Re: Mistake in postrm preventing functioning of newer package (stable/testing/unstable)
On Fri, Sep 09, 2011 at 12:54:32PM +0200, David Paleino wrote:
> On Fri, 9 Sep 2011 12:42:11 +0200, sean finney wrote:
> > On Fri, Sep 09, 2011 at 11:37:41AM +0200, David Paleino wrote:
> > > I just wanted to (try to) fix it for the biggest number of people possible.
> > > Maybe I'll just leave stable as-is, and try the solution I described within
> > > testing/unstable :/
> > Another option, as awful as it sounds, is to nuke (or surgically modify)
> > the postrm script of the old package from the preinst of the new package.
> > You would need to support that until the current stable was rotated out
> > through oldstable. Yes, awful. But I've seen it done before...
> I didn't think to that, and it seems the only solution to ensure clean
> upgrade paths.
> I'll use the 'old-version' parameter passed to preinst on upgrade, and check
> if it's less than the current version in sid.
That is the good start.
> Will someone kick me hard if I go this way? :)
I am no expert but I have seen and imitated some last resort solutions.
When maintaioner decides to nuke any modifiable files (conffile), those
package script which uses md5sum or similar to ensure that the nuked
files are the old package defual ones. (Otherwise, they make some back
up of them.)
You should lern from existing ptractice. Please do:
$ grep -e "md5sum" /var/lib/dpkg/info/*.p*