Backport PR pandas-dev#64035: BUG: don't mark pyarrow-string to numpy array as read-only unnecessarily

Author: jorisvandenbossche

doc/source/whatsnew/v3.0.1.rst

@@ -14,6 +14,7 @@ including other versions of pandas.
 Fixed regressions
 ~~~~~~~~~~~~~~~~~
 - Fixed regression in :meth:`DataFrame.groupby` and :meth:`Series.groupby` when grouping on categorical data with NA values, ``observed=False``, and ``dropna=True`` (:issue:`52445`)
+- Fixed regression when calling ``numpy.random``'s ``permutation()`` on a (pyarrow-backed) string :class:`Series` (:issue:`63935`)
 
 .. ---------------------------------------------------------------------------
 .. _whatsnew_301.bugs:

pandas/core/dtypes/astype.py

@@ -277,6 +277,15 @@ def astype_is_view(dtype: DtypeObj, new_dtype: DtypeObj) -> bool:
         return False
 
     elif is_string_dtype(dtype) and is_string_dtype(new_dtype):
+        from pandas.core.arrays.string_ import StringDtype
+
+        if (
+            isinstance(dtype, StringDtype)
+            and dtype.storage == "pyarrow"
+            and new_dtype == "object"
+        ):
+            # for conversion of pyarrow array to numpy object array -> always a copy
+            return False
         # Potentially! a view when converting from object to string
         return True
 

pandas/tests/arrays/string_/test_string.py

@@ -610,3 +610,14 @@ def test_numpy_array_ufunc(dtype, box):
             expected = pd.Series(["aa", "bbbb", "cccccc"])
 
     tm.assert_equal(result, expected)
+
+
+@pytest.mark.parametrize("box", [pd.Series, pd.array])
+def test_numpy_random_permute(dtype, box):
+    # https://github.com/pandas-dev/pandas/issues/63935
+    arr = box(["a", "bb", "ccc"], dtype=dtype)
+
+    rng = np.random.default_rng(2)
+    result = rng.permutation(arr)
+    assert isinstance(result, np.ndarray)
+    assert sorted(result.tolist()) == ["a", "bb", "ccc"]

pandas/tests/copy_view/test_array.py

@@ -180,6 +180,19 @@ def test_dataframe_array_string_dtype():
     assert arr.flags.writeable is False
 
 
+def test_series_array_string_dtype(any_string_dtype):
+    ser = Series(["a", "b"], dtype=any_string_dtype)
+    arr = np.asarray(ser)
+    if any_string_dtype == "string" and any_string_dtype.storage == "pyarrow":
+        # for pyarrow strings, the numpy arrays is not a view, so also does
+        # not need to be read-only (https://github.com/pandas-dev/pandas/pull/64035)
+        assert not np.shares_memory(arr, get_array(ser))
+        assert arr.flags.writeable is True
+    else:
+        assert np.shares_memory(arr, get_array(ser))
+        assert arr.flags.writeable is False
+
+
 def test_dataframe_multiple_numpy_dtypes():
     df = DataFrame({"a": [1, 2, 3], "b": 1.5})
     arr = np.asarray(df)