[FIX] Replace statsmodels.robust.scale.mad with scipy.stats.median_abs_deviation to avoid enabling the GIL

[pvlov]

Hi! Since this is a small diff and I couldn't find a CONTRIBUTING.md, I went ahead and opened a PR. I hope you don't mind!

I wanted to use meegkit in a project that uses a free-threaded Python interpreter, but I ran into the following warning:

<frozen importlib._bootstrap>:491: RuntimeWarning: 
The global interpreter lock (GIL) has been enabled to load module 'statsmodels.robust._qn', 
which has not declared that it can run safely without the GIL. 
To override this behavior and keep the GIL disabled (at your own risk), run with PYTHON_GIL=0 or -Xgil=0.

You can replicate this warning (using uv) with this command:

uv run --no-project --python 3.14t --with statsmodels python -c "import statsmodels.robust.scale"

I checked where/how statsmodels is used and found only one occurrence: statsmodels.robust.scale.mad. This PR replaces it with the equivalent scipy.stats.median_abs_deviation. This makes the package usable in free-threaded contexts while also dropping the statsmodels dependency entirely.

Additionally, I added a test using the built-in sys._is_gil_enabled check to confirm the GIL stays disabled, and added 3.14 and 3.14t to the CI python matrix. Let me know if you want to keep this or if I should drop this.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 83.28%. Comparing base (6b6c1d7) to head (444bcb3).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #126      +/-   ##
==========================================
- Coverage   83.29%   83.28%   -0.01%     
==========================================
  Files          25       25              
  Lines        2861     2860       -1     
==========================================
- Hits         2383     2382       -1     
  Misses        478      478              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[nbara]

Hi @pvlov thanks for this! Perfectly fine to open PRs here :)

[Copilot]

Pull request overview

This PR removes the statsmodels dependency (which can re-enable the GIL under free-threaded CPython) by replacing the one usage of statsmodels.robust.scale.mad with scipy.stats.median_abs_deviation, and adds CI/test coverage to ensure importing meegkit does not re-enable the GIL on free-threaded builds.

Changes:

• Replace statsmodels.robust.scale.mad with scipy.stats.median_abs_deviation in ASR window cleaning logic.
• Drop statsmodels from requirements.txt.
• Add a free-threaded-only test plus extend CI to run on 3.14 and 3.14t.

Reviewed changes

Copilot reviewed 3 out of 4 changed files in this pull request and generated 2 comments.

File	Description
meegkit/asr.py	Replaces MAD implementation to avoid importing statsmodels (prevents triggering GIL enablement).
requirements.txt	Removes the statsmodels runtime dependency.
tests/test_freethreading.py	Adds a subprocess-based import test to verify GIL remains disabled on free-threaded CPython.
.github/workflows/testing.yml	Expands the Python matrix to include 3.14 and 3.14t.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[pvlov]

I know LLMs like their defensive coding but I wanted to make sure I'm not making any false assumptions, so I dug into cpython a bit.

First, if we have a look at sysconfig's documentation we can see that get_config_var will return genuine python ints for boolean values:

>>> import sysconfig
>>> sysconfig.get_config_var('Py_ENABLE_SHARED')
0
...

Next, if we check the implementation of the sysconfig module in cpython, we can see that it does genuinely parse "int-like" values into python ints, unless it is forced to stay a str as per the set _ALWAYS_STR, which does not apply to Py_GIL_DISABLED. Further, undefs for config values set the config_var to 0 and for cypthons builds the build value for true is 1 so this is fine as well. So imo this is unnecessarily defensive.