Fix invalid contextual inference for generic call arguments

[rheard]

Fixes:

• Assigning to intermediate variable changes type checking results #19304
• Inferred attribute type feeds back as call context, rejecting its own defining assignment #21809
• error: "key" to "max" has incompatible type #21104
• Parameter TypeVar unexpectedly becomes optional when unpacking call results #20219
• Argument "key" to "sorted" has incompatible type "Callable…" #19259
• bogus complaint when checking Callable arguments #18689
• mypy incorrectly complains about incompatible type error #18267
• False positive [operator] on union of dicts with heterogeneous value types #18236
• functools.reduce over sets becomes unacceptable when used in a larger expression #17694
• Function's argument type inference is affected by the functions return type when in a comprehension #17025
• MyPy reports errors when performing nested element access using a TypeVar-parameterized generic type #16953
• Generic type inferred wrong; curious difference betweer aiter(f(x)) and c = f(x); aiter(c) #16376
• Generic matching error when tuple-unpacking result #16310
• set.__and__ false-positive involving generic inference #15462
• Spurious error when adding two list #15087
• error in set unions #12760
• False positive for function that takes a Callable as a parameter and returns a Tuple of Union #10281
• Changing just the return type annotation causes incompatible argument error #10105
• Collapsed Optionals in TypeVar cause weird error messages #8478
• Covariant generic type context results in confusing message with __add__ #5971
• Mypy is still too eager about outer context #5874
• error in argument when using chained function-call #5311
• Apparent false positive when using reduce with dictionaries #4150

Mypy can use the outer return context to specialize a generic callable before it has checked whether the actual arguments still fit the specialized signature.

This can produce false positives when the contextual type is wider than the actual argument supports. For example, in the issue repro the outer Iterable[int | str] context pushes Vec.__add__ toward accepting Vec[int | str], but the actual argument is Vec[int], and Vec is invariant.

This PR keeps the existing contextual inference path, but adds a guard before committing to the context-specialized callable. If the specialized argument types are not compatible with the actual arguments, mypy falls back to ordinary argument inference, but only when that ordinary inferred return type still satisfies the original outer context.

The fallback is intentionally narrow. Constructors and special synthetic signatures keep the old behavior, since broader fallback caused regressions in existing tests, including testAbstractTypeInADict and constructor/type-alias cases that started reporting Never-based expected types.

Regression tests cover:

• invariant generic argument widening through Vec.__add__
• a min(..., key=...)-style callback case where return context widens the item type too far

[randolf-scholz]

You may want to check the results in sterliakov/mypy-issues#309

Your PR could probably fix a bunch of other issues like my attempt did back then (randolf-scholz#4). Though your approach seems much simpler. Best of luck!

[github-actions]

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

kornia (https://github.com/kornia/kornia)
+ kornia/models/efficient_vit/nn/act.py:36: error: Unused "type: ignore" comment  [unused-ignore]

spark (https://github.com/apache/spark)
+ python/pyspark/ml/util.py:1169: error: Unused "type: ignore" comment  [unused-ignore]
+ python/pyspark/ml/classification.py:3990: error: Unused "type: ignore" comment  [unused-ignore]

prefect (https://github.com/PrefectHQ/prefect)
- src/prefect/utilities/callables/__init__.py:717: error: Incompatible types in assignment (expression has type "list[None]", variable has type "list[expr | None]")  [assignment]
- src/prefect/utilities/callables/__init__.py:717: note: "list" is invariant -- see https://mypy.readthedocs.io/en/stable/common_issues.html#variance
- src/prefect/utilities/callables/__init__.py:717: note: Consider using "Sequence" instead, which is covariant
- src/prefect/utilities/callables/__init__.py:719: error: Unsupported operand types for + ("list[None]" and "list[expr | None]")  [operator]
+ src/prefect/server/events/jinja_filters.py:86: error: Unused "type: ignore" comment  [unused-ignore]
- src/prefect/task_runners.py:425: error: Argument 1 to "submit" of "Executor" has incompatible type "Callable[[Callable[_P, _T], **_P], _T]"; expected "def (Callable[[Task[P, R], UUID | None, TaskRun | None, dict[str, Any] | None, PrefectFuture[Any] | Any | Iterable[PrefectFuture[Any] | Any] | None, Literal['state', 'result'], dict[str, set[RunInput]] | None, dict[str, Any] | None], R | State[Any] | None], /, **Any) -> Any | State[Any] | None"  [arg-type]
+ src/prefect/task_runners.py:425: error: Argument 1 to "submit" of "Executor" has incompatible type "Callable[[Callable[_P, _T], **_P], _T]"; expected "def [P] (Callable[[Task[P, R], UUID | None, TaskRun | None, dict[str, Any] | None, PrefectFuture[Any] | Any | Iterable[PrefectFuture[Any] | Any] | None, Literal['state', 'result'], dict[str, set[RunInput]] | None, dict[str, Any] | None], R | State[Any] | None], /, **Any) -> Any | State[Any] | None"  [arg-type]
- src/prefect/flow_engine.py:2338: error: Unsupported operand types for | ("dict[str, str]" and "dict[str, str | None]")  [operator]

