REF: remove ensure_datetime64ns, ensure_timedelta64ns (pandas-dev#47219)

Author: jbrockmendel

pandas/_libs/tslibs/conversion.pyi

@@ -10,17 +10,7 @@ from pandas._typing import npt
 DT64NS_DTYPE: np.dtype
 TD64NS_DTYPE: np.dtype
 
-class OutOfBoundsTimedelta(ValueError): ...
-
 def precision_from_unit(
     unit: str,
 ) -> tuple[int, int]: ...  # (int64_t, _)
-def ensure_datetime64ns(
-    arr: np.ndarray,  # np.ndarray[datetime64[ANY]]
-    copy: bool = ...,
-) -> np.ndarray: ...  # np.ndarray[datetime64ns]
-def ensure_timedelta64ns(
-    arr: np.ndarray,  # np.ndarray[timedelta64[ANY]]
-    copy: bool = ...,
-) -> np.ndarray: ...  # np.ndarray[timedelta64ns]
 def localize_pydatetime(dt: datetime, tz: tzinfo | None) -> datetime: ...

pandas/_libs/tslibs/conversion.pyx

@@ -193,45 +193,6 @@ cdef inline int64_t get_datetime64_nanos(object val) except? -1:
     return ival
 
 
-def ensure_datetime64ns(arr: ndarray, copy: bool = True):
-    """
-    Ensure a np.datetime64 array has dtype specifically 'datetime64[ns]'
-
-    Parameters
-    ----------
-    arr : ndarray
-    copy : bool, default True
-
-    Returns
-    -------
-    ndarray with dtype datetime64[ns]
-    """
-    if (<object>arr).dtype.byteorder == ">":
-        # GH#29684 we incorrectly get OutOfBoundsDatetime if we dont swap
-        dtype = arr.dtype
-        arr = arr.astype(dtype.newbyteorder("<"))
-
-    return astype_overflowsafe(arr, DT64NS_DTYPE, copy=copy)
-
-
-def ensure_timedelta64ns(arr: ndarray, copy: bool = True):
-    """
-    Ensure a np.timedelta64 array has dtype specifically 'timedelta64[ns]'
-
-    Parameters
-    ----------
-    arr : ndarray
-    copy : bool, default True
-
-    Returns
-    -------
-    ndarray[timedelta64[ns]]
-    """
-    assert arr.dtype.kind == "m", arr.dtype
-
-    return astype_overflowsafe(arr, dtype=TD64NS_DTYPE, copy=copy)
-
-
 # ----------------------------------------------------------------------
 # _TSObject Conversion
 

pandas/_libs/tslibs/np_datetime.pyx

@@ -319,6 +319,10 @@ cpdef ndarray astype_overflowsafe(
             "datetime64/timedelta64 values and dtype must have a unit specified"
         )
 
+    if (<object>values).dtype.byteorder == ">":
+        # GH#29684 we incorrectly get OutOfBoundsDatetime if we dont swap
+        values = values.astype(values.dtype.newbyteorder("<"))
+
     if from_unit == to_unit:
         # Check this before allocating result for perf, might save some memory
         if copy:

pandas/_libs/tslibs/period.pyx

@@ -47,6 +47,7 @@ from pandas._libs.tslibs.np_datetime cimport (
     NPY_DATETIMEUNIT,
     NPY_FR_D,
     NPY_FR_us,
+    astype_overflowsafe,
     check_dts_bounds,
     dt64_to_dtstruct,
     dtstruct_to_dt64,
@@ -71,7 +72,7 @@ from pandas._libs.tslibs.timedeltas cimport (
     is_any_td_scalar,
 )
 
-from pandas._libs.tslibs.conversion import ensure_datetime64ns
+from pandas._libs.tslibs.conversion import DT64NS_DTYPE
 
 from pandas._libs.tslibs.dtypes cimport (
     FR_ANN,
@@ -981,7 +982,7 @@ def periodarr_to_dt64arr(const int64_t[:] periodarr, int freq):
             dta = periodarr.base.view("M8[h]")
         elif freq == FR_DAY:
             dta = periodarr.base.view("M8[D]")
-        return ensure_datetime64ns(dta)
+        return astype_overflowsafe(dta, dtype=DT64NS_DTYPE)
 
 
 cdef void get_asfreq_info(int from_freq, int to_freq,

pandas/core/arrays/datetimes.py

@@ -25,7 +25,7 @@
     NaTType,
     Resolution,
     Timestamp,
-    conversion,
+    astype_overflowsafe,
     fields,
     get_resolution,
     iNaT,
@@ -2169,7 +2169,7 @@ def _sequence_to_dt64ns(
         # tz-naive DatetimeArray or ndarray[datetime64]
         data = getattr(data, "_ndarray", data)
         if data.dtype != DT64NS_DTYPE:
-            data = conversion.ensure_datetime64ns(data)
+            data = astype_overflowsafe(data, dtype=DT64NS_DTYPE)
             copy = False
 
         if tz is not None:

