BUG: Fix copy semantics in __array__

This fixes the semantics of ``__array__``.  While rejecting ``copy=False``
is pretty harmless, ``copy=True`` should never have been ignored and is
dangerous.

Author: seberg

pandas/core/arrays/arrow/array.py

@@ -667,7 +667,15 @@ def __array__(
         self, dtype: NpDtype | None = None, copy: bool | None = None
     ) -> np.ndarray:
         """Correctly construct numpy arrays when passed to `np.asarray()`."""
-        return self.to_numpy(dtype=dtype)
+        if copy is False:
+            # TODO: By using `zero_copy_only` it may be possible to implement this
+            raise ValueError(
+                "Unable to avoid copy while creating an array as requested."
+            )
+        elif copy is None:
+            copy = False  # The NumPy copy=False meaning is different here.
+
+        return self.to_numpy(dtype=dtype, copy=copy)
 
     def __invert__(self) -> Self:
         # This is a bit wise op for integer types

pandas/core/arrays/categorical.py

@@ -1663,7 +1663,7 @@ def __array__(
             Specifies the the dtype for the array.
 
         copy : bool or None, optional
-            Unused.
+            See :func:`numpy.asarray`.
 
         Returns
         -------
@@ -1686,13 +1686,21 @@ def __array__(
         >>> np.asarray(cat)
         array(['a', 'b'], dtype=object)
         """
+        if copy is False:
+            raise ValueError(
+                "Unable to avoid copy while creating an array as requested."
+            )
+
+        # TODO: using asarray_func because NumPy 1.x doesn't support copy=None
+        asarray_func = np.asarray if copy is None else np.array
+
         ret = take_nd(self.categories._values, self._codes)
         if dtype and np.dtype(dtype) != self.categories.dtype:
-            return np.asarray(ret, dtype)
+            return asarray_func(ret, dtype)
         # When we're a Categorical[ExtensionArray], like Interval,
         # we need to ensure __array__ gets all the way to an
         # ndarray.
-        return np.asarray(ret)
+        return asarray_func(ret)
 
     def __array_ufunc__(self, ufunc: np.ufunc, method: str, *inputs, **kwargs):
         # for binary ops, use our custom dunder methods

pandas/core/arrays/datetimelike.py

@@ -359,7 +359,14 @@ def __array__(
     ) -> np.ndarray:
         # used for Timedelta/DatetimeArray, overwritten by PeriodArray
         if is_object_dtype(dtype):
+            if copy is False:
+                raise ValueError(
+                    "Unable to avoid copy while creating an array as requested."
+                )
             return np.array(list(self), dtype=object)
+
+        if copy is True:
+            return np.array(self._ndarray, dtype=dtype)
         return self._ndarray
 
     @overload

pandas/core/arrays/interval.py

@@ -1606,6 +1606,11 @@ def __array__(
         Return the IntervalArray's data as a numpy array of Interval
         objects (with dtype='object')
         """
+        if copy is False:
+            raise ValueError(
+                "Unable to avoid copy while creating an array as requested."
+            )
+
         left = self._left
         right = self._right
         mask = self.isna()

pandas/core/arrays/masked.py

@@ -581,7 +581,14 @@ def __array__(
         the array interface, return my values
         We return an object array here to preserve our scalar values
         """
-        return self.to_numpy(dtype=dtype)
+        if copy is False:
+            raise ValueError(
+                "Unable to avoid copy while creating an array as requested."
+            )
+
+        if copy is None:
+            copy = False  # The NumPy copy=False meaning is different here.
+        return self.to_numpy(dtype=dtype, copy=copy)
 
     _HANDLED_TYPES: tuple[type, ...]
 

pandas/core/arrays/numpy_.py

@@ -150,6 +150,9 @@ def dtype(self) -> NumpyEADtype:
     def __array__(
         self, dtype: NpDtype | None = None, copy: bool | None = None
     ) -> np.ndarray:
+        if copy is not None:
+            # Note: branch avoids `copy=None` for NumPy 1.x support
+            return np.asarray(self._ndarray, dtype=dtype, copy=copy)
         return np.asarray(self._ndarray, dtype=dtype)
 
     def __array_ufunc__(self, ufunc: np.ufunc, method: str, *inputs, **kwargs):

pandas/core/arrays/period.py

@@ -389,6 +389,11 @@ def freqstr(self) -> str:
     def __array__(
         self, dtype: NpDtype | None = None, copy: bool | None = None
     ) -> np.ndarray:
+        if copy is False:
+            raise ValueError(
+                "Unable to avoid copy while creating an array as requested."
+            )
+
         if dtype == "i8":
             return self.asi8
         elif dtype == bool:

pandas/core/arrays/sparse/array.py

@@ -547,6 +547,11 @@ def from_spmatrix(cls, data: spmatrix) -> Self:
     def __array__(
         self, dtype: NpDtype | None = None, copy: bool | None = None
     ) -> np.ndarray:
+        if copy is False:
+            raise ValueError(
+                "Unable to avoid copy while creating an array as requested."
+            )
+
         fill_value = self.fill_value
 
         if self.sp_index.ngaps == 0:

pandas/core/generic.py

@@ -2015,8 +2015,17 @@ def __array__(
         self, dtype: npt.DTypeLike | None = None, copy: bool | None = None
     ) -> np.ndarray:
         values = self._values
-        arr = np.asarray(values, dtype=dtype)
-        if astype_is_view(values.dtype, arr.dtype) and self._mgr.is_single_block:
+        if copy is None:
+            # Note: branch avoids `copy=None` for NumPy 1.x support
+            arr = np.asarray(values, dtype=dtype)
+        else:
+            arr = np.asarray(values, dtype=dtype, copy=copy)
+
+        if (
+            copy is not False
+            and astype_is_view(values.dtype, arr.dtype)
+            and self._mgr.is_single_block
+        ):
             # Check if both conversions can be done without a copy
             if astype_is_view(self.dtypes.iloc[0], values.dtype) and astype_is_view(
                 values.dtype, arr.dtype

pandas/core/indexes/base.py

@@ -908,7 +908,11 @@ def __array__(self, dtype=None, copy=None) -> np.ndarray:
         """
         The array interface, return my values.
         """
-        return np.asarray(self._data, dtype=dtype)
+        if copy is None:
+            # Note, that the if branch exists for NumPy 1.x support
+            return np.asarray(self._data, dtype=dtype)
+
+        return np.asarray(self._data, dtype=dtype, copy=copy)
 
     def __array_ufunc__(self, ufunc: np.ufunc, method: str_t, *inputs, **kwargs):
         if any(isinstance(other, (ABCSeries, ABCDataFrame)) for other in inputs):