ruff/crates/ty_python_semantic/resources/mdtest/protocols.md

Protocols

Note

See also:

• The typing specification section on protocols
• The many protocol conformance tests provided by the Typing Council for type checkers
• Mypy's documentation and tests for protocols

Most types in Python are nominal types: a fully static nominal type X is only a subtype of another fully static nominal type Y if the class X is a subclass of the class Y. typing.Protocol (or its backport, typing_extensions.Protocol) can be used to define structural types, on the other hand: a type which is defined by its properties and behaviour.

Defining a protocol

[environment]
python-version = "3.12"

A protocol is defined by inheriting from the Protocol class, which is annotated as an instance of _SpecialForm in typeshed's stubs.

from typing import Protocol

class MyProtocol(Protocol): ...

reveal_type(MyProtocol.__mro__)  # revealed: tuple[<class 'MyProtocol'>, typing.Protocol, typing.Generic, <class 'object'>]

Just like for any other class base, it is an error for Protocol to appear multiple times in a class's bases:

class Foo(Protocol, Protocol): ...  # error: [duplicate-base]

reveal_type(Foo.__mro__)  # revealed: tuple[<class 'Foo'>, Unknown, <class 'object'>]

Protocols can also be generic, either by including Generic[] in the bases list, subscripting Protocol directly in the bases list, using PEP-695 type parameters, or some combination of the above:

from typing import TypeVar, Generic

T = TypeVar("T")

class Bar0(Protocol[T]):
    x: T

class Bar1(Protocol[T], Generic[T]):
    x: T

class Bar2[T](Protocol):
    x: T

# error: [invalid-generic-class] "Cannot both inherit from subscripted `typing.Protocol` and use PEP 695 type variables"
class Bar3[T](Protocol[T]):
    x: T

It's an error to include both bare Protocol and subscripted Protocol[] in the bases list simultaneously:

class DuplicateBases(Protocol, Protocol[T]):  # error: [duplicate-base]
    x: T

# revealed: tuple[<class 'DuplicateBases[Unknown]'>, Unknown, <class 'object'>]
reveal_type(DuplicateBases.__mro__)

The introspection helper typing(_extensions).is_protocol can be used to verify whether a class is a protocol class or not:

from typing_extensions import is_protocol

reveal_type(is_protocol(MyProtocol))  # revealed: Literal[True]
reveal_type(is_protocol(Bar0))  # revealed: Literal[True]
reveal_type(is_protocol(Bar1))  # revealed: Literal[True]
reveal_type(is_protocol(Bar2))  # revealed: Literal[True]
reveal_type(is_protocol(Bar3))  # revealed: Literal[True]

class NotAProtocol: ...

reveal_type(is_protocol(NotAProtocol))  # revealed: Literal[False]

A type checker should follow the typeshed stubs if a non-class is passed in, and typeshed's stubs indicate that the argument passed in must be an instance of type.

# We could also reasonably infer `Literal[False]` here, but it probably doesn't matter that much:
# error: [invalid-argument-type]
reveal_type(is_protocol("not a class"))  # revealed: bool

For a class to be considered a protocol class, it must have Protocol directly in its bases tuple: it is not sufficient for it to have Protocol in its MRO.

class SubclassOfMyProtocol(MyProtocol): ...

# revealed: tuple[<class 'SubclassOfMyProtocol'>, <class 'MyProtocol'>, typing.Protocol, typing.Generic, <class 'object'>]
reveal_type(SubclassOfMyProtocol.__mro__)

reveal_type(is_protocol(SubclassOfMyProtocol))  # revealed: Literal[False]

A protocol class may inherit from other protocols, however, as long as it re-inherits from Protocol:

class SubProtocol(MyProtocol, Protocol): ...

reveal_type(is_protocol(SubProtocol))  # revealed: Literal[True]

class OtherProtocol(Protocol):
    some_attribute: str

class ComplexInheritance(SubProtocol, OtherProtocol, Protocol): ...

# revealed: tuple[<class 'ComplexInheritance'>, <class 'SubProtocol'>, <class 'MyProtocol'>, <class 'OtherProtocol'>, typing.Protocol, typing.Generic, <class 'object'>]
reveal_type(ComplexInheritance.__mro__)

reveal_type(is_protocol(ComplexInheritance))  # revealed: Literal[True]

If Protocol is present in the bases tuple, all other bases in the tuple must be protocol classes, or TypeError is raised at runtime when the class is created.

# error: [invalid-protocol] "Protocol class `Invalid` cannot inherit from non-protocol class `NotAProtocol`"
class Invalid(NotAProtocol, Protocol): ...

# revealed: tuple[<class 'Invalid'>, <class 'NotAProtocol'>, typing.Protocol, typing.Generic, <class 'object'>]
reveal_type(Invalid.__mro__)

# error: [invalid-protocol] "Protocol class `AlsoInvalid` cannot inherit from non-protocol class `NotAProtocol`"
class AlsoInvalid(MyProtocol, OtherProtocol, NotAProtocol, Protocol): ...

# revealed: tuple[<class 'AlsoInvalid'>, <class 'MyProtocol'>, <class 'OtherProtocol'>, <class 'NotAProtocol'>, typing.Protocol, typing.Generic, <class 'object'>]
reveal_type(AlsoInvalid.__mro__)

But two exceptions to this rule are object and Generic:

from typing import TypeVar, Generic

T = TypeVar("T")

# Note: pyright and pyrefly do not consider this to be a valid `Protocol` class,
# but mypy does (and has an explicit test for this behaviour). Mypy was the
# reference implementation for PEP-544, and its behaviour also matches the CPython
# runtime, so we choose to follow its behaviour here rather than that of the other
# type checkers.
class Fine(Protocol, object): ...

reveal_type(Fine.__mro__)  # revealed: tuple[<class 'Fine'>, typing.Protocol, typing.Generic, <class 'object'>]

class StillFine(Protocol, Generic[T], object): ...
class EvenThis[T](Protocol, object): ...
class OrThis(Protocol[T], Generic[T]): ...
class AndThis(Protocol[T], Generic[T], object): ...

And multiple inheritance from a mix of protocol and non-protocol classes is fine as long as Protocol itself is not in the bases list:

class FineAndDandy(MyProtocol, OtherProtocol, NotAProtocol): ...