colour (https://github.com/colour-science/colour)
- colour/io/luts/lut.py:1419: error: Argument 1 to "tile" has incompatible type "Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str] | None"; expected "_SupportsArray[dtype[signedinteger[_8Bit] | signedinteger[_16Bit] | signedinteger[_32Bit] | signedinteger[_64Bit] | unsignedinteger[_8Bit] | unsignedinteger[_16Bit] | unsignedinteger[_32Bit] | unsignedinteger[_64Bit] | Any]] | _NestedSequence[_SupportsArray[dtype[signedinteger[_8Bit] | signedinteger[_16Bit] | signedinteger[_32Bit] | signedinteger[_64Bit] | unsignedinteger[_8Bit] | unsignedinteger[_16Bit] | unsignedinteger[_32Bit] | unsignedinteger[_64Bit] | Any]]]"  [arg-type]
+ colour/io/luts/lut.py:1433: error: No overload variant of "pad" matches argument types "ndarray[Any, Any]", "tuple[int, Any | signedinteger[_8Bit] | signedinteger[_16Bit] | signedinteger[_32Bit] | signedinteger[_64Bit] | unsignedinteger[_8Bit] | unsignedinteger[_16Bit] | unsignedinteger[_32Bit] | unsignedinteger[_64Bit]]", "str", "float"  [call-overload]
+ colour/io/luts/lut.py:1433: note: Possible overload variants:
+ colour/io/luts/lut.py:1433: note:     def [ShapeT: tuple[int, ...], DTypeT: dtype[Any]] pad(array: ndarray[ShapeT, DTypeT], pad_width: _SupportsArray[dtype[integer[Any]]] | _NestedSequence[_SupportsArray[dtype[integer[Any]]]] | int | _NestedSequence[int] | dict[int, int] | dict[int, tuple[int, int]] | dict[int, int | tuple[int, int]], mode: Literal['constant', 'edge', 'linear_ramp', 'maximum', 'mean', 'median', 'minimum', 'reflect', 'symmetric', 'wrap', 'empty'] = ..., *, stat_length: _SupportsArray[dtype[integer[Any]]] | _NestedSequence[_SupportsArray[dtype[integer[Any]]]] | int | _NestedSequence[int] | None = ..., constant_values: Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str] = ..., end_values: Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str] = ..., reflect_type: Literal['odd', 'even'] = ...) -> ndarray[ShapeT, DTypeT]
+ colour/io/luts/lut.py:1433: note:     def [ScalarT: generic[Any]] pad(array: _SupportsArray[dtype[ScalarT]] | _NestedSequence[_SupportsArray[dtype[ScalarT]]], pad_width: _SupportsArray[dtype[integer[Any]]] | _NestedSequence[_SupportsArray[dtype[integer[Any]]]] | int | _NestedSequence[int] | dict[int, int] | dict[int, tuple[int, int]] | dict[int, int | tuple[int, int]], mode: Literal['constant', 'edge', 'linear_ramp', 'maximum', 'mean', 'median', 'minimum', 'reflect', 'symmetric', 'wrap', 'empty'] = ..., *, stat_length: _SupportsArray[dtype[integer[Any]]] | _NestedSequence[_SupportsArray[dtype[integer[Any]]]] | int | _NestedSequence[int] | None = ..., constant_values: Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str] = ..., end_values: Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str] = ..., reflect_type: Literal['odd', 'even'] = ...) -> ndarray[tuple[Any, ...], dtype[ScalarT]]
+ colour/io/luts/lut.py:1433: note:     def pad(array: Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str], pad_width: _SupportsArray[dtype[integer[Any]]] | _NestedSequence[_SupportsArray[dtype[integer[Any]]]] | int | _NestedSequence[int] | dict[int, int] | dict[int, tuple[int, int]] | dict[int, int | tuple[int, int]], mode: Literal['constant', 'edge', 'linear_ramp', 'maximum', 'mean', 'median', 'minimum', 'reflect', 'symmetric', 'wrap', 'empty'] = ..., *, stat_length: _SupportsArray[dtype[integer[Any]]] | _NestedSequence[_SupportsArray[dtype[integer[Any]]]] | int | _NestedSequence[int] | None = ..., constant_values: Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str] = ..., end_values: Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str] = ..., reflect_type: Literal['odd', 'even'] = ...) -> ndarray[tuple[Any, ...], dtype[Any]]
+ colour/io/luts/lut.py:1433: note:     def [ShapeT: tuple[int, ...], DTypeT: dtype[Any]] pad(array: ndarray[ShapeT, DTypeT], pad_width: _SupportsArray[dtype[integer[Any]]] | _NestedSequence[_SupportsArray[dtype[integer[Any]]]] | int | _NestedSequence[int] | dict[int, int] | dict[int, tuple[int, int]] | dict[int, int | tuple[int, int]], mode: _ModeFunc, **kwargs: Any) -> ndarray[ShapeT, DTypeT]
+ colour/io/luts/lut.py:1433: note:     def [ScalarT: generic[Any]] pad(array: _SupportsArray[dtype[ScalarT]] | _NestedSequence[_SupportsArray[dtype[ScalarT]]], pad_width: _SupportsArray[dtype[integer[Any]]] | _NestedSequence[_SupportsArray[dtype[integer[Any]]]] | int | _NestedSequence[int] | dict[int, int] | dict[int, tuple[int, int]] | dict[int, int | tuple[int, int]], mode: _ModeFunc, **kwargs: Any) -> ndarray[tuple[Any, ...], dtype[ScalarT]]
+ colour/io/luts/lut.py:1433: note:     def pad(array: Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str], pad_width: _SupportsArray[dtype[integer[Any]]] | _NestedSequence[_SupportsArray[dtype[integer[Any]]]] | int | _NestedSequence[int] | dict[int, int] | dict[int, tuple[int, int]] | dict[int, int | tuple[int, int]], mode: _ModeFunc, **kwargs: Any) -> ndarray[tuple[Any, ...], dtype[Any]]
- colour/io/luts/lut.py:1955: error: Argument 1 to "tile" has incompatible type "Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str] | None"; expected "_SupportsArray[dtype[signedinteger[_8Bit] | signedinteger[_16Bit] | signedinteger[_32Bit] | signedinteger[_64Bit] | unsignedinteger[_8Bit] | unsignedinteger[_16Bit] | unsignedinteger[_32Bit] | unsignedinteger[_64Bit]]] | _NestedSequence[_SupportsArray[dtype[signedinteger[_8Bit] | signedinteger[_16Bit] | signedinteger[_32Bit] | signedinteger[_64Bit] | unsignedinteger[_8Bit] | unsignedinteger[_16Bit] | unsignedinteger[_32Bit] | unsignedinteger[_64Bit]]]]"  [arg-type]
+ colour/io/luts/lut.py:1955: error: Argument 1 to "tile" has incompatible type "Buffer | _SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]] | complex | bytes | str | _NestedSequence[complex | bytes | str] | None"; expected "_SupportsArray[dtype[Any]] | _NestedSequence[_SupportsArray[dtype[Any]]]"  [arg-type]

