✨ stats: casually invent a way to do overload attributes (it even w…

…orks on mypy)

scipy-stubs/stats/_distribution_infrastructure.pyi

@@ -2,8 +2,8 @@
 # pyright: reportUnannotatedClassAttribute=false
 
 import abc
-from collections.abc import Mapping, Sequence, Set as AbstractSet
-from typing import Any, ClassVar, Final, Generic, Literal as L, TypeAlias, overload
+from collections.abc import Callable, Mapping, Sequence, Set as AbstractSet
+from typing import Any, ClassVar, Final, Generic, Literal as L, Protocol, TypeAlias, overload, type_check_only
 from typing_extensions import LiteralString, Never, Self, TypeVar, override
 
 import numpy as np
@@ -61,6 +61,24 @@ _DrawProportions: TypeAlias = tuple[onp.ToFloat, onp.ToFloat, onp.ToFloat, onp.T
 _CDist: TypeAlias = ContinuousDistribution[np.floating[Any], _ShapeT0]
 _CDist0: TypeAlias = ContinuousDistribution[_FloatingT, tuple[()]]
 
+@type_check_only
+class _ParameterField(Protocol[_FloatingT_co, _ShapeT0_co]):
+    # This actually works (even on mypy)!
+    @overload
+    def __get__(
+        self: _ParameterField[_FloatingT, tuple[()]],
+        instance: object,
+        owner: type | None = None,
+        /,
+    ) -> _FloatingT: ...
+    @overload
+    def __get__(
+        self: _ParameterField[_FloatingT, _ShapeT1],
+        instance: object,
+        owner: type | None = None,
+        /,
+    ) -> onp.ArrayND[_FloatingT, _ShapeT1]: ...
+
 ###
 
 _null: Final[_Null] = ...
@@ -233,7 +251,7 @@ class ContinuousDistribution(_BaseDistribution[_FloatingT_co, _ShapeT0_co], Gene
     def __sub__(self, lshift: onp.ToFloat, /) -> ShiftedScaledDistribution[Self, _FloatingT_co, _ShapeT0_co]: ...
     def __mul__(self, scale: onp.ToFloat, /) -> ShiftedScaledDistribution[Self, _FloatingT_co, _ShapeT0_co]: ...
     def __truediv__(self, iscale: onp.ToFloat, /) -> ShiftedScaledDistribution[Self, _FloatingT_co, _ShapeT0_co]: ...
-    def __pow__(self, exp: onp.ToInt, /) -> MonotonicTransformedDistribution[Self, _FloatingT_co, _ShapeT0_co]: ...
+    def __pow__(self, exp: onp.ToInt, /) -> MonotonicTransformedDistribution[Self, _ShapeT0_co]: ...
     __radd__ = __add__
     __rsub__ = __sub__
     __rmul__ = __mul__
@@ -284,7 +302,7 @@ class ContinuousDistribution(_BaseDistribution[_FloatingT_co, _ShapeT0_co], Gene
     @overload
     def llf(self, sample: onp.ToFloat | onp.ToFloatND, /, *, axis: AnyShape | None = -1) -> _Float | onp.ArrayND[_Float]: ...
 
-    #
+_ElementwiseFunction: TypeAlias = Callable[[onp.ArrayND[np.float64]], onp.ArrayND[_FloatingT]]
 
 # 7 years of asking and >400 upvotes, but still no higher-kinded typing support: https://github.com/python/typing/issues/548
 class TransformedDistribution(
@@ -302,18 +320,38 @@ class TransformedDistribution(
     ) -> None: ...
 
 class MonotonicTransformedDistribution(
-    TransformedDistribution[_CDistT_co, _FloatingT_co, _ShapeT0_co],
-    Generic[_CDistT_co, _FloatingT_co, _ShapeT0_co],
+    TransformedDistribution[_CDistT_co, np.float64, _ShapeT0_co],
+    Generic[_CDistT_co, _ShapeT0_co],
 ):
-    # TODO(jorenham)
-    ...
+    _g: Final[_ElementwiseFunction]
+    _h: Final[_ElementwiseFunction]
+    _dh: Final[_ElementwiseFunction]
+    _logdh: Final[_ElementwiseFunction]
+    _increasing: Final[bool]
+    _repr_pattern: Final[str]
+
+    def __init__(
+        self: MonotonicTransformedDistribution[_CDist[_ShapeT0], _ShapeT0],
+        X: _CDistT_co,
+        /,
+        *args: Never,
+        g: _ElementwiseFunction,
+        h: _ElementwiseFunction,
+        dh: _ElementwiseFunction,
+        logdh: _ElementwiseFunction | None = None,
+        increasing: bool = True,
+        repr_pattern: str | None = None,
+        tol: opt.Just[float] | _Null = ...,
+        validation_policy: _ValidationPolicy = None,
+        cache_policy: _CachePolicy = None,
+    ) -> None: ...
 
 class TruncatedDistribution(
     TransformedDistribution[_CDistT_co, _FloatingT_co, _ShapeT0_co],
     Generic[_CDistT_co, _FloatingT_co, _ShapeT0_co],
 ):
-    lb: _FloatingT_co | onp.ArrayND[_FloatingT_co, _ShapeT0_co]
-    ub: _FloatingT_co | onp.ArrayND[_FloatingT_co, _ShapeT0_co]
+    lb: _ParameterField[_FloatingT_co, _ShapeT0_co]
+    ub: _ParameterField[_FloatingT_co, _ShapeT0_co]
 
     @overload
     def __init__(