Imports in one file affect another file

[l0b0]

• a detailed description of the bug or problem you are having
• output of pip list from the virtual environment you are using
• pytest and operating system versions
• minimal example if possible

$ uname --kernel-name --kernel-release --kernel-version --machine --processor --hardware-platform --operating-system
Linux 5.11.7-arch1-1 #1 SMP PREEMPT Wed, 17 Mar 2021 16:59:58 +0000 x86_64 unknown unknown GNU/Linux
$ pip list
Package    Version
---------- -------
attrs      20.3.0
iniconfig  1.1.1
packaging  20.9
pip        21.0.1
pluggy     0.13.1
py         1.10.0
pyparsing  2.4.7
pytest     6.2.2
setuptools 49.2.1
toml       0.10.2

Production code in prod.py:

from random import randint


class C:
    value = randint(0, 999_999_999_999)


if __name__ == "__main__":
    print(C().value)

Verify production code randomness:

$ python prod.py 
237983317672
$ python prod.py 
218559153549

Setup:

cd "$(mktemp --directory)"
python -m venv .venv
. .venv/bin/activate
pip install --upgrade pip
pip install pytest

test_unmocked.py:

from prod import C


def test_set_at_import_time():
    assert C().value == C().value

test_mocked.py:

from unittest.mock import patch


with patch("random.randint", return_value=4):
    from prod import C


def test_uses_randint():
    assert C().value == 4

Actual breakage:

$ pytest test_mocked.py test_unmocked.py 
=========================================== test session starts ============================================
platform linux -- Python 3.9.2, pytest-6.2.2, py-1.10.0, pluggy-0.13.1
rootdir: /tmp/tmp.ucg7uFfMbW
collected 2 items                                                                                          

test_mocked.py .                                                                                     [ 50%]
test_unmocked.py .                                                                                   [100%]

============================================ 2 passed in 0.02s =============================================
$ pytest test_unmocked.py test_mocked.py 
=========================================== test session starts ============================================
platform linux -- Python 3.9.2, pytest-6.2.2, py-1.10.0, pluggy-0.13.1
rootdir: /tmp/tmp.ucg7uFfMbW
collected 2 items                                                                                          

test_unmocked.py .                                                                                   [ 50%]
test_mocked.py F                                                                                     [100%]

================================================= FAILURES =================================================
____________________________________________ test_uses_randint _____________________________________________

    def test_uses_randint():
>       assert C().value == 4
E       assert 187030247152 == 4
E        +  where 187030247152 = <prod.C object at 0x7f033375ac40>.value
E        +    where <prod.C object at 0x7f033375ac40> = C()

test_mocked.py:9: AssertionError
========================================= short test summary info ==========================================
FAILED test_mocked.py::test_uses_randint - assert 187030247152 == 4
======================================= 1 failed, 1 passed in 0.04s ========================================

Basically, by running test_unmocked.py first test_mocked.py gets an unmocked randint. That is, the mocking in test_mocked.py has no effect.

Detected because it caused a test without any mocks to fail with an error message mentioning mocks after working around a pynamodb issue where I can't change a field to be a @property.

[The-Compiler]

This isn't an issue in pytest:

$ python3 -c "import test_mocked, prod; print(prod.C().value)"  
4

I'm converting this to a discussion.

[mrbean-bremen]

This would happen the same way with unittest.
The problem is that C.value is read the first time the prod module is loaded. Reimporting C does not help, because it is already cached.
If you want to use a mocked version you have to remove it from the module cache before importing it, e.g. something like:

import sys
from unittest.mock import patch

with patch("random.randint", return_value=4):
    if 'C' in sys.modules:
        del sys.modules['C']
    from mocked_import.prod import C

    def test_uses_randint():
        assert C().value == 4

You may have to do the same for the other test.

[l0b0]

Thank you! We're definitely in "use some other approach" territory in that case.

[RonnyPfannschmidt]

a simple way to handle it is to patch C.value to something specified

a correct way would be not ever use random values in defaults stored at the module level, and always have specific utilities or named constructors to create configured instances with special needs (having magic random numbers at constants is always a hazard waiting for a heisenbug)