On Thu, Nov 27, 2014 at 14:22 -0800, Nikolaus Rath wrote:

Carl Meyer <carl-faUO1KlGllLR7s880joybQ@public.gmane.org> writes:

...

Hi Nikolaus,

On 11/26/2014 01:21 PM, Nikolaus Rath wrote:

...

I have a parametrized fixture foo, and some test cases that use it:

@pytest.fixture(params=(1,2,3,4,5)) def foo(request): # ... return foo_f

def test_1(foo): # ...

def test_2(foo): # ....

Now, I'd like to add an additional test, but running it only makes sense for some of the fixture parameters. So far I've solved this by providing the parameter value as a fixture attribute and skipping "unsupported" values:

@pytest.fixture(params=(1,2,3,4,5)) def foo(request): # ... foo_f.param = request.param return foo_f

def test_3(foo): if foo.param not in (1,4,5): raise pytest.skip("doesn't make sense")

This works, but I don't like it very much because it means that the test suite can never be executed without skipping some tests. I'd rather reserve skipping for cases where the test could in principle be executed, but for some reason cannot work at the moment (e.g. because there's no internet, or a required utility is not installed).

Is there a way to solve the above scenario without skipping tests?

Ideally, I'd be able to do something like this:

@only_for_param((1,4,5)) def test_3(foo): ...

I think in order to do this you'll need to use the `pytest_generate_tests` hook: http://pytest.org/latest/parametrize.html#pytest-generate-tests

Thanks for the link! The following seems to work nicely:

#!/usr/bin/env python3 import pytest

@pytest.fixture() def bar(request): s = 'bar-%d' % request.param print('preparing', s) return s

def test_one(bar): pass

def test_two(bar): pass test_two.test_with = (2,3)

def pytest_generate_tests(metafunc): params = (1,2,3,4) if not 'bar' in metafunc.fixturenames: return

test_with = getattr(metafunc.function, 'test_with', None) if test_with: params = test_with metafunc.parametrize("bar", params, indirect=True)

Unfortunately, it seems there are several issues with parametrizing fixtures these way that are probably going to bite me as things get more complex:

https://bitbucket.org/hpk42/pytest/issue/635 https://bitbucket.org/hpk42/pytest/issue/634 https://bitbucket.org/hpk42/pytest/issue/531 https://bitbucket.org/hpk42/pytest/issue/519

(Maybe mentioning them here gives them more attention :-).

Heh, you are touching the edges of what is currently possible and implemented with respect to parametrization, including what are probably real bugs. For my own uses of parametrization those edges usually did not play a role yet IIRC. If you need help in trying to fix things yourself let me know. If you don't want to tackle that it's fine as well, maybe somebody else gets to it. In any case, thanks for creating the detailed issues! best, holger