Delays in package processing
Kevin Kofler
kevin.kofler at chello.at
Fri Dec 21 11:41:54 UTC 2007
Michael Schwendt <mschwendt.tmp0701.nospam <at> arcor.de> writes:
> *I* believe we flood our users with too many rushed/untested updates. It
> feels more and more like a rolling release
A rolling release isn't necessarily a bad thing. The problem is not new
software, it's unreliable software. Getting new stable software (such as
upstream bugfix releases) in is a good thing. With Fedora, you get bugfixes and
sometimes even new features very quickly, with a distribution doing security
updates only (e.g. Debian stable) or almost (e.g. RHEL/CentOS), you often have
to wait for months if not years to get a fix for your bug. Regressions are
usually less of a problem than unfixed bugs: if there's a regression, I can
rollback to the last working version, if there's an unfixed bug, there's
nothing to upgrade or downgrade to.
> which is not too far away from Rawhide.
I separated out that part of the sentence because I disagree with the logic
here. Are you seeing a major X.Org X11 upgrade with some regressions in F7/F8
updates? KDE 4 RCs? A rewritten GDM? Yet all this stuff is in Rawhide.
Maintainers _know_ what kind of upgrades are not stable enough and/or change
too much to push as updates to stable releases. So, sure, the updates
are "rolling", but they're a lot more reliable than Rawhide.
The main reason I like Fedora is because the releases are stable, yet up to
date. I think we're doing a good job of separating the risky updates (->
Rawhide only) from the bugfix and/or riskless enhancement ones (-> updates).
> Certainly. Look at the size of the updates repository and also consider
> the number of packages, which have superseded eachother. Those users, who
> don't install a fresh Fedora release during the first two weeks, get to
> see several hundred updates the first time they run an update tool.
That's a feature.
> And after installing so many updates, they see regressions.
Our problem here is that updates-testing doesn't get enough actual testing, not
the updates per se. And I think the occasional regression, while annoying, is
not as bad as sitting on hundreds of bugs and leaving users with an unusable
system for months.
> Plus packages that more often than necessary depend on eachother, because
> once again a "minor version update" of some library broke ABI compatibility
> and requires subsequent rebuilds of other packages.
IMHO that's a non-issue.
Kevin Kofler
More information about the fedora-devel-list
mailing list