DEPR: Raise FutureWarning about raising an error in __array__ when copy=False cannot be honored (#60395)

KevsterAmp · jorisvandenbossche · web-flow · commit e90bb0eb1170 · 2025-01-13T09:32:06.000+01:00
Co-authored-by: Joris Van den Bossche &lt;jorisvandenbossche@gmail.com&gt;
diff --git a/doc/source/whatsnew/v2.3.0.rst b/doc/source/whatsnew/v2.3.0.rst
@@ -34,7 +34,7 @@ Other enhancements
 
 - The semantics for the ``copy`` keyword in ``__array__`` methods (i.e. called
   when using ``np.array()`` or ``np.asarray()`` on pandas objects) has been
-  updated to work correctly with NumPy >= 2 (:issue:`57739`)
+  updated to raise FutureWarning with NumPy >= 2 (:issue:`60340`)
 - The :meth:`~Series.sum` reduction is now implemented for ``StringDtype`` columns (:issue:`59853`)
 -
 
diff --git a/pandas/core/arrays/arrow/array.py b/pandas/core/arrays/arrow/array.py
@@ -12,6 +12,7 @@
     cast,
 )
 import unicodedata
+import warnings
 
 import numpy as np
 
@@ -28,6 +29,7 @@
     pa_version_under13p0,
 )
 from pandas.util._decorators import doc
+from pandas.util._exceptions import find_stack_level
 from pandas.util._validators import validate_fillna_kwargs
 
 from pandas.core.dtypes.cast import (
@@ -663,9 +665,15 @@ def __array__(
     ) -> np.ndarray:
         """Correctly construct numpy arrays when passed to `np.asarray()`."""
         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."
+            warnings.warn(
+                "Starting with NumPy 2.0, the behavior of the 'copy' keyword has "
+                "changed and passing 'copy=False' raises an error when returning "
+                "a zero-copy NumPy array is not possible. pandas will follow "
+                "this behavior starting with pandas 3.0.\nThis conversion to "
+                "NumPy requires a copy, but 'copy=False' was passed. Consider "
+                "using 'np.asarray(..)' instead.",
+                FutureWarning,
+                stacklevel=find_stack_level(),
             )
         elif copy is None:
             # `to_numpy(copy=False)` has the meaning of NumPy `copy=None`.
diff --git a/pandas/core/arrays/categorical.py b/pandas/core/arrays/categorical.py
@@ -1672,8 +1672,15 @@ def __array__(
         array(['a', 'b'], dtype=object)
         """
         if copy is False:
-            raise ValueError(
-                "Unable to avoid copy while creating an array as requested."
+            warnings.warn(
+                "Starting with NumPy 2.0, the behavior of the 'copy' keyword has "
+                "changed and passing 'copy=False' raises an error when returning "
+                "a zero-copy NumPy array is not possible. pandas will follow "
+                "this behavior starting with pandas 3.0.\nThis conversion to "
+                "NumPy requires a copy, but 'copy=False' was passed. Consider "
+                "using 'np.asarray(..)' instead.",
+                FutureWarning,
+                stacklevel=find_stack_level(),
             )
 
         ret = take_nd(self.categories._values, self._codes)
diff --git a/pandas/core/arrays/datetimelike.py b/pandas/core/arrays/datetimelike.py
@@ -359,9 +359,17 @@ def __array__(
         # 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."
+                warnings.warn(
+                    "Starting with NumPy 2.0, the behavior of the 'copy' keyword has "
+                    "changed and passing 'copy=False' raises an error when returning "
+                    "a zero-copy NumPy array is not possible. pandas will follow this "
+                    "behavior starting with pandas 3.0.\nThis conversion to NumPy "
+                    "requires a copy, but 'copy=False' was passed. Consider using "
+                    "'np.asarray(..)' instead.",
+                    FutureWarning,
+                    stacklevel=find_stack_level(),
                 )
+
             return np.array(list(self), dtype=object)
 
         if copy is True:
diff --git a/pandas/core/arrays/interval.py b/pandas/core/arrays/interval.py
@@ -42,6 +42,7 @@
 from pandas.compat.numpy import function as nv
 from pandas.errors import IntCastingNaNError
 from pandas.util._decorators import Appender
+from pandas.util._exceptions import find_stack_level
 
