BUG: .get_indexer_non_unique() must return an array of ints (pandas-dev#44084)

GH#44084 boils down to the following.

According to the docs `.get_indexer_non_unique()` is supposed to return
"integers from 0 to n - 1 indicating that the index at these positions matches
the corresponding target values".  However, for an index that is non unique and
non monotonic it returns a boolean mask.  That is because it uses `.get_loc()`
which for non unique, non monotonic indexes returns a boolean mask.

This patch catches that case and converts the boolean mask from `.get_loc()`
into the corresponding array of integers if the index is not unique and not
monotonic.

Author: johannes-mueller

pandas/core/indexes/interval.py

@@ -727,6 +727,8 @@ def _get_indexer_pointwise(
                 if isinstance(locs, slice):
                     # Only needed for get_indexer_non_unique
                     locs = np.arange(locs.start, locs.stop, locs.step, dtype="intp")
+                elif not self.is_unique and not self.is_monotonic:
+                    locs = np.where(locs)[0]
                 locs = np.array(locs, ndmin=1)
             except KeyError:
                 missing.append(i)

pandas/tests/indexes/interval/test_indexing.py

@@ -373,6 +373,16 @@ def test_get_indexer_with_nans(self):
         expected = np.array([0, 1], dtype=np.intp)
         tm.assert_numpy_array_equal(result, expected)
 
+    def test_get_index_non_unique_non_monotonic(self):
+        # GH#44084
+        index = IntervalIndex.from_tuples(
+            [(0.0, 1.0), (1.0, 2.0), (0.0, 1.0), (1.0, 2.0)]
+        )
+
+        result, _ = index.get_indexer_non_unique([Interval(1.0, 2.0)])
+        expected = np.array([1, 3])
+        tm.assert_numpy_array_equal(result, expected)
+
 
 class TestSliceLocs:
     def test_slice_locs_with_interval(self):