BUG: df.loc[i] with Categorical column containing NaN

Author: jbrockmendel

doc/source/whatsnew/v3.0.0.rst

@@ -947,6 +947,7 @@ Indexing
 - Bug in :meth:`Series.__setitem__` when assigning boolean series with boolean indexer will raise ``LossySetitemError`` (:issue:`57338`)
 - Bug in printing :attr:`Index.names` and :attr:`MultiIndex.levels` would not escape single quotes (:issue:`60190`)
 - Bug in reindexing of :class:`DataFrame` with :class:`PeriodDtype` columns in case of consolidated block (:issue:`60980`, :issue:`60273`)
+- Bug in :meth:`DataFrame.loc.__getitem__` and :meth:`DataFrame.iloc.__getitem__` with a :class:`CategoricalDtype` column with integer categories raising when trying to index a row containing a ``NaN`` entry (:issue:`58954`)
 - Bug in :meth:`Index.__getitem__` incorrectly raising with a 0-dim ``np.ndarray`` key (:issue:`55601`)
 
 Missing

pandas/core/internals/managers.py

@@ -50,6 +50,7 @@
     is_list_like,
 )
 from pandas.core.dtypes.dtypes import (
+    CategoricalDtype,
     DatetimeTZDtype,
     ExtensionDtype,
     SparseDtype,
@@ -1138,7 +1139,24 @@ def fast_xs(self, loc: int) -> SingleBlockManager:
             # Such assignment may incorrectly coerce NaT to None
             # result[blk.mgr_locs] = blk._slice((slice(None), loc))
             for i, rl in enumerate(blk.mgr_locs):
-                result[rl] = blk.iget((i, loc))
+                item = blk.iget((i, loc))
+                if (
+                    result.dtype.kind in "iub"
+                    and lib.is_float(item)
+                    and isna(item)
+                    and isinstance(blk.dtype, CategoricalDtype)
+                ):
+                    # GH#58954 caused bc interleaved_dtype is wrong for Categorical
+                    # TODO(GH#38240) this will be unnecessary
+                    # Note that doing this in a try/except would work for the
+                    #  integer case, but not for bool, which will cast the NaN
+                    #  entry to True.
+                    if result.dtype.kind == "b":
+                        new_dtype = object
+                    else:
+                        new_dtype = np.float64
+                    result = result.astype(new_dtype)
+                result[rl] = item
 
         if isinstance(dtype, ExtensionDtype):
             cls = dtype.construct_array_type()

pandas/tests/indexing/test_categorical.py

@@ -571,3 +571,25 @@ def test_getitem_categorical_with_nan(self):
         df = DataFrame(ser)
         assert df.loc[np.nan, 0] == 2
         assert df.loc[np.nan][0] == 2
+
+    def test_getitem_row_categorical_with_nan(self):
+        # GH#58954
+        df = DataFrame({"a": [1, 2], "b": CategoricalIndex([1, None])})
+
+        res = df.iloc[1]
+        expected = Series([2, np.nan], index=df.columns, name=1)
+        tm.assert_series_equal(res, expected)
+
+        res = df.loc[1]
+        tm.assert_series_equal(res, expected)
+
+    def test_getitem_row_categorical_with_nan_bool(self):
+        # GH#58954
+        df = DataFrame({"a": [True, False], "b": CategoricalIndex([False, None])})
+
+        res = df.iloc[1]
+        expected = Series([False, np.nan], index=df.columns, dtype=object, name=1)
+        tm.assert_series_equal(res, expected)
+
+        res = df.loc[1]
+        tm.assert_series_equal(res, expected)