[Numpy-discussion] detecting shared data

[Stefan van der Walt]

On Thu, Apr 12, 2007 at 11:29:31AM -0400, Anne Archibald wrote:
> > Failures may be expressed using
> >
> > NumpyTestCase.failIf(self, expr, msg=None)
> 
> That's not quite what I mean. There are situations, with the current
> code, that it gets the answer wrong (i.e., claims arrays may share
> memory when they don't). I know, and it's okay, and if it doesn't
> there's a bug, but in view of possible future enhancements, I don't
> want to signal an actual failure if it starts working. I do want to
> test it though, so I was hoping there was a way to express "I expect
> this test to fail, notify me if it doesn't, but don't call it a
> failure if it starts working".

If the test is supposed to pass but currently fails, we can always add
it using level=50 or so.  That way, most people who run the tests will
not see the failure, but the devs may still choose to run them.  We
could even display a warning when running the test suite with such a
high level.

Would such a scheme cause problems for anyone?  An alternative would
be to rework the test gatherer to filter tests based on some flag.

Cheers
Stéfan