pandera (https://github.com/pandera-dev/pandera)
- tests/ibis/test_ibis_dtypes.py:81: error: Unsupported operand types for + ("list[type[String] | type[_PhysicalNumber]]" and "list[type[Map]]")  [operator]
- tests/ibis/test_ibis_dtypes.py:92: error: Unsupported operand types for + ("list[type[String] | type[_PhysicalNumber]]" and "list[type[Map]]")  [operator]

artigraph (https://github.com/artigraph/artigraph)
+ src/arti/graphs/__init__.py:181: error: Unused "type: ignore" comment  [unused-ignore]

pydantic (https://github.com/pydantic/pydantic)
- pydantic/main.py:126: error: Dict entry 1 has incompatible type "str": "Callable[[BaseModel, str, Any], dict[str, Any] | tuple[dict[str, Any], dict[str, Any] | None, set[str]] | Any]"; expected "str": "Callable[[BaseModel, str, Any], None]"  [dict-item]

rotki (https://github.com/rotki/rotki)
+ rotkehlchen/chain/evm/decoding/stakedao/decoder.py:316: error: Unused "type: ignore" comment  [unused-ignore]
+ rotkehlchen/tasks/historical_balances.py:184: error: Incompatible return value type (got "list[tuple[Bucket, EventDirection]]", expected "list[tuple[Bucket, Literal[EventDirection.IN, EventDirection.OUT]]]")  [return-value]
+ rotkehlchen/tasks/historical_balances.py:185: error: Unused "type: ignore" comment  [unused-ignore]
- rotkehlchen/tests/api/test_calendar.py:156: error: Unsupported operand types for | ("dict[str, list[dict[str, str]]]" and "dict[str, int]")  [operator]
- rotkehlchen/tests/api/test_calendar.py:168: error: Unsupported operand types for | ("dict[str, str]" and "dict[str, int]")  [operator]
- rotkehlchen/tests/api/test_calendar.py:180: error: Unsupported operand types for | ("dict[str, str]" and "dict[str, int]")  [operator]
- rotkehlchen/tests/api/test_calendar.py:192: error: Unsupported operand types for | ("dict[str, str]" and "dict[str, int]")  [operator]
- rotkehlchen/tests/api/test_calendar.py:229: error: Unsupported operand types for | ("dict[str, list[dict[str, ChecksumAddress]]]" and "dict[str, int]")  [operator]

scipy (https://github.com/scipy/scipy)
- scipy/stats/tests/test_sampling.py:155: error: Unsupported operand types for + ("list[tuple[tuple[int, ...], type[ValueError], str] | tuple[tuple[int, int], type[UNURANError], str]]" and "list[tuple[tuple[float, float], type[ValueError], str]]")  [operator]

sympy (https://github.com/sympy/sympy)
+ sympy/core/add.py:396: error: Unused "type: ignore" comment  [unused-ignore]
+ sympy/polys/compatibility.py:402: error: Unused "type: ignore" comment  [unused-ignore]

