Switch default string storage from python to pyarrow (if installed) also for NA-variant of the StringDtype

Author: jorisvandenbossche

pandas/core/arrays/string_.py

@@ -154,16 +154,11 @@ def __init__(
     ) -> None:
         # infer defaults
         if storage is None:
-            if na_value is not libmissing.NA:
-                storage = get_option("mode.string_storage")
-                if storage == "auto":
-                    if HAS_PYARROW:
-                        storage = "pyarrow"
-                    else:
-                        storage = "python"
-            else:
-                storage = get_option("mode.string_storage")
-                if storage == "auto":
+            storage = get_option("mode.string_storage")
+            if storage == "auto":
+                if HAS_PYARROW:
+                    storage = "pyarrow"
+                else:
                     storage = "python"
 
         if storage == "pyarrow_numpy":

pandas/tests/arrays/string_/test_string_arrow.py

@@ -4,6 +4,7 @@
 import numpy as np
 import pytest
 
+from pandas.compat import HAS_PYARROW
 import pandas.util._test_decorators as td
 
 import pandas as pd
@@ -26,10 +27,10 @@ def test_eq_all_na():
     tm.assert_extension_array_equal(result, expected)
 
 
-def test_config(string_storage, using_infer_string):
+def test_config(string_storage):
     # with the default string_storage setting
     # always "python" at the moment
-    assert StringDtype().storage == "python"
+    assert StringDtype().storage == "pyarrow" if HAS_PYARROW else "python"
 
     with pd.option_context("string_storage", string_storage):
         assert StringDtype().storage == string_storage