# revealed: tuple[<class 'FineAndDandy'>, <class 'MyProtocol'>, <class 'OtherProtocol'>, typing.Protocol, typing.Generic, <class 'NotAProtocol'>, <class 'object'>]
reveal_type(FineAndDandy.__mro__)

But if Protocol is not present in the bases list, the resulting class doesn't count as a protocol class anymore:

reveal_type(is_protocol(FineAndDandy))  # revealed: Literal[False]

A class does not have to inherit from a protocol class in order for it to be considered a subtype of that protocol (more on that below). However, classes that explicitly inherit from a protocol class are understood as subtypes of that protocol, the same as with nominal types:

from ty_extensions import static_assert, is_subtype_of, is_assignable_to

static_assert(is_subtype_of(SubclassOfMyProtocol, MyProtocol))
static_assert(is_assignable_to(SubclassOfMyProtocol, MyProtocol))

static_assert(is_subtype_of(SubProtocol, MyProtocol))
static_assert(is_assignable_to(SubProtocol, MyProtocol))

static_assert(is_subtype_of(ComplexInheritance, SubProtocol))
static_assert(is_assignable_to(ComplexInheritance, SubProtocol))

static_assert(is_subtype_of(ComplexInheritance, OtherProtocol))
static_assert(is_assignable_to(ComplexInheritance, SubProtocol))

static_assert(is_subtype_of(FineAndDandy, MyProtocol))
static_assert(is_assignable_to(FineAndDandy, MyProtocol))

static_assert(is_subtype_of(FineAndDandy, OtherProtocol))
static_assert(is_assignable_to(FineAndDandy, OtherProtocol))

Note, however, that Protocol itself is not a type, so it is an error to pass it to is_subtype_of or is_assignable_to:

is_subtype_of(MyProtocol, Protocol)  # error: [invalid-type-form]
is_assignable_to(MyProtocol, Protocol)  # error: [invalid-type-form]

And it is also an error to use Protocol in type expressions:

# fmt: off

def f(
    x: Protocol,  # error: [invalid-type-form] "`typing.Protocol` is not allowed in type expressions"
    y: type[Protocol],  # TODO: should emit `[invalid-type-form]` here too
):
    reveal_type(x)  # revealed: Unknown

    # TODO: should be `type[Unknown]`
    reveal_type(y)  # revealed: @Todo(unsupported type[X] special form)

# fmt: on

Nonetheless, Protocol can still be used as the second argument to issubclass() at runtime:

# Could also be `Literal[True]`, but `bool` is fine:
reveal_type(issubclass(MyProtocol, Protocol))  # revealed: bool

typing.Protocol versus typing_extensions.Protocol

typing.Protocol and its backport in typing_extensions should be treated as exactly equivalent.

import typing
import typing_extensions
from ty_extensions import static_assert, is_equivalent_to, TypeOf

static_assert(is_equivalent_to(TypeOf[typing.Protocol], TypeOf[typing_extensions.Protocol]))
static_assert(is_equivalent_to(int | str | TypeOf[typing.Protocol], TypeOf[typing_extensions.Protocol] | str | int))

class Foo(typing.Protocol):
    x: int

class Bar(typing_extensions.Protocol):
    x: int

static_assert(typing_extensions.is_protocol(Foo))
static_assert(typing_extensions.is_protocol(Bar))
static_assert(is_equivalent_to(Foo, Bar))

The same goes for typing.runtime_checkable and typing_extensions.runtime_checkable:

@typing_extensions.runtime_checkable
class RuntimeCheckableFoo(typing.Protocol):
    x: int

@typing.runtime_checkable
class RuntimeCheckableBar(typing_extensions.Protocol):
    x: int

static_assert(typing_extensions.is_protocol(RuntimeCheckableFoo))
static_assert(typing_extensions.is_protocol(RuntimeCheckableBar))
static_assert(is_equivalent_to(RuntimeCheckableFoo, RuntimeCheckableBar))

# These should not error because the protocols are decorated with `@runtime_checkable`
isinstance(object(), RuntimeCheckableFoo)
isinstance(object(), RuntimeCheckableBar)

However, we understand that they are not necessarily the same symbol at the same memory address at runtime -- these reveal bool rather than Literal[True] or Literal[False], which would be incorrect:

reveal_type(typing.Protocol is typing_extensions.Protocol)  # revealed: bool
reveal_type(typing.Protocol is not typing_extensions.Protocol)  # revealed: bool

Calls to protocol classes

Neither Protocol, nor any protocol class, can be directly instantiated:

[environment]
python-version = "3.12"

from typing_extensions import Protocol, reveal_type

# error: [call-non-callable]
reveal_type(Protocol())  # revealed: Unknown

class MyProtocol(Protocol):
    x: int

# error: [call-non-callable] "Cannot instantiate class `MyProtocol`"
reveal_type(MyProtocol())  # revealed: MyProtocol

class GenericProtocol[T](Protocol):
    x: T

# error: [call-non-callable] "Cannot instantiate class `GenericProtocol`"
reveal_type(GenericProtocol[int]())  # revealed: GenericProtocol[int]

But a non-protocol class can be instantiated, even if it has Protocol in its MRO:

class SubclassOfMyProtocol(MyProtocol): ...

reveal_type(SubclassOfMyProtocol())  # revealed: SubclassOfMyProtocol

class SubclassOfGenericProtocol[T](GenericProtocol[T]): ...

reveal_type(SubclassOfGenericProtocol[int]())  # revealed: SubclassOfGenericProtocol[int]

And as a corollary, type[MyProtocol] can also be called:

def f(x: type[MyProtocol]):
    reveal_type(x())  # revealed: MyProtocol

Members of a protocol

A protocol defines an interface through its members: if a protocol Foo has members X and Y, a type Bar can only be a subtype of Foo if inhabitants of Bar also have attributes X and Y.

A protocol class defines its members through declarations in the class body. The members of a protocol can be introspected using the function typing.get_protocol_members, which is backported via typing_extensions.

from typing_extensions import Protocol, get_protocol_members

class Foo(Protocol):
    x: int

    @property
    def y(self) -> str:
        return "y"

    @property
    def z(self) -> int:
        return 42

    @z.setter
    def z(self, z: int) -> None: ...
    def method_member(self) -> bytes:
        return b"foo"

# TODO: actually a frozenset (requires support for legacy generics)
reveal_type(get_protocol_members(Foo))  # revealed: tuple[Literal["method_member"], Literal["x"], Literal["y"], Literal["z"]]

Certain special attributes and methods are not considered protocol members at runtime, and should not be considered protocol members by type checkers either:

class Lumberjack(Protocol):
    __slots__ = ()
    __match_args__ = ()
    x: int

    def __new__(cls, x: int) -> "Lumberjack":
        return object.__new__(cls)

    def __init__(self, x: int) -> None:
        self.x = x

# TODO: actually a frozenset
reveal_type(get_protocol_members(Lumberjack))  # revealed: tuple[Literal["x"]]

A sub-protocol inherits and extends the members of its superclass protocol(s):

class Bar(Protocol):
    spam: str

class Baz(Bar, Protocol):
    ham: memoryview

# TODO: actually a frozenset
reveal_type(get_protocol_members(Baz))  # revealed: tuple[Literal["ham"], Literal["spam"]]

class Baz2(Bar, Foo, Protocol): ...