scikit-learn (https://github.com/scikit-learn/scikit-learn)
- sklearn/preprocessing/tests/test_polynomial.py:630: error: Unsupported operand types for + ("list[ABCMeta | None]" and "list[ABCMeta]")  [operator]
- sklearn/preprocessing/tests/test_polynomial.py:701: error: Unsupported operand types for + ("list[ABCMeta | None]" and "list[ABCMeta]")  [operator]
- sklearn/preprocessing/tests/test_function_transformer.py:124: error: Unsupported operand types for + ("list[ABCMeta | None]" and "list[ABCMeta]")  [operator]
- sklearn/metrics/tests/test_pairwise.py:946: error: Unsupported operand types for + ("list[overloaded function]" and "list[ABCMeta]")  [operator]
- sklearn/metrics/tests/test_pairwise.py:951: error: Unsupported operand types for + ("list[overloaded function]" and "list[ABCMeta]")  [operator]
- sklearn/metrics/tests/test_pairwise.py:964: error: Unsupported operand types for + ("list[overloaded function]" and "list[ABCMeta]")  [operator]
- sklearn/metrics/tests/test_pairwise.py:1052: error: Unsupported operand types for + ("list[overloaded function]" and "list[ABCMeta]")  [operator]
- sklearn/metrics/tests/test_pairwise.py:1057: error: Unsupported operand types for + ("list[overloaded function]" and "list[ABCMeta]")  [operator]
- sklearn/metrics/tests/test_pairwise.py:1083: error: Unsupported operand types for + ("list[overloaded function]" and "list[ABCMeta]")  [operator]
- sklearn/metrics/tests/test_pairwise.py:1107: error: Unsupported operand types for + ("list[overloaded function]" and "list[ABCMeta]")  [operator]
- sklearn/metrics/tests/test_pairwise.py:1112: error: Unsupported operand types for + ("list[overloaded function]" and "list[ABCMeta]")  [operator]
- sklearn/metrics/tests/test_pairwise.py:1138: error: Unsupported operand types for + ("list[overloaded function]" and "list[ABCMeta]")  [operator]
- sklearn/_loss/tests/test_loss.py:224: error: Unsupported operand types for + ("list[tuple[BaseLoss, list[float], list[float]]]" and "list[tuple[BaseLoss, object, list[Any]]]")  [operator]
- sklearn/_loss/tests/test_loss.py:236: error: Unsupported operand types for + ("list[tuple[BaseLoss, list[float], list[float]]]" and "list[tuple[BaseLoss, list[float], object]]")  [operator]
- sklearn/feature_selection/tests/test_variance_threshold.py:14: error: Unsupported operand types for + ("list[ABCMeta | None]" and "list[ABCMeta]")  [operator]
- sklearn/feature_selection/tests/test_variance_threshold.py:31: error: Unsupported operand types for + ("list[None]" and "list[ABCMeta]")  [operator]
- sklearn/feature_selection/tests/test_variance_threshold.py:47: error: Unsupported operand types for + ("list[ABCMeta | None]" and "list[ABCMeta]")  [operator]
- sklearn/feature_selection/tests/test_variance_threshold.py:61: error: Unsupported operand types for + ("list[ABCMeta | None]" and "list[ABCMeta]")  [operator]
- sklearn/utils/tests/test_sparsefuncs.py:451: error: No overload variant of "hstack" matches argument types "list[Any]", "str"  [call-overload]
- sklearn/utils/tests/test_sparsefuncs.py:451: note: Possible overload variants:
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def [T] hstack(blocks: Sequence[_CanStack[T]], format: None = ..., dtype: None = ...) -> T
