BUG: Fix GroupBy aggregate coersion of outputs inconsistency for pyarrow dtypes

pandas/core/arrays/arrow/array.py

@@ -185,6 +185,7 @@ def floordiv_compat(
         ArrayLike,
         AxisInt,
         Dtype,
+        DtypeObj,
         FillnaOptions,
         InterpolateOptions,
         Iterator,
@@ -313,6 +314,18 @@ def __init__(self, values: pa.Array | pa.ChunkedArray) -> None:
             )
         self._dtype = ArrowDtype(self._pa_array.type)
 
+    @classmethod
+    def _from_scalars(cls, scalars, dtype: DtypeObj) -> Self:
+        try:
+            pa_array = cls._from_sequence(scalars, dtype=dtype)
+        except pa.ArrowNotImplementedError:
+            # _from_scalars should only raise ValueError or TypeError.
+            raise ValueError
+
+        if lib.infer_dtype(scalars, skipna=True) != lib.infer_dtype(pa_array, skipna=True):
+            raise ValueError
+        return pa_array
+
     @classmethod
     def _from_sequence(
         cls, scalars, *, dtype: Dtype | None = None, copy: bool = False

pandas/core/arrays/string_arrow.py

@@ -53,6 +53,7 @@
     from pandas._typing import (
         ArrayLike,
         Dtype,
+        DtypeObj,
         NpDtype,
         Self,
         npt,
@@ -180,6 +181,12 @@ def __len__(self) -> int:
         """
         return len(self._pa_array)
 
+    @classmethod
+    def _from_scalars(cls, scalars, dtype: DtypeObj) -> Self:
+        if lib.infer_dtype(scalars, skipna=True) not in ["string", "empty"]:
+            raise ValueError
+        return cls._from_sequence(scalars, dtype=dtype)
+
     @classmethod
     def _from_sequence(
         cls, scalars, *, dtype: Dtype | None = None, copy: bool = False