python-kerberos and test suite

[Martin Pool]

2009/3/23 Bob Tanner <tanner at real-time.com>:
> I see my .bzr.log contains:
>
>  failed to import kerberos lib: No module named kerberos
>
> I've been doing the release exclusively on Ubuntu hardy.
>
> I see there's a python-kerberos in hardy, but it's for python-2.4.
>
> I've been testing with python-2.5, so I back ported jaunty's python-kerberos
> to my hardy pbuilder, but this got me thinking, did I just compromise my
> test environment?
>
> I reverted to my non-python-kerberos pbuilder test area, until I get 1.13.1
> out the door.
>
> I might be overly caution about the back-port thing, so posting here just to
> make sure a back-port of a package isn't an issue for bzr, test validation,
> and bzr releases?

Hi,

It's not necessarily a problem.  I don't think anything from
python-kerberos is copied into bzr during the dist process, so you
won't have packaged the release.

At worst, your testing may give unrepresentative data: hypothetically
bzr works on hardy with your backported kerberos but doesn't work with
python2.4-kerberos, or doesn't work with jaunty's python2.6-kerberos.
However, there are so many possibly combinations that we don't ask the
release manager to test them all: we've seen bugs specific to python
version, all different external libraries, filesystem, OS and release,
processor architecture, etc.  Even if you were willing to, I don't
think it's rational to serialize the release on trying to cover them;
the point of having RCs and frequent releases is so that we can get
and respond to broader feedback.

You should try to have things be relatively stable so if you do have a
problem, you have the best chance to reproduce it.

Also, it's good to test in representative environments, all other
things being equal.  If most people on hardy don't have
python2.5-kerberos, they'll care more about how bzr copes when it's
not there.  In this case there's an obvious tradeoff against coverage.

So overall, don't worry.

-- 
Martin <http://launchpad.net/~mbp/>