GH-46546: [CI][Dev][Python] Use pre-commit for numpydoc

[kou]

Rationale for this change

We want to migrate to pre-commit from archery lint.

What changes are included in this PR?

Use pre-commit for numpydoc but this doesn't support Cython files.

Are these changes tested?

Yes.

Are there any user-facing changes?

No.

• GitHub Issue: [CI][Dev][Python] Use pre-commit for numpydoc #46546

[github-actions]

⚠️ GitHub issue #46546 has been automatically assigned in GitHub to PR creator.

[kou]

archery lint has our own Cython integration code:

arrow/dev/archery/archery/lang/python.py

Lines 115 to 256 in 1943911

	class NumpyDoc:
	IGNORE_VALIDATION_ERRORS_FOR_TYPE = {
	# Enum function signatures should never be documented
	EnumMeta: ["PR01"]
	}
	def __init__(self, symbols=None):
	if not have_numpydoc:
	raise RuntimeError(
	'Numpydoc is not available, install with command: '
	'pip install numpydoc==1.1.0'
	)
	self.symbols = set(symbols or {'pyarrow'})
	def traverse(self, fn, obj, from_package):
	"""Apply a function on publicly exposed API components.
	Recursively iterates over the members of the passed object. It omits
	any '_' prefixed and thirdparty (non pyarrow) symbols.
	Parameters
	----------
	fn : callable
	A function to apply on all traversed objects.
	obj : Any
	The object to start from.
	from_package : string
	Predicate to only consider objects from this package.
	"""
	todo = [obj]
	seen = set()
	while todo:
	obj = todo.pop()
	if obj in seen:
	continue
	else:
	seen.add(obj)
	fn(obj)
	for name in dir(obj):
	if name.startswith('_'):
	continue
	member = getattr(obj, name)
	module = _get_module(member)
	if module is None or not module.startswith(from_package):
	continue
	# Is it a Cython-generated method? If so, try to detect
	# whether it only has a implicitly-generated docstring,
	# and no user-defined docstring following it.
	# The generated docstring would lack description of method
	# parameters and therefore fail Numpydoc validation.
	if hasattr(member, '__objclass__'):
	doc = getattr(member, '__doc__', None)
	# The Cython-generated docstring would be a one-liner,
	# such as "ReadOptions.equals(self, ReadOptions other)".
	if (doc and '\n' not in doc and f'.{name}(' in doc):
	continue
	todo.append(member)
	@contextmanager
	def _apply_patches(self):
	"""
	Patch Docstring class to bypass loading already loaded python objects.
	"""
	orig_load_obj = Docstring._load_obj
	orig_signature = inspect.signature
	@staticmethod
	def _load_obj(obj):
	# By default it expects a qualname and import the object, but we
	# have already loaded object after the API traversal.
	if isinstance(obj, str):
	return orig_load_obj(obj)
	else:
	return obj
	def signature(obj):
	# inspect.signature tries to parse __text_signature__ if other
	# properties like __signature__ doesn't exists, but cython
	# doesn't set that property despite that embedsignature cython
	# directive is set. The only way to inspect a cython compiled
	# callable's signature to parse it from __doc__ while
	# embedsignature directive is set during the build phase.
	# So path inspect.signature function to attempt to parse the first
	# line of callable.__doc__ as a signature.
	try:
	return orig_signature(obj)
	except Exception as orig_error:
	try:
	return inspect_signature(obj)
	except Exception:
	raise orig_error
	try:
	Docstring._load_obj = _load_obj
	inspect.signature = signature
	yield
	finally:
	Docstring._load_obj = orig_load_obj
	inspect.signature = orig_signature
	def validate(self, from_package='', allow_rules=None,
	disallow_rules=None):
	results = []
	def callback(obj):
	try:
	result = validate(obj)
	except OSError as e:
	symbol = f"{_get_module(obj, default='')}.{obj.__name__}"
	logger.warning(f"Unable to validate `{symbol}` due to `{e}`")
	return
	errors = []
	for errcode, errmsg in result.get('errors', []):
	if allow_rules and errcode not in allow_rules:
	continue
	if disallow_rules and errcode in disallow_rules:
	continue
	if any(isinstance(obj, obj_type) and errcode in errcode_list
	for obj_type, errcode_list
	in NumpyDoc.IGNORE_VALIDATION_ERRORS_FOR_TYPE.items()):
	continue
	errors.append((errcode, errmsg))
	if len(errors):
	result['errors'] = errors
	results.append((obj, result))
	with self._apply_patches():
	for symbol in self.symbols:
	try:
	obj = Docstring._load_obj(symbol)
	except (ImportError, AttributeError):
	print(f'{symbol} is not available for import')
	else:
	self.traverse(callback, obj, from_package=from_package)
	return results

But we can't use it with numpydoc lint.

Do we need the Cython integration?

[hiroyuki-sato]

JFYI: other files contains $ character like ?^python/pyarrow/interchange/from_dataframe\.py$

[kou]

Good catch!
I've added $.

[kou]

@pitrou Is Cython docstring validation by numpydoc important for us?

It was implemented by #12698 only for us. If it's important, can we implement it in upstream instead of implementing it in Archery?

[pitrou]

@pitrou Is Cython docstring validation by numpydoc important for us?

We want to validate docstrings regardless of how the functions are generated, yes.

It was implemented by #12698 only for us. If it's important, can we implement it in upstream instead of implementing it in Archery?

We can probably do that. However, we may have to wait for cython/cython#6904 to be released first.

cc @raulcd

[kou]

OK. I've opened #46641 for it.

I'll merge this if nobody objects it. Because we can remove Archery based lint from our lint CI job by merging this. It will solve #46528.

[pitrou]

The problem is we don't know if numpydoc will accept a contribution for it.

We can still use Archery-based lint for numpydoc. I don't think there's any urgency.