Describe the Bug
After bumping Pyrefly from 1.0 to 1.1 in optype, a bunch of type-tests started failing for the numpy<2.2 jobs (see e.g. https://github.com/jorenham/optype/actions/runs/27886459886/job/82522566360). This also broadly affects scipy-stubs, which heavily depends on these relevant optype types. But I'm guessing that this also causes issues for other projects that support these older NumPy versions.
When an overloaded method has an overload whose self is annotated with a TypeVar, Pyrefly 1.1 drops that overload when the method is bound on an instance. With it gone, the type no longer satisfies any Protocol that structurally needs it:
from typing import Any, Protocol, Self, overload
class C1:
@overload
def f[S](self, /) -> Self: ...
@overload
def f(self, x: int, /) -> int: ...
class C2:
@overload
def f[S](self: S, /) -> S: ...
@overload
def f(self, x: int, /) -> int: ...
class P(Protocol):
def f(self, /) -> object: ...
x: P = C1() # ✔️
x: P = C2() # ❌ `C2` is not assignable to `P` [bad-assignment]The error message shows that first overload disappeared:
ERROR sandbox.pyi:19:8-12: `C2` is not assignable to `P` [[bad-assignment](https://pyrefly.org/en/docs/error-kinds/#bad-assignment)]
`C2.f` has type `Overload[
(self: C2, x: int, /) -> int
]`, which is not assignable to `(self: C2, /) -> object`, the type of `P.f`
And indeed, without that overload, C2 would be assignable to P.
This doesn't apply to non-overloaded methods, so e.g. def f[S](self: S) -> S is fine.
It also seems to be order-independent, so swapping the overloads doesn't change anything.
I verified that this is still the case on main.
Oh and the reason this only affects numpy<2.2 is because, starting from 2.2, numpy.generic.__array__ uses Self instead of a self-bound TypeVar.
Related (but verified to be different) issues:
Sandbox Link
https://pyrefly.org/sandbox/?project=N4IgZglgNgpgziAXKOBDAdgEwEYHsAeAdAA4CeSIIANCGQE4xhSmEAuuAtlBQMQAEABVIMmpPmix58fAMa50kAOYBXOqlYR5hADrp%2BAZRgw%2BAC1aticRAHpr9Rs0K46i6zHTXMuGXGtyFECpqGvLWfGDOfKgAbqjQqNiwsvJKquqa6Hy4xCHocDroZKwm8gC00TB0cBl8ALx82iAAzIQAjABMjboA2pV0znAAurrK6BAcxM6sMJilmBAMMhoVdXwA5IHozjBruluspQwAjsoLM6UA1jCkpagyMvBwq2sA7qh06Lvo1CASOAQkUgQChgfocPisUjECDoRR8caTOisPgAQXQpCogn67DkUExhigYExuAqdCguFQmF0uhkUFQcCeAGFWohdHx2XwAAIkyrkylsjmYRjhbr6QYACjgMEJmOsAEo%2BKUAHx8AlgRB8QhagXs7mkvlUzKC4VgSXSol8fAamGsWUK5Xw9CsDVago0ukMviM9qso26nlkimGjl8IVgEVis2EjX6O2KlX6F3av1cgMGnWhk1Ri1Wx22vjy%2BN5pNu9C0%2BlPATigTY7y4KBy30hsPhbNxh24bAAKxgSxL1PQuYEq2Z4rluiHI-aY90IAAvjQ7ssYAAxaAwCh-KSA4FzoA
(Only applicable for extension issues) IDE Information
No response
Describe the Bug
After bumping Pyrefly from
1.0to1.1in optype, a bunch of type-tests started failing for thenumpy<2.2jobs (see e.g. https://github.com/jorenham/optype/actions/runs/27886459886/job/82522566360). This also broadly affects scipy-stubs, which heavily depends on these relevant optype types. But I'm guessing that this also causes issues for other projects that support these older NumPy versions.When an overloaded method has an overload whose
selfis annotated with aTypeVar, Pyrefly 1.1 drops that overload when the method is bound on an instance. With it gone, the type no longer satisfies anyProtocolthat structurally needs it:The error message shows that first overload disappeared:
And indeed, without that overload,
C2would be assignable toP.This doesn't apply to non-overloaded methods, so e.g.
def f[S](self: S) -> Sis fine.It also seems to be order-independent, so swapping the overloads doesn't change anything.
I verified that this is still the case on
main.Oh and the reason this only affects
numpy<2.2is because, starting from 2.2,numpy.generic.__array__usesSelfinstead of a self-boundTypeVar.Related (but verified to be different) issues:
__new__or__call__overloads #3878Sandbox Link
https://pyrefly.org/sandbox/?project=N4IgZglgNgpgziAXKOBDAdgEwEYHsAeAdAA4CeSIIANCGQE4xhSmEAuuAtlBQMQAEABVIMmpPmix58fAMa50kAOYBXOqlYR5hADrp%2BAZRgw%2BAC1aticRAHpr9Rs0K46i6zHTXMuGXGtyFECpqGvLWfGDOfKgAbqjQqNiwsvJKquqa6Hy4xCHocDroZKwm8gC00TB0cBl8ALx82iAAzIQAjABMjboA2pV0znAAurrK6BAcxM6sMJilmBAMMhoVdXwA5IHozjBruluspQwAjsoLM6UA1jCkpagyMvBwq2sA7qh06Lvo1CASOAQkUgQChgfocPisUjECDoRR8caTOisPgAQXQpCogn67DkUExhigYExuAqdCguFQmF0uhkUFQcCeAGFWohdHx2XwAAIkyrkylsjmYRjhbr6QYACjgMEJmOsAEo%2BKUAHx8AlgRB8QhagXs7mkvlUzKC4VgSXSol8fAamGsWUK5Xw9CsDVago0ukMviM9qso26nlkimGjl8IVgEVis2EjX6O2KlX6F3av1cgMGnWhk1Ri1Wx22vjy%2BN5pNu9C0%2BlPATigTY7y4KBy30hsPhbNxh24bAAKxgSxL1PQuYEq2Z4rluiHI-aY90IAAvjQ7ssYAAxaAwCh-KSA4FzoA
(Only applicable for extension issues) IDE Information
No response