# TODO: actually a frozenset
# revealed: tuple[Literal["method_member"], Literal["spam"], Literal["x"], Literal["y"], Literal["z"]]
reveal_type(get_protocol_members(Baz2))

Protocol members in statically known branches

The list of protocol members does not include any members declared in branches that are statically known to be unreachable:

[environment]
python-version = "3.9"

import sys
from typing_extensions import Protocol, get_protocol_members

class Foo(Protocol):
    if sys.version_info >= (3, 10):
        a: int
        b = 42
        def c(self) -> None: ...
    else:
        d: int
        e = 56
        def f(self) -> None: ...

# TODO: actually a frozenset
reveal_type(get_protocol_members(Foo))  # revealed: tuple[Literal["d"], Literal["e"], Literal["f"]]

Invalid calls to get_protocol_members()

Calling get_protocol_members on a non-protocol class raises an error at runtime:

[environment]
python-version = "3.12"

from typing_extensions import Protocol, get_protocol_members

class NotAProtocol: ...

get_protocol_members(NotAProtocol)  # error: [invalid-argument-type]

class AlsoNotAProtocol(NotAProtocol, object): ...

get_protocol_members(AlsoNotAProtocol)  # error: [invalid-argument-type]

The original class object must be passed to the function; a specialised version of a generic version does not suffice:

class GenericProtocol[T](Protocol): ...

get_protocol_members(GenericProtocol[int])  # TODO: should emit a diagnostic here (https://github.com/astral-sh/ruff/issues/17549)

Subtyping of protocols with attribute members

In the following example, the protocol class HasX defines an interface such that any other fully static type can be said to be a subtype of HasX if all inhabitants of that other type have a mutable x attribute of type int:

[environment]
python-version = "3.12"

from typing import Protocol
from ty_extensions import static_assert, is_assignable_to, is_subtype_of

class HasX(Protocol):
    x: int

class Foo:
    x: int

static_assert(is_subtype_of(Foo, HasX))
static_assert(is_assignable_to(Foo, HasX))

class FooSub(Foo): ...

static_assert(is_subtype_of(FooSub, HasX))
static_assert(is_assignable_to(FooSub, HasX))

class Bar:
    x: str

# TODO: these should pass
static_assert(not is_subtype_of(Bar, HasX))  # error: [static-assert-error]
static_assert(not is_assignable_to(Bar, HasX))  # error: [static-assert-error]

class Baz:
    y: int

static_assert(not is_subtype_of(Baz, HasX))
static_assert(not is_assignable_to(Baz, HasX))

Note that declaring an attribute member on a protocol mandates that the attribute must be mutable. A type with a read-only x property does not satisfy the HasX interface; nor does a type with a Final x attribute. The type of the attribute must also be treated as invariant due to the attribute's mutability:

from typing import Final

class A:
    @property
    def x(self) -> int:
        return 42

# TODO: these should pass
static_assert(not is_subtype_of(A, HasX))  # error: [static-assert-error]
static_assert(not is_assignable_to(A, HasX))  # error: [static-assert-error]

class B:
    x: Final = 42

# TODO: these should pass
static_assert(not is_subtype_of(A, HasX))  # error: [static-assert-error]
static_assert(not is_assignable_to(A, HasX))  # error: [static-assert-error]

class IntSub(int): ...

class C:
    x: IntSub

# due to invariance, a type is only a subtype of `HasX`
# if its `x` attribute is of type *exactly* `int`:
# a subclass of `int` does not satisfy the interface
#
# TODO: these should pass
static_assert(not is_subtype_of(C, HasX))  # error: [static-assert-error]
static_assert(not is_assignable_to(C, HasX))  # error: [static-assert-error]

All attributes on frozen dataclasses and namedtuples are immutable, so instances of these classes can never be considered to inhabit a protocol that declares a mutable-attribute member:

from dataclasses import dataclass
from typing import NamedTuple

@dataclass
class MutableDataclass:
    x: int

static_assert(is_subtype_of(MutableDataclass, HasX))
static_assert(is_assignable_to(MutableDataclass, HasX))

@dataclass(frozen=True)
class ImmutableDataclass:
    x: int

# TODO: these should pass
static_assert(not is_subtype_of(ImmutableDataclass, HasX))  # error: [static-assert-error]
static_assert(not is_assignable_to(ImmutableDataclass, HasX))  # error: [static-assert-error]

class NamedTupleWithX(NamedTuple):
    x: int

# TODO: these should pass
static_assert(not is_subtype_of(NamedTupleWithX, HasX))  # error: [static-assert-error]
static_assert(not is_assignable_to(NamedTupleWithX, HasX))  # error: [static-assert-error]

However, a type with a read-write property x does satisfy the HasX protocol. The HasX protocol only specifies what the type of x should be when accessed from instances; instances of XProperty in the below example have a mutable attribute x of type int:

class XProperty:
    _x: int

    @property
    def x(self) -> int:
        return self._x

    @x.setter
    def x(self, x: int) -> None:
        self._x = x**2

static_assert(is_subtype_of(XProperty, HasX))
static_assert(is_assignable_to(XProperty, HasX))

Attribute members on protocol classes are allowed to have default values, just like instance attributes on other classes. Similar to nominal classes, attributes with defaults can be accessed on the class object itself and any explicit subclasses of the protocol class. It cannot be assumed to exist on the meta-type of any arbitrary inhabitant of the protocol type, however; an implicit subtype of the protocol will not necessarily have a default value for the instance attribute provided in its class body:

class HasXWithDefault(Protocol):
    x: int = 42

reveal_type(HasXWithDefault.x)  # revealed: int

class ExplicitSubclass(HasXWithDefault): ...

reveal_type(ExplicitSubclass.x)  # revealed: int

def f(arg: HasXWithDefault):
    # TODO: should emit `[unresolved-reference]` and reveal `Unknown`
    reveal_type(type(arg).x)  # revealed: int

Assignments in a class body of a protocol -- of any kind -- are not permitted by ty unless the symbol being assigned to is also explicitly declared in the protocol's class body. Note that this is stricter validation of protocol members than many other type checkers currently apply (as of 2025/04/21).

The reason for this strict validation is that undeclared variables in the class body would lead to an ambiguous interface being declared by the protocol.

from typing_extensions import TypeAlias, get_protocol_members

class MyContext:
    def __enter__(self) -> int:
        return 42

    def __exit__(self, *args) -> None: ...

class LotsOfBindings(Protocol):
    a: int
    a = 42  # this is fine, since `a` is declared in the class body
    b: int = 56  # this is also fine, by the same principle

    type c = str  # this is very strange but I can't see a good reason to disallow it
    d: TypeAlias = bytes  # same here

    class Nested: ...  # also weird, but we should also probably allow it
    class NestedProtocol(Protocol): ...  # same here...
    e = 72  # TODO: this should error with `[invalid-protocol]` (`e` is not declared)

    f, g = (1, 2)  # TODO: this should error with `[invalid-protocol]` (`f` and `g` are not declared)

    h: int = (i := 3)  # TODO: this should error with `[invalid-protocol]` (`i` is not declared)

    for j in range(42):  # TODO: this should error with `[invalid-protocol]` (`j` is not declared)
        pass

    with MyContext() as k:  # TODO: this should error with `[invalid-protocol]` (`k` is not declared)
        pass

    match object():
        case l:  # TODO: this should error with `[invalid-protocol]` (`l` is not declared)
            ...

