Repository navigation
Tests always skipped when dependency version is present #318
Description
__doctest_requires__ causes tests to be skipped whenever a dependency version requirement is present, even if the dependency is present and meets the requirement.
The code was behaving correctly in 1.5.0. The bug was present in 1.6.0. The bug is still present in 1.7.0.
To reproduce, save code fragments shown below as demo.py and pyproject.toml into an empty project root directory, then run the following commands from that directory:
- Create virtual environment:
python -m venv .venv - Install dependencies:
.\.venv\Scripts\pip.exe install -e . - Show NumPy version:
.\.venv\Scripts\pip.exe show numpy - Read the output to verify NumPy >= 2.0 is installed.
- Run
doctest: .\.venv\Scripts\python.exe -m pytest demo.py --doctest-modules -v - Observe the output depending on the version of pytest-doctestplus:
1.5.0:demo.py::demo.my_function PASSED
1.6.0:demo.py::demo.my_function SKIPPED (could not import 'numpy>=2.0': No module named 'numpy>=2')
1.7.0:demo.py::demo.my_function SKIPPED (could not import 'numpy>=2.0': No module named 'numpy>=2') - Upgrade pytest-doctestplus to 1.6.0 and run the test again, following the steps
mentioned above. - Upgrade pytest-doctestplus to 1.7.0 and run the test again, following the steps
mentioned above.
#demo.py
__doctest_requires__ = {("my_function",): ["numpy>=2.0"]}
def my_function():
"""Simple function to demonstrate the bug.
>>> import numpy as np
>>> np.array([1, 2, 3])
array([1, 2, 3])
"""
pass#pyproject.toml
[project]
name = "pytest-doctestplus-bug-demo"
version = "0.1.0"
description = "Demonstrates pytest-doctestplus 1.7.0 bug with version constraints"
requires-python = ">=3.9"
dependencies = [
"numpy>=2.0",
"pytest-doctestplus==1.5.0",
"pytest>=7.0",
]
[build-system]
requires = ["setuptools>=45", "wheel"]
build-backend = "setuptools.build_meta"
[tool.pytest.ini_options]
addopts = "--doctest-modules --doctest-plus"
Activity
KostaIlic2 commentedon Jan 20, 2026
I have looked at the problem with help of LLM tools. I have a potential fix and a regression test improvement that covers the scenarios from the bug. Would you like me to post a PR?
pllim commentedon Jan 20, 2026
In 1.6.0, these 2 PRs were merged and could be suspect?
pllim commentedon Jan 20, 2026
Perhaps run a test locally first to make sure LLM suggestion actually works? If unit tests pass, then PR is fine. Thanks.
pllim commentedon Jan 20, 2026
Did you try to bisect it first?
bsipocz commentedon Jan 20, 2026
Before you unleash LLM on this, it would be really useful to see an actually failing. Here are some guidance of how to write such examples; ideally they are just a test function (I certainly will not download attached zip files): https://matthewrocklin.com/minimal-bug-reports/
KostaIlic2 commentedon Jan 20, 2026
Dear @bsipocz and @pllim, Thank you for your prompt responses and clear direction.
I rewrote the report per your request to avoid a zip file.
I mentioned LLM to be upfront about it. I reviewed the changes made by the LLM, made adjustments myself, and then had another level of coworker review. Changes to the core file are about 10 lines of code, and changes to the test are about 30 lines of code.
I did not track down the PR that introduced the bug.
KostaIlic2 commentedon Jan 21, 2026
To follow up on my earlier response, I have run updated test locally. Without the
plugin.pycode change, the test fails. With the change, the test passes.I bisected and found the bug was introduced in PR #312 (commit 3188e5b).
I have created a draft PR #319 .
Do you have any expectations from contributors beyond updates to the code, the tests, and the changelog?