TST/BUG (string dtype): Fix and adjust indexes string tests (#59544)

Co-authored-by: Joris Van den Bossche <[email protected]>

Author: phofl

pandas/core/construction.py

@@ -609,7 +609,10 @@ def sanitize_array(
                 dtype = StringDtype(na_value=np.nan)
                 subarr = dtype.construct_array_type()._from_sequence(data, dtype=dtype)
 
-            if subarr is data and copy:
+            if (
+                subarr is data
+                or (subarr.dtype == "str" and subarr.dtype.storage == "python")  # type: ignore[union-attr]
+            ) and copy:
                 subarr = subarr.copy()
 
         else:

pandas/core/indexes/base.py

@@ -506,7 +506,8 @@ def __new__(
 
         elif is_ea_or_datetimelike_dtype(dtype):
             # non-EA dtype indexes have special casting logic, so we punt here
-            pass
+            if isinstance(data, (set, frozenset)):
+                data = list(data)
 
         elif is_ea_or_datetimelike_dtype(data_dtype):
             pass
@@ -6995,6 +6996,9 @@ def insert(self, loc: int, item) -> Index:
             #  We cannot keep the same dtype, so cast to the (often object)
             #  minimal shared dtype before doing the insert.
             dtype = self._find_common_type_compat(item)
+            if dtype == self.dtype:
+                # EA's might run into recursion errors if loc is invalid
+                raise
             return self.astype(dtype).insert(loc, item)
 
         if arr.dtype != object or not isinstance(

pandas/tests/indexes/base_class/test_setops.py

@@ -3,8 +3,6 @@
 import numpy as np
 import pytest
 
-from pandas._config import using_string_dtype
-
 import pandas as pd
 from pandas import (
     Index,
@@ -233,7 +231,6 @@ def test_tuple_union_bug(self, method, expected, sort):
         expected = Index(expected)
         tm.assert_index_equal(result, expected)
 
-    @pytest.mark.xfail(using_string_dtype(), reason="TODO(infer_string)", strict=False)
     @pytest.mark.parametrize("first_list", [["b", "a"], []])
     @pytest.mark.parametrize("second_list", [["a", "b"], []])
     @pytest.mark.parametrize(
@@ -243,6 +240,7 @@ def test_tuple_union_bug(self, method, expected, sort):
     def test_union_name_preservation(
         self, first_list, second_list, first_name, second_name, expected_name, sort
     ):
+        expected_dtype = object if not first_list or not second_list else "str"
         first = Index(first_list, name=first_name)
         second = Index(second_list, name=second_name)
         union = first.union(second, sort=sort)
@@ -253,7 +251,7 @@ def test_union_name_preservation(
             expected = Index(sorted(vals), name=expected_name)
             tm.assert_index_equal(union, expected)
         else:
-            expected = Index(vals, name=expected_name)
+            expected = Index(vals, name=expected_name, dtype=expected_dtype)
             tm.assert_index_equal(union.sort_values(), expected.sort_values())
 
     @pytest.mark.parametrize(

pandas/tests/indexes/test_base.py

@@ -76,9 +76,6 @@ def test_constructor_casting(self, index):
         tm.assert_contains_all(arr, new_index)
         tm.assert_index_equal(index, new_index)
 
-    @pytest.mark.xfail(
-        using_string_dtype() and not HAS_PYARROW, reason="TODO(infer_string)"
-    )
     def test_constructor_copy(self, using_infer_string):
         index = Index(list("abc"), name="name")
         arr = np.array(index)
@@ -346,11 +343,6 @@ def test_constructor_empty_special(self, empty, klass):
     def test_view_with_args(self, index):
         index.view("i8")
 
-    @pytest.mark.xfail(
-        using_string_dtype() and not HAS_PYARROW,
-        reason="TODO(infer_string)",
-        strict=False,
-    )
     @pytest.mark.parametrize(
         "index",
         [
@@ -367,7 +359,8 @@ def test_view_with_args_object_array_raises(self, index):
             msg = "When changing to a larger dtype"
             with pytest.raises(ValueError, match=msg):
                 index.view("i8")
-        elif index.dtype == "string":
+        elif index.dtype == "str" and not index.dtype.storage == "python":
+            # TODO(infer_string): Make the errors consistent
             with pytest.raises(NotImplementedError, match="i8"):
                 index.view("i8")
         else:

pandas/tests/indexes/test_old_base.py

@@ -6,10 +6,7 @@
 import numpy as np
 import pytest
 
-from pandas._config import using_string_dtype
-
 from pandas._libs.tslibs import Timestamp
-from pandas.compat import HAS_PYARROW
 
 from pandas.core.dtypes.common import (
     is_integer_dtype,
@@ -28,6 +25,7 @@
     PeriodIndex,
     RangeIndex,
     Series,
+    StringDtype,
     TimedeltaIndex,
     isna,
     period_range,
@@ -233,7 +231,6 @@ def test_logical_compat(self, simple_index):
             with pytest.raises(TypeError, match=msg):
                 idx.any()
 
-    @pytest.mark.xfail(using_string_dtype(), reason="TODO(infer_string)", strict=False)
     def test_repr_roundtrip(self, simple_index):
         if isinstance(simple_index, IntervalIndex):
             pytest.skip(f"Not a valid repr for {type(simple_index).__name__}")
@@ -250,11 +247,6 @@ def test_repr_max_seq_item_setting(self, simple_index):
             repr(idx)
             assert "..." not in str(idx)
 
-    @pytest.mark.xfail(
-        using_string_dtype() and not HAS_PYARROW,
-        reason="TODO(infer_string)",
-        strict=False,
-    )
     @pytest.mark.filterwarnings(r"ignore:PeriodDtype\[B\] is deprecated:FutureWarning")
     def test_ensure_copied_data(self, index):
         # Check the "copy" argument of each Index.__new__ is honoured
@@ -302,7 +294,9 @@ def test_ensure_copied_data(self, index):
                 tm.assert_numpy_array_equal(
                     index._values._mask, result._values._mask, check_same="same"
                 )
-            elif index.dtype == "string[python]":
+            elif (
+                isinstance(index.dtype, StringDtype) and index.dtype.storage == "python"
+            ):
                 assert np.shares_memory(index._values._ndarray, result._values._ndarray)
                 tm.assert_numpy_array_equal(
                     index._values._ndarray, result._values._ndarray, check_same="same"
@@ -432,11 +426,7 @@ def test_insert_base(self, index):
             result = trimmed.insert(0, index[0])
         assert index[0:4].equals(result)
 
-    @pytest.mark.skipif(
-        using_string_dtype(),
-        reason="completely different behavior, tested elsewher",
-    )
-    def test_insert_out_of_bounds(self, index):
+    def test_insert_out_of_bounds(self, index, using_infer_string):
         # TypeError/IndexError matches what np.insert raises in these cases
 
         if len(index) > 0:
@@ -448,6 +438,12 @@ def test_insert_out_of_bounds(self, index):
             msg = "index (0|0.5) is out of bounds for axis 0 with size 0"
         else:
             msg = "slice indices must be integers or None or have an __index__ method"
+
+        if using_infer_string and (
+            index.dtype == "string" or index.dtype == "category"  # noqa: PLR1714
+        ):
+            msg = "loc must be an integer between"
+
         with pytest.raises(err, match=msg):
             index.insert(0.5, "foo")