Bug#690172: gcc-4.7-base: adding Breaks: gcc-4.4-base ( 4.4.7) ?

2013-03-02 Thread Julien Cristau
Control: severity -1 normal

On Sat, Feb 16, 2013 at 15:49:49 +0100, Andreas Beckmann wrote:

 Control: severity -1 serious
 
 On 2013-01-17 16:10, Andreas Beckmann wrote:
  Hi,
  
  how are the chances of getting this fix in sid and wheezy?
  
  I just verified that this really fixes the upgrade paths involving
  gnustep-back0.18 and friends.
 
 Raising the severity as this causes several incomplete upgrades from
 squeeze = wheezy. After having done a lot of upgrade tests with a
 fixed gcc-4.7-base I haven't seen any more issues due to ancient
 packages from src:gcc-X.Y. (But there are still enough other packages
 that need to be fixed in some way, too.)
 
Incomplete upgrades are by no means serious.

Cheers,
Julien


signature.asc
Description: Digital signature


Bug#690172: gcc-4.7-base: adding Breaks: gcc-4.4-base ( 4.4.7) ?

2013-02-16 Thread Andreas Beckmann
Control: severity -1 serious

On 2013-01-17 16:10, Andreas Beckmann wrote:
 Hi,
 
 how are the chances of getting this fix in sid and wheezy?
 
 I just verified that this really fixes the upgrade paths involving
 gnustep-back0.18 and friends.

Raising the severity as this causes several incomplete upgrades from
squeeze = wheezy. After having done a lot of upgrade tests with a
fixed gcc-4.7-base I haven't seen any more issues due to ancient
packages from src:gcc-X.Y. (But there are still enough other packages
that need to be fixed in some way, too.)

After having looked into apt's scoring a bit, most of these paths are
caused by #699759 apt: score computation may prefer obsolete installed
packages over their successors and a small fix to the score computation
solves this (by no longer doing something like
score[pkg1] += abs(some_negative_score_from_pkg2)
which makes packages with scores -1 and -2 more important than packages
with score 0).
But as upgrades need to work with squeeze's apt (and that cannot be
changed), we need to work around this in wheezy, usually using
high-score packages.

Andreas


-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org



Bug#690172: gcc-4.7-base: adding Breaks: gcc-4.4-base ( 4.4.7) ?

2013-01-17 Thread Andreas Beckmann
Hi,

how are the chances of getting this fix in sid and wheezy?

I just verified that this really fixes the upgrade paths involving
gnustep-back0.18 and friends.

Andreas


-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org



Bug#690172: gcc-4.7-base: adding Breaks: gcc-4.4-base ( 4.4.7) ?

2012-10-10 Thread Andreas Beckmann
Package: gcc-4.7-base
Version: 4.7.2-4
Severity: whishlist

Hi Matthias,

breaking the old versions of gcj-4.4-base seems to work fine in sid
(#677582). Thanks again for adding these.

Now I have discovered some more upgrade issues involving some old
libgnustep*, libobjc2 and gcc-4.4-base. apt chooses again to keep the
old (squeeze) version of gcc-4.4-base instead of removing libobjc2.

I think such incomplete upgrades of gcc-4.4-base could be fixed
(better prevented) by adding another Breaks to gcc-4.7-base.
But I'm not sure which versions to break:
* gcc-4.4-base ( 4.4.7)  # because 4.4.7 is in wheezy
* gcc-4.4-base ( 4.4.6-9~)   # because we break gcj-4.4-base ( 4.4.6-9~)
* gcc-4.4-base ( 4.4.5-13~)  # because libobjc2 was dropped in 4.4.5-13
and haven't made any tests with such a change so far.

Adding such a Breaks would have solved #677582 as well.

IMO distupgrades from squeeze should not involve keeping back
upgradable packages. Packages that are no longer available are allowed
to stay installed only if they don't block any other package from being
upgraded. And unfortunately we can't fix squeeze's apt to make better
choices.


Andreas

PS: I think my first approach of adding the
  Breaks: gcj-4.4-base ( something)
in gcc-4.4-base stopped working at the time gcc-4.4-base got the
Priority downgraded from required to optional, which shuffled the
scores given by apt.
And these score changes may be the reason for the new libgnustep* related
problems, too.

Andreas


-- 
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org