CoW: Return read-only array in Index.values

doc/source/whatsnew/v2.1.0.rst

@@ -19,6 +19,7 @@ Enhancements
 Copy-on-Write improvements
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
 
+- Calling :meth:`Index.values` will now return a read-only NumPy array (:issue:`53704`)
 - Setting a :class:`Series` into a :class:`DataFrame` now creates a lazy instead of a deep copy (:issue:`53142`)
 
 .. _whatsnew_210.enhancements.enhancement2:
@@ -400,7 +401,7 @@ Strings
 
 Interval
 ^^^^^^^^
--
+- :meth:`pd.IntervalIndex.get_indexer` and :meth:`pd.IntervalIndex.get_indexer_nonunique` raising if ``target`` is read-only array (:issue:`53703`)
 -
 
 Indexing

pandas/_libs/intervaltree.pxi.in

@@ -125,7 +125,7 @@ cdef class IntervalTree(IntervalMixin):
         sort_order = self.left_sorter
         return is_monotonic(sort_order, False)[0]
 
-    def get_indexer(self, scalar_t[:] target) -> np.ndarray:
+    def get_indexer(self, ndarray[scalar_t, ndim=1] target) -> np.ndarray:
         """Return the positions corresponding to unique intervals that overlap
         with the given array of scalar targets.
         """
@@ -153,7 +153,7 @@ cdef class IntervalTree(IntervalMixin):
             old_len = result.data.n
         return result.to_array().astype('intp')
 
-    def get_indexer_non_unique(self, scalar_t[:] target):
+    def get_indexer_non_unique(self, ndarray[scalar_t, ndim=1] target):
         """Return the positions corresponding to intervals that overlap with
         the given array of scalar targets. Non-unique positions are repeated.
         """

pandas/conftest.py

@@ -685,7 +685,7 @@ def index_with_missing(request):
     # GH 35538. Use deep copy to avoid illusive bug on np-dev
     # GHA pipeline that writes into indices_dict despite copy
     ind = indices_dict[request.param].copy(deep=True)
-    vals = ind.values
+    vals = ind.values.copy()
     if request.param in ["tuples", "mi-with-dt64tz-level", "multi"]:
         # For setting missing values in the top level of MultiIndex
         vals = ind.tolist()

pandas/core/indexes/base.py

@@ -22,7 +22,10 @@
 
 import numpy as np
 
-from pandas._config import get_option
+from pandas._config import (
+    get_option,
+    using_copy_on_write,
+)
 
 from pandas._libs import (
     NaT,
@@ -5055,6 +5058,12 @@ def values(self) -> ArrayLike:
         >>> idx.values
         array([1, 2, 3])
         """
+        if using_copy_on_write():
+            data = self._data
+            if isinstance(data, np.ndarray):
+                data = data.view()
+                data.flags.writeable = False
+            return data
         return self._data
 
     @cache_readonly

pandas/core/indexes/datetimelike.py

@@ -18,6 +18,8 @@
 
 import numpy as np
 
+from pandas._config import using_copy_on_write
+
 from pandas._libs import (
     NaT,
     Timedelta,
@@ -451,7 +453,11 @@ def _with_freq(self, freq):
     @property
     def values(self) -> np.ndarray:
         # NB: For Datetime64TZ this is lossy
-        return self._data._ndarray
+        data = self._data._ndarray
+        if using_copy_on_write():
+            data = data.view()
+            data.flags.writeable = False
+        return data
 
     @doc(DatetimeIndexOpsMixin.shift)
     def shift(self, periods: int = 1, freq=None) -> Self:

pandas/tests/copy_view/index/test_datetimeindex.py

@@ -54,3 +54,12 @@ def test_datetimeindex_isocalendar(using_copy_on_write):
     ser.iloc[0] = Timestamp("2020-12-31")
     if using_copy_on_write:
         tm.assert_index_equal(df.index, expected)
+
+
+def test_index_values(using_copy_on_write):
+    idx = date_range("2019-12-31", periods=3, freq="D")
+    result = idx.values
+    if using_copy_on_write:
+        assert result.flags.writeable is False
+    else:
+        assert result.flags.writeable is True

pandas/tests/copy_view/index/test_index.py

@@ -153,3 +153,12 @@ def test_infer_objects(using_copy_on_write):
     view_.iloc[0, 0] = "aaaa"
     if using_copy_on_write:
         tm.assert_index_equal(idx, expected, check_names=False)
+
+
+def test_index_values(using_copy_on_write):
+    idx = Index([1, 2, 3])
+    result = idx.values
+    if using_copy_on_write:
+        assert result.flags.writeable is False
+    else:
+        assert result.flags.writeable is True

pandas/tests/copy_view/test_setitem.py

@@ -1,4 +1,5 @@
 import numpy as np
+import pytest
 
 from pandas import (
     DataFrame,
@@ -59,17 +60,25 @@ def test_set_column_with_index(using_copy_on_write):
     assert not np.shares_memory(get_array(df, "c"), idx.values)
 
     # and thus modifying the index does not modify the DataFrame
-    idx.values[0] = 0
-    tm.assert_series_equal(df["c"], Series([1, 2, 3], name="c"))
+    if using_copy_on_write:
+        with pytest.raises(ValueError, match="assignment"):
+            idx.values[0] = 0
+    else:
+        idx.values[0] = 0
+        tm.assert_series_equal(df["c"], Series([1, 2, 3], name="c"))
 
     idx = RangeIndex(1, 4)
     arr = idx.values
 
     df["d"] = idx
 
     assert not np.shares_memory(get_array(df, "d"), arr)
-    arr[0] = 0
-    tm.assert_series_equal(df["d"], Series([1, 2, 3], name="d"))
+    if using_copy_on_write:
+        with pytest.raises(ValueError, match="assignment"):
+            arr[0] = 0
+    else:
+        arr[0] = 0
+        tm.assert_series_equal(df["d"], Series([1, 2, 3], name="d"))
 
 
 def test_set_columns_with_dataframe(using_copy_on_write):

pandas/tests/indexes/interval/test_indexing.py

@@ -424,6 +424,18 @@ def test_get_indexer_interval_index(self, box):
         expected = np.array([-1, -1, -1], dtype=np.intp)
         tm.assert_numpy_array_equal(actual, expected)
 
+    def test_get_indexer_read_only(self):
+        idx = interval_range(start=0, end=5)
+        arr = np.array([1, 2])
+        arr.flags.writeable = False
+        result = idx.get_indexer(arr)
+        expected = np.array([0, 1])
+        tm.assert_numpy_array_equal(result, expected, check_dtype=False)
+
+        result = idx.get_indexer_non_unique(arr)[0]
+        expected = np.array([0, 1])
+        tm.assert_numpy_array_equal(result, expected, check_dtype=False)
+
 
 class TestSliceLocs:
     def test_slice_locs_with_interval(self):

pandas/tests/indexes/numeric/test_numeric.py

@@ -327,7 +327,7 @@ def test_constructor_from_list_no_dtype(self):
         index = Index([1, 2, 3])
         assert index.dtype == np.int64
 
-    def test_constructor(self, dtype):
+    def test_constructor(self, dtype, using_copy_on_write):
         index_cls = Index
 
         # scalar raise Exception
@@ -347,8 +347,12 @@ def test_constructor(self, dtype):
         val = arr[0] + 3000
 
         # this should not change index
-        arr[0] = val
-        assert new_index[0] != val
+        if not using_copy_on_write:
+            arr[0] = val
+            assert new_index[0] != val
+        else:
+            with pytest.raises(ValueError, match="assignment"):
+                arr[0] = val
 
         if dtype == np.int64:
             # pass list, coerce fine

pandas/tests/io/test_parquet.py

@@ -433,8 +433,12 @@ def test_read_columns(self, engine):
             df, engine, expected=expected, read_kwargs={"columns": ["string"]}
         )
 
-    def test_write_index(self, engine):
+    def test_write_index(self, engine, using_copy_on_write, request):
         check_names = engine != "fastparquet"
+        if using_copy_on_write and engine == "fastparquet":
+            request.node.add_marker(
+                pytest.mark.xfail(reason="fastparquet write into index")
+            )
 
         df = pd.DataFrame({"A": [1, 2, 3]})
         check_round_trip(df, engine)
@@ -1213,12 +1217,14 @@ def test_error_on_using_partition_cols_and_partition_on(
                 partition_cols=partition_cols,
             )
 
+    @pytest.mark.skipif(using_copy_on_write(), reason="fastparquet writes into Index")
     def test_empty_dataframe(self, fp):
         # GH #27339
         df = pd.DataFrame()
         expected = df.copy()
         check_round_trip(df, fp, expected=expected)
 
+    @pytest.mark.skipif(using_copy_on_write(), reason="fastparquet writes into Index")
     def test_timezone_aware_index(self, fp, timezone_aware_date_list):
         idx = 5 * [timezone_aware_date_list]
 
@@ -1328,6 +1334,7 @@ def test_invalid_dtype_backend(self, engine):
             with pytest.raises(ValueError, match=msg):
                 read_parquet(path, dtype_backend="numpy")
 
+    @pytest.mark.skipif(using_copy_on_write(), reason="fastparquet writes into Index")
     def test_empty_columns(self, fp):
         # GH 52034
         df = pd.DataFrame(index=pd.Index(["a", "b", "c"], name="custom name"))