Testing that warnings are redirected with logging.captureWarnings

[emlys]

I am trying to test a context manager that sets up logging for my application and redirects warnings to logging with logging.captureWarnings(True). I want to assert that a warning is not raised, but is logged instead. Here is a minimal example:

import logging
import warnings

def test_log_warning():
    logging.captureWarnings(True)
    warnings.warn('foo')

with default filterwarnings configuration, it works as expected. the warning is redirected to logging and not raised:

$ pytest -o log_cli=true -s tests/test_logging.py 
==================================================================================== test session starts ====================================================================================
platform darwin -- Python 3.10.12, pytest-7.4.0, pluggy-1.2.0
rootdir: /Users/emily/invest
plugins: subtests-0.11.0
collected 1 item                                                                                                                                                                            

tests/test_logging.py::test_log_warning 
--------------------------------------------------------------------------------------- live log call ---------------------------------------------------------------------------------------
WARNING  py.warnings:warnings.py:109 /Users/emily/invest/tests/test_logging.py:6: UserWarning: foo
  warnings.warn('foo')

PASSED

===================================================================================== 1 passed in 0.01s =====================================================================================

with filterwarnings=error, the warning is raised, not logged, and the test fails. This is counterintuitive - it seems that pytest's configuration is affecting the behavior of the code under test:

$ pytest -o log_cli=true -o filterwarnings=error -s tests/test_logging.py 
==================================================================================== test session starts ====================================================================================
platform darwin -- Python 3.10.12, pytest-7.4.0, pluggy-1.2.0
rootdir: /Users/emily/invest
plugins: subtests-0.11.0
collected 1 item                                                                                                                                                                            

tests/test_logging.py::test_log_warning FAILED

========================================================================================= FAILURES ==========================================================================================
_____________________________________________________________________________________ test_log_warning ______________________________________________________________________________________

    def test_log_warning():
        logging.captureWarnings(True)
>       warnings.warn('foo')
E       UserWarning: foo

tests/test_logging.py:6: UserWarning
================================================================================== short test summary info ==================================================================================
FAILED tests/test_logging.py::test_log_warning - UserWarning: foo
===================================================================================== 1 failed in 0.06s =====================================================================================

It seems that pytest's filterwarnings configuration is implemented using the warnings filter, which overrides logging.captureWarnings and affects the code being tested. Is it possible to test the functionality of logging.captureWarnings with pytest?

[RonnyPfannschmidt]

as showwarning is replaced by logging, the code works "as intended" but not "as expected/desired
warning filtering happens before the showing

so in order to test/examine warnings as always shown, its necessary to use one of the warning module utilities to add a temporary additional filter to let things be

[emlys]

I would like to test both of these things:

• a warning is written to logging
• a warning is not issued as usual

By temporarily modifying the filter:

with warnings.catch_warnings():
    warnings.simplefilter('default')
    logging.captureWarnings(True)
    warnings.warn('foo')
    # assert that the warning was written to the log
    # assert that the warning was not emitted as usual

I can test that warnings are written to logging, but I can no longer verify that the warning is not issued. It seems that it's impossible to assert both of these conditions in pytest.

[RonnyPfannschmidt]

Both are possibly,but there's little benefits to testing it as it's a stdlib functionality

The Test would involve an awful amount of Messing around with logging and warnings internals