Backport PR pandas-dev#56152: BUG: translate losing object dtype with new string dtype

Author: phofl

doc/source/whatsnew/v2.1.4.rst

@@ -25,7 +25,7 @@ Bug fixes
 - Bug in :meth:`Index.__getitem__` returning wrong result for Arrow dtypes and negative stepsize (:issue:`55832`)
 - Fixed bug in :meth:`DataFrame.__setitem__` casting :class:`Index` with object-dtype to PyArrow backed strings when ``infer_string`` option is set (:issue:`55638`)
 - Fixed bug in :meth:`Index.insert` casting object-dtype to PyArrow backed strings when ``infer_string`` option is set (:issue:`55638`)
--
+- Fixed bug in :meth:`Series.str.translate` losing object dtype when string option is set (:issue:`56152`)
 
 .. ---------------------------------------------------------------------------
 .. _whatsnew_214.other:

pandas/core/strings/accessor.py

@@ -259,6 +259,7 @@ def _wrap_result(
         fill_value=np.nan,
         returns_string: bool = True,
         returns_bool: bool = False,
+        dtype=None,
     ):
         from pandas import (
             Index,
@@ -379,29 +380,29 @@ def cons_row(x):
                     out = out.get_level_values(0)
                 return out
             else:
-                return Index(result, name=name)
+                return Index(result, name=name, dtype=dtype)
         else:
             index = self._orig.index
             # This is a mess.
-            dtype: DtypeObj | str | None
+            _dtype: DtypeObj | str | None = dtype
             vdtype = getattr(result, "dtype", None)
             if self._is_string:
                 if is_bool_dtype(vdtype):
-                    dtype = result.dtype
+                    _dtype = result.dtype
                 elif returns_string:
-                    dtype = self._orig.dtype
+                    _dtype = self._orig.dtype
                 else:
-                    dtype = vdtype
-            else:
-                dtype = vdtype
+                    _dtype = vdtype
+            elif vdtype is not None:
+                _dtype = vdtype
 
             if expand:
                 cons = self._orig._constructor_expanddim
-                result = cons(result, columns=name, index=index, dtype=dtype)
+                result = cons(result, columns=name, index=index, dtype=_dtype)
             else:
                 # Must be a Series
                 cons = self._orig._constructor
-                result = cons(result, name=name, index=index, dtype=dtype)
+                result = cons(result, name=name, index=index, dtype=_dtype)
             result = result.__finalize__(self._orig, method="str")
             if name is not None and result.ndim == 1:
                 # __finalize__ might copy over the original name, but we may
@@ -2317,7 +2318,8 @@ def translate(self, table):
         dtype: object
         """
         result = self._data.array._str_translate(table)
-        return self._wrap_result(result)
+        dtype = object if self._data.dtype == "object" else None
+        return self._wrap_result(result, dtype=dtype)
 
     @forbid_nonstring_types(["bytes"])
     def count(self, pat, flags: int = 0):

pandas/tests/strings/test_find_replace.py

@@ -5,6 +5,7 @@
 import pytest
 
 from pandas.errors import PerformanceWarning
+import pandas.util._test_decorators as td
 
 import pandas as pd
 from pandas import (
@@ -893,7 +894,10 @@ def test_find_nan(any_string_dtype):
 # --------------------------------------------------------------------------------------
 
 
-def test_translate(index_or_series, any_string_dtype):
+@pytest.mark.parametrize(
+    "infer_string", [False, pytest.param(True, marks=td.skip_if_no("pyarrow"))]
+)
+def test_translate(index_or_series, any_string_dtype, infer_string):
     obj = index_or_series(
         ["abcdefg", "abcc", "cdddfg", "cdefggg"], dtype=any_string_dtype
     )