To someone with power to push packages on Fedora 21
kevin.kofler at chello.at
Fri Oct 23 02:51:43 UTC 2015
Sérgio Basto wrote:
> Kevin, Sounds like the experiences with an update with 50 packages like
> a KDE , but most cases are just a leaf package, which a "works for me"
> is a good feedback, for me .
Even for a leaf package, it does not work. The example of 10 "works for me"
and 1 "deletes all my data" applies even to one single application.
(Consider, for example, a new version of a circuit simulator that corrupts
the circuit description if and only if a certain component is used in it.
And no, I don't have a concrete update in mind there, it is just an example.
This kind of issues is not all that rare in leaf applications.) Of course,
kernel updates, library updates and grouped updates are even more likely to
run into this kind of issues (and there have been MANY broken kernel and
library updates pushed due to autokarma! And then the silly update stability
policies delayed the FIX for the regression), but applications are by no
ways immune to them.
Autokarma also used to not understand upgrade paths at all and happily break
them. Now in Bodhi 2, there is at least an option that SHOULD fix this, no
idea whether it actually works. (To be honest, I'd rather not trust it,
considering how buggy Bodhi's automagic has been so far. But at least it's
> Conclusion bodhi auto-karma are in the correct proportions , and
> should /can be adjust for popular packages .
I disagree. IMHO, enabling autokarma is ALWAYS a mistake. (Well, actually,
there is one valid case to use it, and that is when you actually think the
update can go stable NOW, but Bodhi won't let you due to the silly update
stability policies. But it's the ONLY case, and it wouldn't exist either if
it weren't for silly bureaucracy. And even there, beware of the upgrade
> What I'm trying point out, after a general availability of a release,
> I'm a beta tester, I disable update-testings repo and after some weeks I
> do a: dnf list extras, I see many packages that still in update-testing
> and yes I can make dnf distro-sync, and I can downgrade a bunch of
> packages but isn't it more logical push that packages to stable ?, that
> way all users (beta testers and others) as the same experience
Complain to the maintainers abusing updates-testing then. It is not a
dumping ground for broken packages. If the package works, push it to stable.
If it doesn't work, fix it. If you can't fix it in a timely manner, unpush
it. But most importantly DON'T rely on a computer doing the decision for
you! (Maintainers relying on autokarma are actually the main reason packages
linger in testing forever to begin with. A responsible maintainer sees there
is no negative feedback for 1 or 2 weeks and just pushes the darn thing.)
More information about the devel