Fix negative n for str.replace with arrow string (pandas-dev#56406)

doc/source/whatsnew/v2.2.0.rst

@@ -577,6 +577,7 @@ Strings
 - Bug in :func:`pandas.api.types.is_string_dtype` while checking object array with no elements is of the string dtype (:issue:`54661`)
 - Bug in :meth:`DataFrame.apply` failing when ``engine="numba"`` and columns or index have ``StringDtype`` (:issue:`56189`)
 - Bug in :meth:`Series.__mul__` for :class:`ArrowDtype` with ``pyarrow.string`` dtype and ``string[pyarrow]`` for the pyarrow backend (:issue:`51970`)
+- Bug in :meth:`Series.str.replace` when ``n < 0`` for :class:`ArrowDtype` with ``pyarrow.string`` (:issue:`56404`)
 - Bug in :meth:`Series.str.startswith` and :meth:`Series.str.endswith` with arguments of type ``tuple[str, ...]`` for ``string[pyarrow]`` (:issue:`54942`)
 
 Interval

pandas/core/arrays/arrow/array.py

@@ -2155,7 +2155,15 @@ def _str_replace(
             )
 
         func = pc.replace_substring_regex if regex else pc.replace_substring
-        result = func(self._pa_array, pattern=pat, replacement=repl, max_replacements=n)
+        # https://github.com/apache/arrow/issues/39149
+        # GH 56404, unexpected behavior with negative max_replacements with pyarrow.
+        pa_max_replacements = None if n < 0 else n
+        result = func(
+            self._pa_array,
+            pattern=pat,
+            replacement=repl,
+            max_replacements=pa_max_replacements,
+        )
         return type(self)(result)
 
     def _str_repeat(self, repeats: int | Sequence[int]):

pandas/tests/extension/test_arrow.py

@@ -1776,6 +1776,14 @@ def test_str_replace(pat, repl, n, regex, exp):
     tm.assert_series_equal(result, expected)
 
 
+def test_str_replace_negative_n():
+    # GH 56404
+    ser = pd.Series(["abc", "aaaaaa"], dtype=ArrowDtype(pa.string()))
+    actual = ser.str.replace("a", "", -3, True)
+    expected = pd.Series(["bc", ""], dtype=ArrowDtype(pa.string()))
+    tm.assert_series_equal(expected, actual)
+
+
 def test_str_repeat_unsupported():
     ser = pd.Series(["abc", None], dtype=ArrowDtype(pa.string()))
     with pytest.raises(NotImplementedError, match="repeat is not"):