Re: errors during recent dist-upgrades
On Wed, 10 Nov 1999, Raphael Hertzog wrote:
> Le Wed, Nov 10, 1999 at 10:23:20AM -0700, Jason Gunthorpe écrivait:
> > First off, if not installing those packages causes other upgraded packages
> > to not work then it is a bug in those other packages, plain and simple.
>
> Not it doesn't cause any problem but if perl-5.005 is available, it's in
> order to be used ... :-)
?? I don't understand what you are saying here.. Kieth very clearly had a
case where something perly didn't work without perl-5.005 that obviosly
needed a better dependency. That is a bug.
> > As for the rest of it, I don't know - this mess of depends that perl has
> > become is really really complicated I'd have to sit down and figure out if
> > any of the problems are really bugs in APT or not.
>
> The problem is that perl-5.005 conflicts with perl which was
> important in slink.
Yes that is part of the problem, the other part(s) are all derived from
that and some (all? I don't know) of them can be fixed in APT.
You know it is really really difficult to best guess the best fit set of
packages to satisify any given goal - in fact there are something like
2^3000 unique combinations of package states [ok, in practice there are
many less, but still too many]! That is alot of choice! It
is really really hard to par that down to something that is workable.
Obviosly APT does not brute force a solution, and in fact the algorithm it
uses is iterative and becomes less likely to converge on the 'best'
solution the more complicated the interrelations are.
The new Perl cases are really stretching things, you have a critical
package being renamed [perl -> somethingerother] you have new packages
being installed, old ones being removed and alot of this is removed 2-3
packages from the real point of contention!
IMHO the way potato is right now, two things are true:
1) People not using APT are going to have one nasty upgrade
2) Old APTs just won't cut it
We have never before had this level of complication in the dependency
tree(s)!
Note: This only applies to APTs automatic problem resolver which is used
to compute the new packages to install and the old packages to
remove during dist-upgrade and install options, dselect isn't nearly
as badly effected by this aside from generating lots of dependency
resolution screens.
Jason
Reply to: