attrs.py

"""Plugin for supporting the attrs library (http://www.attrs.org)"""

from __future__ import annotations

from collections import defaultdict
from functools import reduce
from typing import Final, Iterable, List, Mapping, cast
from typing_extensions import Literal

import mypy.plugin  # To avoid circular imports.
from mypy.applytype import apply_generic_arguments
from mypy.errorcodes import LITERAL_REQ
from mypy.expandtype import expand_type, expand_type_by_instance
from mypy.exprtotype import TypeTranslationError, expr_to_unanalyzed_type
from mypy.meet import meet_types
from mypy.messages import format_type_bare
from mypy.nodes import (
    ARG_NAMED,
    ARG_NAMED_OPT,
    ARG_OPT,
    ARG_POS,
    MDEF,
    Argument,
    AssignmentStmt,
    CallExpr,
    Context,
    Decorator,
    Expression,
    FuncDef,
    IndexExpr,
    JsonDict,
    LambdaExpr,
    ListExpr,
    MemberExpr,
    NameExpr,
    OverloadedFuncDef,
    PlaceholderNode,
    RefExpr,
    SymbolTableNode,
    TempNode,
    TupleExpr,
    TypeApplication,
    TypeInfo,
    TypeVarExpr,
    Var,
    is_class_var,
)
from mypy.plugin import SemanticAnalyzerPluginInterface
from mypy.plugins.common import (
    _get_argument,
    _get_bool_argument,
    _get_decorator_bool_argument,
    add_attribute_to_class,
    add_method_to_class,
    deserialize_and_fixup_type,
)
from mypy.server.trigger import make_wildcard_trigger
from mypy.typeops import get_type_vars, make_simplified_union, map_type_from_supertype
from mypy.types import (
    AnyType,
    CallableType,
    FunctionLike,
    Instance,
    LiteralType,
    NoneType,
    Overloaded,
    ProperType,
    TupleType,
    Type,
    TypeOfAny,
    TypeType,
    TypeVarType,
    UninhabitedType,
    UnionType,
    get_proper_type,
)
from mypy.typevars import fill_typevars
from mypy.util import unmangle

# The names of the different functions that create classes or arguments.
attr_class_makers: Final = {"attr.s", "attr.attrs", "attr.attributes"}
attr_dataclass_makers: Final = {"attr.dataclass"}
attr_frozen_makers: Final = {"attr.frozen", "attrs.frozen"}
attr_define_makers: Final = {"attr.define", "attr.mutable", "attrs.define", "attrs.mutable"}
attr_attrib_makers: Final = {"attr.ib", "attr.attrib", "attr.attr", "attr.field", "attrs.field"}
attr_optional_converters: Final = {"attr.converters.optional", "attrs.converters.optional"}

SELF_TVAR_NAME: Final = "_AT"
MAGIC_ATTR_NAME: Final = "__attrs_attrs__"
MAGIC_ATTR_CLS_NAME_TEMPLATE: Final = "__{}_AttrsAttributes__"  # The tuple subclass pattern.
ATTRS_INIT_NAME: Final = "__attrs_init__"


class Converter:
    """Holds information about a `converter=` argument"""

    def __init__(self, init_type: Type | None = None, ret_type: Type | None = None) -> None:
        self.init_type = init_type
        self.ret_type = ret_type


class Attribute:
    """The value of an attr.ib() call."""

    def __init__(
        self,
        name: str,
        alias: str | None,
        info: TypeInfo,
        has_default: bool,
        init: bool,
        kw_only: bool,
        converter: Converter | None,
        context: Context,
        init_type: Type | None,
    ) -> None:
        self.name = name
        self.alias = alias
        self.info = info
        self.has_default = has_default
        self.init = init
        self.kw_only = kw_only
        self.converter = converter
        self.context = context
        self.init_type = init_type

    def argument(self, ctx: mypy.plugin.ClassDefContext) -> Argument:
        """Return this attribute as an argument to __init__."""
        assert self.init
        init_type: Type | None = None
        if self.converter:
            if self.converter.init_type:
                init_type = self.converter.init_type
                if init_type and self.init_type and self.converter.ret_type:
                    # The converter return type should be the same type as the attribute type.
                    # Copy type vars from attr type to converter.
                    converter_vars = get_type_vars(self.converter.ret_type)
                    init_vars = get_type_vars(self.init_type)
                    if converter_vars and len(converter_vars) == len(init_vars):
                        variables = {
                            binder.id: arg for binder, arg in zip(converter_vars, init_vars)
                        }
                        init_type = expand_type(init_type, variables)
            else:
                ctx.api.fail("Cannot determine __init__ type from converter", self.context)
                init_type = AnyType(TypeOfAny.from_error)
        else:  # There is no converter, the init type is the normal type.
            init_type = self.init_type or self.info[self.name].type

        unannotated = False
        if init_type is None:
            unannotated = True
            # Convert type not set to Any.
            init_type = AnyType(TypeOfAny.unannotated)
        else:
            proper_type = get_proper_type(init_type)
            if isinstance(proper_type, AnyType):
                if proper_type.type_of_any == TypeOfAny.unannotated:
                    unannotated = True

        if unannotated and ctx.api.options.disallow_untyped_defs:
            # This is a compromise.  If you don't have a type here then the
            # __init__ will be untyped. But since the __init__ is added it's
            # pointing at the decorator. So instead we also show the error in the
            # assignment, which is where you would fix the issue.
            node = self.info[self.name].node
            assert node is not None
            ctx.api.msg.need_annotation_for_var(node, self.context)

        if self.kw_only:
            arg_kind = ARG_NAMED_OPT if self.has_default else ARG_NAMED
        else:
            arg_kind = ARG_OPT if self.has_default else ARG_POS

        # Attrs removes leading underscores when creating the __init__ arguments.
        name = self.alias or self.name.lstrip("_")
        return Argument(Var(name, init_type), init_type, None, arg_kind)

    def serialize(self) -> JsonDict:
        """Serialize this object so it can be saved and restored."""
        return {
            "name": self.name,
            "alias": self.alias,
            "has_default": self.has_default,
            "init": self.init,
            "kw_only": self.kw_only,
            "has_converter": self.converter is not None,
            "converter_init_type": self.converter.init_type.serialize()
            if self.converter and self.converter.init_type
            else None,
            "context_line": self.context.line,
            "context_column": self.context.column,
            "init_type": self.init_type.serialize() if self.init_type else None,
        }

    @classmethod
    def deserialize(
        cls, info: TypeInfo, data: JsonDict, api: SemanticAnalyzerPluginInterface
    ) -> Attribute:
        """Return the Attribute that was serialized."""
        raw_init_type = data["init_type"]
        init_type = deserialize_and_fixup_type(raw_init_type, api) if raw_init_type else None
        raw_converter_init_type = data["converter_init_type"]
        converter_init_type = (
            deserialize_and_fixup_type(raw_converter_init_type, api)
            if raw_converter_init_type
            else None
        )

        return Attribute(
            data["name"],
            data["alias"],
            info,
            data["has_default"],
            data["init"],
            data["kw_only"],
            Converter(converter_init_type) if data["has_converter"] else None,
            Context(line=data["context_line"], column=data["context_column"]),
            init_type,
        )

    def expand_typevar_from_subtype(self, sub_type: TypeInfo) -> None:
        """Expands type vars in the context of a subtype when an attribute is inherited
        from a generic super type."""
        if self.init_type:
            self.init_type = map_type_from_supertype(self.init_type, sub_type, self.info)
        else:
            self.init_type = None


def _determine_eq_order(ctx: mypy.plugin.ClassDefContext) -> bool:
    """
    Validate the combination of *cmp*, *eq*, and *order*. Derive the effective
    value of order.
    """
    cmp = _get_decorator_optional_bool_argument(ctx, "cmp")
    eq = _get_decorator_optional_bool_argument(ctx, "eq")
    order = _get_decorator_optional_bool_argument(ctx, "order")

    if cmp is not None and any((eq is not None, order is not None)):
        ctx.api.fail('Don\'t mix "cmp" with "eq" and "order"', ctx.reason)

    # cmp takes precedence due to bw-compatibility.
    if cmp is not None:
        return cmp

    # If left None, equality is on and ordering mirrors equality.
    if eq is None:
        eq = True

    if order is None:
        order = eq

    if eq is False and order is True:
        ctx.api.fail("eq must be True if order is True", ctx.reason)

    return order


def _get_decorator_optional_bool_argument(
    ctx: mypy.plugin.ClassDefContext, name: str, default: bool | None = None
) -> bool | None:
    """Return the Optional[bool] argument for the decorator.

    This handles both @decorator(...) and @decorator.
    """
    if isinstance(ctx.reason, CallExpr):
        attr_value = _get_argument(ctx.reason, name)
        if attr_value:
            if isinstance(attr_value, NameExpr):
                if attr_value.fullname == "builtins.True":
                    return True
                if attr_value.fullname == "builtins.False":
                    return False
                if attr_value.fullname == "builtins.None":
                    return None
            ctx.api.fail(
                f'"{name}" argument must be a True, False, or None literal',
                ctx.reason,
                code=LITERAL_REQ,
            )
            return default
        return default
    else:
        return default


def attr_tag_callback(ctx: mypy.plugin.ClassDefContext) -> None:
    """Record that we have an attrs class in the main semantic analysis pass.

    The later pass implemented by attr_class_maker_callback will use this
    to detect attrs classes in base classes.
    """
    # The value is ignored, only the existence matters.
    ctx.cls.info.metadata["attrs_tag"] = {}


def attr_class_maker_callback(
    ctx: mypy.plugin.ClassDefContext,
    auto_attribs_default: bool | None = False,
    frozen_default: bool = False,
    slots_default: bool = False,
) -> bool:
    """Add necessary dunder methods to classes decorated with attr.s.

    attrs is a package that lets you define classes without writing dull boilerplate code.

    At a quick glance, the decorator searches the class body for assignments of `attr.ib`s (or
    annotated variables if auto_attribs=True), then depending on how the decorator is called,
    it will add an __init__ or all the compare methods.
    For frozen=True it will turn the attrs into properties.

    See https://www.attrs.org/en/stable/how-does-it-work.html for information on how attrs works.

    If this returns False, some required metadata was not ready yet and we need another
    pass.
    """
    info = ctx.cls.info

    init = _get_decorator_bool_argument(ctx, "init", True)
    frozen = _get_frozen(ctx, frozen_default)
    order = _determine_eq_order(ctx)
    slots = _get_decorator_bool_argument(ctx, "slots", slots_default)

    auto_attribs = _get_decorator_optional_bool_argument(ctx, "auto_attribs", auto_attribs_default)
    kw_only = _get_decorator_bool_argument(ctx, "kw_only", False)
    match_args = _get_decorator_bool_argument(ctx, "match_args", True)

    for super_info in ctx.cls.info.mro[1:-1]:
        if "attrs_tag" in super_info.metadata and "attrs" not in super_info.metadata:
            # Super class is not ready yet. Request another pass.
            return False

    attributes = _analyze_class(ctx, auto_attribs, kw_only)

    # Check if attribute types are ready.
    for attr in attributes:
        node = info.get(attr.name)
        if node is None:
            # This name is likely blocked by some semantic analysis error that
            # should have been reported already.
            _add_empty_metadata(info)
            return True

    _add_attrs_magic_attribute(ctx, [(attr.name, info[attr.name].type) for attr in attributes])
    if slots:
        _add_slots(ctx, attributes)
    if match_args and ctx.api.options.python_version[:2] >= (3, 10):
        # `.__match_args__` is only added for python3.10+, but the argument
        # exists for earlier versions as well.
        _add_match_args(ctx, attributes)

    # Save the attributes so that subclasses can reuse them.
    ctx.cls.info.metadata["attrs"] = {
        "attributes": [attr.serialize() for attr in attributes],
        "frozen": frozen,
    }

    adder = MethodAdder(ctx)
    # If  __init__ is not being generated, attrs still generates it as __attrs_init__ instead.
    _add_init(ctx, attributes, adder, "__init__" if init else ATTRS_INIT_NAME)
    if order:
        _add_order(ctx, adder)
    if frozen:
        _make_frozen(ctx, attributes)

    return True


def _get_frozen(ctx: mypy.plugin.ClassDefContext, frozen_default: bool) -> bool:
    """Return whether this class is frozen."""
    if _get_decorator_bool_argument(ctx, "frozen", frozen_default):
        return True
    # Subclasses of frozen classes are frozen so check that.
    for super_info in ctx.cls.info.mro[1:-1]:
        if "attrs" in super_info.metadata and super_info.metadata["attrs"]["frozen"]:
            return True
    return False


def _analyze_class(
    ctx: mypy.plugin.ClassDefContext, auto_attribs: bool | None, kw_only: bool
) -> list[Attribute]:
    """Analyze the class body of an attr maker, its parents, and return the Attributes found.

    auto_attribs=True means we'll generate attributes from type annotations also.
    auto_attribs=None means we'll detect which mode to use.
    kw_only=True means that all attributes created here will be keyword only args in __init__.
    """
    own_attrs: dict[str, Attribute] = {}
    if auto_attribs is None:
        auto_attribs = _detect_auto_attribs(ctx)

    # Walk the body looking for assignments and decorators.
    for stmt in ctx.cls.defs.body:
        if isinstance(stmt, AssignmentStmt):
            for attr in _attributes_from_assignment(ctx, stmt, auto_attribs, kw_only):
                # When attrs are defined twice in the same body we want to use the 2nd definition
                # in the 2nd location. So remove it from the OrderedDict.
                # Unless it's auto_attribs in which case we want the 2nd definition in the
                # 1st location.
                if not auto_attribs and attr.name in own_attrs:
                    del own_attrs[attr.name]
                own_attrs[attr.name] = attr
        elif isinstance(stmt, Decorator):
            _cleanup_decorator(stmt, own_attrs)

    for attribute in own_attrs.values():
        # Even though these look like class level assignments we want them to look like
        # instance level assignments.
        if attribute.name in ctx.cls.info.names:
            node = ctx.cls.info.names[attribute.name].node
            if isinstance(node, PlaceholderNode):
                # This node is not ready yet.
                continue
            assert isinstance(node, Var)
            node.is_initialized_in_class = False

    # Traverse the MRO and collect attributes from the parents.
    taken_attr_names = set(own_attrs)
    super_attrs = []
    for super_info in ctx.cls.info.mro[1:-1]:
        if "attrs" in super_info.metadata:
            # Each class depends on the set of attributes in its attrs ancestors.
            ctx.api.add_plugin_dependency(make_wildcard_trigger(super_info.fullname))

            for data in super_info.metadata["attrs"]["attributes"]:
                # Only add an attribute if it hasn't been defined before.  This
                # allows for overwriting attribute definitions by subclassing.
                if data["name"] not in taken_attr_names:
                    a = Attribute.deserialize(super_info, data, ctx.api)
                    a.expand_typevar_from_subtype(ctx.cls.info)
                    super_attrs.append(a)
                    taken_attr_names.add(a.name)
    attributes = super_attrs + list(own_attrs.values())

    # Check the init args for correct default-ness.  Note: This has to be done after all the
    # attributes for all classes have been read, because subclasses can override parents.
    last_default = False

    for i, attribute in enumerate(attributes):
        if not attribute.init:
            continue

        if attribute.kw_only:
            # Keyword-only attributes don't care whether they are default or not.
            continue

        # If the issue comes from merging different classes, report it
        # at the class definition point.
        context = attribute.context if i >= len(super_attrs) else ctx.cls

        if not attribute.has_default and last_default:
            ctx.api.fail("Non-default attributes not allowed after default attributes.", context)
        last_default |= attribute.has_default

    return attributes


def _add_empty_metadata(info: TypeInfo) -> None:
    """Add empty metadata to mark that we've finished processing this class."""
    info.metadata["attrs"] = {"attributes": [], "frozen": False}


def _detect_auto_attribs(ctx: mypy.plugin.ClassDefContext) -> bool:
    """Return whether auto_attribs should be enabled or disabled.

    It's disabled if there are any unannotated attribs()
    """
    for stmt in ctx.cls.defs.body:
        if isinstance(stmt, AssignmentStmt):
            for lvalue in stmt.lvalues:
                lvalues, rvalues = _parse_assignments(lvalue, stmt)

                if len(lvalues) != len(rvalues):
                    # This means we have some assignment that isn't 1 to 1.
                    # It can't be an attrib.
                    continue

                for lhs, rvalue in zip(lvalues, rvalues):
                    # Check if the right hand side is a call to an attribute maker.
                    if (
                        isinstance(rvalue, CallExpr)
                        and isinstance(rvalue.callee, RefExpr)
                        and rvalue.callee.fullname in attr_attrib_makers
                        and not stmt.new_syntax
                    ):
                        # This means we have an attrib without an annotation and so
                        # we can't do auto_attribs=True
                        return False
    return True


def _attributes_from_assignment(
    ctx: mypy.plugin.ClassDefContext, stmt: AssignmentStmt, auto_attribs: bool, kw_only: bool
) -> Iterable[Attribute]:
    """Return Attribute objects that are created by this assignment.

    The assignments can look like this:
        x = attr.ib()
        x = y = attr.ib()
        x, y = attr.ib(), attr.ib()
    or if auto_attribs is enabled also like this:
        x: type
        x: type = default_value
        x: type = attr.ib(...)
    """
    for lvalue in stmt.lvalues:
        lvalues, rvalues = _parse_assignments(lvalue, stmt)

        if len(lvalues) != len(rvalues):
            # This means we have some assignment that isn't 1 to 1.
            # It can't be an attrib.
            continue

        for lhs, rvalue in zip(lvalues, rvalues):
            # Check if the right hand side is a call to an attribute maker.
            if (
                isinstance(rvalue, CallExpr)
                and isinstance(rvalue.callee, RefExpr)
                and rvalue.callee.fullname in attr_attrib_makers
            ):
                attr = _attribute_from_attrib_maker(ctx, auto_attribs, kw_only, lhs, rvalue, stmt)
                if attr:
                    yield attr
            elif auto_attribs and stmt.type and stmt.new_syntax and not is_class_var(lhs):
                yield _attribute_from_auto_attrib(ctx, kw_only, lhs, rvalue, stmt)


def _cleanup_decorator(stmt: Decorator, attr_map: dict[str, Attribute]) -> None:
    """Handle decorators in class bodies.

    `x.default` will set a default value on x
    `x.validator` and `x.default` will get removed to avoid throwing a type error.
    """
    remove_me = []
    for func_decorator in stmt.decorators:
        if (
            isinstance(func_decorator, MemberExpr)
            and isinstance(func_decorator.expr, NameExpr)
            and func_decorator.expr.name in attr_map
        ):
            if func_decorator.name == "default":
                attr_map[func_decorator.expr.name].has_default = True

            if func_decorator.name in ("default", "validator"):
                # These are decorators on the attrib object that only exist during
                # class creation time.  In order to not trigger a type error later we
                # just remove them.  This might leave us with a Decorator with no
                # decorators (Emperor's new clothes?)
                # TODO: It would be nice to type-check these rather than remove them.
                #       default should be Callable[[], T]
                #       validator should be Callable[[Any, 'Attribute', T], Any]
                #       where T is the type of the attribute.
                remove_me.append(func_decorator)
    for dec in remove_me:
        stmt.decorators.remove(dec)


def _attribute_from_auto_attrib(
    ctx: mypy.plugin.ClassDefContext,
    kw_only: bool,
    lhs: NameExpr,
    rvalue: Expression,
    stmt: AssignmentStmt,
) -> Attribute:
    """Return an Attribute for a new type assignment."""
    name = unmangle(lhs.name)
    # `x: int` (without equal sign) assigns rvalue to TempNode(AnyType())
    has_rhs = not isinstance(rvalue, TempNode)
    sym = ctx.cls.info.names.get(name)
    init_type = sym.type if sym else None
    return Attribute(name, None, ctx.cls.info, has_rhs, True, kw_only, None, stmt, init_type)


def _attribute_from_attrib_maker(
    ctx: mypy.plugin.ClassDefContext,
    auto_attribs: bool,
    kw_only: bool,
    lhs: NameExpr,
    rvalue: CallExpr,
    stmt: AssignmentStmt,
) -> Attribute | None:
    """Return an Attribute from the assignment or None if you can't make one."""
    if auto_attribs and not stmt.new_syntax:
        # auto_attribs requires an annotation on *every* attr.ib.
        assert lhs.node is not None
        ctx.api.msg.need_annotation_for_var(lhs.node, stmt)
        return None

    if len(stmt.lvalues) > 1:
        ctx.api.fail("Too many names for one attribute", stmt)
        return None

    # This is the type that belongs in the __init__ method for this attrib.
    init_type = stmt.type

    # Read all the arguments from the call.
    init = _get_bool_argument(ctx, rvalue, "init", True)
    # Note: If the class decorator says kw_only=True the attribute is ignored.
    # See https://github.com/python-attrs/attrs/issues/481 for explanation.
    kw_only |= _get_bool_argument(ctx, rvalue, "kw_only", False)

    # TODO: Check for attr.NOTHING
    attr_has_default = bool(_get_argument(rvalue, "default"))
    attr_has_factory = bool(_get_argument(rvalue, "factory"))

    if attr_has_default and attr_has_factory:
        ctx.api.fail('Can\'t pass both "default" and "factory".', rvalue)
    elif attr_has_factory:
        attr_has_default = True

    # If the type isn't set through annotation but is passed through `type=` use that.
    type_arg = _get_argument(rvalue, "type")
    if type_arg and not init_type:
        try:
            un_type = expr_to_unanalyzed_type(type_arg, ctx.api.options, ctx.api.is_stub_file)
        except TypeTranslationError:
            ctx.api.fail("Invalid argument to type", type_arg)
        else:
            init_type = ctx.api.anal_type(un_type)
            if init_type and isinstance(lhs.node, Var) and not lhs.node.type:
                # If there is no annotation, add one.
                lhs.node.type = init_type
                lhs.is_inferred_def = False

    # Note: convert is deprecated but works the same as converter.
    converter = _get_argument(rvalue, "converter")
    convert = _get_argument(rvalue, "convert")
    if convert and converter:
        ctx.api.fail('Can\'t pass both "convert" and "converter".', rvalue)
    elif convert:
        ctx.api.fail("convert is deprecated, use converter", rvalue)
        converter = convert
    converter_info = _parse_converter(ctx, converter)

    # Custom alias might be defined:
    alias = None
    alias_expr = _get_argument(rvalue, "alias")
    if alias_expr:
        alias = ctx.api.parse_str_literal(alias_expr)
        if alias is None:
            ctx.api.fail(
                '"alias" argument to attrs field must be a string literal',
                rvalue,
                code=LITERAL_REQ,
            )
    name = unmangle(lhs.name)
    return Attribute(
        name, alias, ctx.cls.info, attr_has_default, init, kw_only, converter_info, stmt, init_type
    )


def _parse_converter(
    ctx: mypy.plugin.ClassDefContext, converter_expr: Expression | None
) -> Converter | None:
    """Return the Converter object from an Expression."""
    # TODO: Support complex converters, e.g. lambdas, calls, etc.
    if not converter_expr:
        return None
    converter_info = Converter()
    if (
        isinstance(converter_expr, CallExpr)
        and isinstance(converter_expr.callee, RefExpr)
        and converter_expr.callee.fullname in attr_optional_converters
        and converter_expr.args
        and converter_expr.args[0]
    ):
        # Special handling for attr.converters.optional(type)
        # We extract the type and add make the init_args Optional in Attribute.argument
        converter_expr = converter_expr.args[0]
        is_attr_converters_optional = True
    else:
        is_attr_converters_optional = False

    converter_type: Type | None = None
    if isinstance(converter_expr, RefExpr) and converter_expr.node:
        if isinstance(converter_expr.node, FuncDef):
            if converter_expr.node.type and isinstance(converter_expr.node.type, FunctionLike):
                converter_type = converter_expr.node.type
            else:  # The converter is an unannotated function.
                converter_info.init_type = AnyType(TypeOfAny.unannotated)
                return converter_info
        elif isinstance(converter_expr.node, OverloadedFuncDef) and is_valid_overloaded_converter(
            converter_expr.node
        ):
            converter_type = converter_expr.node.type
        elif isinstance(converter_expr.node, TypeInfo):
            from mypy.checkmember import type_object_type  # To avoid import cycle.

            converter_type = type_object_type(converter_expr.node, ctx.api.named_type)
    elif (
        isinstance(converter_expr, IndexExpr)
        and isinstance(converter_expr.analyzed, TypeApplication)
        and isinstance(converter_expr.base, RefExpr)
        and isinstance(converter_expr.base.node, TypeInfo)
    ):
        # The converter is a generic type.
        from mypy.checkmember import type_object_type  # To avoid import cycle.

        converter_type = type_object_type(converter_expr.base.node, ctx.api.named_type)
        if isinstance(converter_type, CallableType):
            converter_type = apply_generic_arguments(
                converter_type,
                converter_expr.analyzed.types,
                ctx.api.msg.incompatible_typevar_value,
                converter_type,
            )
        else:
            converter_type = None

    if isinstance(converter_expr, LambdaExpr):
        # TODO: should we send a fail if converter_expr.min_args > 1?
        converter_info.init_type = AnyType(TypeOfAny.unannotated)
        return converter_info

    if not converter_type:
        # Signal that we have an unsupported converter.
        ctx.api.fail(
            "Unsupported converter, only named functions, types and lambdas are currently "
            "supported",
            converter_expr,
        )
        converter_info.init_type = AnyType(TypeOfAny.from_error)
        return converter_info

    converter_type = get_proper_type(converter_type)
    if isinstance(converter_type, CallableType) and converter_type.arg_types:
        converter_info.init_type = converter_type.arg_types[0]
        if not is_attr_converters_optional:
            converter_info.ret_type = converter_type.ret_type
    elif isinstance(converter_type, Overloaded):
        types: list[Type] = []
        for item in converter_type.items:
            # Walk the overloads looking for methods that can accept one argument.
            num_arg_types = len(item.arg_types)
            if not num_arg_types:
                continue
            if num_arg_types > 1 and any(kind == ARG_POS for kind in item.arg_kinds[1:]):
                continue
            types.append(item.arg_types[0])
        # Make a union of all the valid types.
        if types:
            converter_info.init_type = make_simplified_union(types)

    if is_attr_converters_optional and converter_info.init_type:
        # If the converter was attr.converter.optional(type) then add None to
        # the allowed init_type.
        converter_info.init_type = UnionType.make_union([converter_info.init_type, NoneType()])

    return converter_info


def is_valid_overloaded_converter(defn: OverloadedFuncDef) -> bool:
    return all(
        (not isinstance(item, Decorator) or isinstance(item.func.type, FunctionLike))
        for item in defn.items
    )


def _parse_assignments(
    lvalue: Expression, stmt: AssignmentStmt
) -> tuple[list[NameExpr], list[Expression]]:
    """Convert a possibly complex assignment expression into lists of lvalues and rvalues."""
    lvalues: list[NameExpr] = []
    rvalues: list[Expression] = []
    if isinstance(lvalue, (TupleExpr, ListExpr)):
        if all(isinstance(item, NameExpr) for item in lvalue.items):
            lvalues = cast(List[NameExpr], lvalue.items)
        if isinstance(stmt.rvalue, (TupleExpr, ListExpr)):
            rvalues = stmt.rvalue.items
    elif isinstance(lvalue, NameExpr):
        lvalues = [lvalue]
        rvalues = [stmt.rvalue]
    return lvalues, rvalues


def _add_order(ctx: mypy.plugin.ClassDefContext, adder: MethodAdder) -> None:
    """Generate all the ordering methods for this class."""
    bool_type = ctx.api.named_type("builtins.bool")
    object_type = ctx.api.named_type("builtins.object")
    # Make the types be:
    #    AT = TypeVar('AT')
    #    def __lt__(self: AT, other: AT) -> bool
    # This way comparisons with subclasses will work correctly.
    tvd = TypeVarType(
        SELF_TVAR_NAME,
        ctx.cls.info.fullname + "." + SELF_TVAR_NAME,
        id=-1,
        values=[],
        upper_bound=object_type,
        default=AnyType(TypeOfAny.from_omitted_generics),
    )
    self_tvar_expr = TypeVarExpr(
        SELF_TVAR_NAME,
        ctx.cls.info.fullname + "." + SELF_TVAR_NAME,
        [],
        object_type,
        AnyType(TypeOfAny.from_omitted_generics),
    )
    ctx.cls.info.names[SELF_TVAR_NAME] = SymbolTableNode(MDEF, self_tvar_expr)

    args = [Argument(Var("other", tvd), tvd, None, ARG_POS)]
    for method in ["__lt__", "__le__", "__gt__", "__ge__"]:
        adder.add_method(method, args, bool_type, self_type=tvd, tvd=tvd)


def _make_frozen(ctx: mypy.plugin.ClassDefContext, attributes: list[Attribute]) -> None:
    """Turn all the attributes into properties to simulate frozen classes."""
    for attribute in attributes:
        if attribute.name in ctx.cls.info.names:
            # This variable belongs to this class so we can modify it.
            node = ctx.cls.info.names[attribute.name].node
            if not isinstance(node, Var):
                # The superclass attribute was overridden with a non-variable.
                # No need to do anything here, override will be verified during
                # type checking.
                continue
            node.is_property = True
        else:
            # This variable belongs to a super class so create new Var so we
            # can modify it.
            var = Var(attribute.name, attribute.init_type)
            var.info = ctx.cls.info
            var._fullname = f"{ctx.cls.info.fullname}.{var.name}"
            ctx.cls.info.names[var.name] = SymbolTableNode(MDEF, var)
            var.is_property = True


def _add_init(
    ctx: mypy.plugin.ClassDefContext,
    attributes: list[Attribute],
    adder: MethodAdder,
    method_name: Literal["__init__", "__attrs_init__"],
) -> None:
    """Generate an __init__ method for the attributes and add it to the class."""
    # Convert attributes to arguments with kw_only arguments at the end of
    # the argument list
    pos_args = []
    kw_only_args = []
    sym_table = ctx.cls.info.names
    for attribute in attributes:
        if not attribute.init:
            continue
        if attribute.kw_only:
            kw_only_args.append(attribute.argument(ctx))
        else:
            pos_args.append(attribute.argument(ctx))

        # If the attribute is Final, present in `__init__` and has
        # no default, make sure it doesn't error later.
        if not attribute.has_default and attribute.name in sym_table:
            sym_node = sym_table[attribute.name].node
            if isinstance(sym_node, Var) and sym_node.is_final:
                sym_node.final_set_in_init = True
    args = pos_args + kw_only_args
    if all(
        # We use getattr rather than instance checks because the variable.type
        # might be wrapped into a Union or some other type, but even non-Any
        # types reliably track the fact that the argument was not annotated.
        getattr(arg.variable.type, "type_of_any", None) == TypeOfAny.unannotated
        for arg in args
    ):
        # This workaround makes --disallow-incomplete-defs usable with attrs,
        # but is definitely suboptimal as a long-term solution.
        # See https://github.com/python/mypy/issues/5954 for discussion.
        for a in args:
            a.variable.type = AnyType(TypeOfAny.implementation_artifact)
            a.type_annotation = AnyType(TypeOfAny.implementation_artifact)
    adder.add_method(method_name, args, NoneType())


def _add_attrs_magic_attribute(
    ctx: mypy.plugin.ClassDefContext, attrs: list[tuple[str, Type | None]]
) -> None:
    any_type = AnyType(TypeOfAny.explicit)
    attributes_types: list[Type] = [
        ctx.api.named_type_or_none("attr.Attribute", [attr_type or any_type]) or any_type
        for _, attr_type in attrs
    ]
    fallback_type = ctx.api.named_type(
        "builtins.tuple", [ctx.api.named_type_or_none("attr.Attribute", [any_type]) or any_type]
    )

    attr_name = MAGIC_ATTR_CLS_NAME_TEMPLATE.format(ctx.cls.fullname.replace(".", "_"))
    ti = ctx.api.basic_new_typeinfo(attr_name, fallback_type, 0)
    for (name, _), attr_type in zip(attrs, attributes_types):
        var = Var(name, attr_type)
        var._fullname = name
        var.is_property = True
        proper_type = get_proper_type(attr_type)
        if isinstance(proper_type, Instance):
            var.info = proper_type.type
        ti.names[name] = SymbolTableNode(MDEF, var, plugin_generated=True)
    attributes_type = Instance(ti, [])

    # We need to stash the type of the magic attribute so it can be
    # loaded on cached runs.
    ctx.cls.info.names[attr_name] = SymbolTableNode(MDEF, ti, plugin_generated=True)

    add_attribute_to_class(
        ctx.api,
        ctx.cls,
        MAGIC_ATTR_NAME,
        TupleType(attributes_types, fallback=attributes_type),
        fullname=f"{ctx.cls.fullname}.{MAGIC_ATTR_NAME}",
        override_allow_incompatible=True,
        is_classvar=True,
    )


def _add_slots(ctx: mypy.plugin.ClassDefContext, attributes: list[Attribute]) -> None:
    if any(p.slots is None for p in ctx.cls.info.mro[1:-1]):
        # At least one type in mro (excluding `self` and `object`)
        # does not have concrete `__slots__` defined. Ignoring.
        return

    # Unlike `@dataclasses.dataclass`, `__slots__` is rewritten here.
    ctx.cls.info.slots = {attr.name for attr in attributes}

    # Also, inject `__slots__` attribute to class namespace:
    slots_type = TupleType(
        [ctx.api.named_type("builtins.str") for _ in attributes],
        fallback=ctx.api.named_type("builtins.tuple"),
    )
    add_attribute_to_class(api=ctx.api, cls=ctx.cls, name="__slots__", typ=slots_type)


def _add_match_args(ctx: mypy.plugin.ClassDefContext, attributes: list[Attribute]) -> None:
    if (
        "__match_args__" not in ctx.cls.info.names
        or ctx.cls.info.names["__match_args__"].plugin_generated
    ):
        str_type = ctx.api.named_type("builtins.str")
        match_args = TupleType(
            [
                str_type.copy_modified(last_known_value=LiteralType(attr.name, fallback=str_type))
                for attr in attributes
                if not attr.kw_only and attr.init
            ],
            fallback=ctx.api.named_type("builtins.tuple"),
        )
        add_attribute_to_class(api=ctx.api, cls=ctx.cls, name="__match_args__", typ=match_args)


class MethodAdder:
    """Helper to add methods to a TypeInfo.

    ctx: The ClassDefCtx we are using on which we will add methods.
    """

    # TODO: Combine this with the code build_namedtuple_typeinfo to support both.

    def __init__(self, ctx: mypy.plugin.ClassDefContext) -> None:
        self.ctx = ctx
        self.self_type = fill_typevars(ctx.cls.info)

    def add_method(
        self,
        method_name: str,
        args: list[Argument],
        ret_type: Type,
        self_type: Type | None = None,
        tvd: TypeVarType | None = None,
    ) -> None:
        """Add a method: def <method_name>(self, <args>) -> <ret_type>): ... to info.

        self_type: The type to use for the self argument or None to use the inferred self type.
        tvd: If the method is generic these should be the type variables.
        """
        self_type = self_type if self_type is not None else self.self_type
        add_method_to_class(
            self.ctx.api, self.ctx.cls, method_name, args, ret_type, self_type, tvd
        )


def _get_attrs_init_type(typ: Instance) -> CallableType | None:
    """
    If `typ` refers to an attrs class, get the type of its initializer method.
    """
    magic_attr = typ.type.get(MAGIC_ATTR_NAME)
    if magic_attr is None or not magic_attr.plugin_generated:
        return None
    init_method = typ.type.get_method("__init__") or typ.type.get_method(ATTRS_INIT_NAME)
    if not isinstance(init_method, FuncDef) or not isinstance(init_method.type, CallableType):
        return None
    return init_method.type


def _fail_not_attrs_class(ctx: mypy.plugin.FunctionSigContext, t: Type, parent_t: Type) -> None:
    t_name = format_type_bare(t, ctx.api.options)
    if parent_t is t:
        msg = (
            f'Argument 1 to "evolve" has a variable type "{t_name}" not bound to an attrs class'
            if isinstance(t, TypeVarType)
            else f'Argument 1 to "evolve" has incompatible type "{t_name}"; expected an attrs class'
        )
    else:
        pt_name = format_type_bare(parent_t, ctx.api.options)
        msg = (
            f'Argument 1 to "evolve" has type "{pt_name}" whose item "{t_name}" is not bound to an attrs class'
            if isinstance(t, TypeVarType)
            else f'Argument 1 to "evolve" has incompatible type "{pt_name}" whose item "{t_name}" is not an attrs class'
        )

    ctx.api.fail(msg, ctx.context)


def _get_expanded_attr_types(
    ctx: mypy.plugin.FunctionSigContext,
    typ: ProperType,
    display_typ: ProperType,
    parent_typ: ProperType,
) -> list[Mapping[str, Type]] | None:
    """
    For a given type, determine what attrs classes it can be: for each class, return the field types.
    For generic classes, the field types are expanded.
    If the type contains Any or a non-attrs type, returns None; in the latter case, also reports an error.
    """
    if isinstance(typ, AnyType):
        return None
    elif isinstance(typ, UnionType):
        ret: list[Mapping[str, Type]] | None = []
        for item in typ.relevant_items():
            item = get_proper_type(item)
            item_types = _get_expanded_attr_types(ctx, item, item, parent_typ)
            if ret is not None and item_types is not None:
                ret += item_types
            else:
                ret = None  # but keep iterating to emit all errors
        return ret
    elif isinstance(typ, TypeVarType):
        return _get_expanded_attr_types(
            ctx, get_proper_type(typ.upper_bound), display_typ, parent_typ
        )
    elif isinstance(typ, Instance):
        init_func = _get_attrs_init_type(typ)
        if init_func is None:
            _fail_not_attrs_class(ctx, display_typ, parent_typ)
            return None
        init_func = expand_type_by_instance(init_func, typ)
        # [1:] to skip the self argument of AttrClass.__init__
        field_names = cast(List[str], init_func.arg_names[1:])
        field_types = init_func.arg_types[1:]
        return [dict(zip(field_names, field_types))]
    else:
        _fail_not_attrs_class(ctx, display_typ, parent_typ)
        return None


def _meet_fields(types: list[Mapping[str, Type]]) -> Mapping[str, Type]:
    """
    "Meet" the fields of a list of attrs classes, i.e. for each field, its new type will be the lower bound.
    """
    field_to_types = defaultdict(list)
    for fields in types:
        for name, typ in fields.items():
            field_to_types[name].append(typ)

    return {
        name: get_proper_type(reduce(meet_types, f_types))
        if len(f_types) == len(types)
        else UninhabitedType()
        for name, f_types in field_to_types.items()
    }


def evolve_function_sig_callback(ctx: mypy.plugin.FunctionSigContext) -> CallableType:
    """
    Generate a signature for the 'attr.evolve' function that's specific to the call site
    and dependent on the type of the first argument.
    """
    if len(ctx.args) != 2:
        # Ideally the name and context should be callee's, but we don't have it in FunctionSigContext.
        ctx.api.fail(f'"{ctx.default_signature.name}" has unexpected type annotation', ctx.context)
        return ctx.default_signature

    if len(ctx.args[0]) != 1:
        return ctx.default_signature  # leave it to the type checker to complain

    inst_arg = ctx.args[0][0]
    inst_type = get_proper_type(ctx.api.get_expression_type(inst_arg))
    inst_type_str = format_type_bare(inst_type, ctx.api.options)

    attr_types = _get_expanded_attr_types(ctx, inst_type, inst_type, inst_type)
    if attr_types is None:
        return ctx.default_signature
    fields = _meet_fields(attr_types)

    return CallableType(
        arg_names=["inst", *fields.keys()],
        arg_kinds=[ARG_POS] + [ARG_NAMED_OPT] * len(fields),
        arg_types=[inst_type, *fields.values()],
        ret_type=inst_type,
        fallback=ctx.default_signature.fallback,
        name=f"{ctx.default_signature.name} of {inst_type_str}",
    )


def fields_function_sig_callback(ctx: mypy.plugin.FunctionSigContext) -> CallableType:
    """Provide the signature for `attrs.fields`."""
    if len(ctx.args) != 1 or len(ctx.args[0]) != 1:
        return ctx.default_signature

    proper_type = get_proper_type(ctx.api.get_expression_type(ctx.args[0][0]))

    # fields(Any) -> Any, fields(type[Any]) -> Any
    if (
        isinstance(proper_type, AnyType)
        or isinstance(proper_type, TypeType)
        and isinstance(proper_type.item, AnyType)
    ):
        return ctx.default_signature

    cls = None
    arg_types = ctx.default_signature.arg_types

    if isinstance(proper_type, TypeVarType):
        inner = get_proper_type(proper_type.upper_bound)
        if isinstance(inner, Instance):
            # We need to work arg_types to compensate for the attrs stubs.
            arg_types = [proper_type]
            cls = inner.type
    elif isinstance(proper_type, CallableType):
        cls = proper_type.type_object()

    if cls is not None and MAGIC_ATTR_NAME in cls.names:
        # This is a proper attrs class.
        ret_type = cls.names[MAGIC_ATTR_NAME].type
        assert ret_type is not None
        return ctx.default_signature.copy_modified(arg_types=arg_types, ret_type=ret_type)

    return ctx.default_signature