CoW: Warn for cases that go through putmask

pandas/core/internals/base.py

@@ -14,7 +14,10 @@
 
 import numpy as np
 
-from pandas._config import using_copy_on_write
+from pandas._config import (
+    using_copy_on_write,
+    warn_copy_on_write,
+)
 
 from pandas._libs import (
     algos as libalgos,
@@ -49,6 +52,11 @@
     )
 
 
+class _AlreadyWarned:
+    def __init__(self):
+        self.warned_already = False
+
+
 class DataManager(PandasObject):
     # TODO share more methods/attributes
 
@@ -203,12 +211,17 @@ def putmask(self, mask, new, align: bool = True) -> Self:
             align_keys = ["mask"]
             new = extract_array(new, extract_numpy=True)
 
+        already_warned = None
+        if warn_copy_on_write():
+            already_warned = _AlreadyWarned()
+
         return self.apply_with_block(
             "putmask",
             align_keys=align_keys,
             mask=mask,
             new=new,
             using_cow=using_copy_on_write(),
+            already_warned=already_warned,
         )
 
     @final

pandas/core/internals/blocks.py

@@ -18,6 +18,7 @@
 from pandas._config import (
     get_option,
     using_copy_on_write,
+    warn_copy_on_write,
 )
 
 from pandas._libs import (
@@ -136,6 +137,29 @@
 _dtype_obj = np.dtype("object")
 
 
+COW_WARNING_GENERAL_MSG = """\
+Setting a value on a view: behaviour will change in pandas 3.0.
+You are mutating a Series or DataFrame object, and currently this mutation will
+also have effect on other Series or DataFrame objects that share data with this
+object. In pandas 3.0 (with Copy-on-Write), updating one Series or DataFrame object
+will never modify another.
+"""
+
+
+COW_WARNING_SETITEM_MSG = """\
+Setting a value on a view: behaviour will change in pandas 3.0.
+Currently, the mutation will also have effect on the object that shares data
+with this object. For example, when setting a value in a Series that was
+extracted from a column of a DataFrame, that DataFrame will also be updated:
+
+    ser = df["col"]
+    ser[0] = 0     <--- in pandas 2, this also updates `df`
+
+In pandas 3.0 (with Copy-on-Write), updating one Series/DataFrame will never
+modify another, and thus in the example above, `df` will not be changed.
+"""
+
+
 def maybe_split(meth: F) -> F:
     """
     If we have a multi-column block, split and operate block-wise.  Otherwise
@@ -1355,7 +1379,9 @@ def setitem(self, indexer, value, using_cow: bool = False) -> Block:
             values[indexer] = casted
         return self
 
-    def putmask(self, mask, new, using_cow: bool = False) -> list[Block]:
+    def putmask(
+        self, mask, new, using_cow: bool = False, already_warned=None
+    ) -> list[Block]:
         """
         putmask the data to the block; it is possible that we may create a
         new dtype of block
@@ -1388,6 +1414,19 @@ def putmask(self, mask, new, using_cow: bool = False) -> list[Block]:
                 return [self.copy(deep=False)]
             return [self]
 
+        if (
+            warn_copy_on_write()
+            and already_warned is not None
+            and not already_warned.warned_already
+        ):
+            if self.refs.has_reference():
+                warnings.warn(
+                    COW_WARNING_SETITEM_MSG,
+                    FutureWarning,
+                    stacklevel=find_stack_level(),
+                )
+                already_warned.warned_already = True
+
         try:
             casted = np_can_hold_element(values.dtype, new)
 
@@ -2020,7 +2059,9 @@ def where(
         return [nb]
 
     @final
-    def putmask(self, mask, new, using_cow: bool = False) -> list[Block]:
+    def putmask(
+        self, mask, new, using_cow: bool = False, already_warned=None
+    ) -> list[Block]:
         """
         See Block.putmask.__doc__
         """
@@ -2038,6 +2079,19 @@ def putmask(self, mask, new, using_cow: bool = False) -> list[Block]:
                 return [self.copy(deep=False)]
             return [self]
 
+        if (
+            warn_copy_on_write()
+            and already_warned is not None
+            and not already_warned.warned_already
+        ):
+            if self.refs.has_reference():
+                warnings.warn(
+                    COW_WARNING_SETITEM_MSG,
+                    FutureWarning,
+                    stacklevel=find_stack_level(),
+                )
+                already_warned.warned_already = True
+
         self = self._maybe_copy(using_cow, inplace=True)
         values = self.values
         if values.ndim == 2:

pandas/core/internals/managers.py

@@ -72,6 +72,8 @@
     interleaved_dtype,
 )
 from pandas.core.internals.blocks import (
+    COW_WARNING_GENERAL_MSG,
+    COW_WARNING_SETITEM_MSG,
     Block,
     NumpyBlock,
     ensure_block_shape,
@@ -100,29 +102,6 @@
     from pandas.api.extensions import ExtensionArray
 
 
-COW_WARNING_GENERAL_MSG = """\
-Setting a value on a view: behaviour will change in pandas 3.0.
-You are mutating a Series or DataFrame object, and currently this mutation will
-also have effect on other Series or DataFrame objects that share data with this
-object. In pandas 3.0 (with Copy-on-Write), updating one Series or DataFrame object
-will never modify another.
-"""
-
-
-COW_WARNING_SETITEM_MSG = """\
-Setting a value on a view: behaviour will change in pandas 3.0.
-Currently, the mutation will also have effect on the object that shares data
-with this object. For example, when setting a value in a Series that was
-extracted from a column of a DataFrame, that DataFrame will also be updated:
-
-    ser = df["col"]
-    ser[0] = 0     <--- in pandas 2, this also updates `df`
-
-In pandas 3.0 (with Copy-on-Write), updating one Series/DataFrame will never
-modify another, and thus in the example above, `df` will not be changed.
-"""
-
-
 class BaseBlockManager(DataManager):
     """
     Core internal data structure to implement DataFrame, Series, etc.

pandas/tests/copy_view/test_clip.py

@@ -5,12 +5,16 @@
 from pandas.tests.copy_view.util import get_array
 
 
-def test_clip_inplace_reference(using_copy_on_write):
+def test_clip_inplace_reference(using_copy_on_write, warn_copy_on_write):
     df = DataFrame({"a": [1.5, 2, 3]})
     df_copy = df.copy()
     arr_a = get_array(df, "a")
     view = df[:]
-    df.clip(lower=2, inplace=True)
+    if warn_copy_on_write:
+        with tm.assert_cow_warning():
+            df.clip(lower=2, inplace=True)
+    else:
+        df.clip(lower=2, inplace=True)
 
     if using_copy_on_write:
         assert not np.shares_memory(get_array(df, "a"), arr_a)

pandas/tests/copy_view/test_indexing.py

@@ -367,10 +367,11 @@ def test_subset_set_with_mask(backend, using_copy_on_write, warn_copy_on_write):
 
     mask = subset > 3
 
-    # TODO(CoW-warn) should warn -> mask is a DataFrame, which ends up going through
-    # DataFrame._where(..., inplace=True)
-    if using_copy_on_write or warn_copy_on_write:
+    if using_copy_on_write:
         subset[mask] = 0
+    elif warn_copy_on_write:
+        with tm.assert_cow_warning():
+            subset[mask] = 0
     else:
         with pd.option_context("chained_assignment", "warn"):
             with tm.assert_produces_warning(SettingWithCopyWarning):
@@ -867,18 +868,8 @@ def test_series_subset_set_with_indexer(
         and indexer.dtype.kind == "i"
     ):
         warn = FutureWarning
-    is_mask = (
-        indexer_si is tm.setitem
-        and isinstance(indexer, np.ndarray)
-        and indexer.dtype.kind == "b"
-    )
     if warn_copy_on_write:
-        # TODO(CoW-warn) should also warn for setting with mask
-        # -> Series.__setitem__ with boolean mask ends up using Series._set_values
-        # or Series._where depending on value being set
-        with tm.assert_cow_warning(
-            not is_mask, raise_on_extra_warnings=warn is not None
-        ):
+        with tm.assert_cow_warning(raise_on_extra_warnings=warn is not None):
             indexer_si(subset)[indexer] = 0
     else:
         with tm.assert_produces_warning(warn, match=msg):

pandas/tests/copy_view/test_methods.py

@@ -1407,11 +1407,12 @@ def test_items(using_copy_on_write, warn_copy_on_write):
 
 
 @pytest.mark.parametrize("dtype", ["int64", "Int64"])
-def test_putmask(using_copy_on_write, dtype):
+def test_putmask(using_copy_on_write, dtype, warn_copy_on_write):
     df = DataFrame({"a": [1, 2], "b": 1, "c": 2}, dtype=dtype)
     view = df[:]
     df_orig = df.copy()
-    df[df == df] = 5
+    with tm.assert_cow_warning(warn_copy_on_write):
+        df[df == df] = 5
 
     if using_copy_on_write:
         assert not np.shares_memory(get_array(view, "a"), get_array(df, "a"))
@@ -1445,15 +1446,21 @@ def test_putmask_aligns_rhs_no_reference(using_copy_on_write, dtype):
 @pytest.mark.parametrize(
     "val, exp, warn", [(5.5, True, FutureWarning), (5, False, None)]
 )
-def test_putmask_dont_copy_some_blocks(using_copy_on_write, val, exp, warn):
+def test_putmask_dont_copy_some_blocks(
+    using_copy_on_write, val, exp, warn, warn_copy_on_write
+):
     df = DataFrame({"a": [1, 2], "b": 1, "c": 1.5})
     view = df[:]
     df_orig = df.copy()
     indexer = DataFrame(
         [[True, False, False], [True, False, False]], columns=list("abc")
     )
-    with tm.assert_produces_warning(warn, match="incompatible dtype"):
-        df[indexer] = val
+    if warn_copy_on_write:
+        with tm.assert_cow_warning():
+            df[indexer] = val
+    else:
+        with tm.assert_produces_warning(warn, match="incompatible dtype"):
+            df[indexer] = val
 
     if using_copy_on_write:
         assert not np.shares_memory(get_array(view, "a"), get_array(df, "a"))
@@ -1785,13 +1792,17 @@ def test_update_frame(using_copy_on_write, warn_copy_on_write):
         tm.assert_frame_equal(view, expected)
 
 
-def test_update_series(using_copy_on_write):
+def test_update_series(using_copy_on_write, warn_copy_on_write):
     ser1 = Series([1.0, 2.0, 3.0])
     ser2 = Series([100.0], index=[1])
     ser1_orig = ser1.copy()
     view = ser1[:]
 
-    ser1.update(ser2)
+    if warn_copy_on_write:
+        with tm.assert_cow_warning():
+            ser1.update(ser2)
+    else:
+        ser1.update(ser2)
 
     expected = Series([1.0, 100.0, 3.0])
     tm.assert_series_equal(ser1, expected)

pandas/tests/copy_view/test_replace.py

@@ -279,14 +279,18 @@ def test_replace_categorical(using_copy_on_write, val):
 
 
 @pytest.mark.parametrize("method", ["where", "mask"])
-def test_masking_inplace(using_copy_on_write, method):
+def test_masking_inplace(using_copy_on_write, method, warn_copy_on_write):
     df = DataFrame({"a": [1.5, 2, 3]})
     df_orig = df.copy()
     arr_a = get_array(df, "a")
     view = df[:]
 
     method = getattr(df, method)
-    method(df["a"] > 1.6, -1, inplace=True)
+    if warn_copy_on_write:
+        with tm.assert_cow_warning():
+            method(df["a"] > 1.6, -1, inplace=True)
+    else:
+        method(df["a"] > 1.6, -1, inplace=True)
 
     if using_copy_on_write:
         assert not np.shares_memory(get_array(df, "a"), arr_a)

pandas/tests/frame/methods/test_replace.py

@@ -717,7 +717,7 @@ def test_replace_value_is_none(self, datetime_frame):
         datetime_frame.iloc[0, 0] = orig_value
         datetime_frame.iloc[1, 0] = orig2
 
-    def test_replace_for_new_dtypes(self, datetime_frame):
+    def test_replace_for_new_dtypes(self, datetime_frame, warn_copy_on_write):
         # dtypes
         tsframe = datetime_frame.copy().astype(np.float32)
         tsframe.loc[tsframe.index[:5], "A"] = np.nan
@@ -732,7 +732,11 @@ def test_replace_for_new_dtypes(self, datetime_frame):
         tsframe.loc[tsframe.index[:5], "B"] = -1e8
 
         b = tsframe["B"]
-        b[b == -1e8] = np.nan
+        if warn_copy_on_write:
+            with tm.assert_cow_warning():
+                b[b == -1e8] = np.nan
+        else:
+            b[b == -1e8] = np.nan
         tsframe["B"] = b
         msg = "DataFrame.fillna with 'method' is deprecated"
         with tm.assert_produces_warning(FutureWarning, match=msg):

pandas/tests/indexing/test_chaining_and_caching.py

@@ -164,9 +164,8 @@ def test_setitem_chained_setfault(self, using_copy_on_write, warn_copy_on_write)
                 df.response[mask] = "none"
             tm.assert_frame_equal(df, DataFrame({"response": data}))
         else:
-            # TODO(CoW-warn) should warn
-            # with tm.assert_cow_warning(warn_copy_on_write):
-            df.response[mask] = "none"
+            with tm.assert_cow_warning(warn_copy_on_write):
+                df.response[mask] = "none"
             tm.assert_frame_equal(df, DataFrame({"response": mdata}))
 
         recarray = np.rec.fromarrays([data], names=["response"])
@@ -177,9 +176,8 @@ def test_setitem_chained_setfault(self, using_copy_on_write, warn_copy_on_write)
                 df.response[mask] = "none"
             tm.assert_frame_equal(df, DataFrame({"response": data}))
         else:
-            # TODO(CoW-warn) should warn
-            # with tm.assert_cow_warning(warn_copy_on_write):
-            df.response[mask] = "none"
+            with tm.assert_cow_warning(warn_copy_on_write):
+                df.response[mask] = "none"
             tm.assert_frame_equal(df, DataFrame({"response": mdata}))
 
         df = DataFrame({"response": data, "response1": data})
@@ -190,9 +188,8 @@ def test_setitem_chained_setfault(self, using_copy_on_write, warn_copy_on_write)
                 df.response[mask] = "none"
             tm.assert_frame_equal(df, df_original)
         else:
-            # TODO(CoW-warn) should warn
-            # with tm.assert_cow_warning(warn_copy_on_write):
-            df.response[mask] = "none"
+            with tm.assert_cow_warning(warn_copy_on_write):
+                df.response[mask] = "none"
             tm.assert_frame_equal(df, DataFrame({"response": mdata, "response1": data}))
 
         # GH 6056
@@ -203,7 +200,6 @@ def test_setitem_chained_setfault(self, using_copy_on_write, warn_copy_on_write)
                 df["A"].iloc[0] = np.nan
             expected = DataFrame({"A": ["foo", "bar", "bah", "foo", "bar"]})
         else:
-            # TODO(CoW-warn) custom warning message
             with tm.assert_cow_warning(warn_copy_on_write):
                 df["A"].iloc[0] = np.nan
             expected = DataFrame({"A": [np.nan, "bar", "bah", "foo", "bar"]})