TST (string dtype): change any_string_dtype fixture to use actual dtype instances (pandas-dev#59345)

* TST (string dtype): change any_string_dtype fixture to use actual dtype instances

* avoid pyarrow import error during test collection

* fix dtype equality in case pyarrow is not installed

* keep using mode.string_storage as default for NA variant + more xfails

* fix test_series_string_inference_storage_definition

* remove no longer necessary xfails

---------

Co-authored-by: Matthew Roeschke <[email protected]>

Author: jorisvandenbossche

pandas/conftest.py

@@ -1306,20 +1306,33 @@ def object_dtype(request):
 
 @pytest.fixture(
     params=[
-        "object",
-        "string[python]",
-        pytest.param("string[pyarrow]", marks=td.skip_if_no("pyarrow")),
-        pytest.param("string[pyarrow_numpy]", marks=td.skip_if_no("pyarrow")),
-    ]
+        np.dtype("object"),
+        ("python", pd.NA),
+        pytest.param(("pyarrow", pd.NA), marks=td.skip_if_no("pyarrow")),
+        pytest.param(("pyarrow", np.nan), marks=td.skip_if_no("pyarrow")),
+    ],
+    ids=[
+        "string=object",
+        "string=string[python]",
+        "string=string[pyarrow]",
+        "string=str[pyarrow]",
+    ],
 )
 def any_string_dtype(request):
     """
     Parametrized fixture for string dtypes.
     * 'object'
-    * 'string[python]'
-    * 'string[pyarrow]'
+    * 'string[python]' (NA variant)
+    * 'string[pyarrow]' (NA variant)
+    * 'str' (NaN variant, with pyarrow)
     """
-    return request.param
+    if isinstance(request.param, np.dtype):
+        return request.param
+    else:
+        # need to instantiate the StringDtype here instead of in the params
+        # to avoid importing pyarrow during test collection
+        storage, na_value = request.param
+        return pd.StringDtype(storage, na_value)
 
 
 @pytest.fixture(params=tm.DATETIME64_DTYPES)

pandas/core/arrays/string_.py

@@ -124,7 +124,7 @@ def __init__(
     ) -> None:
         # infer defaults
         if storage is None:
-            if using_string_dtype():
+            if using_string_dtype() and na_value is not libmissing.NA:
                 storage = "pyarrow"
             else:
                 storage = get_option("mode.string_storage")
@@ -162,7 +162,9 @@ def __eq__(self, other: object) -> bool:
                 return True
             try:
                 other = self.construct_from_string(other)
-            except TypeError:
+            except (TypeError, ImportError):
+                # TypeError if `other` is not a valid string for StringDtype
+                # ImportError if pyarrow is not installed for "string[pyarrow]"
                 return False
         if isinstance(other, type(self)):
             return self.storage == other.storage and self.na_value is other.na_value

pandas/tests/arrays/categorical/test_constructors.py

@@ -743,7 +743,6 @@ def test_interval(self):
         tm.assert_numpy_array_equal(cat.codes, expected_codes)
         tm.assert_index_equal(cat.categories, idx)
 
-    @pytest.mark.xfail(using_string_dtype(), reason="TODO(infer_string)")
     def test_categorical_extension_array_nullable(self, nulls_fixture):
         # GH:
         arr = pd.arrays.StringArray._from_sequence(

pandas/tests/copy_view/test_astype.py

@@ -98,7 +98,6 @@ def test_astype_numpy_to_ea():
     assert np.shares_memory(get_array(ser), get_array(result))
 
 
-@pytest.mark.xfail(using_string_dtype(), reason="TODO(infer_string)")
 @pytest.mark.parametrize(
     "dtype, new_dtype", [("object", "string"), ("string", "object")]
 )
@@ -116,7 +115,6 @@ def test_astype_string_and_object(using_copy_on_write, dtype, new_dtype):
     tm.assert_frame_equal(df, df_orig)
 
 
-@pytest.mark.xfail(using_string_dtype(), reason="TODO(infer_string)")
 @pytest.mark.parametrize(
     "dtype, new_dtype", [("object", "string"), ("string", "object")]
 )

pandas/tests/dtypes/test_common.py

@@ -3,8 +3,6 @@
 import numpy as np
 import pytest
 
-from pandas._config import using_string_dtype
-
 import pandas.util._test_decorators as td
 
 from pandas.core.dtypes.astype import astype_array
@@ -130,7 +128,6 @@ def test_dtype_equal(name1, dtype1, name2, dtype2):
         assert not com.is_dtype_equal(dtype1, dtype2)
 
