BUG: casting dt64/td64 in DataFrame.reindex

doc/source/whatsnew/v1.3.0.rst

@@ -335,6 +335,7 @@ Indexing
 - Bug in :meth:`DataFrame.__setitem__` raising ``ValueError`` when setting multiple values to duplicate columns (:issue:`15695`)
 - Bug in :meth:`DataFrame.loc`, :meth:`Series.loc`, :meth:`DataFrame.__getitem__` and :meth:`Series.__getitem__` returning incorrect elements for non-monotonic :class:`DatetimeIndex` for string slices (:issue:`33146`)
 - Bug in :meth:`DataFrame.reindex` and :meth:`Series.reindex` with timezone aware indexes raising ``TypeError`` for ``method="ffill"`` and ``method="bfill"`` and specified ``tolerance`` (:issue:`38566`)
+- Bug in :meth:`DataFrame.reindex` with ``datetime64[ns]`` or ``timedelta64[ns]`` incorrectly casting to integers when the ``fill_value`` requires casting to object dtype (:issue:`39755`)
 - Bug in :meth:`DataFrame.__setitem__` raising ``ValueError`` with empty :class:`DataFrame` and specified columns for string indexer and non empty :class:`DataFrame` to set (:issue:`38831`)
 - Bug in :meth:`DataFrame.loc.__setitem__` raising ValueError when expanding unique column for :class:`DataFrame` with duplicate columns (:issue:`38521`)
 - Bug in :meth:`DataFrame.iloc.__setitem__` and :meth:`DataFrame.loc.__setitem__` with mixed dtypes when setting with a dictionary value (:issue:`38335`)

pandas/core/algorithms.py

@@ -1388,6 +1388,9 @@ def wrapper(arr, indexer, out, fill_value=np.nan):
 
 def _convert_wrapper(f, conv_dtype):
     def wrapper(arr, indexer, out, fill_value=np.nan):
+        if conv_dtype == object:
+            # GH#39755 avoid casting dt64/td64 to integers
+            arr = ensure_wrapped_if_datetimelike(arr)
         arr = arr.astype(conv_dtype)
         f(arr, indexer, out, fill_value=fill_value)
 

pandas/tests/frame/methods/test_reindex.py

@@ -933,3 +933,38 @@ def test_reindex_empty(self, src_idx, cat_idx):
         result = df.reindex(columns=cat_idx)
         expected = DataFrame(index=["K"], columns=cat_idx, dtype="f8")
         tm.assert_frame_equal(result, expected)
+
+    @pytest.mark.parametrize("dtype", ["m8[ns]", "M8[ns]"])
+    def test_reindex_datetimelike_to_object(self, dtype):
+        # GH#39755 dont cast dt64/td64 to ints
+        mi = MultiIndex.from_product([list("ABCDE"), range(2)])
+
+        dti = date_range("2016-01-01", periods=10)
+        fv = np.timedelta64("NaT", "ns")
+        if dtype == "m8[ns]":
+            dti = dti - dti[0]
+            fv = np.datetime64("NaT", "ns")
+
+        ser = Series(dti, index=mi)
+        ser[::3] = pd.NaT
+
+        df = ser.unstack()
+
+        index = df.index.append(Index([1]))
+        columns = df.columns.append(Index(["foo"]))
+
+        res = df.reindex(index=index, columns=columns, fill_value=fv)
+
+        expected = DataFrame(
+            {
+                0: df[0].tolist() + [fv],
+                1: df[1].tolist() + [fv],
+                "foo": np.array(["NaT"] * 6, dtype=fv.dtype),
+            },
+            index=index,
+        )
+        assert (res.dtypes[[0, 1]] == object).all()
+        assert res.iloc[0, 0] is pd.NaT
+        assert res.iloc[-1, 0] is fv
+        assert res.iloc[-1, 1] is fv
+        tm.assert_frame_equal(res, expected)