GH1159 Disallow StringMethods for Series and Index when subtype is not str

Author: loicdiridollou

pandas-stubs/core/frame.pyi

@@ -256,9 +256,9 @@ class _LocIndexerFrame(_LocIndexer, Generic[_T]):
             ]
             | None
         ),
-    ) -> Series: ...
+    ) -> UnknownSeries: ...
     @overload
-    def __getitem__(self, idx: tuple[Scalar, slice]) -> Series | _T: ...
+    def __getitem__(self, idx: tuple[Scalar, slice]) -> UnknownSeries | _T: ...
     @overload
     def __setitem__(
         self,
@@ -288,13 +288,13 @@ class _LocIndexerFrame(_LocIndexer, Generic[_T]):
 if sys.version_info >= (3, 12):
     class _GetItemHack:
         @overload
-        def __getitem__(self, key: Scalar | tuple[Hashable, ...]) -> Series: ...  # type: ignore[overload-overlap] # pyright: ignore[reportOverlappingOverload]
+        def __getitem__(self, key: Scalar | tuple[Hashable, ...]) -> UnknownSeries: ...  # type: ignore[overload-overlap] # pyright: ignore[reportOverlappingOverload]
         @overload
         def __getitem__(  # type: ignore[overload-overlap] # pyright: ignore[reportOverlappingOverload]
             self, key: Iterable[Hashable] | slice
         ) -> Self: ...
         @overload
-        def __getitem__(self, key: Hashable) -> Series: ...
+        def __getitem__(self, key: Hashable) -> UnknownSeries: ...
 
 else:
     class _GetItemHack:

pandas-stubs/core/series.pyi

@@ -245,6 +245,54 @@ _ListLike: TypeAlias = (
     ArrayLike | dict[_str, np.ndarray] | Sequence[S1] | IndexOpsMixin[S1]
 )
 
+class _StrMethods:
+    @overload
+    def __get__(self, instance: Series[str], owner: Any) -> StringMethods[
+        Series[str],
+        DataFrame,
+        Series[bool],
+        Series[list[str]],
+        Series[int],
+        Series[bytes],
+        Series[str],
+        Series[type[object]],
+    ]: ...
+    @overload
+    def __get__(self, instance: Series[bytes], owner: Any) -> StringMethods[
+        Series[bytes],
+        DataFrame,
+        Series[bool],
+        Series[list[str]],
+        Series[int],
+        Series[bytes],
+        Series[str],
+        Series[type[object]],
+    ]: ...
+    @overload
+    def __get__(self, instance: Series[list[str]], owner: Any) -> StringMethods[
+        Series[list[str]],
+        DataFrame,
+        Series[bool],
+        Series[list[str]],
+        Series[int],
+        Series[bytes],
+        Series[str],
+        Series[type[object]],
+    ]: ...
+    @overload
+    def __get__(self, instance: Series[S1], owner: Any) -> NoReturn: ...
+    @overload
+    def __get__(self, instance: UnknownSeries, owner: Any) -> StringMethods[
+        Series,
+        DataFrame,
+        Series[bool],
+        Series[list[str]],
+        Series[int],
+        Series[bytes],
+        Series[str],
+        Series[type[object]],
+    ]: ...
+
 class Series(IndexOpsMixin[S1], NDFrame):
     __hash__: ClassVar[None]
 
@@ -1170,19 +1218,7 @@ class Series(IndexOpsMixin[S1], NDFrame):
         copy: _bool = ...,
     ) -> Series[S1]: ...
     def to_period(self, freq: _str | None = ..., copy: _bool = ...) -> DataFrame: ...
-    @property
-    def str(
-        self,
-    ) -> StringMethods[
-        Self,
-        DataFrame,
-        Series[bool],
-        Series[list[str]],
-        Series[int],
-        Series[bytes],
-        Series[str],
-        Series[type[object]],
-    ]: ...
+    str: _StrMethods
     @property
     def dt(self) -> CombinedDatetimelikeProperties: ...
     @property

pyproject.toml

@@ -35,7 +35,7 @@ types-pytz = ">= 2022.1.1"
 numpy = ">= 1.23.5"
 
 [tool.poetry.group.dev.dependencies]
-mypy = "1.15.0"
+mypy = "1.16.0"
 pandas = "2.2.3"
 pyarrow = ">=10.0.1"
 pytest = ">=7.1.2"

tests/test_frame.py

@@ -3073,15 +3073,15 @@ def test_set_columns() -> None:
     # https://github.com/python/mypy/issues/3004
     # pyright accepts this, so we only type check for pyright,
     # and also test the code with pytest
-    df.columns = ["c", "d"]  # type: ignore[assignment]
-    df.columns = [1, 2]  # type: ignore[assignment]
-    df.columns = [1, "a"]  # type: ignore[assignment]
-    df.columns = np.array([1, 2])  # type: ignore[assignment]
-    df.columns = pd.Series([1, 2])  # type: ignore[assignment]
-    df.columns = np.array([1, "a"])  # type: ignore[assignment]
-    df.columns = pd.Series([1, "a"])  # type: ignore[assignment]
-    df.columns = (1, 2)  # type: ignore[assignment]
-    df.columns = (1, "a")  # type: ignore[assignment]
+    df.columns = ["c", "d"]
+    df.columns = [1, 2]
+    df.columns = [1, "a"]
+    df.columns = np.array([1, 2])
+    df.columns = pd.Series([1, 2])
+    df.columns = np.array([1, "a"])
+    df.columns = pd.Series([1, "a"])
+    df.columns = (1, 2)
+    df.columns = (1, "a")
     if TYPE_CHECKING_INVALID_USAGE:
         df.columns = "abc"  # type: ignore[assignment] # pyright: ignore[reportAttributeAccessIssue]
 
@@ -4373,8 +4373,8 @@ def test_hashable_args() -> None:
     # https://github.com/python/mypy/issues/3004
     # pyright accepts this, so we only type check for pyright,
     # and also test the code with pytest
-    df.columns = test  # type: ignore[assignment]
-    df.columns = ["test"]  # type: ignore[assignment]
+    df.columns = test
+    df.columns = ["test"]
 
     testDict = {"test": 1}
     with ensure_clean() as path:

tests/test_string_accessors.py

@@ -6,6 +6,7 @@
 from typing_extensions import assert_type
 
 from tests import (
+    TYPE_CHECKING_INVALID_USAGE,
     check,
     np_ndarray_bool,
 )
@@ -411,3 +412,10 @@ def test_index_overloads_extract():
         pd.Index,
         object,
     )
+
+
+def test_series_unknown():
+    if TYPE_CHECKING_INVALID_USAGE:
+        s = pd.Series([1, 2, 3])
+        s.str.startswith("a")  # type:ignore[attr-defined]
+        s.str.slice(2, 4)  # type:ignore[attr-defined]