Fix overloads with a generic self: dropped when matching a Protocol

[jorenham]

Summary

filter_overloads_by_self_type dropped any overload whose self: mentions the overload's own type params, since is_subset_eq(receiver, self) checks against a rigid, unsolvable variable. Substitute those params with Any first, so the receiver is matched against a gradual self:.
An overload whose self: mentions its own type parameters is matched gradually by substituting those parameters with Any (self: Arr[S, T] -> Arr[Any, Any]), which is exact for the realistic single-occurrence case and only over-accepts the presumably niche degenerate repeated-parameter case, (self: C[Z, Z]), same as mypy and pyright.

Fixes #3974

Test Plan

Regression tests added

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

jax (https://github.com/google/jax)
+ ERROR jax/_src/lax/linalg.py:1066:74-78: Argument `list[signedinteger[_32Bit]]` is not assignable to parameter `broadcast_dimensions` with type `Sequence[int]` in function `jax._src.lax.lax.broadcast_in_dim` [bad-argument-type]
+ ERROR jax/_src/lax/linalg.py:1068:63-67: Argument `list[signedinteger[_32Bit]]` is not assignable to parameter `broadcast_dimensions` with type `Sequence[int]` in function `jax._src.lax.lax.broadcast_in_dim` [bad-argument-type]

scipy-stubs (https://github.com/scipy/scipy-stubs)
+ ERROR tests/sparse/test_csr.pyi:196:1-2: Unused `# pyrefly: ignore` comment for code(s): assert-type [unused-ignore]

DateType (https://github.com/glyph/DateType)
- ERROR src/datetype/test/test_datetype.py:49:38-54: `Time[None]` is not assignable to `NaiveTime` [bad-assignment]

colour (https://github.com/colour-science/colour)
- ERROR colour/temperature/cie_d.py:113:21-118:14: No matching overload found for function `scipy.optimize._minimize.minimize` called with arguments: ((CCT: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]], xy: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]) -> DTypeFloat, x0=list[int], args=tuple[Any], **dict[str, str | dict[str, float]]) [no-matching-overload]
+ ERROR colour/temperature/cie_d.py:113:21-118:14: No matching overload found for function `scipy.optimize._minimize.minimize` called with arguments: ((CCT: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]], xy: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]) -> DTypeFloat, x0=list[int], args=tuple[ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]], **dict[str, str | dict[str, float]]) [no-matching-overload]
- ERROR colour/temperature/hernandez1999.py:176:21-181:14: No matching overload found for function `scipy.optimize._minimize.minimize` called with arguments: ((xy: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]], CCT: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]) -> DTypeFloat, x0=Unknown, args=tuple[Any], **dict[str, str | dict[str, float]]) [no-matching-overload]
+ ERROR colour/temperature/hernandez1999.py:176:21-181:14: No matching overload found for function `scipy.optimize._minimize.minimize` called with arguments: ((xy: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]], CCT: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]) -> DTypeFloat, x0=Unknown, args=tuple[ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]], **dict[str, str | dict[str, float]]) [no-matching-overload]
- ERROR colour/temperature/kang2002.py:112:21-117:14: No matching overload found for function `scipy.optimize._minimize.minimize` called with arguments: ((CCT: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]], xy: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]) -> DTypeFloat, x0=list[int], args=tuple[Any], **dict[str, str | dict[str, float]]) [no-matching-overload]
+ ERROR colour/temperature/kang2002.py:112:21-117:14: No matching overload found for function `scipy.optimize._minimize.minimize` called with arguments: ((CCT: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]], xy: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]) -> DTypeFloat, x0=list[int], args=tuple[ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]], **dict[str, str | dict[str, float]]) [no-matching-overload]
- ERROR colour/temperature/krystek1985.py:117:21-122:14: No matching overload found for function `scipy.optimize._minimize.minimize` called with arguments: ((CCT: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]], uv: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]) -> DTypeFloat, x0=list[int], args=tuple[Any], **dict[str, str | dict[str, float]]) [no-matching-overload]
+ ERROR colour/temperature/krystek1985.py:117:21-122:14: No matching overload found for function `scipy.optimize._minimize.minimize` called with arguments: ((CCT: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]], uv: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]) -> DTypeFloat, x0=list[int], args=tuple[ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]], **dict[str, str | dict[str, float]]) [no-matching-overload]
- ERROR colour/temperature/mccamy1992.py:160:21-165:14: No matching overload found for function `scipy.optimize._minimize.minimize` called with arguments: ((xy: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]], CCT: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]) -> DTypeFloat, x0=Unknown, args=tuple[Any], **dict[str, str | dict[str, float]]) [no-matching-overload]
+ ERROR colour/temperature/mccamy1992.py:160:21-165:14: No matching overload found for function `scipy.optimize._minimize.minimize` called with arguments: ((xy: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]], CCT: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]) -> DTypeFloat, x0=Unknown, args=tuple[ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]], **dict[str, str | dict[str, float]]) [no-matching-overload]
- ERROR colour/temperature/planck1900.py:120:21-125:14: No matching overload found for function `scipy.optimize._minimize.minimize` called with arguments: ((CCT: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]], uv: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]) -> DTypeFloat, x0=list[int], args=tuple[Any], **dict[str, str | dict[str, float]]) [no-matching-overload]
+ ERROR colour/temperature/planck1900.py:120:21-125:14: No matching overload found for function `scipy.optimize._minimize.minimize` called with arguments: ((CCT: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]], uv: ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]) -> DTypeFloat, x0=list[int], args=tuple[ndarray[tuple[Any, ...], dtype[float64 | floating[_16Bit] | floating[_32Bit]]]], **dict[str, str | dict[str, float]]) [no-matching-overload]

[jorenham]

• jax (src): two new true positives are now reported (previously false negatives). list(ndarray) now infers the precise scalar element type (via ndarray.__iter__'s generic-self overload, previously dropped), correctly flagging a list[np.int32] passed where Sequence[int] is expected.
• scipy-stubs (src): this was indeed a false positive that is now resolved.
• DateType (src): also looks like a resovled false positive that (they seem to be using a lot of self: Self in overloads, and there's a protocol involved here).
• colour: all that changed here are the reported error messages; so +0

So all primer changes are good news.

[NathanTempest]

Approving and thanks, @jorenham for the clean fix.

filter_overloads_by_self_type was matching the receiver against a self (still-unbound type params) so is_subset_eq compared against rigid unsolvable variables and silently dropped overloads that should apply. Substituting just those params with implicit Any before the applicability check is the right direction here.

Only the overload's own tparams are gradualized but the concrete parts of the self: are preserved The Any-substituted type is only used for the keep/drop decision; the surviving overload keeps its real self: Arr[S2, T2] -> Arr[S2, T2], which is then solved precisely against the receiver. So this increases precision rather than loosening it.

[meta-codesync]

@NathanTempest has imported this pull request. If you are a Meta employee, you can view this in D111462642.