Inconsistencies between `__init__` and `__new__`

[erictraut]

I just implemented a new check in pyright called reportInconsistentConstructor that detects inconsistencies between the __init__ and __new__ methods. It's disabled by default but can be enabled in the configuration. It's in pyright 1.1.207, which I just published.

The check is pretty conservative, but it does identify three places in typeshed that look like legitimate problems.

1. The class WeakMethod in weakref.pyi defines a __new__ that is incompatible with the __init__ declared in its parent class ReferenceType.
2. The class _TokenType in pygments/token.pyi defines an __init__ that is incompatible with the __new__ in tuple[str].
3. The class CascadeOptions in SQLAlchemy/sqlalchemy/orm/util.pyi defines a __new__ that is incompatible with the __init__ in its parent class frozenset[Any].

If/when these are addressed, we might want to update pyright to 1.1.207 and enable the new diagnostic check to prevent future inconsistencies from being introduced.

[Akuli]

For problem 2, I think we should just delete _TokenType.__init__. You aren't supposed to instantiate _TokenType yourself anyway.

The implementation is:

class _TokenType(tuple):
    ...
    def __init__(self, *args):
        # no need to call super.__init__
        self.subtypes = set()

[srittau]

Sounds like a good plan.

[AlexWaygood]

Relatedly (but slightly tangentially), can anybody explain to me why classmethod.__new__ and classmethod.__init__ have different signatures in builtins? I've never really understood why, and there's no comment clarifying it:

typeshed/stdlib/builtins.pyi

Lines 117 to 118 in 5d20635

	def __init__(self: staticmethod[_R], __f: Callable[..., _R]) -> None: ...
	def __new__(cls: type[Self], *args: Any, **kwargs: Any) -> Self: ...

(Same issue for staticmethod.)