On Tue, Oct 21, 2003 at 07:39:26AM +1000, Brian May wrote:
> On Mon, Oct 20, 2003 at 02:02:22PM -0500, Steve Langasek wrote:
> > - Installing the new version of heimdal in testing breaks the version of
> > postgresql currently in testing. The version of postgresql in
> > unstable no longer depends on heimdal (it uses MIT Kerberos instead),
> > so it doesn't have to go in at the same time as the above three
> > packages -- it can go in before them, when it's ready.
> errr.... Are you sure?
> If you are looking at bug like #216290 and #205592, then the same
> version will be broken in testing, too.
From update_excuses.html:
# heimdal (0.5.2-2 to 0.6-3)
* Maintainer: Brian May
* 21 days old (needed 10 days)
* heimdal (source) is (less) buggy! (1 <= 1)
* Valid candidate
* Depends: heimdal krb4
216290 is therefore not a blocking bug, because there's no net increase in
the bugginess of the package.
205592 is no longer an RC bug (in unstable), because the one case where
the lack of versioned symbols was materially affecting other packages
has been eliminated now that postgresql has been recompiled against MIT
Kerberos.
> > - The new version of coreutils (aside from currently being out-of-date
> > on m68k, which should be remedied soon) cannot be installed in testing
> > until krb4 is updated there, because it's an Essential: yes package
> > with a dependency on libacl1, which conflicts with
> > libacl1-kerberos4kth, with the net result that libacl1-kerberos4kth
> > would be uninstallable.
> I thought this problem had been fixed???
> Or maybe the version in unstable has been fixed, but not yet entered
> testing?
Correct, it's fixed in unstable -- and needs to also be fixed in testing
before the new coreutils package can be allowed into testing.
> > Please consider this an invitation to a mini-freeze for these packages,
> > and also an invitation to help watch over one another's packages over
> > the next few weeks to facilitate a smooth transition. Ideally, all of
> > these packages will be eligible for testing in a little over two weeks;
> > but one bad upload of a package with a 10-day waiting period could
> > easily throw that off.
> Bug #216290 still needs to be fixed.
> This means recompiling both Heimdal and krb4 with versioned symbols.
216290, AFAICT, does not involve versioned library symbols; it's a bug
regarding a versioned dependency on debhelper. It's questionable
whether that bug really merits an RC severity, since the lack of
versioning will never bother the buildd's on whose behalf the
Build-Depends field was added. (A missing Build-Dep would be a problem,
but this is not that.)
In any case, as noted above, there's no net increase in the RC bug
count, so it doesn't have to be fixed immediately. There is time for
another heimdal upload to fix this bug before perl will be ready, if no
other RC problems are introduced.
As for the actual versioned symbols question, I still think this is a
good idea to implement for future-proofing; and I have a fresh patch for
you to add version script support to the new version of libtool, but
when I noticed that my test case for this (and accompanying RC bug) had
disappeared, it fell a ways on my priority list. ;)
--
Steve Langasek
postmodern programmer
Attachment:
pgpEf2vMAbYDi.pgp
Description: PGP signature