BUG: unstack with sort=False fails when used with the level parameter (…

…#54987)

Assign new codes to labels when sort=False. This is done so that the data appears to be already sorted,
fixing the bug.

pandas/core/reshape/reshape.py

@@ -162,8 +162,18 @@ def _indexer_and_to_sort(
     ]:
         v = self.level
 
-        codes = list(self.index.codes)
         levs = list(self.index.levels)
+        codes = list(self.index.codes)
+
+        if not self.sort:
+            # Create new codes considering that labels are already sorted
+            codes = [list(code) for code in codes]
+            ids_code = [
+                ({y: x for x, y in enumerate(sorted(set(code), key=code.index))}, code)
+                for code in codes
+            ]
+            codes = [np.array([d[x] for x in code]) for d, code in ids_code]
+
         to_sort = codes[:v] + codes[v + 1 :] + [codes[v]]
         sizes = tuple(len(x) for x in levs[:v] + levs[v + 1 :] + [levs[v]])
 
@@ -174,25 +184,33 @@ def _indexer_and_to_sort(
         return indexer, to_sort
 
     @cache_readonly
-    def sorted_labels(self) -> list[np.ndarray]:
+    def labels(self) -> list[np.ndarray]:
         indexer, to_sort = self._indexer_and_to_sort
         if self.sort:
             return [line.take(indexer) for line in to_sort]
         return to_sort
 
-    def _make_sorted_values(self, values: np.ndarray) -> np.ndarray:
+    @cache_readonly
+    def sorted_labels(self) -> list[np.ndarray]:
         if self.sort:
-            indexer, _ = self._indexer_and_to_sort
+            return self.labels
+
+        v = self.level
+        codes = list(self.index.codes)
+        to_sort = codes[:v] + codes[v + 1 :] + [codes[v]]
+        return to_sort
 
-            sorted_values = algos.take_nd(values, indexer, axis=0)
-            return sorted_values
-        return values
+    def _make_sorted_values(self, values: np.ndarray) -> np.ndarray:
+        indexer, _ = self._indexer_and_to_sort
+        sorted_values = algos.take_nd(values, indexer, axis=0)
+        return sorted_values
 
     def _make_selectors(self):
         new_levels = self.new_index_levels
 
         # make the mask
-        remaining_labels = self.sorted_labels[:-1]
+        remaining_labels = self.labels[:-1]
+        choosen_labels = self.labels[-1]
         level_sizes = tuple(len(x) for x in new_levels)
 
         comp_index, obs_ids = get_compressed_ids(remaining_labels, level_sizes)
@@ -202,7 +220,7 @@ def _make_selectors(self):
         stride = self.index.levshape[self.level] + self.lift
         self.full_shape = ngroups, stride
 
-        selector = self.sorted_labels[-1] + stride * comp_index + self.lift
+        selector = choosen_labels + stride * comp_index + self.lift
         mask = np.zeros(np.prod(self.full_shape), dtype=bool)
         mask.put(selector, True)
 

pandas/tests/frame/test_stack_unstack.py

@@ -1318,6 +1318,21 @@ def test_unstack_sort_false(frame_or_series, dtype):
         [("two", "z", "b"), ("two", "y", "a"), ("one", "z", "b"), ("one", "y", "a")]
     )
     obj = frame_or_series(np.arange(1.0, 5.0), index=index, dtype=dtype)
+
+    result = obj.unstack(level=0, sort=False)
+
+    if frame_or_series is DataFrame:
+        expected_columns = MultiIndex.from_tuples([(0, "two"), (0, "one")])
+    else:
+        expected_columns = ["two", "one"]
+    expected = DataFrame(
+        [[1.0, 3.0], [2.0, 4.0]],
+        index=MultiIndex.from_tuples([("z", "b"), ("y", "a")]),
+        columns=expected_columns,
+        dtype=dtype,
+    )
+    tm.assert_frame_equal(result, expected)
+
     result = obj.unstack(level=-1, sort=False)
 
     if frame_or_series is DataFrame: