[SciPy-dev] Making numpy test work in scikits

[David Cournapeau]

Matthieu Brucher wrote:
>
>     It is intended if you read the docstring (I totally missed it too, at
>     first). If level is negative, then it is equivalent to returning
>     all the
>     test corresponding to abs(level) in a test suite, without running
>     them.
>
>
> Yes, I read it, but it does not find tests at all for instance :
>
>   Found 0 tests for Components.IO.ipb.tests.test_ipbreader
>   Found 0 tests for Components.IO.nifti.tests.test_niftiimage
>
> instead of :
>
>   Found 8 tests for Components.IO.ipb.tests.test_ipbreader
>   Found 3 tests for Components.IO.nifti.tests.test_niftiimage
>
>  for a positive level.
I also have similar problems: for example, using testall does not work 
at all for me in some cases. The problem is setuptools is already kind 
of magic (many things going on in your back), and numpy testing 
facilities feel a bit too voodoo for me too (uses a lot of 
introspection, dynamic loading of test scripts and source parsing to 
build test suites).

I kind of gave up on this one: I think I will convert all my tests to 
standard unittest, and build my own, simpler test runner: the only 
feature I need from NumpyTest class is the ability to find tests in test 
directories. I don't think it worth trying to make NumpyTest work as I 
want just to avoid building TestSuite by hand.

David