test warnings in docstrings without monkey-patching warnings.showwarning

[dimpase]

In order to test warnings in docstrings of functions and classes, it appears necessary to monkey-patch warnings.showwarning,
e.g. as follows

# c.py
import warnings
dowarn = warnings.showwarning
def showwarn(message, category, filename, lineno, file=None, line=None):
    import sys
    dowarn(message, category, filename, lineno, sys.stdout, line)

warnings.showwarning = showwarn # use stdout instead of stderr


class tst:
    """
    >>> import warnings
    >>> warnings.warn("foo")
    <doctest ...: UserWarning: foo
      warnings.warn("foo")
    """
    pass

Then pytest --doctest-modules -p no:warnings c.py passes - otherwise it does not get any output of warn() call, and the test fails.

Is there a more standard way to accomplish this? Obviously monkey-patching of the standard library is not what one should normally do.

[The-Compiler]

I'm not too familiar with doctests, but what about turning warnings into errors (pytest -Werror or filterwarnings = error in the config), and then testing for the exception?

class tst:
    """
    >>> import warnings
    >>> warnings.warn("foo")
    Traceback (most recent call last):
        ...
    UserWarning: foo
    """
    pass

[dimpase]

I'm not too familiar with doctests, but what about turning warnings into errors (pytest -Werror or filterwarnings = error in the config), and then testing for the exception?

We have many thousands of doctests, many of these are meant to produce deprecation warnings, experimental status warnings, etc etc. It seems this kind of change would require massive changes in our sources

[The-Compiler]

You could also use warnings.catch_warnings() and then check the warnings programatically. Also see the relevant Python docs around testing warnings.