Dependencies between xunit-style setups and the other fixtures

[stanislavlevin]

#4091 introduced a nice feature employed xunit-style setups and unittest definitions as internal fixtures.

But unfortunately, it looks like there is no way to make a strict order of fixture setup as It made in case of regular ones. Yes, sure, setup*/teardown* have
their own fixture names, but they are internal, for example:

_Class__pytest_setup_class
_UnitTestCase__pytest_class_setup

I couldn't rely on them since they are not documented and could be different from release to release.

The second problem that even if I drop xunit/unittest setups I can't forbid their usage by Pytest means.

[RonnyPfannschmidt]

you could make a pytest plugin that raises errors when xunit style setup is used

[stanislavlevin]

I thought about this.
At first look, I don't see how I can catch this. But since you talk about such a possibility, now I believe there is a good chance to do it.
@RonnyPfannschmidt , could you give me a little hint about a reliable way to catch xunit/unittest?

[stanislavlevin]

Example of implementation of a such plugin (based on attributes of item.cls):
forbidden.py.txt

Produces output:

module: test_f.py
        setup_module
        setup_function
        class: TestClass
                setup_class
                setup_method

!!!!!!!!!!! _pytest.outcomes.Exit: xunit style setups and unittest are not allowed !!!!!!!!!!!!
==================================== no tests ran in 0.11s ===================================

[RonnyPfannschmidt]

@stanislavlevin that looks like a good start, if you use the results of the makeitem hooks and use item.warn, you can have to the point warnings for the classes/modules that use xunit style setups

to be on the safe side, please take into account that setup_foo fixtures are allowed anf replace xunit with fixture style

[stanislavlevin]

@RonnyPfannschmidt, thank you very much for the clues.
Unfortunately, pytest_pycollect_makeitem doesn't work as I expect or my expectations are wrong.

For example,
test class:

import unittest

class TestClass:
    def test_1(self):
        pass

class TestCase(unittest.TestCase):
    def test_method1(self):
        pass

@pytest.hookimpl(hookwrapper=True)
def pytest_pycollect_makeitem(collector, name, obj):
    outcome = yield
    res = outcome.get_result()
    if res is not None:
        if isinstance(res, list):
            for n in res:
                print(n.name)
        else:
            print(type(res))

Run only one test class test_unittest.py::TestClass.
Outcome:

collecting ... <class '_pytest.python.Class'>
<class '_pytest.unittest.UnitTestCase'>
test_1
collected 1 item

I thought that the result produced by this hook will be filtered by collection (I mean only test_1 in the outcome).

Updated version of plugin:

import sys
import pytest

forb_mod_scope = [
    'setup_module',
    'setup_function',
    'teardown_module',
    'teardown_function',
]

forb_cls_scope = [
    'setup_class',
    'setup_method',
    'teardown_class',
    'teardown_method',
]

def pytest_collection_finish(session):
    unit_test = sys.modules.get('unittest', None)
    unit_test_cls = getattr(unit_test, "TestCase", None)

    for item in session.items:
        cls = getattr(item, 'cls', None)
        subclassed = False
        if unit_test_cls is not None and cls is not None:
             subclassed = issubclass(cls, unit_test_cls)
        if subclassed:
            item.warn(pytest.PytestWarning("unittest is deprecated"))
            continue

        def xunit_depr_warn(item, attr, names):
            for n in names:
                obj = getattr(item, attr, None)
                method = getattr(obj, n, None)
                fixtured = hasattr(method, '__pytest_wrapped__')
                if method is not None and not fixtured:
                    item.warn(pytest.PytestWarning("xunit style is deprecated\n"))
 
        xunit_depr_warn(item, 'module', forb_mod_scope)
        xunit_depr_warn(item, 'cls', forb_cls_scope)