Duplicate fixtures in complex projects

[hunse]

Related to #979, I think, or at least a similar error.

I have the following directory structure:

project/amodule/tests/conftest.py
project/amodule/tests/test_amodule.py
project/bmodule/tests/conftest.py
project/bmodule/tests/test_bmodule.py

where both the conftest files contain

def pytest_generate_tests(metafunc):
    if 'myfixture' in metafunc.fixturenames:
        metafunc.parametrize("myfixture", ['a', 'b', 'c'])

and both the test_*.py files contain

def test_generate_tests_fixtures(myfixture):
    print(myfixture)

When I run all the tests with py.test project from the root directory (the parent of project), I get the following error

================================================ test session starts ================================================
platform linux2 -- Python 2.7.6, pytest-2.8.1.dev1, py-1.4.30, pluggy-0.3.1
rootdir: /data/eric/src/pytest, inifile: tox.ini
collected 3 items / 1 errors 

project/bmodule/tests/test_bmodule.py ...

====================================================== ERRORS =======================================================
______________________________ ERROR collecting project/amodule/tests/test_amodule.py _______________________________
project/amodule/tests/conftest.py:3: in pytest_generate_tests
    metafunc.parametrize("myfixture", ['a', 'b', 'c'])
_pytest/python.py:1002: in parametrize
    param_index)
_pytest/python.py:842: in setmulti
    self._checkargnotcontained(arg)
_pytest/python.py:825: in _checkargnotcontained
    raise ValueError("duplicate %r" %(arg,))
E   ValueError: duplicate 'myfixture'
========================================= 3 passed, 1 error in 0.08 seconds =========================================

However, if I delete both the tests subdirectories so everything is up one level, then it works fine.

[hunse]

By the way, this is a regression (or at least a change in behaviour), which according to git bisect was caused by 32165d8.

[nicoddemus]

Thanks @hunse! 😄

[Zac-HD]

Closing in favor of #3966, which is more recent and more heavily discussed.