BUG: MultiIndex.join losing dtype

doc/source/whatsnew/v2.0.0.rst

@@ -701,6 +701,7 @@ MultiIndex
 - Bug in :meth:`MultiIndex.union` not sorting when sort=None and index contains missing values (:issue:`49010`)
 - Bug in :meth:`MultiIndex.append` not checking names for equality (:issue:`48288`)
 - Bug in :meth:`MultiIndex.symmetric_difference` losing extension array (:issue:`48607`)
+- Bug in :meth:`MultiIndex.join` losing dtypes when :class:`MultiIndex` has duplicates and is non-monotonic (:issue:`49830`)
 - Bug in :meth:`MultiIndex.putmask` losing extension array (:issue:`49830`)
 - Bug in :meth:`MultiIndex.value_counts` returning a :class:`Series` indexed by flat index of tuples instead of a :class:`MultiIndex` (:issue:`49558`)
 -

pandas/core/indexes/base.py

@@ -4587,22 +4587,9 @@ def _join_non_unique(
         )
         mask = left_idx == -1
 
-        join_array = self._values.take(left_idx)
-        right = other._values.take(right_idx)
-
-        if isinstance(join_array, np.ndarray):
-            # error: Argument 3 to "putmask" has incompatible type
-            # "Union[ExtensionArray, ndarray[Any, Any]]"; expected
-            # "Union[_SupportsArray[dtype[Any]], _NestedSequence[
-            # _SupportsArray[dtype[Any]]], bool, int, float, complex,
-            # str, bytes, _NestedSequence[Union[bool, int, float,
-            # complex, str, bytes]]]"
-            np.putmask(join_array, mask, right)  # type: ignore[arg-type]
-        else:
-            join_array._putmask(mask, right)
-
-        join_index = self._wrap_joined_index(join_array, other)
-
+        join_idx = self.take(left_idx)
+        right = other.take(right_idx)
+        join_index = join_idx.putmask(mask, right)
         return join_index, left_idx, right_idx
 
     @final

pandas/tests/indexes/multi/test_join.py

@@ -225,3 +225,34 @@ def test_join_multi_with_nan():
         index=MultiIndex.from_product([["A"], [1.0, 2.0]], names=["id1", "id2"]),
     )
     tm.assert_frame_equal(result, expected)
+
+
+def test_join_dtypes(any_numeric_ea_dtype):
+    # GH#49830
+    midx = MultiIndex.from_arrays([Series([1, 2], dtype=any_numeric_ea_dtype), [3, 4]])
+    midx2 = MultiIndex.from_arrays(
+        [Series([1, 0, 0], dtype=any_numeric_ea_dtype), [3, 4, 4]]
+    )
+    result = midx.join(midx2, how="outer")
+    expected = MultiIndex.from_arrays(
+        [Series([0, 0, 1, 2], dtype=any_numeric_ea_dtype), [4, 4, 3, 4]]
+    )
+    tm.assert_index_equal(result, expected)
+
+
+def test_join_dtypes_all_nan(any_numeric_ea_dtype):
+    # GH#49830
+    midx = MultiIndex.from_arrays(
+        [Series([1, 2], dtype=any_numeric_ea_dtype), [np.nan, np.nan]]
+    )
+    midx2 = MultiIndex.from_arrays(
+        [Series([1, 0, 0], dtype=any_numeric_ea_dtype), [np.nan, np.nan, np.nan]]
+    )
+    result = midx.join(midx2, how="outer")
+    expected = MultiIndex.from_arrays(
+        [
+            Series([0, 0, 1, 2], dtype=any_numeric_ea_dtype),
+            [np.nan, np.nan, np.nan, np.nan],
+        ]
+    )
+    tm.assert_index_equal(result, expected)