# TODO: actually a frozenset
# revealed: tuple[Literal["Nested"], Literal["NestedProtocol"], Literal["a"], Literal["b"], Literal["c"], Literal["d"], Literal["e"], Literal["f"], Literal["g"], Literal["h"], Literal["i"], Literal["j"], Literal["k"], Literal["l"]]
reveal_type(get_protocol_members(LotsOfBindings))

Attribute members are allowed to have assignments in methods on the protocol class, just like non-protocol classes. Unlike other classes, however, instance attributes that are not declared in the class body are disallowed. This is mandated by the spec:

Additional attributes only defined in the body of a method by assignment via self are not allowed. The rationale for this is that the protocol class implementation is often not shared by subtypes, so the interface should not depend on the default implementation.

class Foo(Protocol):
    x: int
    y: str

    def __init__(self) -> None:
        self.x = 42  # fine
        self.a = 56  # TODO: should emit diagnostic
        self.b: int = 128  # TODO: should emit diagnostic

    def non_init_method(self) -> None:
        self.y = 64  # fine
        self.c = 72  # TODO: should emit diagnostic

# Note: the list of members does not include `a`, `b` or `c`,
# as none of these attributes is declared in the class body.
#
# TODO: actually a frozenset
reveal_type(get_protocol_members(Foo))  # revealed: tuple[Literal["non_init_method"], Literal["x"], Literal["y"]]

If a member is declared in a superclass of a protocol class, it is fine for it to be assigned to in the sub-protocol class without a redeclaration:

class Super(Protocol):
    x: int

class Sub(Super, Protocol):
    x = 42  # no error here, since it's declared in the superclass

# TODO: actually frozensets
reveal_type(get_protocol_members(Super))  # revealed: tuple[Literal["x"]]
reveal_type(get_protocol_members(Sub))  # revealed: tuple[Literal["x"]]

If a protocol has 0 members, then all other types are assignable to it, and all fully static types are subtypes of it:

from typing import Protocol

class UniversalSet(Protocol): ...

static_assert(is_assignable_to(object, UniversalSet))
static_assert(is_subtype_of(object, UniversalSet))

Which means that UniversalSet here is in fact an equivalent type to object:

from ty_extensions import is_equivalent_to

static_assert(is_equivalent_to(UniversalSet, object))

object is a subtype of certain other protocols too. Since all fully static types (whether nominal or structural) are subtypes of object, these protocols are also subtypes of object; and this means that these protocols are also equivalent to UniversalSet and object:

class SupportsStr(Protocol):
    def __str__(self) -> str: ...

static_assert(is_equivalent_to(SupportsStr, UniversalSet))
static_assert(is_equivalent_to(SupportsStr, object))

class SupportsClass(Protocol):
    @property
    def __class__(self) -> type: ...

static_assert(is_equivalent_to(SupportsClass, UniversalSet))
static_assert(is_equivalent_to(SupportsClass, SupportsStr))
static_assert(is_equivalent_to(SupportsClass, object))

If a protocol contains members that are not defined on object, then that protocol will (like all types in Python) still be assignable to object, but object will not be assignable to that protocol:

static_assert(is_assignable_to(HasX, object))
static_assert(is_subtype_of(HasX, object))
static_assert(not is_assignable_to(object, HasX))
static_assert(not is_subtype_of(object, HasX))

But object is the only fully static nominal type that a protocol type can ever be assignable to or a subtype of:

static_assert(not is_assignable_to(HasX, Foo))
static_assert(not is_subtype_of(HasX, Foo))

Equivalence of protocols

Two protocols are considered equivalent types if they specify the same interface, even if they have different names:

from typing import Protocol
from ty_extensions import is_equivalent_to, static_assert

class HasX(Protocol):
    x: int

class AlsoHasX(Protocol):
    x: int

static_assert(is_equivalent_to(HasX, AlsoHasX))

And unions containing equivalent protocols are recognised as equivalent, even when the order is not identical:

class HasY(Protocol):
    y: str

class AlsoHasY(Protocol):
    y: str

class A: ...
class B: ...

static_assert(is_equivalent_to(A | HasX | B | HasY, B | AlsoHasY | AlsoHasX | A))

Protocols are considered equivalent if their members are equivalent, even if those members are differently ordered unions:

class C: ...

class UnionProto1(Protocol):
    x: A | B | C

class UnionProto2(Protocol):
    x: C | A | B

static_assert(is_equivalent_to(UnionProto1, UnionProto2))
static_assert(is_equivalent_to(UnionProto1 | A | B, B | UnionProto2 | A))

Intersections of protocols

An intersection of two protocol types X and Y is equivalent to a protocol type Z that inherits from both X and Y:

from typing import Protocol
from ty_extensions import Intersection, static_assert, is_equivalent_to

class HasX(Protocol):
    x: int

class HasY(Protocol):
    y: str

class HasXAndYProto(HasX, HasY, Protocol): ...

# TODO: this should pass
static_assert(is_equivalent_to(HasXAndYProto, Intersection[HasX, HasY]))  # error: [static-assert-error]

But this is only true if the subclass has Protocol in its explicit bases (otherwise, it is a nominal type rather than a structural type):

class HasXAndYNominal(HasX, HasY): ...

static_assert(not is_equivalent_to(HasXAndYNominal, Intersection[HasX, HasY]))

A protocol type X and a nominal type Y can be inferred as disjoint types if Y is a @final type and Y does not satisfy the interface declared by X. But if Y is not @final, then this does not hold true, since a subclass of Y could always provide additional methods or attributes that would lead to it satisfying X's interface:

from typing import final
from ty_extensions import is_disjoint_from

class NotFinalNominal: ...

@final
class FinalNominal: ...

static_assert(not is_disjoint_from(NotFinalNominal, HasX))
static_assert(is_disjoint_from(FinalNominal, HasX))

def _(arg1: Intersection[HasX, NotFinalNominal], arg2: Intersection[HasX, FinalNominal]):
    reveal_type(arg1)  # revealed: HasX & NotFinalNominal
    reveal_type(arg2)  # revealed: Never

Satisfying a protocol's interface

A type does not have to be an Instance type in order to be a subtype of a protocol. Other protocols can be a subtype of a protocol, as can ModuleLiteral types, ClassLiteral types, and others. Another protocol can be a subtype of HasX either through "explicit" (nominal) inheritance from HasX, or by specifying a superset of HasX's interface:

module.py:

x: int = 42

main.py:

import module
from typing import Protocol
from ty_extensions import is_subtype_of, is_assignable_to, static_assert, TypeOf

class HasX(Protocol):
    x: int

# TODO: this should pass
static_assert(is_subtype_of(TypeOf[module], HasX))  # error: [static-assert-error]
static_assert(is_assignable_to(TypeOf[module], HasX))

class ExplicitProtocolSubtype(HasX, Protocol):
    y: int

static_assert(is_subtype_of(ExplicitProtocolSubtype, HasX))
static_assert(is_assignable_to(ExplicitProtocolSubtype, HasX))

class ImplicitProtocolSubtype(Protocol):
    x: int
    y: str

static_assert(is_subtype_of(ImplicitProtocolSubtype, HasX))
static_assert(is_assignable_to(ImplicitProtocolSubtype, HasX))

class Meta(type):
    x: int

class UsesMeta(metaclass=Meta): ...

# TODO: these should pass
static_assert(is_subtype_of(UsesMeta, HasX))  # error: [static-assert-error]
static_assert(is_assignable_to(UsesMeta, HasX))  # error: [static-assert-error]

ClassVar attribute members

If a protocol ClassVarX has a ClassVar attribute member x with type int, this indicates that a readable x attribute must be accessible on any inhabitant of ClassVarX, and that a readable x attribute must also be accessible on the type of that inhabitant:

classvars.py:

from typing import ClassVar, Protocol
from ty_extensions import is_subtype_of, is_assignable_to, static_assert

class ClassVarXProto(Protocol):
    x: ClassVar[int]

def f(obj: ClassVarXProto):
    reveal_type(obj.x)  # revealed: int
    reveal_type(type(obj).x)  # revealed: int
    obj.x = 42  # error: [invalid-attribute-access] "Cannot assign to ClassVar `x` from an instance of type `ClassVarXProto`"

class InstanceAttrX:
    x: int

# TODO: these should pass
static_assert(not is_assignable_to(InstanceAttrX, ClassVarXProto))  # error: [static-assert-error]
static_assert(not is_subtype_of(InstanceAttrX, ClassVarXProto))  # error: [static-assert-error]

class PropertyX:
    @property
    def x(self) -> int:
        return 42

# TODO: these should pass
static_assert(not is_assignable_to(PropertyX, ClassVarXProto))  # error: [static-assert-error]
static_assert(not is_subtype_of(PropertyX, ClassVarXProto))  # error: [static-assert-error]

class ClassVarX:
    x: ClassVar[int] = 42

static_assert(is_assignable_to(ClassVarX, ClassVarXProto))
static_assert(is_subtype_of(ClassVarX, ClassVarXProto))

This is mentioned by the spec and tested in the conformance suite as something that must be supported by type checkers:

To distinguish between protocol class variables and protocol instance variables, the special ClassVar annotation should be used.

Subtyping of protocols with property members

A read-only property on a protocol can be satisfied by a mutable attribute, a read-only property, a read/write property, a Final attribute, or a ClassVar attribute:

from typing import ClassVar, Final, Protocol
from ty_extensions import is_subtype_of, is_assignable_to, static_assert

class HasXProperty(Protocol):
    @property
    def x(self) -> int: ...

class XAttr:
    x: int

static_assert(is_subtype_of(XAttr, HasXProperty))
static_assert(is_assignable_to(XAttr, HasXProperty))

class XReadProperty:
    @property
    def x(self) -> int:
        return 42

static_assert(is_subtype_of(XReadProperty, HasXProperty))
static_assert(is_assignable_to(XReadProperty, HasXProperty))

class XReadWriteProperty:
    @property
    def x(self) -> int:
        return 42

    @x.setter
    def x(self, val: int) -> None: ...

static_assert(is_subtype_of(XReadWriteProperty, HasXProperty))
static_assert(is_assignable_to(XReadWriteProperty, HasXProperty))

class XClassVar:
    x: ClassVar[int] = 42

static_assert(is_subtype_of(XClassVar, HasXProperty))
static_assert(is_assignable_to(XClassVar, HasXProperty))

class XFinal:
    x: Final = 42

static_assert(is_subtype_of(XFinal, HasXProperty))
static_assert(is_assignable_to(XFinal, HasXProperty))

A read-only property on a protocol, unlike a mutable attribute, is covariant: XSub in the below example satisfies the HasXProperty interface even though the type of the x attribute on XSub is a subtype of int rather than being exactly int.

class MyInt(int): ...

class XSub:
    x: MyInt

static_assert(is_subtype_of(XSub, HasXProperty))
static_assert(is_assignable_to(XSub, HasXProperty))

A read/write property on a protocol, where the getter returns the same type that the setter takes, is equivalent to a normal mutable attribute on a protocol.

class HasMutableXProperty(Protocol):
    @property
    def x(self) -> int: ...
    @x.setter
    def x(self, val: int) -> None: ...

class XAttr:
    x: int

static_assert(is_subtype_of(XAttr, HasXProperty))
static_assert(is_assignable_to(XAttr, HasXProperty))

class XReadProperty:
    @property
    def x(self) -> int:
        return 42

# TODO: these should pass
static_assert(not is_subtype_of(XReadProperty, HasXProperty))  # error: [static-assert-error]
static_assert(not is_assignable_to(XReadProperty, HasXProperty))  # error: [static-assert-error]

class XReadWriteProperty:
    @property
    def x(self) -> int:
        return 42

    @x.setter
    def x(self, val: int) -> None: ...

static_assert(is_subtype_of(XReadWriteProperty, HasXProperty))
static_assert(is_assignable_to(XReadWriteProperty, HasXProperty))

class XSub:
    x: MyInt

# TODO: should pass
static_assert(not is_subtype_of(XSub, HasXProperty))  # error: [static-assert-error]
static_assert(not is_assignable_to(XSub, HasXProperty))  # error: [static-assert-error]

A protocol with a read/write property x is exactly equivalent to a protocol with a mutable attribute x. Both are subtypes of a protocol with a read-only prooperty x:

from ty_extensions import is_equivalent_to

class HasMutableXAttr(Protocol):
    x: int

# TODO: should pass
static_assert(is_equivalent_to(HasMutableXAttr, HasMutableXProperty))  # error: [static-assert-error]

static_assert(is_subtype_of(HasMutableXAttr, HasXProperty))
static_assert(is_assignable_to(HasMutableXAttr, HasXProperty))

static_assert(is_subtype_of(HasMutableXProperty, HasXProperty))
static_assert(is_assignable_to(HasMutableXProperty, HasXProperty))

A read/write property on a protocol, where the setter accepts a subtype of the type returned by the getter, can be satisfied by a mutable attribute of any type bounded by the upper bound of the getter-returned type and the lower bound of the setter-accepted type.

This follows from the principle that a type X can only be a subtype of a given protocol if the X's behaviour is a superset of the behaviour specified by the interface declared by the protocol. In the below example, the behaviour of an instance of XAttr is a superset of the behaviour specified by the protocol HasAsymmetricXProperty. The protocol specifies that reading an x attribute on the instance must resolve to an instance of int or a subclass thereof, and XAttr satisfies this requirement. The protocol also specifies that you must be able to assign instances of MyInt to the x attribute, and again this is satisfied by XAttr: on instances of XAttr, you can assign any instance of int to the x attribute, and thus by extension you can assign any instance of IntSub to the x attribute, since any instance of IntSub is an instance of int:

class HasAsymmetricXProperty(Protocol):
    @property
    def x(self) -> int: ...
    @x.setter
    def x(self, val: MyInt) -> None: ...

class XAttr:
    x: int

static_assert(is_subtype_of(XAttr, HasAsymmetricXProperty))
static_assert(is_assignable_to(XAttr, HasAsymmetricXProperty))

The end conclusion of this is that the getter-returned type of a property is always covariant and the setter-accepted type is always contravariant. The combination of these leads to invariance for a regular mutable attribute, where the implied getter-returned and setter-accepted types are the same.

class XAttrSub:
    x: MyInt

static_assert(is_subtype_of(XAttrSub, HasAsymmetricXProperty))
static_assert(is_assignable_to(XAttrSub, HasAsymmetricXProperty))

class MyIntSub(MyInt):
    pass

class XAttrSubSub:
    x: MyIntSub

# TODO: should pass
static_assert(not is_subtype_of(XAttrSubSub, HasAsymmetricXProperty))  # error: [static-assert-error]
static_assert(not is_assignable_to(XAttrSubSub, HasAsymmetricXProperty))  # error: [static-assert-error]

An asymmetric property on a protocol can also be satisfied by an asymmetric property on a nominal class whose getter and setter types satisfy the covariant and contravariant requirements, respectively.

class XAsymmetricProperty:
    @property
    def x(self) -> MyInt:
        return MyInt(0)

    @x.setter
    def x(self, x: int) -> None: ...

static_assert(is_subtype_of(XAsymmetricProperty, HasAsymmetricXProperty))
static_assert(is_assignable_to(XAsymmetricProperty, HasAsymmetricXProperty))

A custom descriptor attribute on the nominal class will also suffice:

class Descriptor:
    def __get__(self, instance, owner) -> MyInt:
        return MyInt(0)

    def __set__(self, value: int) -> None: ...

class XCustomDescriptor:
    x: Descriptor = Descriptor()

static_assert(is_subtype_of(XCustomDescriptor, HasAsymmetricXProperty))
static_assert(is_assignable_to(XCustomDescriptor, HasAsymmetricXProperty))

Moreover, a read-only property on a protocol can be satisfied by a nominal class that defines a __getattr__ method returning a suitable type. A read/write property can be satisfied by a nominal class that defines a __getattr__ method returning a suitable type and a __setattr__ method accepting a suitable type:

class HasGetAttr:
    def __getattr__(self, attr: str) -> int:
        return 42

static_assert(is_subtype_of(HasGetAttr, HasXProperty))
static_assert(is_assignable_to(HasGetAttr, HasXProperty))

# TODO: these should pass
static_assert(not is_subtype_of(HasGetAttr, HasMutableXAttr))  # error: [static-assert-error]
static_assert(not is_subtype_of(HasGetAttr, HasMutableXAttr))  # error: [static-assert-error]

class HasGetAttrWithUnsuitableReturn:
    def __getattr__(self, attr: str) -> tuple[int, int]:
        return (1, 2)

# TODO: these should pass
static_assert(not is_subtype_of(HasGetAttrWithUnsuitableReturn, HasXProperty))  # error: [static-assert-error]
static_assert(not is_assignable_to(HasGetAttrWithUnsuitableReturn, HasXProperty))  # error: [static-assert-error]

class HasGetAttrAndSetAttr:
    def __getattr__(self, attr: str) -> MyInt:
        return MyInt(0)

    def __setattr__(self, attr: str, value: int) -> None: ...

static_assert(is_subtype_of(HasGetAttrAndSetAttr, HasXProperty))
static_assert(is_assignable_to(HasGetAttrAndSetAttr, HasXProperty))

# TODO: these should pass
static_assert(is_subtype_of(HasGetAttrAndSetAttr, XAsymmetricProperty))  # error: [static-assert-error]
static_assert(is_assignable_to(HasGetAttrAndSetAttr, XAsymmetricProperty))  # error: [static-assert-error]

Narrowing of protocols

By default, a protocol class cannot be used as the second argument to isinstance() or issubclass(), and a type checker must emit an error on such calls. However, we still narrow the type inside these branches (this matches the behaviour of other type checkers):

from typing_extensions import Protocol, reveal_type

class HasX(Protocol):
    x: int

def f(arg: object, arg2: type):
    if isinstance(arg, HasX):  # error: [invalid-argument-type]
        reveal_type(arg)  # revealed: HasX
    else:
        reveal_type(arg)  # revealed: ~HasX

    if issubclass(arg2, HasX):  # error: [invalid-argument-type]
        reveal_type(arg2)  # revealed: type[HasX]
    else:
        reveal_type(arg2)  # revealed: type & ~type[HasX]

A protocol class decorated with @typing(_extensions).runtime_checkable can be used as the second argument to isisinstance() at runtime:

from typing import runtime_checkable

@runtime_checkable
class RuntimeCheckableHasX(Protocol):
    x: int

def f(arg: object):
    if isinstance(arg, RuntimeCheckableHasX):  # no error!
        reveal_type(arg)  # revealed: RuntimeCheckableHasX
    else:
        reveal_type(arg)  # revealed: ~RuntimeCheckableHasX

but in order for a protocol class to be used as the second argument to issubclass(), it must satisfy two conditions:

1. It must be decorated with @runtime_checkable
2. It must only have method members (protocols with attribute members are not permitted)

@runtime_checkable
class OnlyMethodMembers(Protocol):
    def method(self) -> None: ...

def f(arg1: type, arg2: type):
    if issubclass(arg1, RuntimeCheckableHasX):  # TODO: should emit an error here (has non-method members)
        reveal_type(arg1)  # revealed: type[RuntimeCheckableHasX]
    else:
        reveal_type(arg1)  # revealed: type & ~type[RuntimeCheckableHasX]

    if issubclass(arg2, OnlyMethodMembers):  # no error!
        reveal_type(arg2)  # revealed: type[OnlyMethodMembers]
    else:
        reveal_type(arg2)  # revealed: type & ~type[OnlyMethodMembers]

Truthiness of protocol instance

An instance of a protocol type generally has ambiguous truthiness:

from typing import Protocol

class Foo(Protocol):
    x: int

def f(foo: Foo):
    reveal_type(bool(foo))  # revealed: bool

But this is not the case if the protocol has a __bool__ method member that returns Literal[True] or Literal[False]:

from typing import Literal

class Truthy(Protocol):
    def __bool__(self) -> Literal[True]: ...

class FalsyFoo(Foo, Protocol):
    def __bool__(self) -> Literal[False]: ...

class FalsyFooSubclass(FalsyFoo, Protocol):
    y: str

def g(a: Truthy, b: FalsyFoo, c: FalsyFooSubclass):
    reveal_type(bool(a))  # revealed: Literal[True]
    reveal_type(bool(b))  # revealed: Literal[False]
    reveal_type(bool(c))  # revealed: Literal[False]

It is not sufficient for a protocol to have a callable __bool__ instance member that returns Literal[True] for it to be considered always truthy. Dunder methods are looked up on the class rather than the instance. If a protocol X has an instance-attribute __bool__ member, it is unknowable whether that attribute can be accessed on the type of an object that satisfies X's interface:

from typing import Callable

class InstanceAttrBool(Protocol):
    __bool__: Callable[[], Literal[True]]

def h(obj: InstanceAttrBool):
    reveal_type(bool(obj))  # revealed: bool

Fully static protocols; gradual protocols

A protocol is only fully static if all of its members are fully static:

from typing import Protocol, Any
from ty_extensions import is_fully_static, static_assert

class FullyStatic(Protocol):
    x: int

class NotFullyStatic(Protocol):
    x: Any

static_assert(is_fully_static(FullyStatic))
static_assert(not is_fully_static(NotFullyStatic))

Non-fully-static protocols do not participate in subtyping or equivalence, only assignability and gradual equivalence:

from ty_extensions import is_subtype_of, is_assignable_to, is_equivalent_to, is_gradual_equivalent_to

class NominalWithX:
    x: int = 42

static_assert(is_assignable_to(NominalWithX, FullyStatic))
static_assert(is_assignable_to(NominalWithX, NotFullyStatic))

static_assert(not is_subtype_of(FullyStatic, NotFullyStatic))
static_assert(is_assignable_to(FullyStatic, NotFullyStatic))

static_assert(not is_subtype_of(NotFullyStatic, FullyStatic))
static_assert(is_assignable_to(NotFullyStatic, FullyStatic))

static_assert(not is_subtype_of(NominalWithX, NotFullyStatic))
static_assert(is_assignable_to(NominalWithX, NotFullyStatic))

static_assert(is_subtype_of(NominalWithX, FullyStatic))

static_assert(is_equivalent_to(FullyStatic, FullyStatic))
static_assert(not is_equivalent_to(NotFullyStatic, NotFullyStatic))

static_assert(is_gradual_equivalent_to(FullyStatic, FullyStatic))
static_assert(is_gradual_equivalent_to(NotFullyStatic, NotFullyStatic))

class AlsoNotFullyStatic(Protocol):
    x: Any

static_assert(not is_equivalent_to(NotFullyStatic, AlsoNotFullyStatic))
static_assert(is_gradual_equivalent_to(NotFullyStatic, AlsoNotFullyStatic))

Empty protocols are fully static; this follows from the fact that an empty protocol is equivalent to the nominal type object (as described above):

class Empty(Protocol): ...

static_assert(is_fully_static(Empty))

A method member is only considered fully static if all its parameter annotations and its return annotation are fully static:

class FullyStaticMethodMember(Protocol):
    def method(self, x: int) -> str: ...

class DynamicParameter(Protocol):
    def method(self, x: Any) -> str: ...

class DynamicReturn(Protocol):
    def method(self, x: int) -> Any: ...

static_assert(is_fully_static(FullyStaticMethodMember))

# TODO: these should pass
static_assert(not is_fully_static(DynamicParameter))  # error: [static-assert-error]
static_assert(not is_fully_static(DynamicReturn))  # error: [static-assert-error]

The typing spec states:

If any parameters of a protocol method are not annotated, then their types are assumed to be Any

Thus, a partially unannotated method member can also not be considered to be fully static:

class NoParameterAnnotation(Protocol):
    def method(self, x) -> str: ...

class NoReturnAnnotation(Protocol):
    def method(self, x: int): ...

# TODO: these should pass
static_assert(not is_fully_static(NoParameterAnnotation))  # error: [static-assert-error]
static_assert(not is_fully_static(NoReturnAnnotation))  # error: [static-assert-error]

Callable protocols

An instance of a protocol type is callable if the protocol defines a __call__ method:

from typing import Protocol

class CallMeMaybe(Protocol):
    def __call__(self, x: int) -> str: ...

def f(obj: CallMeMaybe):
    reveal_type(obj(42))  # revealed: str
    obj("bar")  # error: [invalid-argument-type]

An instance of a protocol like this can be assignable to a Callable type, but only if it has the right signature:

from typing import Callable
from ty_extensions import is_subtype_of, is_assignable_to, static_assert

static_assert(is_subtype_of(CallMeMaybe, Callable[[int], str]))
static_assert(is_assignable_to(CallMeMaybe, Callable[[int], str]))
static_assert(not is_subtype_of(CallMeMaybe, Callable[[str], str]))
static_assert(not is_assignable_to(CallMeMaybe, Callable[[str], str]))
static_assert(not is_subtype_of(CallMeMaybe, Callable[[CallMeMaybe, int], str]))
static_assert(not is_assignable_to(CallMeMaybe, Callable[[CallMeMaybe, int], str]))

def g(obj: Callable[[int], str], obj2: CallMeMaybe, obj3: Callable[[str], str]):
    obj = obj2
    obj3 = obj2  # error: [invalid-assignment]

By the same token, a Callable type can also be assignable to a protocol-instance type if the signature implied by the Callable type is assignable to the signature of the __call__ method specified by the protocol:

from ty_extensions import TypeOf

class Foo(Protocol):
    def __call__(self, x: int, /) -> str: ...

static_assert(is_subtype_of(Callable[[int], str], Foo))
static_assert(is_assignable_to(Callable[[int], str], Foo))

# TODO: these should pass
static_assert(not is_subtype_of(Callable[[str], str], Foo))  # error: [static-assert-error]
static_assert(not is_assignable_to(Callable[[str], str], Foo))  # error: [static-assert-error]
static_assert(not is_subtype_of(Callable[[CallMeMaybe, int], str], Foo))  # error: [static-assert-error]
static_assert(not is_assignable_to(Callable[[CallMeMaybe, int], str], Foo))  # error: [static-assert-error]

def h(obj: Callable[[int], str], obj2: Foo, obj3: Callable[[str], str]):
    obj2 = obj

    # TODO: we should emit [invalid-assignment] here because the signature of `obj3` is not assignable
    # to the declared type of `obj2`
    obj2 = obj3

def satisfies_foo(x: int) -> str:
    return "foo"

static_assert(is_subtype_of(TypeOf[satisfies_foo], Foo))
static_assert(is_assignable_to(TypeOf[satisfies_foo], Foo))

Protocols are never singleton types, and are never single-valued types

It might be possible to have a singleton protocol-instance type...?

For example, WeirdAndWacky in the following snippet only has a single possible inhabitant: None! It is thus a singleton type. However, going out of our way to recognise it as such is probably not worth it. Such cases should anyway be exceedingly rare and/or contrived.

from typing import Protocol, Callable
from ty_extensions import is_singleton, is_single_valued

class WeirdAndWacky(Protocol):
    @property
    def __class__(self) -> Callable[[], None]: ...

reveal_type(is_singleton(WeirdAndWacky))  # revealed: Literal[False]
reveal_type(is_single_valued(WeirdAndWacky))  # revealed: Literal[False]

Integration test: typing.SupportsIndex and typing.Sized

typing.SupportsIndex and typing.Sized are two protocols that are very commonly used in the wild.

from typing import SupportsIndex, Sized, Literal

def one(some_int: int, some_literal_int: Literal[1], some_indexable: SupportsIndex):
    a: SupportsIndex = some_int
    b: SupportsIndex = some_literal_int
    c: SupportsIndex = some_indexable

def two(some_list: list, some_tuple: tuple[int, str], some_sized: Sized):
    a: Sized = some_list
    b: Sized = some_tuple
    c: Sized = some_sized

Recursive protocols

Properties

from __future__ import annotations

from typing import Protocol, Any
from ty_extensions import is_fully_static, static_assert, is_assignable_to, is_subtype_of, is_equivalent_to

class RecursiveFullyStatic(Protocol):
    parent: RecursiveFullyStatic
    x: int

class RecursiveNonFullyStatic(Protocol):
    parent: RecursiveNonFullyStatic
    x: Any

static_assert(is_fully_static(RecursiveFullyStatic))
static_assert(not is_fully_static(RecursiveNonFullyStatic))

static_assert(not is_subtype_of(RecursiveFullyStatic, RecursiveNonFullyStatic))
static_assert(not is_subtype_of(RecursiveNonFullyStatic, RecursiveFullyStatic))

static_assert(is_assignable_to(RecursiveNonFullyStatic, RecursiveNonFullyStatic))
static_assert(is_assignable_to(RecursiveFullyStatic, RecursiveNonFullyStatic))
static_assert(is_assignable_to(RecursiveNonFullyStatic, RecursiveFullyStatic))

class AlsoRecursiveFullyStatic(Protocol):
    parent: AlsoRecursiveFullyStatic
    x: int

static_assert(is_equivalent_to(AlsoRecursiveFullyStatic, RecursiveFullyStatic))

class RecursiveOptionalParent(Protocol):
    parent: RecursiveOptionalParent | None

static_assert(is_fully_static(RecursiveOptionalParent))

static_assert(is_assignable_to(RecursiveOptionalParent, RecursiveOptionalParent))

static_assert(is_assignable_to(RecursiveNonFullyStatic, RecursiveOptionalParent))
static_assert(not is_assignable_to(RecursiveOptionalParent, RecursiveNonFullyStatic))

class Other(Protocol):
    z: str

def _(rec: RecursiveFullyStatic, other: Other):
    reveal_type(rec.parent.parent.parent)  # revealed: RecursiveFullyStatic

    rec.parent.parent.parent = rec
    rec = rec.parent.parent.parent

    rec.parent.parent.parent = other  # error: [invalid-assignment]
    other = rec.parent.parent.parent  # error: [invalid-assignment]

class Foo(Protocol):
    @property
    def x(self) -> "Foo": ...

class Bar(Protocol):
    @property
    def x(self) -> "Bar": ...

# TODO: this should pass
# error: [static-assert-error]
static_assert(is_equivalent_to(Foo, Bar))

Nested occurrences of self-reference

Make sure that we handle self-reference correctly, even if the self-reference appears deeply nested within the type of a protocol member:

[environment]
python-version = "3.12"

from __future__ import annotations

from typing import Protocol, Callable
from ty_extensions import Intersection, Not, is_fully_static, is_assignable_to, is_equivalent_to, static_assert

class C: ...

class GenericC[T](Protocol):
    pass

class Recursive(Protocol):
    direct: Recursive

    union: None | Recursive

    intersection1: Intersection[C, Recursive]
    intersection2: Intersection[C, Not[Recursive]]

    t: tuple[int, tuple[str, Recursive]]

    callable1: Callable[[int], Recursive]
    callable2: Callable[[Recursive], int]

    subtype_of: type[Recursive]

    generic: GenericC[Recursive]

    def method(self, x: Recursive) -> Recursive: ...

    nested: Recursive | Callable[[Recursive | Recursive, tuple[Recursive, Recursive]], Recursive | Recursive]

static_assert(is_fully_static(Recursive))
static_assert(is_equivalent_to(Recursive, Recursive))
static_assert(is_assignable_to(Recursive, Recursive))

def _(r: Recursive):
    reveal_type(r.direct)  # revealed: Recursive
    reveal_type(r.union)  # revealed: None | Recursive
    reveal_type(r.intersection1)  # revealed: C & Recursive
    reveal_type(r.intersection2)  # revealed: C & ~Recursive
    reveal_type(r.t)  # revealed: tuple[int, tuple[str, Recursive]]
    reveal_type(r.callable1)  # revealed: (int, /) -> Recursive
    reveal_type(r.callable2)  # revealed: (Recursive, /) -> int
    reveal_type(r.subtype_of)  # revealed: type[Recursive]
    reveal_type(r.generic)  # revealed: GenericC[Recursive]
    reveal_type(r.method(r))  # revealed: Recursive
    reveal_type(r.nested)  # revealed: Recursive | ((Recursive, tuple[Recursive, Recursive], /) -> Recursive)

    reveal_type(r.method(r).callable1(1).direct.t[1][1])  # revealed: Recursive

Regression test: narrowing with self-referential protocols

This snippet caused us to panic on an early version of the implementation for protocols.

from typing import Protocol

class A(Protocol):
    def x(self) -> "B | A": ...

class B(Protocol):
    def y(self): ...

obj = something_unresolvable  # error: [unresolved-reference]
reveal_type(obj)  # revealed: Unknown
if isinstance(obj, (B, A)):
    reveal_type(obj)  # revealed: (Unknown & B) | (Unknown & A)

TODO

Add tests for:

• More tests for protocols inside type[]. Spec reference.
• Protocols with instance-method members, including:
  • Protocols with methods that have parameters or the return type unannotated
  • Protocols with methods that have parameters or the return type annotated with Any
• Protocols with @classmethod and @staticmethod
• Assignability of non-instance types to protocols with instance-method members (e.g. a class-literal type can be a subtype of Sized if its metaclass has a __len__ method)
• Protocols with methods that have annotated self parameters. Spec reference.
• Protocols with overloaded method members
• super() on nominal subtypes (explicit and implicit) of protocol classes
• Recursive protocols
• Generic protocols
• Non-generic protocols with function-scoped generic methods
• Protocols with instance attributes annotated with Callable (can a nominal type with a method satisfy that protocol, and if so in what cases?)
• Protocols decorated with @final
• Equivalence and subtyping between Callable types and protocols that define __call__