Lennart Regebro a écrit :
On Sun, Apr 18, 2010 at 10:37, Christophe Combelles <ccomb@free.fr> wrote:
Depending on setuptools for tests in another evil thing. We should not assume a *setup* tool to be a testrunner, and we should not depend on the behaviour of setuptools to write tests.
I'm not sure what you mean there, setuptools doesn't behave in any particular way when running tests.
ok, not a lot, there is currently one single particular behaviour, which is adding tests from the result of the "additional_tests" function in a module. That probably won't hurt.
Setuptools is already doing too many things and Tarek is taking care (with distutils2 and distribute) to cleany separate the functionalities, such as installing, distributing, etc.
Yes, but that work is not done yet. And one thing that's needed to be able to run the tests under Python 3 is building it. So it's hard to separate the building and the running of testsm as a Python 3 capable testrunner would need to run setup.py build first anyway. Which means you depends on setuptools/distribute whichever way you turn.
Testing is not related to configuring nor installing, and "python setup.py test" is no more meaningful than "python setup.py makethecoffee".
It might seem so, but it is unfortunately not true when considering Python 3.