Modify get_indexer_non_unique to account for nan keys - Bug pandas-dev#52234

Author: ltauro

pandas/_libs/index.pyx

@@ -360,6 +360,8 @@ cdef class IndexEngine:
         values = self.values
         stargets = set(targets)
 
+        na_in_stargets = any([checknull(t) for t in stargets])
+
         n = len(values)
         n_t = len(targets)
         if n > 10_000:
@@ -374,7 +376,7 @@ cdef class IndexEngine:
         if (
                 stargets and
                 len(stargets) < 5 and
-                not any([checknull(t) for t in stargets]) and
+                not na_in_stargets and
                 self.is_monotonic_increasing
         ):
             # if there are few enough stargets and the index is monotonically
@@ -396,7 +398,7 @@ cdef class IndexEngine:
             # otherwise, map by iterating through all items in the index
 
             # short-circuit na check
-            if values.dtype == object:
+            if na_in_stargets:
                 check_na_values = True
                 # keep track of nas in values
                 found_nas = set()

pandas/tests/indexes/base_class/test_indexing.py

@@ -37,6 +37,32 @@ def test_get_indexer_non_unique_dtype_mismatch(self):
         tm.assert_numpy_array_equal(np.array([-1], dtype=np.intp), indexes)
         tm.assert_numpy_array_equal(np.array([0], dtype=np.intp), missing)
 
+    def test_get_indexer_non_unique_int_index(self):
+        indexes, missing = Index([np.nan, 100, 200, 100]).get_indexer_non_unique(
+            Index([np.nan])
+        )
+        tm.assert_numpy_array_equal(np.array([0], dtype=np.intp), indexes)
+        tm.assert_numpy_array_equal(np.array([], dtype=np.intp), missing)
+
+        indexes, missing = Index([np.nan, 100, 200, 100]).get_indexer_non_unique(
+            Index([np.nan, 100])
+        )
+        tm.assert_numpy_array_equal(np.array([0, 1, 3], dtype=np.intp), indexes)
+        tm.assert_numpy_array_equal(np.array([], dtype=np.intp), missing)
+
+    def test_get_indexer_non_unique_float_index(self):
+        indexes, missing = Index([np.nan, 100.0, 200.0, 100.0]).get_indexer_non_unique(
+            Index([np.nan])
+        )
+        tm.assert_numpy_array_equal(np.array([0], dtype=np.intp), indexes)
+        tm.assert_numpy_array_equal(np.array([], dtype=np.intp), missing)
+
+        indexes, missing = Index([np.nan, 100.0, 200.0, 100.0]).get_indexer_non_unique(
+            Index([np.nan, 100.0])
+        )
+        tm.assert_numpy_array_equal(np.array([0, 1, 3], dtype=np.intp), indexes)
+        tm.assert_numpy_array_equal(np.array([], dtype=np.intp), missing)
+
 
 class TestGetLoc:
     @pytest.mark.slow  # to_flat_index takes a while