         be good to have ucf (or a subset of it) as an essential package,
> >         since this error happens a lot.
> 
> So assuming that we're stuck with the current implementation for a bit where
> ucf is not part of essential, I do wonder if checking whether ucf is
> installed is actually the correct thing to do in postrm purge.  The state
> prior to purge is defined as "config files"; the difference between config
> files state and "purged" is whether there are still config files left on the
> system.  If the package can't actually succeed in removing its config files
> because ucf is not installed, isn't it *correct* for the postrm purge
> command to fail?  I.e., I think it's more of a bug to allow dpkg to put the
> package into "purged" state leaving orphaned config files behind than it is
> for postrm purge to fail and leave the package in "config files" state.

Hm. I don't use ucf on my own packages (yet), so my understanding is a
bit hazy, but if I have understood correctly, the actual config file is
removed with rm anyway, and ucf is needed on purge only to remove the
config file also from ucf's history data. Thus, only running ucf if it's
there should be the right thing.

Manoj, could you comment on this?

-- 
Yet another password: just say no.