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def [ScalarT: number[Any, Any] | numpy.bool[builtins.bool]] hstack(blocks: Sequence[sparray[ScalarT, tuple[int, int]]], format: Literal['bsr'], dtype: None = ...) -> bsr_array[ScalarT]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def [ScalarT: number[Any, Any] | numpy.bool[builtins.bool]] hstack(blocks: Sequence[sparray[ScalarT, tuple[int, int]]], format: Literal['coo'], dtype: None = ...) -> coo_array[ScalarT, tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def [ScalarT: number[Any, Any] | numpy.bool[builtins.bool]] hstack(blocks: Sequence[sparray[ScalarT, tuple[int, int]]], format: Literal['csc'], dtype: None = ...) -> csc_array[ScalarT]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def [ScalarT: number[Any, Any] | numpy.bool[builtins.bool]] hstack(blocks: Sequence[sparray[ScalarT, tuple[int, int]]], format: Literal['csr'], dtype: None = ...) -> csr_array[ScalarT, tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def [ScalarT: number[Any, Any] | numpy.bool[builtins.bool]] hstack(blocks: Sequence[sparray[ScalarT, tuple[int, int]]], format: Literal['dia'], dtype: None = ...) -> dia_array[ScalarT]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def [ScalarT: number[Any, Any] | numpy.bool[builtins.bool]] hstack(blocks: Sequence[sparray[ScalarT, tuple[int, int]]], format: Literal['dok'], dtype: None = ...) -> dok_array[ScalarT, tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def [ScalarT: number[Any, Any] | numpy.bool[builtins.bool]] hstack(blocks: Sequence[sparray[ScalarT, tuple[int, int]]], format: Literal['lil'], dtype: None = ...) -> lil_array[ScalarT]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def [T] hstack(blocks: Sequence[_CanStackAs[numpy.bool[builtins.bool], T]], format: None = ..., *, dtype: type[builtins.bool] | type[numpy.bool[builtins.bool]] | dtype[numpy.bool[builtins.bool]] | HasDType[dtype[numpy.bool[builtins.bool]]] | Literal['bool', 'bool_', 'b1', '|b1', '?']) -> T
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['bsr'], dtype: type[builtins.bool] | type[numpy.bool[builtins.bool]] | dtype[numpy.bool[builtins.bool]] | HasDType[dtype[numpy.bool[builtins.bool]]] | Literal['bool', 'bool_', 'b1', '|b1', '?']) -> bsr_array[numpy.bool[builtins.bool]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['coo'], dtype: type[builtins.bool] | type[numpy.bool[builtins.bool]] | dtype[numpy.bool[builtins.bool]] | HasDType[dtype[numpy.bool[builtins.bool]]] | Literal['bool', 'bool_', 'b1', '|b1', '?']) -> coo_array[numpy.bool[builtins.bool], tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['csc'], dtype: type[builtins.bool] | type[numpy.bool[builtins.bool]] | dtype[numpy.bool[builtins.bool]] | HasDType[dtype[numpy.bool[builtins.bool]]] | Literal['bool', 'bool_', 'b1', '|b1', '?']) -> csc_array[numpy.bool[builtins.bool]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['csr'], dtype: type[builtins.bool] | type[numpy.bool[builtins.bool]] | dtype[numpy.bool[builtins.bool]] | HasDType[dtype[numpy.bool[builtins.bool]]] | Literal['bool', 'bool_', 'b1', '|b1', '?']) -> csr_array[numpy.bool[builtins.bool], tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['dia'], dtype: type[builtins.bool] | type[numpy.bool[builtins.bool]] | dtype[numpy.bool[builtins.bool]] | HasDType[dtype[numpy.bool[builtins.bool]]] | Literal['bool', 'bool_', 'b1', '|b1', '?']) -> dia_array[numpy.bool[builtins.bool]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['dok'], dtype: type[builtins.bool] | type[numpy.bool[builtins.bool]] | dtype[numpy.bool[builtins.bool]] | HasDType[dtype[numpy.bool[builtins.bool]]] | Literal['bool', 'bool_', 'b1', '|b1', '?']) -> dok_array[numpy.bool[builtins.bool], tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['lil'], dtype: type[builtins.bool] | type[numpy.bool[builtins.bool]] | dtype[numpy.bool[builtins.bool]] | HasDType[dtype[numpy.bool[builtins.bool]]] | Literal['bool', 'bool_', 'b1', '|b1', '?']) -> lil_array[numpy.bool[builtins.bool]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def [T] hstack(blocks: Sequence[_CanStackAs[signedinteger[_32Bit | _64Bit], T]], format: None = ..., *, dtype: type[JustInt] | type[signedinteger[_64Bit]] | dtype[signedinteger[_64Bit]] | HasDType[dtype[signedinteger[_64Bit]]] | Literal['int_', 'int', 'intp', 'n', '<n', '>n']) -> T
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['bsr'], dtype: type[JustInt] | type[signedinteger[_64Bit]] | dtype[signedinteger[_64Bit]] | HasDType[dtype[signedinteger[_64Bit]]] | Literal['int_', 'int', 'intp', 'n', '<n', '>n']) -> bsr_array[signedinteger[_32Bit | _64Bit]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['coo'], dtype: type[JustInt] | type[signedinteger[_64Bit]] | dtype[signedinteger[_64Bit]] | HasDType[dtype[signedinteger[_64Bit]]] | Literal['int_', 'int', 'intp', 'n', '<n', '>n']) -> coo_array[signedinteger[_32Bit | _64Bit], tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['csc'], dtype: type[JustInt] | type[signedinteger[_64Bit]] | dtype[signedinteger[_64Bit]] | HasDType[dtype[signedinteger[_64Bit]]] | Literal['int_', 'int', 'intp', 'n', '<n', '>n']) -> csc_array[signedinteger[_32Bit | _64Bit]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['csr'], dtype: type[JustInt] | type[signedinteger[_64Bit]] | dtype[signedinteger[_64Bit]] | HasDType[dtype[signedinteger[_64Bit]]] | Literal['int_', 'int', 'intp', 'n', '<n', '>n']) -> csr_array[signedinteger[_32Bit | _64Bit], tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['dia'], dtype: type[JustInt] | type[signedinteger[_64Bit]] | dtype[signedinteger[_64Bit]] | HasDType[dtype[signedinteger[_64Bit]]] | Literal['int_', 'int', 'intp', 'n', '<n', '>n']) -> dia_array[signedinteger[_32Bit | _64Bit]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['dok'], dtype: type[JustInt] | type[signedinteger[_64Bit]] | dtype[signedinteger[_64Bit]] | HasDType[dtype[signedinteger[_64Bit]]] | Literal['int_', 'int', 'intp', 'n', '<n', '>n']) -> dok_array[signedinteger[_32Bit | _64Bit], tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['lil'], dtype: type[JustInt] | type[signedinteger[_64Bit]] | dtype[signedinteger[_64Bit]] | HasDType[dtype[signedinteger[_64Bit]]] | Literal['int_', 'int', 'intp', 'n', '<n', '>n']) -> lil_array[signedinteger[_32Bit | _64Bit]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def [T] hstack(blocks: Sequence[_CanStackAs[float64, T]], format: None = ..., *, dtype: type[JustFloat] | type[floating[_64Bit]] | dtype[floating[_64Bit]] | HasDType[dtype[floating[_64Bit]]] | Literal['float64', 'double', 'float', 'f8', '<f8', '>f8', 'd']) -> T
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['bsr'], dtype: type[JustFloat] | type[floating[_64Bit]] | dtype[floating[_64Bit]] | HasDType[dtype[floating[_64Bit]]] | Literal['float64', 'double', 'float', 'f8', '<f8', '>f8', 'd']) -> bsr_array[float64]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['coo'], dtype: type[JustFloat] | type[floating[_64Bit]] | dtype[floating[_64Bit]] | HasDType[dtype[floating[_64Bit]]] | Literal['float64', 'double', 'float', 'f8', '<f8', '>f8', 'd']) -> coo_array[float64, tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['csc'], dtype: type[JustFloat] | type[floating[_64Bit]] | dtype[floating[_64Bit]] | HasDType[dtype[floating[_64Bit]]] | Literal['float64', 