Open Stack

On 20 February 2018 at 04:39, Andrey Kurilin <andr.kurilin at gmail.com> wrote:
> Can someone explain me the reason for including "tests" module into
> packages?

Namespacing the tests makes the test ids unique which is very helpful
for aggregating test data as we do. Including that in the tar.gz that
is uploaded to PyPI is pretty standard - a) its how you can verify
that what you downloaded works in your context (and no, going to git
is not a good answer there because that means you now need the entire
ecosystem of tools to build man pages etc etc etc). and b) its
providing the full source of the thing we're releasing. Thats you
know, how F/LOSS works.

It should be possible, if we care to, to exclude the tests from wheels
made from those distributions, which would make the footprint for
binary usage smaller - I have no strong opinion on whether thats wise
or not, but I will say I can't see a use case for needing the tests in
that scenario.

Similarly whether those tests should be included in Linux distribution
packages or not is a debate I don't care to enter - but again I don't
see a use case: binary distributions are presumed to be integration
tested by the distributor, not the consumers.

I don't think importing code from another packages 'tests' module is
wrong or right - python is very much a consenting-adults language -
but I do think that in OpenStack with the sheer number of people
involved we should set very clear guidance; and I'd suggest that
saying its not supported is a good default: if folk want to offer a
contract where something can be imported they can always put it in a
different package.

In summary:
 - moving 'tests' to the root is a poor idea, please don't do it - we
had this debate back in 2011 or so and nothing has changed that I can
see.
 - we can, and perhaps should, exclude $package.tests from wheels to
save bandwidth (but *not* from .tar.gz).
 - linux distributions should IMO follow what we put in wheels.

-Rob