astroid/brain/brain_namedtuple_enum.py

# Licensed under the LGPL: https://www.gnu.org/licenses/old-licenses/lgpl-2.1.en.html
# For details: https://github.com/pylint-dev/astroid/blob/main/LICENSE
# Copyright (c) https://github.com/pylint-dev/astroid/blob/main/CONTRIBUTORS.txt

"""Astroid hooks for the Python standard library."""

from __future__ import annotations

import functools
import keyword
from collections.abc import Iterator
from textwrap import dedent
from typing import Final

import astroid
from astroid import arguments, bases, nodes, util
from astroid.builder import AstroidBuilder, _extract_single_node, extract_node
from astroid.context import InferenceContext
from astroid.exceptions import (
    AstroidTypeError,
    AstroidValueError,
    InferenceError,
    UseInferenceDefault,
)
from astroid.inference_tip import inference_tip
from astroid.manager import AstroidManager
from astroid.nodes.scoped_nodes.scoped_nodes import SYNTHETIC_ROOT

ENUM_QNAME: Final[str] = "enum.Enum"
TYPING_NAMEDTUPLE_QUALIFIED: Final = {
    "typing.NamedTuple",
    "typing_extensions.NamedTuple",
}
TYPING_NAMEDTUPLE_BASENAMES: Final = {
    "NamedTuple",
    "typing.NamedTuple",
    "typing_extensions.NamedTuple",
}


def _infer_first(node, context):
    if isinstance(node, util.UninferableBase):
        raise UseInferenceDefault
    try:
        value = next(node.infer(context=context))
    except StopIteration as exc:
        raise InferenceError from exc
    if isinstance(value, util.UninferableBase):
        raise UseInferenceDefault()
    return value


def _find_func_form_arguments(node, context):
    def _extract_namedtuple_arg_or_keyword(  # pylint: disable=inconsistent-return-statements
        position, key_name=None
    ):
        if len(args) > position:
            return _infer_first(args[position], context)
        if key_name and key_name in found_keywords:
            return _infer_first(found_keywords[key_name], context)

    args = node.args
    keywords = node.keywords
    found_keywords = (
        {keyword.arg: keyword.value for keyword in keywords} if keywords else {}
    )

    name = _extract_namedtuple_arg_or_keyword(position=0, key_name="typename")
    names = _extract_namedtuple_arg_or_keyword(position=1, key_name="field_names")
    if name and names:
        return name.value, names

    raise UseInferenceDefault()


def infer_func_form(
    node: nodes.Call,
    base_type: nodes.NodeNG,
    *,
    parent: nodes.NodeNG,
    context: InferenceContext | None = None,
    enum: bool = False,
) -> tuple[nodes.ClassDef, str, list[str]]:
    """Specific inference function for namedtuple or Python 3 enum."""
    # node is a Call node, class name as first argument and generated class
    # attributes as second argument

    # namedtuple or enums list of attributes can be a list of strings or a
    # whitespace-separate string
    try:
        name, names = _find_func_form_arguments(node, context)
        try:
            attributes: list[str] = names.value.replace(",", " ").split()
        except AttributeError as exc:
            # Handle attributes of NamedTuples
            if not enum:
                attributes = []
                fields = _get_namedtuple_fields(node)
                if fields:
                    fields_node = extract_node(fields)
                    attributes = [
                        _infer_first(const, context).value for const in fields_node.elts
                    ]

            # Handle attributes of Enums
            else:
                # Enums supports either iterator of (name, value) pairs
                # or mappings.
                if hasattr(names, "items") and isinstance(names.items, list):
                    attributes = [
                        _infer_first(const[0], context).value
                        for const in names.items
                        if isinstance(const[0], nodes.Const)
                    ]
                elif hasattr(names, "elts"):
                    # Enums can support either ["a", "b", "c"]
                    # or [("a", 1), ("b", 2), ...], but they can't
                    # be mixed.
                    if all(isinstance(const, nodes.Tuple) for const in names.elts):
                        attributes = [
                            _infer_first(const.elts[0], context).value
                            for const in names.elts
                            if isinstance(const, nodes.Tuple)
                        ]
                    else:
                        attributes = [
                            _infer_first(const, context).value for const in names.elts
                        ]
                else:
                    raise AttributeError from exc
                if not attributes:
                    raise AttributeError from exc
    except (AttributeError, InferenceError) as exc:
        raise UseInferenceDefault from exc

    if not enum:
        # namedtuple maps sys.intern(str()) over over field_names
        attributes = [str(attr) for attr in attributes]
        # XXX this should succeed *unless* __str__/__repr__ is incorrect or throws
        # in which case we should not have inferred these values and raised earlier
    attributes = [attr for attr in attributes if " " not in attr]

    # If we can't infer the name of the class, don't crash, up to this point
    # we know it is a namedtuple anyway.
    name = name or "Uninferable"
    # we want to return a Class node instance with proper attributes set
    class_node = nodes.ClassDef(
        name,
        lineno=node.lineno,
        col_offset=node.col_offset,
        end_lineno=node.end_lineno,
        end_col_offset=node.end_col_offset,
        parent=parent,
    )
    class_node.postinit(
        bases=[base_type],
        body=[],
        decorators=None,
    )
    # XXX add __init__(*attributes) method
    for attr in attributes:
        fake_node = nodes.EmptyNode()
        fake_node.parent = class_node
        fake_node.attrname = attr
        class_node.instance_attrs[attr] = [fake_node]
    return class_node, name, attributes


def _has_namedtuple_base(node):
    """Predicate for class inference tip.

    :type node: ClassDef
    :rtype: bool
    """
    return set(node.basenames) & TYPING_NAMEDTUPLE_BASENAMES


def _looks_like(node, name) -> bool:
    func = node.func
    if isinstance(func, nodes.Attribute):
        return func.attrname == name
    if isinstance(func, nodes.Name):
        return func.name == name
    return False


_looks_like_namedtuple = functools.partial(_looks_like, name="namedtuple")
_looks_like_enum = functools.partial(_looks_like, name="Enum")
_looks_like_typing_namedtuple = functools.partial(_looks_like, name="NamedTuple")


def infer_named_tuple(
    node: nodes.Call, context: InferenceContext | None = None
) -> Iterator[nodes.ClassDef]:
    """Specific inference function for namedtuple Call node."""
    tuple_base: nodes.Name = _extract_single_node("tuple")
    class_node, name, attributes = infer_func_form(
        node, tuple_base, parent=SYNTHETIC_ROOT, context=context
    )

    call_site = arguments.CallSite.from_call(node, context=context)
    func = util.safe_infer(
        _extract_single_node("import collections; collections.namedtuple")
    )
    assert isinstance(func, nodes.NodeNG)
    try:
        rename_arg_bool_value = next(
            call_site.infer_argument(func, "rename", context or InferenceContext())
        ).bool_value()
        rename = rename_arg_bool_value is True
    except (InferenceError, StopIteration):
        rename = False

    try:
        attributes = _check_namedtuple_attributes(name, attributes, rename)
    except AstroidTypeError as exc:
        raise UseInferenceDefault("TypeError: " + str(exc)) from exc
    except AstroidValueError as exc:
        raise UseInferenceDefault("ValueError: " + str(exc)) from exc

    replace_args = ", ".join(f"{arg}=None" for arg in attributes)
    field_def = (
        "    {name} = property(lambda self: self[{index:d}], "
        "doc='Alias for field number {index:d}')"
    )
    field_defs = "\n".join(
        field_def.format(name=name, index=index)
        for index, name in enumerate(attributes)
    )
    fake = AstroidBuilder(AstroidManager()).string_build(
        f"""
class {name}(tuple):
    __slots__ = ()
    _fields = {attributes!r}
    def _asdict(self):
        return self.__dict__
    @classmethod
    def _make(cls, iterable, new=tuple.__new__, len=len):
        return new(cls, iterable)
    def _replace(self, {replace_args}):
        return self
    def __getnewargs__(self):
        return tuple(self)
{field_defs}
    """
    )
    class_node.locals["_asdict"] = fake.body[0].locals["_asdict"]
    class_node.locals["_make"] = fake.body[0].locals["_make"]
    class_node.locals["_replace"] = fake.body[0].locals["_replace"]
    class_node.locals["_fields"] = fake.body[0].locals["_fields"]
    for attr in attributes:
        class_node.locals[attr] = fake.body[0].locals[attr]
    # we use UseInferenceDefault, we can't be a generator so return an iterator
    return iter([class_node])


def _get_renamed_namedtuple_attributes(field_names):
    names = list(field_names)
    seen = set()
    for i, name in enumerate(field_names):
        # pylint: disable = too-many-boolean-expressions
        if (
            not all(c.isalnum() or c == "_" for c in name)
            or keyword.iskeyword(name)
            or not name
            or name[0].isdigit()
            or name.startswith("_")
            or name in seen
        ):
            names[i] = "_%d" % i
        seen.add(name)
    return tuple(names)


def _check_namedtuple_attributes(typename, attributes, rename=False):
    attributes = tuple(attributes)
    if rename:
        attributes = _get_renamed_namedtuple_attributes(attributes)

    # The following snippet is derived from the CPython Lib/collections/__init__.py sources
    # <snippet>
    for name in (typename, *attributes):
        if not isinstance(name, str):
            raise AstroidTypeError(
                f"Type names and field names must be strings, not {type(name)!r}"
            )
        if not name.isidentifier():
            raise AstroidValueError(
                "Type names and field names must be valid" + f"identifiers: {name!r}"
            )
        if keyword.iskeyword(name):
            raise AstroidValueError(
                f"Type names and field names cannot be a keyword: {name!r}"
            )

    seen = set()
    for name in attributes:
        if name.startswith("_") and not rename:
            raise AstroidValueError(
                f"Field names cannot start with an underscore: {name!r}"
            )
        if name in seen:
            raise AstroidValueError(f"Encountered duplicate field name: {name!r}")
        seen.add(name)
    # </snippet>

    return attributes


def infer_enum(
    node: nodes.Call, context: InferenceContext | None = None
) -> Iterator[bases.Instance]:
    """Specific inference function for enum Call node."""
    # Raise `UseInferenceDefault` if `node` is a call to a a user-defined Enum.
    try:
        inferred = node.func.infer(context)
    except (InferenceError, StopIteration) as exc:
        raise UseInferenceDefault from exc

    if not any(
        isinstance(item, nodes.ClassDef) and item.qname() == ENUM_QNAME
        for item in inferred
    ):
        raise UseInferenceDefault

    enum_meta = _extract_single_node(
        """
    class EnumMeta(object):
        'docstring'
        def __call__(self, node):
            class EnumAttribute(object):
                name = ''
                value = 0
            return EnumAttribute()
        def __iter__(self):
            class EnumAttribute(object):
                name = ''
                value = 0
            return [EnumAttribute()]
        def __reversed__(self):
            class EnumAttribute(object):
                name = ''
                value = 0
            return (EnumAttribute, )
        def __next__(self):
            return next(iter(self))
        def __getitem__(self, attr):
            class Value(object):
                @property
                def name(self):
                    return ''
                @property
                def value(self):
                    return attr

            return Value()
        __members__ = ['']
    """
    )

    # FIXME arguably, the base here shouldn't be the EnumMeta class definition
    # itself, but a reference (Name) to it. Otherwise, the invariant that all
    # children of a node have that node as their parent is broken.
    class_node = infer_func_form(
        node,
        enum_meta,
        parent=SYNTHETIC_ROOT,
        context=context,
        enum=True,
    )[0]
    return iter([class_node.instantiate_class()])


INT_FLAG_ADDITION_METHODS = """
    def __or__(self, other):
        return {name}(self.value | other.value)
    def __and__(self, other):
        return {name}(self.value & other.value)
    def __xor__(self, other):
        return {name}(self.value ^ other.value)
    def __add__(self, other):
        return {name}(self.value + other.value)
    def __div__(self, other):
        return {name}(self.value / other.value)
    def __invert__(self):
        return {name}(~self.value)
    def __mul__(self, other):
        return {name}(self.value * other.value)
"""


def infer_enum_class(node: nodes.ClassDef) -> nodes.ClassDef:
    """Specific inference for enums."""
    for basename in (b for cls in node.mro() for b in cls.basenames):
        if node.root().name == "enum":
            # Skip if the class is directly from enum module.
            break
        dunder_members = {}
        target_names = set()
        for local, values in node.locals.items():
            if (
                any(not isinstance(value, nodes.AssignName) for value in values)
                or local == "_ignore_"
            ):
                continue