'double', 'float', 'f8', '<f8', '>f8', 'd']) -> csc_array[float64]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['csr'], dtype: type[JustFloat] | type[floating[_64Bit]] | dtype[floating[_64Bit]] | HasDType[dtype[floating[_64Bit]]] | Literal['float64', 'double', 'float', 'f8', '<f8', '>f8', 'd']) -> csr_array[float64, tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['dia'], dtype: type[JustFloat] | type[floating[_64Bit]] | dtype[floating[_64Bit]] | HasDType[dtype[floating[_64Bit]]] | Literal['float64', 'double', 'float', 'f8', '<f8', '>f8', 'd']) -> dia_array[float64]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['dok'], dtype: type[JustFloat] | type[floating[_64Bit]] | dtype[floating[_64Bit]] | HasDType[dtype[floating[_64Bit]]] | Literal['float64', 'double', 'float', 'f8', '<f8', '>f8', 'd']) -> dok_array[float64, tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['lil'], dtype: type[JustFloat] | type[floating[_64Bit]] | dtype[floating[_64Bit]] | HasDType[dtype[floating[_64Bit]]] | Literal['float64', 'double', 'float', 'f8', '<f8', '>f8', 'd']) -> lil_array[float64]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def [T] hstack(blocks: Sequence[_CanStackAs[complex128, T]], format: None = ..., *, dtype: type[JustComplex] | type[complexfloating[_64Bit, _64Bit]] | dtype[complexfloating[_64Bit, _64Bit]] | HasDType[dtype[complexfloating[_64Bit, _64Bit]]] | Literal['complex', 'complex128', 'cdouble', 'c16', '<c16', '>c16', 'D']) -> T
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['bsr'], dtype: type[JustComplex] | type[complexfloating[_64Bit, _64Bit]] | dtype[complexfloating[_64Bit, _64Bit]] | HasDType[dtype[complexfloating[_64Bit, _64Bit]]] | Literal['complex', 'complex128', 'cdouble', 'c16', '<c16', '>c16', 'D']) -> bsr_array[complex128]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['coo'], dtype: type[JustComplex] | type[complexfloating[_64Bit, _64Bit]] | dtype[complexfloating[_64Bit, _64Bit]] | HasDType[dtype[complexfloating[_64Bit, _64Bit]]] | Literal['complex', 'complex128', 'cdouble', 'c16', '<c16', '>c16', 'D']) -> coo_array[complex128, tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['csc'], dtype: type[JustComplex] | type[complexfloating[_64Bit, _64Bit]] | dtype[complexfloating[_64Bit, _64Bit]] | HasDType[dtype[complexfloating[_64Bit, _64Bit]]] | Literal['complex', 'complex128', 'cdouble', 'c16', '<c16', '>c16', 'D']) -> csc_array[complex128]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['csr'], dtype: type[JustComplex] | type[complexfloating[_64Bit, _64Bit]] | dtype[complexfloating[_64Bit, _64Bit]] | HasDType[dtype[complexfloating[_64Bit, _64Bit]]] | Literal['complex', 'complex128', 'cdouble', 'c16', '<c16', '>c16', 'D']) -> csr_array[complex128, tuple[int, int]]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['dia'], dtype: type[JustComplex] | type[complexfloating[_64Bit, _64Bit]] | dtype[complexfloating[_64Bit, _64Bit]] | HasDType[dtype[complexfloating[_64Bit, _64Bit]]] | Literal['complex', 'complex128', 'cdouble', 'c16', '<c16', '>c16', 'D']) -> dia_array[complex128]
- sklearn/utils/tests/test_sparsefuncs.py:451: note:     def hstack(blocks: Sequence[sparray[Any, tuple[Any, ...]]], format: Literal['dok'], dtype: type[JustComplex] | type[complexfloating[_64Bit, _64Bit]] | dtype[complexfloating[_64Bit, _64Bit]] | HasDType[dtype[complexfloating[_64Bit, _64Bit]]] | Literal['complex', 'complex128', 'cdouble', 'c16', '<c16', '>c16', 'D']) -> dok_array[complex128, tuple[int, int]]

... (truncated 54 lines) ...

pandas (https://github.com/pandas-dev/pandas)
+ pandas/tests/dtypes/test_missing.py:861: error: Unused "type: ignore" comment  [unused-ignore]
+ pandas/tests/dtypes/test_missing.py:883: error: Unused "type: ignore" comment  [unused-ignore]
+ pandas/conftest.py:1746: error: Unused "type: ignore" comment  [unused-ignore]

xarray (https://github.com/pydata/xarray)
+ xarray/tests/test_dataarray.py:425: error: Unused "type: ignore" comment  [unused-ignore]

ibis (https://github.com/ibis-project/ibis)
- ibis/expr/operations/udf.py:144: error: Dict entry 1 has incompatible type "str": "InputType"; expected "str": "Argument"  [dict-item]
- ibis/expr/operations/udf.py:147: error: Dict entry 2 has incompatible type "str": "property"; expected "str": "Argument"  [dict-item]
- ibis/expr/operations/udf.py:148: error: Dict entry 3 has incompatible type "str": "FrozenDict[Never, Never]"; expected "str": "Argument"  [dict-item]
- ibis/expr/operations/udf.py:149: error: Dict entry 4 has incompatible type "str": "Namespace"; expected "str": "Argument"  [dict-item]
- ibis/expr/operations/udf.py:150: error: Dict entry 5 has incompatible type "str": "str"; expected "str": "Argument"  [dict-item]
- ibis/expr/operations/udf.py:151: error: Dict entry 6 has incompatible type "str": "str"; expected "str": "Argument"  [dict-item]

dd-trace-py (https://github.com/DataDog/dd-trace-py)
- ddtrace/testing/internal/pytest/plugin.py:1443: error: Unsupported operand types for + ("list[tuple[str, object]]" and "list[tuple[str, bool]]")  [operator]

alectryon (https://github.com/cpitclaudel/alectryon)
- alectryon/core.py:251: error: Argument 3 to "ungroup_by" has incompatible type "Callable[[T], K | None]"; expected "Callable[[T | U], K | None]"  [arg-type]

[rheard]

@randolf-scholz Thanks, that was very helpful! I was not familiar with mypy-issues.

It seems like this PR would fix a bunch of issues. The first issue in the report there, #21809, I actually commented on and already added a unit test to cover that edge case specifically. #5874 seems to be basically complaining about the root cause here.

However there is one thing that concerned me with that check: #16522. It seems the reported error has changed which worries me. The fallback I added was checking that its inferred return type satisfied the outer context, but an ambiguously inferred Never trivially satisfies that because Never is a subtype of everything.

I've added the same ambiguous-uninhabited/erased-type guards that contextual inference already uses, along with a regression test, and reran the checks.