Ruff config enforces almost nothing, inviting out-of-scope lint sweeps #161

Summary

The Ruff setup enforces almost nothing, and the gap is wide enough that contributors reasonably assume the opposite. This produces large "lint cleanup" diffs that CI neither requires nor rejects.

Current state

.github/workflows/python-package-version-test.yml runs:

- name: Lint with ruff
  run: |
    python -m ruff check pyreason/scripts

and pyproject.toml configures only:

[tool.ruff.lint]
# Ignore ambiguous variable name errors (E741) in interpretation files
[tool.ruff.lint.per-file-ignores]
"pyreason/scripts/interpretation/*.py" = ["E741"]

There is no select, so Ruff falls back to its default rule set: E4, E7, E9, F — a narrow slice of pycodestyle plus pyflakes. Everything else is off, including the rules most people expect a Ruff gate to cover:

• I001 import sorting
• UP pyupgrade
• SIM flake8-simplify
• RUF012 mutable class attributes needing ClassVar
• B / BLE flake8-bugbear and blind-except

Two smaller scope gaps in the same command:

• Only pyreason/scripts is linted. pyreason/__init__.py and pyreason/pyreason.py are not, nor are tests/ or examples/.
• Note the empty [tool.ruff.lint] table above [tool.ruff.lint.per-file-ignores] — it currently holds nothing but the comment.

Why it matters

A contributor who runs ruff check --fix locally with a stock editor config gets a very different result from CI, and produces a sweeping diff believing it was required to get the build green.

This happened in #160: an import-sorting, SIM, UP, RUF012 and BLE001 sweep touched ~45 files alongside a ~12-line performance change. None of it was needed — CI was already passing — and it was reverted so the actual optimization could be reviewed. Without a config change the same thing will recur, and any convention applied by hand will drift straight back since nothing holds it in place.

Suggested resolution

Decide explicitly, in either direction — both are defensible, but the current state is neither:

Option A — enforce a real rule set. Add select to [tool.ruff.lint], land the one-time sweep as its own standalone PR (never mixed with behavioral changes), and consider adding ruff format --check. Worth starting conservative, e.g. ["E4", "E7", "E9", "F", "I", "UP"], and growing it, since SIM and B are noisier on numba-heavy code.

Option B — document the narrow gate. Keep the default rules and add a CONTRIBUTING.md note saying only E4/E7/E9/F are enforced and broader autofixes should not be submitted. Cheaper, and honest about what the gate does.

Either way, worth deciding whether the lint command should cover pyreason/ rather than just pyreason/scripts.

One constraint on Option A: interpretation.py and interpretation_parallel.py are asserted byte-identical except the parallel= decorator by tests/unit/dont_disable_jit/test_parallel_consistency.py. Any autofix must land on both files identically or that test fails.