REF: remove how arg from maybe_cast_result (#41000)

Author: jbrockmendel

pandas/core/dtypes/cast.py

@@ -55,7 +55,6 @@
     ensure_str,
     is_bool,
     is_bool_dtype,
-    is_categorical_dtype,
     is_complex,
     is_complex_dtype,
     is_datetime64_dtype,
@@ -79,6 +78,7 @@
     pandas_dtype,
 )
 from pandas.core.dtypes.dtypes import (
+    CategoricalDtype,
     DatetimeTZDtype,
     ExtensionDtype,
     IntervalDtype,
@@ -359,15 +359,15 @@ def trans(x):
     return result
 
 
-def maybe_cast_result(
+def maybe_cast_pointwise_result(
     result: ArrayLike,
     dtype: DtypeObj,
     numeric_only: bool = False,
-    how: str = "",
     same_dtype: bool = True,
 ) -> ArrayLike:
     """
-    Try casting result to a different type if appropriate
+    Try casting result of a pointwise operation back to the original dtype if
+    appropriate.
 
     Parameters
     ----------
@@ -377,8 +377,6 @@ def maybe_cast_result(
         Input Series from which result was calculated.
     numeric_only : bool, default False
         Whether to cast only numerics or datetimes as well.
-    how : str, default ""
-        How the result was computed.
     same_dtype : bool, default True
         Specify dtype when calling _from_sequence
 
@@ -387,12 +385,12 @@ def maybe_cast_result(
     result : array-like
         result maybe casted to the dtype.
     """
-    dtype = maybe_cast_result_dtype(dtype, how)
 
     assert not is_scalar(result)
 
     if isinstance(dtype, ExtensionDtype):
-        if not is_categorical_dtype(dtype) and dtype.kind != "M":
+        if not isinstance(dtype, (CategoricalDtype, DatetimeTZDtype)):
+            # TODO: avoid this special-casing
             # We have to special case categorical so as not to upcast
             # things like counts back to categorical
 

pandas/core/groupby/ops.py

@@ -36,7 +36,7 @@
 from pandas.util._decorators import cache_readonly
 
 from pandas.core.dtypes.cast import (
-    maybe_cast_result,
+    maybe_cast_pointwise_result,
     maybe_cast_result_dtype,
     maybe_downcast_to_dtype,
 )
@@ -798,7 +798,7 @@ def _aggregate_series_pure_python(self, obj: Series, func: F):
             result[label] = res
 
         out = lib.maybe_convert_objects(result, try_float=False)
-        out = maybe_cast_result(out, obj.dtype, numeric_only=True)
+        out = maybe_cast_pointwise_result(out, obj.dtype, numeric_only=True)
 
         return out, counts
 

pandas/core/series.py

@@ -60,7 +60,7 @@
 from pandas.core.dtypes.cast import (
     convert_dtypes,
     maybe_box_native,
-    maybe_cast_result,
+    maybe_cast_pointwise_result,
     validate_numeric_casting,
 )
 from pandas.core.dtypes.common import (
@@ -3070,22 +3070,26 @@ def combine(self, other, func, fill_value=None) -> Series:
             # so do this element by element
             new_index = self.index.union(other.index)
             new_name = ops.get_op_result_name(self, other)
-            new_values = []
-            for idx in new_index:
+            new_values = np.empty(len(new_index), dtype=object)
+            for i, idx in enumerate(new_index):
                 lv = self.get(idx, fill_value)
                 rv = other.get(idx, fill_value)
                 with np.errstate(all="ignore"):
-                    new_values.append(func(lv, rv))
+                    new_values[i] = func(lv, rv)
         else:
             # Assume that other is a scalar, so apply the function for
             # each element in the Series
             new_index = self.index
+            new_values = np.empty(len(new_index), dtype=object)
             with np.errstate(all="ignore"):
-                new_values = [func(lv, other) for lv in self._values]
+                new_values[:] = [func(lv, other) for lv in self._values]
             new_name = self.name
 
-        res_values = sanitize_array(new_values, None)
-        res_values = maybe_cast_result(res_values, self.dtype, same_dtype=False)
+        # try_float=False is to match _aggregate_series_pure_python
+        res_values = lib.maybe_convert_objects(new_values, try_float=False)
+        res_values = maybe_cast_pointwise_result(
+            res_values, self.dtype, same_dtype=False
+        )
         return self._constructor(res_values, index=new_index, name=new_name)
 
     def combine_first(self, other) -> Series: