On 07/11/2013 09:36 AM, Dolph Mathews wrote:
On Thu, Jul 11, 2013 at 6:12 AM, Mark McLoughlin <[email protected]
<mailto:[email protected]>> wrote:
On Wed, 2013-07-10 at 17:07 -0500, Dolph Mathews wrote:
> On Wednesday, July 10, 2013, Sean Dague wrote:
>
> > Yesterday in the very exciting run around to figure out why the
gate was
> > broken, we realized something interesting. Because of the way
the gate
> > process pip requirements (one project at a time), on a current
gate run we
> > actually install and uninstall python-keystoneclient 4 times in
a normal
> > run, flipping back and forth from HEAD to 0.2.5.
> >
> >
http://paste.openstack.org/**show/39880/<http://paste.openstack.org/show/39880/>-
shows what's going on
> >
> > The net of this means that if any of the projects specify a
capped client,
> > it has the potential for preventing that client from being
tested in the
> > gate. This is very possibly part of the reason we ended up with
a broken
> > python-keystoneclient 0.3.0 released.
>
>
> > I think we need to get strict on projects and prevent them from
capping
> > their client requirements. That will also put burden on clients
that they
> > don't break backwards compatibility (which I think was a goal
regardless).
> > However there is probably going to be a bit of pain getting
from where we
> > are today, to this world.
>
>
> Thanks for investigating the underlying issue! I think the same
> policy should apply a bit further to any code we develop and consume
> ourselves as a community (oslo.config, etc). I have no doubt
that's the
> standard we strive for, but it's all too easy to throw a cap into a
> requirements file and forget about it.
I don't think we've ever capped oslo.config anywhere. Got a pointer to
what triggered that concern?
No no, I didn't mean to call you out... just using oslo.config as a
prime example of a non-client project that should (and from my
perspective: does) abide by the same policy.
We should/could be capping oslo.config like this:
oslo.config>=1.1.0,<2.0
because the API stability commitment is that we won't break the API
without bumping the release number to 2.0. I don't anticipate doing 2.0
soon/ever, so I've never pushed ahead with that capping.
I think that capping on the major version number is acceptable, as long
as we require major version bumps to break backwards compatibility...
and don't do major version bumps on a regular basis.
The problem is, the gate has not good way to navigate this ATM. So the
moment someone caps major versions of a client don't get tested.
So let's go with complete uncapping for now, and only deal with the
major version issue if it actually comes up.
-Sean
--
Sean Dague
http://dague.net
_______________________________________________
OpenStack-dev mailing list
[email protected]
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
