Hi, After some time, I realized I might have been too optimistic. Good practice by the package maintainer with proper version dependency in his package may not keep us free from trouble if you have mixed system with apt_preference .
On Sat, Feb 20, 2010 at 03:21:15PM +0100, David Kalnischkies wrote: > Hi Osamu Aoki, > Pinning itself can't effect system stability - It causes > uninstallable packages or strange choices while upgrading, but > your system will not be broken in the "broken-package-state"-sense. ... > The other point is that i consider a not strong enough version dependency > already as an important bug as you will always have at some point a mixed > system - maybe only in between the apt-get dist-upgrade (or similar) run, > but even in this stage all configured packages should work… > > It is out of scope for the apt_preferences manpage to indicate that a package > from unstable maybe has more bugs than one from stable -- but it is absolutely > possible that wrong pinning of stable in a testing environment will cause > similar problems to the reversed environment setup… I realize that version dependency limitations are recorded by a newer package requesting other packages to be newer than some version. For simplicity, let's think Python and related packages as an example. Let's suppose you have extensively mixed system. Let's say you use testing mostly. Now you install one package depending on a new unstable version of Python. Its installation pulls unstable version of Python. Although most programs work fine, this new Python in unstable may contain backward incompatible changes. So If you have such an unfortunate program from testing, it will be broken by this upgrade. (I do not expect Python package to set conflict with all these old version of programs everytime we have such case. Usually, we upload new packages to fix these issues.) Of course, you nor I may not experience this kind of situation by having good instinct, but this problem deserves fair attention. Regards, Osamu -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
