Enable the muff linter (muff check) with a rule set suited to this codebase #39

Background

#37 replaced isort, cercis and flake8 with muff-format, pydoclint and a new set of pre-commit hooks, copied from jsh9/bootstrap2. Only the muff formatter runs:

• The pre-commit config has only the muff-format hook.
• The muff-format tox env runs muff format --diff.
• Nothing runs muff check.

muff.toml contains a full [lint] section copied from bootstrap2 (extend-select with about 50 rule groups), but it is unused.

Before #37, CI also ran three flake8 envs (flake8-basic, flake8-misc, flake8-docstrings). Python linting is therefore reduced to pydoclint, which checks docstrings only. This issue is about adding a linter back.

Current numbers

These were measured with muff 0.15.20 and the current muff.toml on the #37 branch. They cover PySeismoSoil/ only, because muff.toml excludes tests/.

With the full bootstrap2 rule set: 1,597 violations (84 auto-fixable). Most come from rules that conflict with this codebase's conventions and public API:

Proposal

1. Add the hook. Add id: muff-check from https://github.com/jsh9/muff-pre-commit, with args: [--config, muff.toml], to .pre-commit-config.yaml. Optionally also add a muff-check tox env (muff check --config=muff.toml PySeismoSoil) to the CI lint job. If you do, extend the sync-precommit-tox-versions hook so the env's muff pin stays in sync.
2. Ignore the rules that conflict with the codebase's conventions in muff.toml, for example:

   ignore = [
     # ... existing entries ...
     "E501",                                     # muff-format handles code; long strings/URLs remain
     "FBT001", "FBT002", "FBT003",               # boolean positional args are part of the public API
     "N801", "N802", "N803", "N806", "N815", "N816", "N999",  # scientific naming (Vs, Tmax, Vs_Profile, ...)
     "PLR2004",                                  # magic numbers are common in physics formulas
   ]

   That leaves 523 violations. The largest groups are:
   • UP031 (81): % formatting
   • RET504 (39): unnecessary assignment before return
   • ANN001 (38): missing argument annotations
   • PTH118 (29): os.path.join
   • TC002 (26): imports used only for typing
   • PLR6201 (22): literal membership tests
   • PLR0915 (18): too many statements
   • PLR6104 (18): non-augmented assignment
   • RUF100 (18): unused noqa directives
   • PLC0415 (16): import outside top level
   • SLF001 (16): private member access
3. Fix or ignore the rest, rule by rule. Some groups are worth fixing; others (e.g. PTH, ANN, TC, PLR6201) may be style preferences better added to ignore.

Possible first step

Start with roughly what flake8 used to check:

--select E,F,W,B,C4,ERA,Q,T10,ISC,RET,D
--ignore E501,D100,D104,D105,D107,D200,D205,D400

That gives only 55 violations:

• 39 × RET504
• 7 × D420
• 5 × ERA001
• 3 × B905
• 1 × RET506

Then broaden the rule set gradually.

Related cleanups that would surface

• Stale flake8 # noqa codes. muff reports 18 × RUF100 (unused noqa) and 7 × RUF102 (invalid rule code). These are leftovers from the flake8 setup:

  • codes for rules muff doesn't enable, e.g. E226, E501, R506, E731
  • codes muff doesn't know, e.g. LN001, W503, E800

  Most are auto-fixable.

• RUF012 (mutable class default). This flags Site_Factors.Vs30_array, z1_array and PGA_array (lists used as class attributes). They are never mutated, so this is low risk. Switching them to tuples would also require changing the three '... %s' % Site_Factors.Vs30_array error messages, which break with a tuple.

• tests/. It is currently excluded via extend-exclude = ["tests"] (on purpose in Modernize CI pipeline (muff-format, pydoclint, pre-commit); drop Python 3.10 #37), so the tests/* per-file ignores in muff.toml have no effect until that exclusion is removed.

This should be a separate PR from #37: unlike #37, it would change code beyond formatting.