Merge pull request #19503 from charris/backport-19468

MAINT: Add missing dtype overloads for object- and ctypes-based inputs

numpy/__init__.pyi

@@ -1,6 +1,9 @@
 import builtins
 import os
 import sys
+import mmap
+import ctypes as ct
+import array as _array
 import datetime as dt
 from abc import abstractmethod
 from types import TracebackType
@@ -920,7 +923,7 @@ class dtype(Generic[_DTypeScalar_co]):
     # other special cases. Order is sometimes important because of the
     # subtype relationships
     #
-    # bool < int < float < complex
+    # bool < int < float < complex < object
     #
     # so we have to make sure the overloads for the narrowest type is
     # first.
@@ -938,51 +941,54 @@ class dtype(Generic[_DTypeScalar_co]):
     @overload
     def __new__(cls, dtype: Type[bytes], align: bool = ..., copy: bool = ...) -> dtype[bytes_]: ...
 
-    # `unsignedinteger` string-based representations
+    # `unsignedinteger` string-based representations and ctypes
     @overload
-    def __new__(cls, dtype: _UInt8Codes, align: bool = ..., copy: bool = ...) -> dtype[uint8]: ...
+    def __new__(cls, dtype: _UInt8Codes | Type[ct.c_uint8], align: bool = ..., copy: bool = ...) -> dtype[uint8]: ...
     @overload
-    def __new__(cls, dtype: _UInt16Codes, align: bool = ..., copy: bool = ...) -> dtype[uint16]: ...
+    def __new__(cls, dtype: _UInt16Codes | Type[ct.c_uint16], align: bool = ..., copy: bool = ...) -> dtype[uint16]: ...
     @overload
-    def __new__(cls, dtype: _UInt32Codes, align: bool = ..., copy: bool = ...) -> dtype[uint32]: ...
+    def __new__(cls, dtype: _UInt32Codes | Type[ct.c_uint32], align: bool = ..., copy: bool = ...) -> dtype[uint32]: ...
     @overload
-    def __new__(cls, dtype: _UInt64Codes, align: bool = ..., copy: bool = ...) -> dtype[uint64]: ...
+    def __new__(cls, dtype: _UInt64Codes | Type[ct.c_uint64], align: bool = ..., copy: bool = ...) -> dtype[uint64]: ...
     @overload
-    def __new__(cls, dtype: _UByteCodes, align: bool = ..., copy: bool = ...) -> dtype[ubyte]: ...
+    def __new__(cls, dtype: _UByteCodes | Type[ct.c_ubyte], align: bool = ..., copy: bool = ...) -> dtype[ubyte]: ...
     @overload
-    def __new__(cls, dtype: _UShortCodes, align: bool = ..., copy: bool = ...) -> dtype[ushort]: ...
+    def __new__(cls, dtype: _UShortCodes | Type[ct.c_ushort], align: bool = ..., copy: bool = ...) -> dtype[ushort]: ...
     @overload
-    def __new__(cls, dtype: _UIntCCodes, align: bool = ..., copy: bool = ...) -> dtype[uintc]: ...
+    def __new__(cls, dtype: _UIntCCodes | Type[ct.c_uint], align: bool = ..., copy: bool = ...) -> dtype[uintc]: ...
+
+    # NOTE: We're assuming here that `uint_ptr_t == size_t`,
+    # an assumption that does not hold in rare cases (same for `ssize_t`)
     @overload
-    def __new__(cls, dtype: _UIntPCodes, align: bool = ..., copy: bool = ...) -> dtype[uintp]: ...
+    def __new__(cls, dtype: _UIntPCodes | Type[ct.c_void_p] | Type[ct.c_size_t], align: bool = ..., copy: bool = ...) -> dtype[uintp]: ...
     @overload
-    def __new__(cls, dtype: _UIntCodes, align: bool = ..., copy: bool = ...) -> dtype[uint]: ...
+    def __new__(cls, dtype: _UIntCodes | Type[ct.c_ulong], align: bool = ..., copy: bool = ...) -> dtype[uint]: ...
     @overload
-    def __new__(cls, dtype: _ULongLongCodes, align: bool = ..., copy: bool = ...) -> dtype[ulonglong]: ...
+    def __new__(cls, dtype: _ULongLongCodes | Type[ct.c_ulonglong], align: bool = ..., copy: bool = ...) -> dtype[ulonglong]: ...
 
