REGR: Merge raising when left merging on arrow string index (#54895)

Author: phofl

doc/source/whatsnew/v2.1.1.rst

@@ -13,6 +13,7 @@ including other versions of pandas.
 
 Fixed regressions
 ~~~~~~~~~~~~~~~~~
+- Fixed regression in :func:`merge` when merging over a PyArrow string index (:issue:`54894`)
 - Fixed regression in :func:`read_csv` when ``usecols`` is given and ``dtypes`` is a dict for ``engine="python"`` (:issue:`54868`)
 - Fixed regression in :meth:`.GroupBy.get_group` raising for ``axis=1`` (:issue:`54858`)
 - Fixed regression in :meth:`DataFrame.__setitem__` raising ``AssertionError`` when setting a :class:`Series` with a partial :class:`MultiIndex` (:issue:`54875`)

pandas/core/reshape/merge.py

@@ -2437,8 +2437,12 @@ def _factorize_keys(
             length = len(dc.dictionary)
 
             llab, rlab, count = (
-                pc.fill_null(dc.indices[slice(len_lk)], length).to_numpy(),
-                pc.fill_null(dc.indices[slice(len_lk, None)], length).to_numpy(),
+                pc.fill_null(dc.indices[slice(len_lk)], length)
+                .to_numpy()
+                .astype(np.intp, copy=False),
+                pc.fill_null(dc.indices[slice(len_lk, None)], length)
+                .to_numpy()
+                .astype(np.intp, copy=False),
                 len(dc.dictionary),
             )
             if how == "right":

pandas/tests/reshape/merge/test_merge.py

@@ -2947,3 +2947,15 @@ def test_merge_ea_int_and_float_numpy():
 
     result = df2.merge(df1)
     tm.assert_frame_equal(result, expected.astype("float64"))
+
+
+def test_merge_arrow_string_index():
+    # GH#54894
+    pytest.importorskip("pyarrow")
+    left = DataFrame({"a": ["a", "b"]}, dtype="string[pyarrow]")
+    right = DataFrame({"b": 1}, index=Index(["a", "c"], dtype="string[pyarrow]"))
+    result = left.merge(right, left_on="a", right_index=True, how="left")
+    expected = DataFrame(
+        {"a": Series(["a", "b"], dtype="string[pyarrow]"), "b": [1, np.nan]}
+    )
+    tm.assert_frame_equal(result, expected)