Daily property test run failed on Tue May 20 2025

[github-actions]

Run listed here: https://github.com/astral-sh/ty/actions/runs/15136976142

[AlexWaygood]

Minimal repro:

from ty_extensions import static_assert, is_equivalent_to, TypeOf

static_assert(is_equivalent_to(TypeOf[int.bit_length], TypeOf[int.bit_length]))

The assertion now fails on main

The definition for int.bit_length() is about as simple as it gets:

https://github.com/astral-sh/ruff/blob/60b486abce4e756fbd26a586a03d551ba44b9ea8/crates/ty_vendored/vendor/typeshed/stdlib/builtins.pyi#L253

[AlexWaygood]

(Confirmed that this passed before astral-sh/ruff@97058e8 but fails on that commit)

[AlexWaygood]

This also fails, which I think confirms that we do not consider the type equivalent to itself anymore:

type X = TypeOf[int.bit_length]

static_assert(is_equivalent_to(X, X))

[AlexWaygood]

Even simpler:

from ty_extensions import is_equivalent_to, TypeOf

def f(): ...

reveal_type(is_equivalent_to(TypeOf[f], TypeOf[f]))  # revealed: Literal[False] ??

[dcreager]

@carljm, @AlexWaygood and I just discussed this at the PyCon sprint. The issue is that as of astral-sh/ruff#18155, we take the parameter/return type annotations into account when comparing function literals for subtyping/equivalence. If an annotation is not given, we fall back on Unknown, which is gradual. But I did not also update is_fully_static to take this into account. So the subtyping/equivalence checks would return False when encountering the unannotated parameter/return in.

For now we are going to fix the glitch by updating is_fully_static to be consistent with the other methods. A longer term fix is to have separate representations for function literals and specialized functions, similar to the distinction we already make between ClassLiteral and ClassType.