Missing reflected dunder methods from AbstractSet

[meshy]

Problem

Reflected operations (__rand__, __ror__, __rsub__, __rxor__) are missing from AbstractSet and its subclasses.

These are noted as explicit exclusions from tests:

typeshed/stdlib/@tests/stubtest_allowlists/common.txt

Lines 50 to 54 in 9f8f621

	# Adding these reflected dunders to `typing.AbstractSet` causes a large number of false-positives. See #7414.
	_collections_abc.Set.__rand__
	_collections_abc.Set.__ror__
	_collections_abc.Set.__rsub__
	_collections_abc.Set.__rxor__

Example of consequences

@samueljsb and I have defined a protocol (in another project) for an object which has an __rsub__ method, and discovered that set, despite implementing this method, does not adhere to the protocol.

A simplified illustration of this issue is:

from typing import Any, Protocol

class Subtractable(Protocol):
    def __rsub__(self, other: Any) -> Any: ...

_: Subtractable = set()

$ mypy t.py
t.py:6: error: Incompatible types in assignment (expression has type "set[Never]", variable has type "Subtractable")  [assignment]
Found 1 error in 1 file (checked 1 source file)

Demonstrated with:

• mypy 1.19.1 (compiled: yes)
• Python 3.14.2 (main, Dec 5 2025, 16:49:16) [Clang 17.0.0 (clang-1700.4.4.1)]

Investigation

The exclusions reference #7414, which was an attempt to add type hints for them. It was closed without merging because it encountered errors. We believe it encountered errors because the type signatures were wrong. Specifically, it didn't account for the return types of the operations matching the inputs.

For example, instead of:

def __rsub__(self, s: AbstractSet[_T]) -> AbstractSet[_T]: ...

We believe the signature needs to be (in the new syntax):

def __rsub__[T_Set: AbstractSet](self, other: T_Set) -> T_Set: ...

The above can be applied to __rsub__ and __rand__, because those methods return a subset of the input, and therefore will not change its type.

__ror__ and __rxor__ are more complicated because they return a type which is a superset of both inputs. We haven't yet looked into how to solve those.