 
-@pytest.mark.xfail(using_string_dtype(), reason="TODO(infer_string)", strict=False)
 @pytest.mark.parametrize("name,dtype", list(dtypes.items()), ids=lambda x: str(x))
 def test_pyarrow_string_import_error(name, dtype):
     # GH-44276

pandas/tests/io/parser/test_index_col.py

@@ -8,6 +8,8 @@
 import numpy as np
 import pytest
 
+from pandas._config import using_string_dtype
+
 from pandas import (
     DataFrame,
     Index,
@@ -342,6 +344,7 @@ def test_infer_types_boolean_sum(all_parsers):
     tm.assert_frame_equal(result, expected, check_index_type=False)
 
 
+@pytest.mark.xfail(using_string_dtype(), reason="TODO(infer_string)", strict=False)
 @pytest.mark.parametrize("dtype, val", [(object, "01"), ("int64", 1)])
 def test_specify_dtype_for_index_col(all_parsers, dtype, val, request):
     # GH#9435

pandas/tests/series/test_constructors.py

@@ -2138,7 +2138,7 @@ def test_series_string_inference_storage_definition(self):
         # returning the NA string dtype, so expected is changed from
         # "string[pyarrow_numpy]" to "string[pyarrow]"
         pytest.importorskip("pyarrow")
-        expected = Series(["a", "b"], dtype="string[pyarrow]")
+        expected = Series(["a", "b"], dtype="string[python]")
         with pd.option_context("future.infer_string", True):
             result = Series(["a", "b"], dtype="string")
         tm.assert_series_equal(result, expected)

pandas/tests/strings/__init__.py

@@ -2,7 +2,15 @@
 
 import pandas as pd
 
-object_pyarrow_numpy = ("object", "string[pyarrow_numpy]")
+
+def is_object_or_nan_string_dtype(dtype):
+    """
+    Check if string-like dtype is following NaN semantics, i.e. is object
+    dtype or a NaN-variant of the StringDtype.
+    """
+    return (isinstance(dtype, np.dtype) and dtype == "object") or (
+        dtype.na_value is np.nan
+    )
 
 
 def _convert_na_value(ser, expected):

pandas/tests/strings/test_find_replace.py

@@ -14,7 +14,7 @@
 )
 from pandas.tests.strings import (
     _convert_na_value,
-    object_pyarrow_numpy,
+    is_object_or_nan_string_dtype,
 )
 
 # --------------------------------------------------------------------------------------
@@ -34,7 +34,9 @@ def test_contains(any_string_dtype):
     pat = "mmm[_]+"
 
     result = values.str.contains(pat)
-    expected_dtype = "object" if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        "object" if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
     expected = Series(
         np.array([False, np.nan, True, True, False], dtype=np.object_),
         dtype=expected_dtype,
@@ -53,7 +55,9 @@ def test_contains(any_string_dtype):
         dtype=any_string_dtype,
     )
     result = values.str.contains(pat)
-    expected_dtype = np.bool_ if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        np.bool_ if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
     expected = Series(np.array([False, False, True, True]), dtype=expected_dtype)
     tm.assert_series_equal(result, expected)
 
@@ -80,14 +84,18 @@ def test_contains(any_string_dtype):
     pat = "mmm[_]+"
 
     result = values.str.contains(pat)
-    expected_dtype = "object" if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        "object" if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
     expected = Series(
         np.array([False, np.nan, True, True], dtype=np.object_), dtype=expected_dtype
     )
     tm.assert_series_equal(result, expected)
 
     result = values.str.contains(pat, na=False)
-    expected_dtype = np.bool_ if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        np.bool_ if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
     expected = Series(np.array([False, False, True, True]), dtype=expected_dtype)
     tm.assert_series_equal(result, expected)
 
@@ -172,7 +180,9 @@ def test_contains_moar(any_string_dtype):
     )
 
     result = s.str.contains("a")
-    expected_dtype = "object" if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        "object" if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
     expected = Series(
         [False, False, False, True, True, False, np.nan, False, False, True],
         dtype=expected_dtype,
@@ -213,7 +223,9 @@ def test_contains_nan(any_string_dtype):
     s = Series([np.nan, np.nan, np.nan], dtype=any_string_dtype)
 
     result = s.str.contains("foo", na=False)
-    expected_dtype = np.bool_ if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        np.bool_ if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
     expected = Series([False, False, False], dtype=expected_dtype)
     tm.assert_series_equal(result, expected)
 
@@ -231,7 +243,9 @@ def test_contains_nan(any_string_dtype):
     tm.assert_series_equal(result, expected)
 
     result = s.str.contains("foo")
-    expected_dtype = "object" if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        "object" if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
     expected = Series([np.nan, np.nan, np.nan], dtype=expected_dtype)
     tm.assert_series_equal(result, expected)
 
