Bug#177952: apt pinning doesn't work for essential packages
On Thu, 1 Apr 2010, Teodor MICU wrote:
> reopen 177952
> found 177952 0.7.20.2+lenny1
> fixed 177952 0.7.25.3
> thanks
>
> The main point of these old bug reports is that there is no way of
> (willingly) preventing the installation of an essential package.
> In our present time we're confronting with the 'diff' -> 'diffutils'
> transition of an essential package. I have to keep "testing" in my
> sources.list file for a single 'noarch' package (PHP-based
> application) that I want to keep it updated. I have no reason to
> replace 'diff' with 'diffutils' because of this, so I'm looking for a
> solution of preventing an accidental install of 'diffutils' essential
> package.
Please note that the upgrade path from diff to diffutils does not take
in account at all the apt-get feature of "installing essential packages
automatically without asking and without warning at all", so in squeeze
there is a dummy diff package which depends on diffutils.
Does this
echo diff hold | dpkg --set-selections
improve things in your system?
Note: No, I'm not going to drop the dummy diff "because apt-get
already cares about new essential packages". A dummy package is
required to prevent a time window in which /usr/bin/diff would be
unavailable otherwise, as explained in Bug#563895.
BTW: Yet another reason why I think apt does the wrong thing: By
installing essential packages automatically, we are sending the wrong
message to developers, namely, that we don't "need" transition
packages to rename essential packages as apt-get already cares about
that. That's not true. We need them.
Would apt authors please consider documenting the current behaviour at least?
Reply to: