BUG: Index.equals raising with index of tuples that contain NA (pandas-dev#48446)

* BUG: Index.equals raising with index of tuples that contain NA

* Adress review

* Add sanity check

Author: phofl

doc/source/whatsnew/v1.6.0.rst

@@ -169,7 +169,7 @@ Indexing
 
 Missing
 ^^^^^^^
--
+- Bug in :meth:`Index.equals` raising ``TypeError` when :class:`Index` consists of tuples that contain ``NA`` (:issue:`48446`)
 -
 
 MultiIndex

pandas/_libs/lib.pyx

@@ -29,6 +29,8 @@ from cython cimport (
     floating,
 )
 
+from pandas._libs.missing import check_na_tuples_nonequal
+
 import_datetime()
 
 import numpy as np
@@ -636,7 +638,7 @@ def array_equivalent_object(left: object[:], right: object[:]) -> bool:
                 or is_matching_na(x, y, nan_matches_none=True)
             ):
                 return False
-        except ValueError:
+        except (ValueError, TypeError):
             # Avoid raising ValueError when comparing Numpy arrays to other types
             if cnp.PyArray_IsAnyScalar(x) != cnp.PyArray_IsAnyScalar(y):
                 # Only compare scalars to scalars and non-scalars to non-scalars
@@ -645,7 +647,12 @@ def array_equivalent_object(left: object[:], right: object[:]) -> bool:
                   and not (isinstance(x, type(y)) or isinstance(y, type(x)))):
                 # Check if non-scalars have the same type
                 return False
+            elif check_na_tuples_nonequal(x, y):
+                # We have tuples where one Side has a NA and the other side does not
+                # Only condition we may end up with a TypeError
+                return False
             raise
+
     return True
 
 

pandas/_libs/missing.pxd

@@ -5,6 +5,7 @@ from numpy cimport (
 
 
 cpdef bint is_matching_na(object left, object right, bint nan_matches_none=*)
+cpdef bint check_na_tuples_nonequal(object left, object right)
 
 cpdef bint checknull(object val, bint inf_as_na=*)
 cpdef ndarray[uint8_t] isnaobj(ndarray arr, bint inf_as_na=*)

pandas/_libs/missing.pyx

@@ -42,6 +42,37 @@ cdef:
     type cDecimal = Decimal  # for faster isinstance checks
 
 
+cpdef bint check_na_tuples_nonequal(object left, object right):
+    """
+    When we have NA in one of the tuples but not the other we have to check here,
+    because our regular checks fail before with ambigous boolean value.
+
+    Parameters
+    ----------
+    left: Any
+    right: Any
+
+    Returns
+    -------
+    True if we are dealing with tuples that have NA on one side and non NA on
+    the other side.
+
+    """
+    if not isinstance(left, tuple) or not isinstance(right, tuple):
+        return False
+
+    if len(left) != len(right):
+        return False
+
+    for left_element, right_element in zip(left, right):
+        if left_element is C_NA and right_element is not C_NA:
+            return True
+        elif right_element is C_NA and left_element is not C_NA:
+            return True
+
+    return False
+
+
 cpdef bint is_matching_na(object left, object right, bint nan_matches_none=False):
     """
     Check if two scalars are both NA of matching types.

pandas/tests/dtypes/test_missing.py

@@ -538,6 +538,23 @@ def test_array_equivalent_nested():
     assert not array_equivalent(left, right, strict_nan=True)
 
 
+def test_array_equivalent_index_with_tuples():
+    # GH#48446
+    idx1 = pd.Index(np.array([(pd.NA, 4), (1, 1)], dtype="object"))
+    idx2 = pd.Index(np.array([(1, 1), (pd.NA, 4)], dtype="object"))
+    assert not array_equivalent(idx1, idx2)
+    assert not idx1.equals(idx2)
+    assert not array_equivalent(idx2, idx1)
+    assert not idx2.equals(idx1)
+
+    idx1 = pd.Index(np.array([(4, pd.NA), (1, 1)], dtype="object"))
+    idx2 = pd.Index(np.array([(1, 1), (4, pd.NA)], dtype="object"))
+    assert not array_equivalent(idx1, idx2)
+    assert not idx1.equals(idx2)
+    assert not array_equivalent(idx2, idx1)
+    assert not idx2.equals(idx1)
+
+
 @pytest.mark.parametrize(
     "dtype, na_value",
     [