Robert Scheck wrote:
On Fri, 23 Jan 2009, Jeroen van Meeuwen wrote:
> I would definitely say that since they're in the business of making
> customers happy, not us, part of that responsibility is bumping release
> numbers, searching for higher nevra and communication with the EPEL
> maintainers, so that their customers remain happy (rather then confused
> between RHEL and EPEL packages).
thank you. You're somehow the first guy on this thread being knowledged in
basic things of packaging. I'm also a bit shocked, that such basics are not
known to other packagers (aren't some of the people that have shown up here
in the thread even provenpackagers? *shrug*).
Not saying you are right or wrong or whether I agree or disagree with
what you just said here, I really want to emphasize that EPEL seems to
be "on it's own"; Red Hat GSS does not seem to concern itself with
anything in or surrounding EPEL. While this is somewhat to be expected
(we are just another third party repo), I'm sure we'll hear about it if
and when we do override their packages. And, rather then playing the
guilt-question, I'd like to explore opportunities to get this resolved.
Red Hat GSS obviously takes advantage with EPEL, as they can often just
grab a package from EPEL X to be included in RHEL X.Y+1 whenever they
feel like it. I'm pretty sure they put enough QA effort in on such
package, but meanwhile they do not even bother to notify the current
EPEL maintainer(s). I think making this a standard checklist item in
their procedures should help. Even if just sending an (automated?) mail
to <package>-owner(a)fedoraproject.org, saying "this package is now in the
There's a usual scenario: Customer is using RHEL 5.0, was
duplicity package from EPEL, thus pexpect from EPEL popped onto the system;
then he updated to RHEL 5.1, everything fine. Then he updated to RHEL 5.2;
pexpect from EPEL still remains and the pexpect from RHEL is ignored, as
they've both the same (!) ENVRA. So the customer using EPEL before RHEL 5.2
in this case is simply disadvantaged/aggrieved, because he never will get
the pexpect package from RHEL without manual interaction.
On the other end is the issue of EL "customers" still running 4.X or 5.Y
and not upgrading. To maintain package foo in EPEL for 4.X or 5.Y (but
not 4.X+1 or 5.Y+1), we'd need 4.X/5.Y specific branches. However, the
amount of work involved, the infrastructure, the number of volunteers
(or their time), and what-have-you, seems to not add up / be in balance
in order for us to be able to do so.
Personally, I regret that, but I'm not complaining -complaining in my
opinion doesn't help. Since I'd love to see these things resolved, I can
only wait for us to start doing whatever it is we need to do in a manner
that we all think is great and brings peace to the world and then put in
some work to get it done.
This is, why I said, Red Hat broke it, Red Hat has to fix it. EPEL
solve this in a perfect way. And if the customer is still on a RHEL 5.x.y
(where x < 2) tree, he even can't install duplicity because of the missing
dependency to pexpect; so lowering release from -1 to -0 is not all.
I'm sorta curious how many EPEL users out there run into these kind of
situations, and whether it's a real problem for them (or whether it's
just annoying). Can we reach out to the EPEL users somehow?
Jeroen van Meeuwen