 from pandas.core.dtypes.cast import (
     LossySetitemError,
@@ -1575,8 +1576,15 @@ def __array__(
         objects (with dtype='object')
         """
         if copy is False:
-            raise ValueError(
-                "Unable to avoid copy while creating an array as requested."
+            warnings.warn(
+                "Starting with NumPy 2.0, the behavior of the 'copy' keyword has "
+                "changed and passing 'copy=False' raises an error when returning "
+                "a zero-copy NumPy array is not possible. pandas will follow "
+                "this behavior starting with pandas 3.0.\nThis conversion to "
+                "NumPy requires a copy, but 'copy=False' was passed. Consider "
+                "using 'np.asarray(..)' instead.",
+                FutureWarning,
+                stacklevel=find_stack_level(),
             )
 
         left = self._left
diff --git a/pandas/core/arrays/masked.py b/pandas/core/arrays/masked.py
@@ -38,6 +38,7 @@
 )
 from pandas.errors import AbstractMethodError
 from pandas.util._decorators import doc
+from pandas.util._exceptions import find_stack_level
 from pandas.util._validators import validate_fillna_kwargs
 
 from pandas.core.dtypes.base import ExtensionDtype
@@ -604,8 +605,16 @@ def __array__(
             if not self._hasna:
                 # special case, here we can simply return the underlying data
                 return np.array(self._data, dtype=dtype, copy=copy)
-            raise ValueError(
-                "Unable to avoid copy while creating an array as requested."
+
+            warnings.warn(
+                "Starting with NumPy 2.0, the behavior of the 'copy' keyword has "
+                "changed and passing 'copy=False' raises an error when returning "
+                "a zero-copy NumPy array is not possible. pandas will follow "
+                "this behavior starting with pandas 3.0.\nThis conversion to "
+                "NumPy requires a copy, but 'copy=False' was passed. Consider "
+                "using 'np.asarray(..)' instead.",
+                FutureWarning,
+                stacklevel=find_stack_level(),
             )
 
         if copy is None:
diff --git a/pandas/core/arrays/period.py b/pandas/core/arrays/period.py
@@ -415,8 +415,15 @@ def __array__(
                 return np.array(self.asi8, dtype=dtype)
 
         if copy is False:
-            raise ValueError(
-                "Unable to avoid copy while creating an array as requested."
+            warnings.warn(
+                "Starting with NumPy 2.0, the behavior of the 'copy' keyword has "
+                "changed and passing 'copy=False' raises an error when returning "
+                "a zero-copy NumPy array is not possible. pandas will follow "
+                "this behavior starting with pandas 3.0.\nThis conversion to "
+                "NumPy requires a copy, but 'copy=False' was passed. Consider "
+                "using 'np.asarray(..)' instead.",
+                FutureWarning,
+                stacklevel=find_stack_level(),
             )
 
         if dtype == bool:
diff --git a/pandas/core/arrays/sparse/array.py b/pandas/core/arrays/sparse/array.py
@@ -562,8 +562,15 @@ def __array__(
                 return self.sp_values
 
         if copy is False:
-            raise ValueError(
-                "Unable to avoid copy while creating an array as requested."
+            warnings.warn(
+                "Starting with NumPy 2.0, the behavior of the 'copy' keyword has "
+                "changed and passing 'copy=False' raises an error when returning "
+                "a zero-copy NumPy array is not possible. pandas will follow "
+                "this behavior starting with pandas 3.0.\nThis conversion to "
+                "NumPy requires a copy, but 'copy=False' was passed. Consider "
+                "using 'np.asarray(..)' instead.",
+                FutureWarning,
+                stacklevel=find_stack_level(),
             )
 
         fill_value = self.fill_value
diff --git a/pandas/core/generic.py b/pandas/core/generic.py
@@ -2151,9 +2151,16 @@ def __array__(
     ) -> np.ndarray:
         if copy is False and not self._mgr.is_single_block and not self.empty:
             # check this manually, otherwise ._values will already return a copy
-            # and np.array(values, copy=False) will not raise an error
-            raise ValueError(
-                "Unable to avoid copy while creating an array as requested."
+            # and np.array(values, copy=False) will not raise a warning
+            warnings.warn(
+                "Starting with NumPy 2.0, the behavior of the 'copy' keyword has "
+                "changed and passing 'copy=False' raises an error when returning "
+                "a zero-copy NumPy array is not possible. pandas will follow "
+                "this behavior starting with pandas 3.0.\nThis conversion to "
+                "NumPy requires a copy, but 'copy=False' was passed. Consider "
+                "using 'np.asarray(..)' instead.",
+                FutureWarning,
+                stacklevel=find_stack_level(),
             )
         values = self._values
         if copy is None:
diff --git a/pandas/core/indexes/multi.py b/pandas/core/indexes/multi.py
@@ -1314,8 +1314,15 @@ def __array__(self, dtype=None, copy=None) -> np.ndarray:
         """the array interface, return my values"""
         if copy is False:
             # self.values is always a newly construct array, so raise.
-            raise ValueError(
-                "Unable to avoid copy while creating an array as requested."
+            warnings.warn(
+                "Starting with NumPy 2.0, the behavior of the 'copy' keyword has "
+                "changed and passing 'copy=False' raises an error when returning "
+                "a zero-copy NumPy array is not possible. pandas will follow "
+                "this behavior starting with pandas 3.0.\nThis conversion to "
+                "NumPy requires a copy, but 'copy=False' was passed. Consider "
+                "using 'np.asarray(..)' instead.",
+                FutureWarning,
+                stacklevel=find_stack_level(),
             )
         if copy is True:
             # explicit np.array call to ensure a copy is made and unique objects
diff --git a/pandas/tests/arrays/sparse/test_array.py b/pandas/tests/arrays/sparse/test_array.py
@@ -500,8 +500,8 @@ def test_array_interface(arr_data, arr):
         # copy=False semantics are only supported in NumPy>=2.
         return
 
-    # for sparse arrays, copy=False is never allowed
-    with pytest.raises(ValueError, match="Unable to avoid copy while creating"):
+    msg = "Starting with NumPy 2.0, the behavior of the 'copy' keyword has changed"
+    with tm.assert_produces_warning(FutureWarning, match=msg):
         np.array(arr, copy=False)
 
     # except when there are actually no sparse filled values
diff --git a/pandas/tests/base/test_conversion.py b/pandas/tests/base/test_conversion.py
@@ -378,8 +378,8 @@ def test_to_numpy(arr, expected, zero_copy, index_or_series_or_array):
         return
 
     if not zero_copy:
-        with pytest.raises(ValueError, match="Unable to avoid copy while creating"):
-            # An error is always acceptable for `copy=False`
+        msg = "Starting with NumPy 2.0, the behavior of the 'copy' keyword has changed"
+        with tm.assert_produces_warning(FutureWarning, match=msg):
             np.array(thing, copy=False)
 
     else:
diff --git a/pandas/tests/copy_view/test_array.py b/pandas/tests/copy_view/test_array.py
@@ -187,7 +187,8 @@ def test_dataframe_multiple_numpy_dtypes():
     if np_version_gt2:
         # copy=False semantics are only supported in NumPy>=2.
 
-        with pytest.raises(ValueError, match="Unable to avoid copy while creating"):
+        msg = "Starting with NumPy 2.0, the behavior of the 'copy' keyword has changed"
+        with pytest.raises(FutureWarning, match=msg):
             arr = np.array(df, copy=False)
 
     arr = np.array(df, copy=True)
diff --git a/pandas/tests/extension/base/interface.py b/pandas/tests/extension/base/interface.py
@@ -1,3 +1,5 @@
+import warnings
+
 import numpy as np
 import pytest
 
@@ -82,15 +84,27 @@ def test_array_interface_copy(self, data):
             # copy=False semantics are only supported in NumPy>=2.
             return
 
-        try:
+        warning_raised = False
+        msg = "Starting with NumPy 2.0, the behavior of the 'copy' keyword has changed"
+        with warnings.catch_warnings(record=True) as w:
+            warnings.simplefilter("always")
             result_nocopy1 = np.array(data, copy=False)
-        except ValueError:
-            # An error is always acceptable for `copy=False`
-            return
-
-        result_nocopy2 = np.array(data, copy=False)
-        # If copy=False was given and did not raise, these must share the same data
-        assert np.may_share_memory(result_nocopy1, result_nocopy2)
+            assert len(w) <= 1
+            if len(w):
+                warning_raised = True
+                assert msg in str(w[0].message)
+
+        with warnings.catch_warnings(record=True) as w:
+            warnings.simplefilter("always")
+            result_nocopy2 = np.array(data, copy=False)
+            assert len(w) <= 1
+            if len(w):
+                warning_raised = True
+                assert msg in str(w[0].message)
+
+        if not warning_raised:
+            # If copy=False was given and did not raise, these must share the same data
+            assert np.may_share_memory(result_nocopy1, result_nocopy2)
 
     def test_is_extension_array_dtype(self, data):
         assert is_extension_array_dtype(data)
diff --git a/pandas/tests/extension/decimal/test_decimal.py b/pandas/tests/extension/decimal/test_decimal.py
@@ -6,6 +6,8 @@
 import numpy as np
 import pytest
 
+from pandas.compat.numpy import np_version_gt2
+
 import pandas as pd
 import pandas._testing as tm
 from pandas.tests.extension import base
@@ -289,6 +291,24 @@ def test_series_repr(self, data):
     def test_unary_ufunc_dunder_equivalence(self, data, ufunc):
         super().test_unary_ufunc_dunder_equivalence(data, ufunc)
 
+    def test_array_interface_copy(self, data):
+        result_copy1 = np.array(data, copy=True)
+        result_copy2 = np.array(data, copy=True)
+        assert not np.may_share_memory(result_copy1, result_copy2)
+        if not np_version_gt2:
+            # copy=False semantics are only supported in NumPy>=2.
+            return
+
+        try:
+            result_nocopy1 = np.array(data, copy=False)
+        except ValueError:
+            # An error is always acceptable for `copy=False`
+            return
+
+        result_nocopy2 = np.array(data, copy=False)
+        # If copy=False was given and did not raise, these must share the same data
+        assert np.may_share_memory(result_nocopy1, result_nocopy2)
+
 
 def test_take_na_value_other_decimal():
     arr = DecimalArray([decimal.Decimal("1.0"), decimal.Decimal("2.0")])
diff --git a/pandas/tests/extension/json/array.py b/pandas/tests/extension/json/array.py
@@ -25,9 +25,12 @@
     TYPE_CHECKING,
     Any,
 )
+import warnings
 
 import numpy as np
 
+from pandas.util._exceptions import find_stack_level
+
 from pandas.core.dtypes.cast import construct_1d_object_array_from_listlike
 from pandas.core.dtypes.common import (
     is_bool_dtype,
@@ -148,8 +151,15 @@ def __ne__(self, other):
 
     def __array__(self, dtype=None, copy=None):
         if copy is False:
-            raise ValueError(
-                "Unable to avoid copy while creating an array as requested."
+            warnings.warn(
+                "Starting with NumPy 2.0, the behavior of the 'copy' keyword has "
+                "changed and passing 'copy=False' raises an error when returning "
+                "a zero-copy NumPy array is not possible. pandas will follow "
+                "this behavior starting with pandas 3.0.\nThis conversion to "
+                "NumPy requires a copy, but 'copy=False' was passed. Consider "
+                "using 'np.asarray(..)' instead.",
+                FutureWarning,
+                stacklevel=find_stack_level(),
             )
 
         if dtype is None:
diff --git a/pandas/tests/indexes/multi/test_conversion.py b/pandas/tests/indexes/multi/test_conversion.py
@@ -47,7 +47,8 @@ def test_array_interface(idx):
         return
 
     # for MultiIndex, copy=False is never allowed
-    with pytest.raises(ValueError, match="Unable to avoid copy while creating"):
+    msg = "Starting with NumPy 2.0, the behavior of the 'copy' keyword has changed"
+    with tm.assert_produces_warning(FutureWarning, match=msg):
         np.array(idx, copy=False)
 
 
diff --git a/pandas/tests/io/test_fsspec.py b/pandas/tests/io/test_fsspec.py
@@ -5,8 +5,6 @@
 
 from pandas._config import using_string_dtype
 
-from pandas.compat import HAS_PYARROW
-
 from pandas import (
     DataFrame,
     date_range,
@@ -170,9 +168,6 @@ def test_excel_options(fsspectest):
     assert fsspectest.test[0] == "read"
 
 
-@pytest.mark.xfail(
-    using_string_dtype() and HAS_PYARROW, reason="TODO(infer_string) fastparquet"
-)
 def test_to_parquet_new_file(cleared_fs, df1):
     """Regression test for writing to a not-yet-existent GCS Parquet file."""
     pytest.importorskip("fastparquet")

Original file line number	Diff line number	Diff line change
`@@ -34,7 +34,7 @@ Other enhancements`
`34`	`34`
`35`	`35`	- The semantics for the ``copy`` keyword in ``__array__`` methods (i.e. called
`36`	`36`	when using ``np.array()`` or ``np.asarray()`` on pandas objects) has been
`37`		- updated to work correctly with NumPy >= 2 (:issue:`57739`)
	`37`	+ updated to raise FutureWarning with NumPy >= 2 (:issue:`60340`)
`38`	`38`	- The :meth:`~Series.sum` reduction is now implemented for ``StringDtype`` columns (:issue:`59853`)
`39`	`39`	`-`
`40`	`40`