ruff.toml

# Kept at the repo root on purpose (unlike config/deny.toml, config/osv-scanner.toml, config/taplo.toml).
# Editors/IDEs (the VS Code Ruff extension, PyCharm) and pre-commit auto-discover ruff config only as a root
# ruff.toml / .ruff.toml / pyproject.toml. ruff has no config-path env var, so moving this under config/
# would silently drop every editor back to ruff's defaults (line-length 88, no isort). The mise tasks could
# pass --config, but the live editor experience can't, so it stays here.
#
# Match the editorconfig line length. Without this, ruff would use its default of 88, so files
# that ruff considered "already formatted" could still trip the editorconfig-check.
line-length = 120

# Exclude componentize-py's generated bindings, which we don't own and don't ship.
# componentize-py regenerates `wit_world` and `componentize_py_async_support`
# under services/ws-modules/wasi-graphics-info/ on every build. Exclude them here
# (on top of ruff's defaults) so the ruff tasks don't need to repeat `--exclude`
# flags, and so they're skipped even when ruff runs with `--no-respect-gitignore`.
extend-exclude = ["componentize_py_async_support", "wit_world"]

[format]
# Force LF so formatting is deterministic across platforms.
# ruff's default (`auto`) writes platform line endings, so the gen:python-* codegen tasks emitted
# CRLF-churned trees when run on Windows -- the sole reason those tasks used to skip there.
line-ending = "lf"

[lint]
# On top of the default rule set (E, F). Codes are kept alphabetically sorted.
#   "ARG" -- flake8-unused-arguments: unused function/method/lambda arguments. This is the local equivalent of
#   DeepSource's PYL-W0613; ruff honours the underscore-prefix convention (`dummy-variable-rgx`), so a param kept
#   only to satisfy a fixed callback signature (the pyo3-runner `init`/`on_text_frame` hooks) is silenced by
#   renaming it `_send` / `_text` rather than deleting it.
#   "B" -- flake8-bugbear: likely bugs (mutable defaults, `zip` without `strict=`, loop-variable shadowing, ...).
#   "C4" -- flake8-comprehensions: unnecessary or unpythonic comprehensions and generator calls.
#   "D" -- pydocstyle (PEP 257) docstring checks; see [lint.pydocstyle].
#   "I" -- isort import sorting. `ruff format` only handles whitespace; the "I" rules give deterministic
#   grouping. datamodel-codegen's ruff formatter doesn't sort imports, so without this the generated
#   messages.py drifts between regens.
#   "N" -- pep8-naming: class / function / constant / exception naming conventions.
#   "PIE" -- flake8-pie: unnecessary or duplicated constructs (redundant passes, dict literals, ...).
#   "PLC" / "PLE" / "PLW" -- pylint conventions / errors / warnings. PLR (its refactor, complexity, and
#   magic-value rules) is deliberately left out as too subjective for the numeric-heavy CV code here.
#   "RUF" -- ruff-native checks (unsorted `__all__`, unused `noqa`, ...).
#   "SIM" -- flake8-simplify: simplifiable conditionals, `return`s, and context managers.
#   "UP" -- pyupgrade: modern syntax for the project's minimum Python.
# Use `extend-select`, not `select`, so we keep the default F401 (unused-import)
# rule that openapi-python-client relies on internally -- it generates
# speculative imports and trims them by re-running ruff with the default rules.
extend-select = ["ARG", "B", "C4", "D", "I", "N", "PIE", "PLC", "PLE", "PLW", "RUF", "SIM", "UP"]

[lint.pydocstyle]
convention = "pep257"

[lint.per-file-ignores]
# Generated clients are exempt from the hand-authoring quality families.
# Their code is machine-emitted by openapi-python-client / datamodel-codegen and regenerated wholesale, so naming,
# modernization, and refactor findings can't be hand-fixed and would break CI on the next regen. The E/F baseline
# stays active, as does "I" (import sorting keeps messages.py stable across regens); "D" was already dropped
# because the docstrings come verbatim from the OpenAPI/JSON-schema spec.
"generated/**" = ["ARG", "B", "C4", "D", "N", "PIE", "PLC", "PLE", "PLW", "RUF", "SIM", "UP"]
# The et-ws-pyo3-runner example modules keep per-process state in module-level globals by design.
# The runner instantiates one copy per process and the contract is deliberately class-free (see echo.py's module
# docstring), so PLW0603's "don't use the global statement" advice doesn't apply.
"services/ws-pyo3-runner/python/**" = ["PLW0603"]
# Test modules: docstrings on test classes/methods are noise, not API surface.
"**/tests/**" = ["D"]