BUG: Preserve extension dtypes in MultiIndex.factorize() #62337

Author: davidjcastrejon

pandas/core/base.py

@@ -1302,7 +1302,22 @@ def factorize(
                 # GH#57517
                 uniques = self[:0]
             else:
-                uniques = self._constructor(uniques)
+                # GH#62337: preserve extension dtypes by reconstructing from original
+                if len(uniques) > 0:
+                    # Map back to original positions to preserve dtypes
+                    unique_positions = np.empty(len(uniques), dtype=np.intp)
+                    seen = {}
+                    pos = 0
+                    for i, code in enumerate(codes):
+                        if code not in seen and code != -1:
+                            unique_positions[pos] = i
+                            seen[code] = pos
+                            pos += 1
+
+                    # Reconstruct uniques from original MultiIndex to preserve dtypes
+                    uniques = self[unique_positions]
+                else:
+                    uniques = self[:0]
         else:
             from pandas import Index
 

pandas/tests/indexes/multi/test_factorize.py

@@ -74,7 +74,7 @@ def test_factorize_preserves_names(self):
 
         codes, uniques = mi.factorize()
 
-        tm.assert_index_equal(uniques.names, mi.names)
+        tm.assert_index_equal(pd.Index(uniques.names), pd.Index(mi.names))
 
     def test_factorize_extension_dtype_with_sort(self):
         # GH#62337: factorize with sort=True should preserve extension dtypes