feedback: revert changes to dtype checks

Author: arw2019

pandas/core/dtypes/cast.py

@@ -23,6 +23,7 @@
 from pandas._libs.tslibs import (
     NaT,
     OutOfBoundsDatetime,
+    Period,
     Timedelta,
     Timestamp,
     conversion,
@@ -53,7 +54,6 @@
     is_datetime64_dtype,
     is_datetime64_ns_dtype,
     is_datetime64tz_dtype,
-    is_datetime_or_timedelta_any_dtype,
     is_datetime_or_timedelta_dtype,
     is_dtype_equal,
     is_extension_array_dtype,
@@ -63,7 +63,6 @@
     is_integer_dtype,
     is_numeric_dtype,
     is_object_dtype,
-    is_period_dtype,
     is_scalar,
     is_sparse,
     is_string_dtype,
@@ -147,7 +146,7 @@ def maybe_downcast_to_dtype(result, dtype: Dtype):
     elif isinstance(result, ABCDataFrame):
         # occurs in pivot_table doctest
         return result
-    elif is_period_dtype(dtype):
+    elif dtype.type is Period:
         from pandas.core.arrays import PeriodArray
 
         with suppress(TypeError):
@@ -184,7 +183,7 @@ def maybe_downcast_to_dtype(result, dtype: Dtype):
 
     # a datetimelike
     # GH12821, iNaT is cast to float
-    if is_datetime_or_timedelta_any_dtype(dtype) and result.dtype.kind in ["i", "f"]:
+    if dtype.kind in ["M", "m"] and result.dtype.kind in ["i", "f"]:
 
         if is_datetime_or_timedelta_dtype(dtype):
             result = result.astype(dtype)

pandas/core/dtypes/common.py

@@ -1067,10 +1067,6 @@ def is_datetime_or_timedelta_dtype(arr_or_dtype) -> bool:
     return _is_dtype_type(arr_or_dtype, classes(np.datetime64, np.timedelta64))
 
 
-def is_datetime_or_timedelta_any_dtype(arr_or_dtype):
-    return is_datetime64_any_dtype(arr_or_dtype) or is_timedelta64_dtype(arr_or_dtype)
-
-
 # This exists to silence numpy deprecation warnings, see GH#29553
 def is_numeric_v_string_like(a, b):
     """