BUG: Make sure correct values are passed to Rolling._on when axis=1

doc/source/whatsnew/v1.0.0.rst

@@ -177,7 +177,7 @@ Groupby/resample/rolling
 ^^^^^^^^^^^^^^^^^^^^^^^^
 
 -
--
+- Bug in :meth:`DataFrame.rolling` not allowing for rolling over datetimes when ``axis=1`` (:issue: `28192`)
 - Bug in :meth:`DataFrame.groupby` not offering selection by column name when ``axis=1`` (:issue:`27614`)
 - Bug in :meth:`DataFrameGroupby.agg` not able to use lambda function with named aggregation (:issue:`27519`)
 

pandas/core/window/rolling.py

@@ -1653,7 +1653,10 @@ def is_datetimelike(self):
     def _on(self):
 
         if self.on is None:
-            return self.obj.index
+            if self.axis == 0:
+                return self.obj.index
+            elif self.axis == 1:
+                return self.obj.columns
         elif isinstance(self.obj, ABCDataFrame) and self.on in self.obj.columns:
             return Index(self.obj[self.on])
         else:

pandas/tests/window/test_rolling.py

@@ -334,3 +334,30 @@ def test_readonly_array(self):
         result = pd.Series(arr).rolling(2).mean()
         expected = pd.Series([np.nan, 2, np.nan, np.nan, 4])
         tm.assert_series_equal(result, expected)
+
+    def test_rolling_datetime(self, axis_frame, tz_naive_fixture):
+        # GH-28192
+        tz = tz_naive_fixture
+        df = pd.DataFrame(
+            {
+                i: [1] * 2
+                for i in pd.date_range("2019-8-01", "2019-08-03", freq="D", tz=tz)
+            }
+        )
+        if axis_frame in [0, "index"]:
+            result = df.T.rolling("2D", axis=axis_frame).sum().T
+        else:
+            result = df.rolling("2D", axis=axis_frame).sum()
+        expected = pd.DataFrame(
+            {
+                **{
+                    i: [1.0] * 2
+                    for i in pd.date_range("2019-8-01", periods=1, freq="D", tz=tz)
+                },
+                **{
+                    i: [2.0] * 2
+                    for i in pd.date_range("2019-8-02", "2019-8-03", freq="D", tz=tz)
+                },
+            }
+        )
+        tm.assert_frame_equal(result, expected)