Package EVR problems in Fedora 2007-10-31

[Michael Schwendt]

On Fri, 02 Nov 2007 11:41:36 +0100, Ralf Corsepius wrote:

> > Why isn't the same update built also for F8?
> > Why do you want to limit the testing to one dist?
>
> E.g. to evaluate for distro specific issues? Remember, that though a
> particular spec might be identical for different distros, it's
> infrastructure underneath might be different.

You didn't answer the questions, especially not because an "identical
spec" does not break any upgrade path. Differences in the build deps
don't either. And changes under the hood can be done with an increase
of the least-significant portion of the EVR, that is to the right of
%{?dist}, for instance. You don't bump V without properly evaluating a
version upgrade, first.

So, let me rephrase the questions:

Why is an update prepared and built for an older dist, which breaks
[read: will break] the upgrade path, without starting this
experimental activity in the latest dist? Why does that dist-specific
test-update get a higher EVR than released packages for the newer
dist(s), such as 'devel'?

For the very rare case that the results of the tests with the old dist
decide on the roadmap of the pkg in the newer dist(s), it still
creates a scenario in which an upgrade path gets broken. Because even
if the test-update remains in updates-testing forever, and in 'devel'
the pkg doesn't move forward, you require your guinea pigs to clean up
their installations manually, if you decide to not push the update
into stable.

> > For some time we've excluded F7 updates-testing from the upgradepath
> > check,
> Which is the correct approach, IMO, because testing will receive
> experimental stuff, which is by no means guaranteed to end up in
> "updates".

Too often it ends up in "updates", and that's not limited to updates
we can observe in Fedora Extras.