Saturday, September 10, 2005

» More thoughts on a "Debian-like" package classification for openSUSE

A few people (including me) have started to discuss on the openSUSE mailing-list what we could do to embrace packages that are made by contributors (like Packman, James Ogley, a lot of smaller "suser-*" repositories on, myself and others) into the openSUSE initiative. There's no consensus yet, we're still far from having discussed enough about it, mostly because we need to have quite a few people from SUSE involved into the thread and they're really busy doing a great job releasing SUSE Linux 10.0. One of the options I have proposed is to have a model where [YaST2/apt/yum] repositories are categorized into 3 levels of "quality" or "stability", namely stable, unstable and testing. Does that remind you of something ? Indeed, that's pretty much known as "the Debian release system" and there has been quite some critics against it. James Ogley has been making the same proposition (more or less) and added "I don't think we should allow ourselves to have a Debian-like interval between releases just because we may adopt a similar model of distro development, the current frequency is about right" I couldn't agree more. What James and I are talking about is not adopting the Debian release concept, we're only talking about that one idea. Note that (AFAICR) it was originally used by BSD. Now, what's wrong with Debian and why are a few people being allergic to that idea ? I think it's mostly related to the fact that although Debian is a brilliant distribution within its scope and goals, it has been having quite a few issues with their release cycles during the last 2-3 years, with major distribution releases for the "stable" branch taking far too much time to be decided and roadmapped, leaving Debian users for very long with a completely outdated "stable" release. Note that this has at least had the advantage of pushing a new distribution out of the ground, namely Ubuntu (and it's derivates, like Kubuntu), that are specifically adressing two of Debian's major "shortcomings" (they only are in a certain context, which is end-users in this case): faster release cycles (as you can read on the Ubuntu website, they make major releases every 6 months) and stress user-friendlyness for the installation. That being said, what James and I have been proposing has nothing to do with that, and it's definately not a consequence of adopting a stable/unstable/testing categorization of contributed packages. First of all, that model doesn't apply in the same way as with Debian, not by any means: the SUSE Linux distribution as produced by the SUSE staff is to be considered a "core distribution" that's absolutely not subject to that categorization. From the openSUSE community's point-of-view, it gives us a strong, stable foundation with 2 or 3 releases every year we can build our specialized distributions or contributed package repositories upon. That categorization idea actually applies to two different criterias: » quality of the packages: making good RPMs is not an easy task and requires a lot of experience, both with RPM itself as with the distribution one is making the packages for (SUSE Linux, in this case) » stability of the package itself: I've found myself quite often into a position where I chose not to package a piece of software because it was still in alpha/beta stage or even worse, it was a CVS snapshot, and I don't want to push a package as an update to hundreds of users who, full of confidence in the level of quality I think I've been able to achieve up to now, pull that newer package and end up with a non-working application - and I suppose you can see how a stable/unstable/testing categorization would apply here All of this just being a summarization of what has already been said on the openSUSE mailing-list, I got to think about it again. While I still believe that this categorization system (not necessarely exactly the above, but similar) could prove to be effective, it certainly is not good enough for end-users. They don't necessarely care about stable/unstable, what levels of quality and trust can be expected from a packager, and only partly take interest in whether that piece of software is to be considered to be stable and tested or just another CVS snapshot that might break. I don't have an answer to that, yet, but we should probably "think beyond the Debian model", as Sonja wrote.


Post a Comment

<< Home