@@ -641,7 +655,9 @@ def test_replace_regex_single_character(regex, any_string_dtype):
 
 def test_match(any_string_dtype):
     # New match behavior introduced in 0.13
-    expected_dtype = "object" if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        "object" if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
 
     values = Series(["fooBAD__barBAD", np.nan, "foo"], dtype=any_string_dtype)
     result = values.str.match(".*(BAD[_]+).*(BAD)")
@@ -696,20 +712,26 @@ def test_match_na_kwarg(any_string_dtype):
     s = Series(["a", "b", np.nan], dtype=any_string_dtype)
 
     result = s.str.match("a", na=False)
-    expected_dtype = np.bool_ if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        np.bool_ if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
     expected = Series([True, False, False], dtype=expected_dtype)
     tm.assert_series_equal(result, expected)
 
     result = s.str.match("a")
-    expected_dtype = "object" if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        "object" if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
     expected = Series([True, False, np.nan], dtype=expected_dtype)
     tm.assert_series_equal(result, expected)
 
 
 def test_match_case_kwarg(any_string_dtype):
     values = Series(["ab", "AB", "abc", "ABC"], dtype=any_string_dtype)
     result = values.str.match("ab", case=False)
-    expected_dtype = np.bool_ if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        np.bool_ if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
     expected = Series([True, True, True, True], dtype=expected_dtype)
     tm.assert_series_equal(result, expected)
 
@@ -725,7 +747,9 @@ def test_fullmatch(any_string_dtype):
         ["fooBAD__barBAD", "BAD_BADleroybrown", np.nan, "foo"], dtype=any_string_dtype
     )
     result = ser.str.fullmatch(".*BAD[_]+.*BAD")
-    expected_dtype = "object" if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        "object" if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
     expected = Series([True, False, np.nan, False], dtype=expected_dtype)
     tm.assert_series_equal(result, expected)
 
@@ -734,7 +758,9 @@ def test_fullmatch_dollar_literal(any_string_dtype):
     # GH 56652
     ser = Series(["foo", "foo$foo", np.nan, "foo$"], dtype=any_string_dtype)
     result = ser.str.fullmatch("foo\\$")
-    expected_dtype = "object" if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        "object" if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
     expected = Series([False, False, np.nan, True], dtype=expected_dtype)
     tm.assert_series_equal(result, expected)
 
@@ -744,14 +770,18 @@ def test_fullmatch_na_kwarg(any_string_dtype):
         ["fooBAD__barBAD", "BAD_BADleroybrown", np.nan, "foo"], dtype=any_string_dtype
     )
     result = ser.str.fullmatch(".*BAD[_]+.*BAD", na=False)
-    expected_dtype = np.bool_ if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        np.bool_ if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
     expected = Series([True, False, False, False], dtype=expected_dtype)
     tm.assert_series_equal(result, expected)
 
 
 def test_fullmatch_case_kwarg(any_string_dtype):
     ser = Series(["ab", "AB", "abc", "ABC"], dtype=any_string_dtype)
-    expected_dtype = np.bool_ if any_string_dtype in object_pyarrow_numpy else "boolean"
+    expected_dtype = (
+        np.bool_ if is_object_or_nan_string_dtype(any_string_dtype) else "boolean"
+    )
 
     expected = Series([True, False, False, False], dtype=expected_dtype)
 
@@ -823,7 +853,9 @@ def test_find(any_string_dtype):
     ser = Series(
         ["ABCDEFG", "BCDEFEF", "DEFGHIJEF", "EFGHEF", "XXXX"], dtype=any_string_dtype
     )
-    expected_dtype = np.int64 if any_string_dtype in object_pyarrow_numpy else "Int64"
+    expected_dtype = (
+        np.int64 if is_object_or_nan_string_dtype(any_string_dtype) else "Int64"
+    )
 
     result = ser.str.find("EF")
     expected = Series([4, 3, 1, 0, -1], dtype=expected_dtype)
@@ -875,7 +907,9 @@ def test_find_nan(any_string_dtype):
     ser = Series(
         ["ABCDEFG", np.nan, "DEFGHIJEF", np.nan, "XXXX"], dtype=any_string_dtype
     )
-    expected_dtype = np.float64 if any_string_dtype in object_pyarrow_numpy else "Int64"
+    expected_dtype = (
+        np.float64 if is_object_or_nan_string_dtype(any_string_dtype) else "Int64"
+    )
 
     result = ser.str.find("EF")
     expected = Series([4, np.nan, 1, np.nan, -1], dtype=expected_dtype)

pandas/tests/strings/test_split_partition.py

@@ -14,7 +14,7 @@
 )
 from pandas.tests.strings import (
     _convert_na_value,
-    object_pyarrow_numpy,
+    is_object_or_nan_string_dtype,
 )
 
 
@@ -384,7 +384,7 @@ def test_split_nan_expand(any_string_dtype):
     # check that these are actually np.nan/pd.NA and not None
     # TODO see GH 18463
     # tm.assert_frame_equal does not differentiate
-    if any_string_dtype in object_pyarrow_numpy:
+    if is_object_or_nan_string_dtype(any_string_dtype):
         assert all(np.isnan(x) for x in result.iloc[1])
     else:
         assert all(x is pd.NA for x in result.iloc[1])