BUG: in DataFrame.reset_index() only call maybe_upcast_putmask with ndarrays

doc/source/whatsnew/v1.2.0.rst

@@ -387,6 +387,7 @@ Indexing
 - Bug in :meth:`Index.sort_values` where, when empty values were passed, the method would break by trying to compare missing values instead of pushing them to the end of the sort order. (:issue:`35584`)
 - Bug in :meth:`Index.get_indexer` and :meth:`Index.get_indexer_non_unique` where int64 arrays are returned instead of intp. (:issue:`36359`)
 - Bug in :meth:`DataFrame.sort_index` where parameter ascending passed as a list on a single level index gives wrong result. (:issue:`32334`)
+- Bug in :meth:`DataFrame.reset_index` was incorrectly raising a ``ValueError`` for input with a :class:`MultiIndex` with missing values in a level with ``Categorical`` dtype (:issue:`24206`)
 
 Missing
 ^^^^^^^

pandas/core/frame.py

@@ -4880,7 +4880,11 @@ def _maybe_casted_values(index, labels=None):
                         values = values._data  # TODO: can we de-kludge yet?
 
                     if mask.any():
-                        values, _ = maybe_upcast_putmask(values, mask, np.nan)
+                        if isinstance(values, np.ndarray):
+                            values, _ = maybe_upcast_putmask(values, mask, np.nan)
+                        else:
+                            # GH24206
+                            values[mask] = np.nan
 
                     if issubclass(values_type, DatetimeLikeArray):
                         values = values_type(values, dtype=values_dtype)

pandas/tests/frame/test_alter_axes.py

@@ -12,10 +12,12 @@
 
 from pandas import (
     Categorical,
+    CategoricalIndex,
     DataFrame,
     DatetimeIndex,
     Index,
     IntervalIndex,
+    MultiIndex,
     Series,
     Timestamp,
     cut,
@@ -171,21 +173,6 @@ def test_assign_columns(self, float_frame):
         tm.assert_series_equal(float_frame["C"], df["baz"], check_names=False)
         tm.assert_series_equal(float_frame["hi"], df["foo2"], check_names=False)
 
-    def test_set_index_preserve_categorical_dtype(self):
-        # GH13743, GH13854
-        df = DataFrame(
-            {
-                "A": [1, 2, 1, 1, 2],
-                "B": [10, 16, 22, 28, 34],
-                "C1": Categorical(list("abaab"), categories=list("bac"), ordered=False),
-                "C2": Categorical(list("abaab"), categories=list("bac"), ordered=True),
-            }
-        )
-        for cols in ["C1", "C2", ["A", "C1"], ["A", "C2"], ["C1", "C2"]]:
-            result = df.set_index(cols).reset_index()
-            result = result.reindex(columns=df.columns)
-            tm.assert_frame_equal(result, df)
-
     def test_rename_signature(self):
         sig = inspect.signature(DataFrame.rename)
         parameters = set(sig.parameters)
@@ -266,3 +253,47 @@ def test_set_reset_index(self):
         df = df.set_index("B")
 
         df = df.reset_index()
+
+
+class TestCategoricalIndex:
+    def test_set_index_preserve_categorical_dtype(self):
+        # GH13743, GH13854
+        df = DataFrame(
+            {
+                "A": [1, 2, 1, 1, 2],
+                "B": [10, 16, 22, 28, 34],
+                "C1": Categorical(list("abaab"), categories=list("bac"), ordered=False),
+                "C2": Categorical(list("abaab"), categories=list("bac"), ordered=True),
+            }
+        )
+        for cols in ["C1", "C2", ["A", "C1"], ["A", "C2"], ["C1", "C2"]]:
+            result = df.set_index(cols).reset_index()
+            result = result.reindex(columns=df.columns)
+            tm.assert_frame_equal(result, df)
+
+    @pytest.mark.parametrize(
+        "codes", ([[0, 0, 1, 1], [0, 1, 0, 1]], [[0, 0, -1, 1], [0, 1, 0, 1]])
+    )
+    def test_reindexing_with_missing_values(self, codes):
+        # GH 24206
+
+        index = MultiIndex(
+            [CategoricalIndex(["A", "B"]), CategoricalIndex(["a", "b"])], codes
+        )
+        data = {"col": range(len(index))}
+        df = DataFrame(data=data, index=index)
+
+        expected = DataFrame(
+            {
+                "level_0": Categorical.from_codes(codes[0], categories=["A", "B"]),
+                "level_1": Categorical.from_codes(codes[1], categories=["a", "b"]),
+                "col": range(4),
+            }
+        )
+
+        res = df.reset_index()
+        tm.assert_frame_equal(res, expected)
+
+        # roundtrip
+        res = expected.set_index(["level_0", "level_1"]).reset_index()
+        tm.assert_frame_equal(res, expected)