DEPR: CategoricalBlock.where casting to object (pandas-dev#29913)

Author: jbrockmendel

doc/source/whatsnew/v1.0.0.rst

@@ -418,6 +418,7 @@ or ``matplotlib.Axes.plot``. See :ref:`plotting.formatters` for more.
 - Removed :meth:`DataFrame.as_blocks`, :meth:`Series.as_blocks`, `DataFrame.blocks`, :meth:`Series.blocks` (:issue:`17656`)
 - :meth:`pandas.Series.str.cat` now defaults to aligning ``others``, using ``join='left'`` (:issue:`27611`)
 - :meth:`pandas.Series.str.cat` does not accept list-likes *within* list-likes anymore (:issue:`27611`)
+- :meth:`Series.where` with ``Categorical`` dtype (or :meth:`DataFrame.where` with ``Categorical`` column) no longer allows setting new categories (:issue:`24114`)
 - :func:`core.internals.blocks.make_block` no longer accepts the "fastpath" keyword(:issue:`19265`)
 - :meth:`Block.make_block_same_class` no longer accepts the "dtype" keyword(:issue:`19434`)
 - Removed the previously deprecated :meth:`ExtensionArray._formatting_values`. Use :attr:`ExtensionArray._formatter` instead. (:issue:`23601`)

pandas/core/internals/blocks.py

@@ -2887,37 +2887,6 @@ def concat_same_type(self, to_concat, placement=None):
  values, placement=placement or slice(0, len(values), 1), ndim=self.ndim
  )
 
- def where(
- self,
- other,
- cond,
- align=True,
- errors="raise",
- try_cast: bool = False,
- axis: int = 0,
- ) -> List["Block"]:
- # TODO(CategoricalBlock.where):
- # This can all be deleted in favor of ExtensionBlock.where once
- # we enforce the deprecation.
- object_msg = (
- "Implicitly converting categorical to object-dtype ndarray. "
- "One or more of the values in 'other' are not present in this "
- "categorical's categories. A future version of pandas will raise "
- "a ValueError when 'other' contains different categories.\n\n"
- "To preserve the current behavior, add the new categories to "
- "the categorical before calling 'where', or convert the "
- "categorical to a different dtype."
- )
- try:
- # Attempt to do preserve categorical dtype.
- result = super().where(other, cond, align, errors, try_cast, axis)
- except (TypeError, ValueError):
- warnings.warn(object_msg, FutureWarning, stacklevel=6)
- result = self.astype(object).where(
- other, cond, align=align, errors=errors, try_cast=try_cast, axis=axis
- )
- return result
-
  def replace(
  self,
  to_replace,

pandas/tests/arrays/categorical/test_indexing.py

@@ -206,13 +206,11 @@ def test_where_other_categorical(self):
  expected = pd.Series(Categorical(["a", "c", "c"], dtype=ser.dtype))
  tm.assert_series_equal(result, expected)
 
- def test_where_warns(self):
+ def test_where_new_category_raises(self):
  ser = pd.Series(Categorical(["a", "b", "c"]))
- with tm.assert_produces_warning(FutureWarning):
- result = ser.where([True, False, True], "d")
-
- expected = pd.Series(np.array(["a", "d", "c"], dtype="object"))
- tm.assert_series_equal(result, expected)
+ msg = "Cannot setitem on a Categorical with a new category"
+ with pytest.raises(ValueError, match=msg):
+ ser.where([True, False, True], "d")
 
  def test_where_ordered_differs_rasies(self):
  ser = pd.Series(
@@ -221,11 +219,8 @@ def test_where_ordered_differs_rasies(self):
  other = Categorical(
  ["b", "c", "a"], categories=["a", "c", "b", "d"], ordered=True
  )
- with tm.assert_produces_warning(FutureWarning):
- result = ser.where([True, False, True], other)
-
- expected = pd.Series(np.array(["a", "c", "c"], dtype=object))
- tm.assert_series_equal(result, expected)
+ with pytest.raises(ValueError, match="without identical categories"):
+ ser.where([True, False, True], other)
 
 
 @pytest.mark.parametrize("index", [True, False])