-    # `signedinteger` string-based representations
+    # `signedinteger` string-based representations and ctypes
     @overload
-    def __new__(cls, dtype: _Int8Codes, align: bool = ..., copy: bool = ...) -> dtype[int8]: ...
+    def __new__(cls, dtype: _Int8Codes | Type[ct.c_int8], align: bool = ..., copy: bool = ...) -> dtype[int8]: ...
     @overload
-    def __new__(cls, dtype: _Int16Codes, align: bool = ..., copy: bool = ...) -> dtype[int16]: ...
+    def __new__(cls, dtype: _Int16Codes | Type[ct.c_int16], align: bool = ..., copy: bool = ...) -> dtype[int16]: ...
     @overload
-    def __new__(cls, dtype: _Int32Codes, align: bool = ..., copy: bool = ...) -> dtype[int32]: ...
+    def __new__(cls, dtype: _Int32Codes | Type[ct.c_int32], align: bool = ..., copy: bool = ...) -> dtype[int32]: ...
     @overload
-    def __new__(cls, dtype: _Int64Codes, align: bool = ..., copy: bool = ...) -> dtype[int64]: ...
+    def __new__(cls, dtype: _Int64Codes | Type[ct.c_int64], align: bool = ..., copy: bool = ...) -> dtype[int64]: ...
     @overload
-    def __new__(cls, dtype: _ByteCodes, align: bool = ..., copy: bool = ...) -> dtype[byte]: ...
+    def __new__(cls, dtype: _ByteCodes | Type[ct.c_byte], align: bool = ..., copy: bool = ...) -> dtype[byte]: ...
     @overload
-    def __new__(cls, dtype: _ShortCodes, align: bool = ..., copy: bool = ...) -> dtype[short]: ...
+    def __new__(cls, dtype: _ShortCodes | Type[ct.c_short], align: bool = ..., copy: bool = ...) -> dtype[short]: ...
     @overload
-    def __new__(cls, dtype: _IntCCodes, align: bool = ..., copy: bool = ...) -> dtype[intc]: ...
+    def __new__(cls, dtype: _IntCCodes | Type[ct.c_int], align: bool = ..., copy: bool = ...) -> dtype[intc]: ...
     @overload
-    def __new__(cls, dtype: _IntPCodes, align: bool = ..., copy: bool = ...) -> dtype[intp]: ...
+    def __new__(cls, dtype: _IntPCodes | Type[ct.c_ssize_t], align: bool = ..., copy: bool = ...) -> dtype[intp]: ...
     @overload
-    def __new__(cls, dtype: _IntCodes, align: bool = ..., copy: bool = ...) -> dtype[int_]: ...
+    def __new__(cls, dtype: _IntCodes | Type[ct.c_long], align: bool = ..., copy: bool = ...) -> dtype[int_]: ...
     @overload
-    def __new__(cls, dtype: _LongLongCodes, align: bool = ..., copy: bool = ...) -> dtype[longlong]: ...
+    def __new__(cls, dtype: _LongLongCodes | Type[ct.c_longlong], align: bool = ..., copy: bool = ...) -> dtype[longlong]: ...
 
-    # `floating` string-based representations
+    # `floating` string-based representations and ctypes
     @overload
     def __new__(cls, dtype: _Float16Codes, align: bool = ..., copy: bool = ...) -> dtype[float16]: ...
     @overload
@@ -992,11 +998,11 @@ class dtype(Generic[_DTypeScalar_co]):
     @overload
     def __new__(cls, dtype: _HalfCodes, align: bool = ..., copy: bool = ...) -> dtype[half]: ...
     @overload
-    def __new__(cls, dtype: _SingleCodes, align: bool = ..., copy: bool = ...) -> dtype[single]: ...
+    def __new__(cls, dtype: _SingleCodes | Type[ct.c_float], align: bool = ..., copy: bool = ...) -> dtype[single]: ...
     @overload
-    def __new__(cls, dtype: _DoubleCodes, align: bool = ..., copy: bool = ...) -> dtype[double]: ...
+    def __new__(cls, dtype: _DoubleCodes | Type[ct.c_double], align: bool = ..., copy: bool = ...) -> dtype[double]: ...
     @overload
-    def __new__(cls, dtype: _LongDoubleCodes, align: bool = ..., copy: bool = ...) -> dtype[longdouble]: ...
+    def __new__(cls, dtype: _LongDoubleCodes | Type[ct.c_longdouble], align: bool = ..., copy: bool = ...) -> dtype[longdouble]: ...
 
     # `complexfloating` string-based representations
     @overload
@@ -1010,21 +1016,21 @@ class dtype(Generic[_DTypeScalar_co]):
     @overload
     def __new__(cls, dtype: _CLongDoubleCodes, align: bool = ..., copy: bool = ...) -> dtype[clongdouble]: ...
 
-    # Miscellaneous string-based representations
+    # Miscellaneous string-based representations and ctypes
     @overload
-    def __new__(cls, dtype: _BoolCodes, align: bool = ..., copy: bool = ...) -> dtype[bool_]: ...
+    def __new__(cls, dtype: _BoolCodes | Type[ct.c_bool], align: bool = ..., copy: bool = ...) -> dtype[bool_]: ...
     @overload
     def __new__(cls, dtype: _TD64Codes, align: bool = ..., copy: bool = ...) -> dtype[timedelta64]: ...
     @overload
     def __new__(cls, dtype: _DT64Codes, align: bool = ..., copy: bool = ...) -> dtype[datetime64]: ...
     @overload
     def __new__(cls, dtype: _StrCodes, align: bool = ..., copy: bool = ...) -> dtype[str_]: ...
     @overload
-    def __new__(cls, dtype: _BytesCodes, align: bool = ..., copy: bool = ...) -> dtype[bytes_]: ...
+    def __new__(cls, dtype: _BytesCodes | Type[ct.c_char], align: bool = ..., copy: bool = ...) -> dtype[bytes_]: ...
     @overload
     def __new__(cls, dtype: _VoidCodes, align: bool = ..., copy: bool = ...) -> dtype[void]: ...
     @overload
-    def __new__(cls, dtype: _ObjectCodes, align: bool = ..., copy: bool = ...) -> dtype[object_]: ...
+    def __new__(cls, dtype: _ObjectCodes | Type[ct.py_object], align: bool = ..., copy: bool = ...) -> dtype[object_]: ...
 
     # dtype of a dtype is the same dtype
     @overload
@@ -1049,14 +1055,22 @@ class dtype(Generic[_DTypeScalar_co]):
         align: bool = ...,
         copy: bool = ...,
     ) -> dtype[Any]: ...
-    # Catchall overload
+    # Catchall overload for void-likes
     @overload
     def __new__(
         cls,
         dtype: _VoidDTypeLike,
         align: bool = ...,
         copy: bool = ...,
     ) -> dtype[void]: ...
+    # Catchall overload for object-likes
+    @overload
+    def __new__(
+        cls,
+        dtype: Type[object],
+        align: bool = ...,
+        copy: bool = ...,
+    ) -> dtype[object_]: ...
 
     @overload
     def __getitem__(self: dtype[void], key: List[str]) -> dtype[void]: ...

numpy/typing/tests/data/fail/dtype.py

@@ -18,5 +18,3 @@ class Test2:
         "field2": (int, 3),
     }
 )
-
-np.dtype[np.float64](np.int64)  # E: Argument 1 to "dtype" has incompatible type

numpy/typing/tests/data/reveal/dtype.py

@@ -1,3 +1,4 @@
+import ctypes as ct
 import numpy as np
 
 dtype_obj: np.dtype[np.str_]
@@ -22,6 +23,15 @@
 reveal_type(np.dtype(bool))  # E: numpy.dtype[numpy.bool_]
 reveal_type(np.dtype(str))  # E: numpy.dtype[numpy.str_]
 reveal_type(np.dtype(bytes))  # E: numpy.dtype[numpy.bytes_]
+reveal_type(np.dtype(object))  # E: numpy.dtype[numpy.object_]
+
+# ctypes
+reveal_type(np.dtype(ct.c_double))  # E: numpy.dtype[{double}]
+reveal_type(np.dtype(ct.c_longlong))  # E: numpy.dtype[{longlong}]
+reveal_type(np.dtype(ct.c_uint32))  # E: numpy.dtype[{uint32}]
+reveal_type(np.dtype(ct.c_bool))  # E: numpy.dtype[numpy.bool_]
+reveal_type(np.dtype(ct.c_char))  # E: numpy.dtype[numpy.bytes_]
+reveal_type(np.dtype(ct.py_object))  # E: numpy.dtype[numpy.object_]
 
 # Special case for None
 reveal_type(np.dtype(None))  # E: numpy.dtype[{double}]