Cleanup some timedelta/datetime column logic

python/cudf/cudf/core/_internals/timezones.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2023, NVIDIA CORPORATION.
+# Copyright (c) 2023-2024, NVIDIA CORPORATION.
 
 import os
 import zoneinfo
@@ -111,7 +111,7 @@ def _find_ambiguous_and_nonexistent(
     tz_data_for_zone = get_tz_data(zone_name)
     transition_times = tz_data_for_zone["transition_times"]
     offsets = tz_data_for_zone["offsets"].astype(
-        f"timedelta64[{data._time_unit}]"
+        f"timedelta64[{data.time_unit}]"
     )
 
     if len(offsets) == 1:  # no transitions
@@ -180,7 +180,7 @@ def localize(
             "Already localized. "
             "Use `tz_convert` to convert between time zones."
         )
-    dtype = pd.DatetimeTZDtype(data._time_unit, zone_name)
+    dtype = pd.DatetimeTZDtype(data.time_unit, zone_name)
     ambiguous, nonexistent = _find_ambiguous_and_nonexistent(data, zone_name)
     localized = cast(
         DatetimeColumn,
@@ -227,7 +227,7 @@ def convert(data: DatetimeTZColumn, zone_name: str) -> DatetimeTZColumn:
         DatetimeTZColumn,
         build_column(
             data=utc_time.base_data,
-            dtype=pd.DatetimeTZDtype(data._time_unit, zone_name),
+            dtype=pd.DatetimeTZDtype(data.time_unit, zone_name),
             mask=utc_time.base_mask,
             size=utc_time.size,
             offset=utc_time.offset,

python/cudf/cudf/core/column/datetime.py

@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 import datetime
+import functools
 import locale
 import re
 from locale import nl_langinfo
@@ -239,6 +240,8 @@ def __init__(
         null_count: Optional[int] = None,
     ):
         dtype = cudf.dtype(dtype)
+        if dtype.kind != "M":
+            raise TypeError(f"{self.dtype} is not a supported datetime type")
 
         if data.size % dtype.itemsize:
             raise ValueError("Buffer size must be divisible by element size")
@@ -254,24 +257,21 @@ def __init__(
             null_count=null_count,
         )
 
-        if self.dtype.type is not np.datetime64:
-            raise TypeError(f"{self.dtype} is not a supported datetime type")
-
-        self._time_unit, _ = np.datetime_data(self.dtype)
-
     def __contains__(self, item: ScalarLike) -> bool:
         try:
-            item_as_dt64 = np.datetime64(item, self._time_unit)
+            item_as_dt64 = np.datetime64(item, self.time_unit)
         except ValueError:
             # If item cannot be converted to datetime type
             # np.datetime64 raises ValueError, hence `item`
             # cannot exist in `self`.
             return False
         return item_as_dt64.astype("int64") in self.as_numerical
 
-    @property
+    @functools.cached_property
     def time_unit(self) -> str:
-        return self._time_unit
+        if isinstance(self.dtype, pd.DatetimeTZDtype):
+            return self.dtype.unit
+        return np.datetime_data(self.dtype)[0]
 
     @property
     def year(self) -> ColumnBase:

python/cudf/cudf/core/column/timedelta.py

@@ -1,8 +1,9 @@
-# Copyright (c) 2020-2023, NVIDIA CORPORATION.
+# Copyright (c) 2020-2024, NVIDIA CORPORATION.
 
 from __future__ import annotations
 
 import datetime
+import functools
 from typing import Any, Optional, Sequence, cast
 
 import numpy as np
@@ -18,13 +19,6 @@
 from cudf.utils.dtypes import np_to_pa_dtype
 from cudf.utils.utils import _all_bools_with_nulls
 
-_dtype_to_format_conversion = {
-    "timedelta64[ns]": "%D days %H:%M:%S",
-    "timedelta64[us]": "%D days %H:%M:%S",
-    "timedelta64[ms]": "%D days %H:%M:%S",
-    "timedelta64[s]": "%D days %H:%M:%S",
-}
-
 _unit_to_nanoseconds_conversion = {
     "ns": 1,
     "us": 1_000,
@@ -86,6 +80,8 @@ def __init__(
         null_count: Optional[int] = None,
     ):
         dtype = cudf.dtype(dtype)
+        if dtype.kind != "m":
+            raise TypeError(f"{self.dtype} is not a supported duration type")
 
         if data.size % dtype.itemsize:
             raise ValueError("Buffer size must be divisible by element size")
@@ -101,14 +97,9 @@ def __init__(
             null_count=null_count,
         )
 
-        if self.dtype.type is not np.timedelta64:
-            raise TypeError(f"{self.dtype} is not a supported duration type")
-
-        self._time_unit, _ = np.datetime_data(self.dtype)
-
     def __contains__(self, item: DatetimeLikeScalar) -> bool:
         try:
-            item = np.timedelta64(item, self._time_unit)
+            item = np.timedelta64(item, self.time_unit)
         except ValueError:
             # If item cannot be converted to duration type
             # np.timedelta64 raises ValueError, hence `item`
@@ -235,16 +226,12 @@ def normalize_binop_value(self, other) -> ColumnBinaryOperand:
             "Cannot perform binary operation on timezone-naive columns"
             " and timezone-aware timestamps."
         )
-        if isinstance(other, pd.Timestamp):
-            if other.tz is not None:
+        if isinstance(other, datetime.datetime):
+            if other.tzinfo is not None:
                 raise NotImplementedError(tz_error_msg)
-            other = other.to_datetime64()
-        elif isinstance(other, pd.Timedelta):
-            other = other.to_timedelta64()
+            other = pd.Timestamp(other).to_datetime64()
         elif isinstance(other, datetime.timedelta):
-            other = np.timedelta64(other)
-        elif isinstance(other, datetime.datetime) and other.tzinfo is not None:
-            raise NotImplementedError(tz_error_msg)
+            other = pd.Timedelta(other).to_timedelta64()
 
         if isinstance(other, np.timedelta64):
             other_time_unit = cudf.utils.dtypes.get_time_unit(other)
@@ -256,7 +243,7 @@ def normalize_binop_value(self, other) -> ColumnBinaryOperand:
             else:
                 common_dtype = determine_out_dtype(self.dtype, other.dtype)
             return cudf.Scalar(other.astype(common_dtype))
-        elif np.isscalar(other):
+        elif is_scalar(other):
             return cudf.Scalar(other)
         return NotImplemented
 
@@ -273,9 +260,9 @@ def as_numerical(self) -> "cudf.core.column.NumericalColumn":
             ),
         )
 
-    @property
+    @functools.cached_property
     def time_unit(self) -> str:
-        return self._time_unit
+        return np.datetime_data(self.dtype)[0]
 
     def fillna(
         self,
@@ -318,9 +305,7 @@ def as_string_column(
         self, dtype: Dtype, format=None, **kwargs
     ) -> "cudf.core.column.StringColumn":
         if format is None:
-            format = _dtype_to_format_conversion.get(
-                self.dtype.name, "%D days %H:%M:%S"
-            )
+            format = "%D days %H:%M:%S"
         if len(self) > 0:
             return string._timedelta_to_str_typecast_functions[
                 cudf.dtype(self.dtype)