            stmt = values[0].statement()
            if isinstance(stmt, nodes.Assign):
                if isinstance(stmt.targets[0], nodes.Tuple):
                    targets = stmt.targets[0].itered()
                else:
                    targets = stmt.targets
            elif isinstance(stmt, nodes.AnnAssign):
                targets = [stmt.target]
            else:
                continue

            inferred_return_value = None
            if stmt.value is not None:
                if isinstance(stmt.value, nodes.Const):
                    if isinstance(stmt.value.value, str):
                        inferred_return_value = repr(stmt.value.value)
                    else:
                        inferred_return_value = stmt.value.value
                else:
                    inferred_return_value = stmt.value.as_string()

            new_targets = []
            for target in targets:
                if isinstance(target, nodes.Starred):
                    continue
                target_names.add(target.name)
                # Replace all the assignments with our mocked class.
                classdef = dedent(
                    """
                class {name}({types}):
                    @property
                    def value(self):
                        return {return_value}
                    @property
                    def _value_(self):
                        return {return_value}
                    @property
                    def name(self):
                        return "{name}"
                    @property
                    def _name_(self):
                        return "{name}"
                """.format(
                        name=target.name,
                        types=", ".join(node.basenames),
                        return_value=inferred_return_value,
                    )
                )
                if "IntFlag" in basename:
                    # Alright, we need to add some additional methods.
                    # Unfortunately we still can't infer the resulting objects as
                    # Enum members, but once we'll be able to do that, the following
                    # should result in some nice symbolic execution
                    classdef += INT_FLAG_ADDITION_METHODS.format(name=target.name)

                fake = AstroidBuilder(
                    AstroidManager(), apply_transforms=False
                ).string_build(classdef)[target.name]
                fake.parent = target.parent
                for method in node.mymethods():
                    fake.locals[method.name] = [method]
                new_targets.append(fake.instantiate_class())
                if stmt.value is None:
                    continue
                dunder_members[local] = fake
            node.locals[local] = new_targets

        # The undocumented `_value2member_map_` member:
        node.locals["_value2member_map_"] = [
            nodes.Dict(
                parent=node,
                lineno=node.lineno,
                col_offset=node.col_offset,
                end_lineno=node.end_lineno,
                end_col_offset=node.end_col_offset,
            )
        ]

        members = nodes.Dict(
            parent=node,
            lineno=node.lineno,
            col_offset=node.col_offset,
            end_lineno=node.end_lineno,
            end_col_offset=node.end_col_offset,
        )
        members.postinit(
            [
                (
                    nodes.Const(k, parent=members),
                    nodes.Name(
                        v.name,
                        parent=members,
                        lineno=v.lineno,
                        col_offset=v.col_offset,
                        end_lineno=v.end_lineno,
                        end_col_offset=v.end_col_offset,
                    ),
                )
                for k, v in dunder_members.items()
            ]
        )
        node.locals["__members__"] = [members]
        # The enum.Enum class itself defines two @DynamicClassAttribute data-descriptors
        # "name" and "value" (which we override in the mocked class for each enum member
        # above). When dealing with inference of an arbitrary instance of the enum
        # class, e.g. in a method defined in the class body like:
        #     class SomeEnum(enum.Enum):
        #         def method(self):
        #             self.name  # <- here
        # In the absence of an enum member called "name" or "value", these attributes
        # should resolve to the descriptor on that particular instance, i.e. enum member.
        # For "value", we have no idea what that should be, but for "name", we at least
        # know that it should be a string, so infer that as a guess.
        if "name" not in target_names:
            code = dedent(
                '''
                @property
                def name(self):
                    """The name of the Enum member.

                    This is a reconstruction by astroid: enums are too dynamic to understand, but we at least
                    know 'name' should be a string, so this is astroid's best guess.
                    """
                    return ''
                '''
            )
            name_dynamicclassattr = AstroidBuilder(AstroidManager()).string_build(code)[
                "name"
            ]
            node.locals["name"] = [name_dynamicclassattr]
        break
    return node


def infer_typing_namedtuple_class(class_node, context: InferenceContext | None = None):
    """Infer a subclass of typing.NamedTuple."""
    # Check if it has the corresponding bases
    annassigns_fields = [
        annassign.target.name
        for annassign in class_node.body
        if isinstance(annassign, nodes.AnnAssign)
    ]
    code = dedent(
        """
    from collections import namedtuple
    namedtuple({typename!r}, {fields!r})
    """
    ).format(typename=class_node.name, fields=",".join(annassigns_fields))
    node = extract_node(code)
    try:
        generated_class_node = next(infer_named_tuple(node, context))
    except StopIteration as e:
        raise InferenceError(node=node, context=context) from e
    for method in class_node.mymethods():
        generated_class_node.locals[method.name] = [method]

    for body_node in class_node.body:
        if isinstance(body_node, nodes.Assign):
            for target in body_node.targets:
                attr = target.name
                generated_class_node.locals[attr] = class_node.locals[attr]
        elif isinstance(body_node, nodes.ClassDef):
            generated_class_node.locals[body_node.name] = [body_node]

    return iter((generated_class_node,))


def infer_typing_namedtuple_function(node, context: InferenceContext | None = None):
    """
    Starting with python3.9, NamedTuple is a function of the typing module.
    The class NamedTuple is build dynamically through a call to `type` during
    initialization of the `_NamedTuple` variable.
    """
    klass = extract_node(
        """
        from typing import _NamedTuple
        _NamedTuple
        """
    )
    return klass.infer(context)


def infer_typing_namedtuple(
    node: nodes.Call, context: InferenceContext | None = None
) -> Iterator[nodes.ClassDef]:
    """Infer a typing.NamedTuple(...) call."""
    # This is essentially a namedtuple with different arguments
    # so we extract the args and infer a named tuple.
    try:
        func = next(node.func.infer())
    except (InferenceError, StopIteration) as exc:
        raise UseInferenceDefault from exc

    if func.qname() not in TYPING_NAMEDTUPLE_QUALIFIED:
        raise UseInferenceDefault

    if len(node.args) != 2:
        raise UseInferenceDefault

    if not isinstance(node.args[1], (nodes.List, nodes.Tuple)):
        raise UseInferenceDefault

    return infer_named_tuple(node, context)


def _get_namedtuple_fields(node: nodes.Call) -> str:
    """Get and return fields of a NamedTuple in code-as-a-string.

    Because the fields are represented in their code form we can
    extract a node from them later on.
    """
    names = []
    container = None
    try:
        container = next(node.args[1].infer())
    except (InferenceError, StopIteration) as exc:
        raise UseInferenceDefault from exc
    # We pass on IndexError as we'll try to infer 'field_names' from the keywords
    except IndexError:
        pass
    if not container:
        for keyword_node in node.keywords:
            if keyword_node.arg == "field_names":
                try:
                    container = next(keyword_node.value.infer())
                except (InferenceError, StopIteration) as exc:
                    raise UseInferenceDefault from exc
                break
    if not isinstance(container, nodes.BaseContainer):
        raise UseInferenceDefault
    for elt in container.elts:
        if isinstance(elt, nodes.Const):
            names.append(elt.as_string())
            continue
        if not isinstance(elt, (nodes.List, nodes.Tuple)):
            raise UseInferenceDefault
        if len(elt.elts) != 2:
            raise UseInferenceDefault
        names.append(elt.elts[0].as_string())

    if names:
        field_names = f"({','.join(names)},)"
    else:
        field_names = ""
    return field_names


def _is_enum_subclass(cls: astroid.ClassDef) -> bool:
    """Return whether cls is a subclass of an Enum."""
    return cls.is_subtype_of("enum.Enum")


def register(manager: AstroidManager) -> None:
    manager.register_transform(
        nodes.Call, inference_tip(infer_named_tuple), _looks_like_namedtuple
    )
    manager.register_transform(nodes.Call, inference_tip(infer_enum), _looks_like_enum)
    manager.register_transform(
        nodes.ClassDef, infer_enum_class, predicate=_is_enum_subclass
    )
    manager.register_transform(
        nodes.ClassDef,
        inference_tip(infer_typing_namedtuple_class),
        _has_namedtuple_base,
    )
    manager.register_transform(
        nodes.FunctionDef,
        inference_tip(infer_typing_namedtuple_function),
        lambda node: node.name == "NamedTuple"
        and getattr(node.root(), "name", None) == "typing",
    )
    manager.register_transform(
        nodes.Call,
        inference_tip(infer_typing_namedtuple),
        _looks_like_typing_namedtuple,
    )