BUG[string]: incorrect index downcast in DataFrame.join

Author: jbrockmendel

doc/source/whatsnew/v3.0.0.rst

@@ -866,6 +866,8 @@ Reshaping
 - Bug in :meth:`concat` where concatenating DataFrame and Series with ``ignore_index = True`` drops the series name (:issue:`60723`, :issue:`56257`)
 - Bug in :func:`melt` where calling with duplicate column names in ``id_vars`` raised a misleading ``AttributeError`` (:issue:`61475`)
 - Bug in :meth:`DataFrame.merge` where user-provided suffixes could result in duplicate column names if the resulting names matched existing columns. Now raises a :class:`MergeError` in such cases. (:issue:`61402`)
+- Bug in :meth:`DataFrame.join` incorrectly downcasting object-dtype indexes (:issue:`??`)
+-
 
 Sparse
 ^^^^^^

pandas/core/reshape/merge.py

@@ -1328,13 +1328,13 @@ def _maybe_add_join_keys(
                 # if we have an all missing left_indexer
                 # make sure to just use the right values or vice-versa
                 if left_indexer is not None and (left_indexer == -1).all():
-                    key_col = Index(rvals)
+                    key_col = Index(rvals, dtype=rvals.dtype, copy=False)
                     result_dtype = rvals.dtype
                 elif right_indexer is not None and (right_indexer == -1).all():
-                    key_col = Index(lvals)
+                    key_col = Index(lvals, dtype=lvals.dtype, copy=False)
                     result_dtype = lvals.dtype
                 else:
-                    key_col = Index(lvals)
+                    key_col = Index(lvals, dtype=lvals.dtype, copy=False)
                     if left_indexer is not None:
                         mask_left = left_indexer == -1
                         key_col = key_col.where(~mask_left, rvals)

pandas/tests/copy_view/test_functions.py

@@ -1,10 +1,6 @@
 import numpy as np
 import pytest
 
-from pandas._config import using_string_dtype
-
-from pandas.compat import HAS_PYARROW
-
 from pandas import (
     DataFrame,
     Index,
@@ -247,13 +243,9 @@ def test_merge_copy_keyword():
     assert np.shares_memory(get_array(df2, "b"), get_array(result, "b"))
 
 
-@pytest.mark.xfail(
-    using_string_dtype() and HAS_PYARROW,
-    reason="TODO(infer_string); result.index infers str dtype while both "
-    "df1 and df2 index are object.",
-)
-def test_join_on_key():
-    df_index = Index(["a", "b", "c"], name="key", dtype=object)
+@pytest.mark.parametrize("dtype", [object, "str"])
+def test_join_on_key(dtype):
+    df_index = Index(["a", "b", "c"], name="key", dtype=dtype)
 
     df1 = DataFrame({"a": [1, 2, 3]}, index=df_index.copy(deep=True))
     df2 = DataFrame({"b": [4, 5, 6]}, index=df_index.copy(deep=True))
@@ -265,7 +257,7 @@ def test_join_on_key():
 
     assert np.shares_memory(get_array(result, "a"), get_array(df1, "a"))
     assert np.shares_memory(get_array(result, "b"), get_array(df2, "b"))
-    assert np.shares_memory(get_array(result.index), get_array(df1.index))
+    assert tm.shares_memory(get_array(result.index), get_array(df1.index))
     assert not np.shares_memory(get_array(result.index), get_array(df2.index))
 
     result.iloc[0, 0] = 0