Hi,
Sylvestre Ledru wrote:
> +++ lapack-3.4.0/debian/changelog 2012-02-19 04:59:35.000000000 -0600
> @@ -1,3 +1,10 @@
> +lapack (3.4.0-1~exp2) experimental; urgency=low
> +
> + * Introduce back liblapack3gf as transitionnal package
> + (Closes: #656039)
> +
> + -- Sylvestre Ledru <sylves...@debian.org> Sun, 19 Feb 2012 11:34:47 +0100
Thanks! At first I was worried by the above description --- why would
adding a dummy package help to fix this bug? --- but the actual patch
looks closer to something that might work.
[...]
> --- lapack-3.4.0/debian/control 2011-12-25 17:55:15.000000000 -0600
> +++ lapack-3.4.0/debian/control 2012-02-19 05:38:18.000000000 -0600
> @@ -13,10 +13,21 @@
> Package: liblapack3
> Architecture: any
> Depends: ${misc:Depends}, ${shlibs:Depends}
> -Provides: liblapack.so.3gf, liblapack.so.3
> +Provides: liblapack.so.3gf, liblapack.so.3, liblapack3gf
> -Conflicts: lapack99, liblapack3gf
> +Conflicts: lapack99, liblapack3gf (<< 3.4.0-1~exp1)
> -Replaces: lapack99, liblapack3gf
> +Replaces: lapack99, liblapack3gf (<< 3.4.0-1~exp1)
This Conflicts could and probably should be a Breaks. liblapack3gf's
prerm (and the upgraded version's preinst, if any) would be run before
liblapack3 gets unpacked, making it safe.
Why isn't the
update-alternatives --remove liblapack.so.3gf
/usr/lib/liblapack/liblapack.so.3
in the updated liblapack3gf's preinst instead of liblapack3's
postinst, by the way?
The Conflicts also seems insufficient, since it doesn't force other
packages such as libatlas3gf-base for which liblapack.so.3gf is still
the master link to be upgraded. Maybe a versioned Breaks against all
packages in squeeze providing liblapack.so.3gf would help, so the
upgrade path would be:
1. Unpack new versions of libraries providing liblapack.so.3gf
en masse. Preinst removes the old liblapack.so.3gf alternative.
2. Configure new versions of libraries providing
liblapack.so.3gf. Postinst adds the
liblapack.so.3/liblapack.so.3gf link group.
> -Description: library of linear algebra routines 3 - shared version
> +Description: Library of linear algebra routines 3 - shared version
I had thought the opposite convention was more common.
Hope that helps,
Jonathan
--
To UNSUBSCRIBE, email to debian-bugs-rc-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