pandas/core/arrays/timedeltas.py

@@ -21,13 +21,11 @@
     Tick,
     Timedelta,
     Timestamp,
+    astype_overflowsafe,
     iNaT,
     to_offset,
 )
-from pandas._libs.tslibs.conversion import (
-    ensure_timedelta64ns,
-    precision_from_unit,
-)
+from pandas._libs.tslibs.conversion import precision_from_unit
 from pandas._libs.tslibs.fields import get_timedelta_field
 from pandas._libs.tslibs.timedeltas import (
     array_to_timedelta64,
@@ -1044,7 +1042,7 @@ def sequence_to_td64ns(
     elif is_timedelta64_dtype(data.dtype):
         if data.dtype != TD64NS_DTYPE:
             # non-nano unit
-            data = ensure_timedelta64ns(data)
+            data = astype_overflowsafe(data, dtype=TD64NS_DTYPE)
             copy = False
 
     else:
@@ -1086,7 +1084,7 @@ def ints_to_td64ns(data, unit="ns"):
         dtype_str = f"timedelta64[{unit}]"
         data = data.view(dtype_str)
 
-        data = ensure_timedelta64ns(data)
+        data = astype_overflowsafe(data, dtype=TD64NS_DTYPE)
 
         # the astype conversion makes a copy, so we can avoid re-copying later
         copy_made = True

pandas/core/dtypes/cast.py

@@ -30,7 +30,7 @@
     OutOfBoundsTimedelta,
     Timedelta,
     Timestamp,
-    conversion,
+    astype_overflowsafe,
 )
 from pandas._libs.tslibs.timedeltas import array_to_timedelta64
 from pandas._typing import (
@@ -1419,10 +1419,10 @@ def sanitize_to_nanoseconds(values: np.ndarray, copy: bool = False) -> np.ndarra
     """
     dtype = values.dtype
     if dtype.kind == "M" and dtype != DT64NS_DTYPE:
-        values = conversion.ensure_datetime64ns(values)
+        values = astype_overflowsafe(values, dtype=DT64NS_DTYPE)
 
     elif dtype.kind == "m" and dtype != TD64NS_DTYPE:
-        values = conversion.ensure_timedelta64ns(values)
+        values = astype_overflowsafe(values, dtype=TD64NS_DTYPE)
 
     elif copy:
         values = values.copy()

pandas/tests/arrays/categorical/test_constructors.py

@@ -751,7 +751,7 @@ def test_from_sequence_copy(self):
 
     @pytest.mark.xfail(
         not IS64 or is_platform_windows(),
-        reason="Incorrectly raising in ensure_datetime64ns",
+        reason="Incorrectly raising in astype_overflowsafe",
     )
     def test_constructor_datetime64_non_nano(self):
         categories = np.arange(10).view("M8[D]")

pandas/tests/indexes/datetimes/test_constructors.py

@@ -13,7 +13,7 @@
 
 from pandas._libs.tslibs import (
     OutOfBoundsDatetime,
-    conversion,
+    astype_overflowsafe,
 )
 from pandas.compat import PY39
 
@@ -975,7 +975,7 @@ def test_index_cast_datetime64_other_units(self):
         arr = np.arange(0, 100, 10, dtype=np.int64).view("M8[D]")
         idx = Index(arr)
 
-        assert (idx.values == conversion.ensure_datetime64ns(arr)).all()
+        assert (idx.values == astype_overflowsafe(arr, dtype=np.dtype("M8[ns]"))).all()
 
     def test_constructor_int64_nocopy(self):
         # GH#1624

pandas/tests/tslibs/test_conversion.py

@@ -6,6 +6,7 @@
 
 from pandas._libs.tslibs import (
     OutOfBoundsTimedelta,
+    astype_overflowsafe,
     conversion,
     iNaT,
     timezones,
@@ -106,7 +107,7 @@ def test_tz_convert_readonly():
 @pytest.mark.parametrize("dtype", ["M8[ns]", "M8[s]"])
 def test_length_zero_copy(dtype, copy):
     arr = np.array([], dtype=dtype)
-    result = conversion.ensure_datetime64ns(arr, copy=copy)
+    result = astype_overflowsafe(arr, copy=copy, dtype=np.dtype("M8[ns]"))
     if copy:
         assert not np.shares_memory(result, arr)
     else:
@@ -119,7 +120,7 @@ def test_length_zero_copy(dtype, copy):
 def test_ensure_datetime64ns_bigendian():
     # GH#29684
     arr = np.array([np.datetime64(1, "ms")], dtype=">M8[ms]")
-    result = conversion.ensure_datetime64ns(arr)
+    result = astype_overflowsafe(arr, dtype=np.dtype("M8[ns]"))
 
     expected = np.array([np.datetime64(1, "ms")], dtype="M8[ns]")
     tm.assert_numpy_array_equal(result, expected)
@@ -129,7 +130,7 @@ def test_ensure_timedelta64ns_overflows():
     arr = np.arange(10).astype("m8[Y]") * 100
     msg = r"Cannot convert 300 years to timedelta64\[ns\] without overflow"
     with pytest.raises(OutOfBoundsTimedelta, match=msg):
-        conversion.ensure_timedelta64ns(arr)
+        astype_overflowsafe(arr, dtype=np.dtype("m8[ns]"))
 
 
 class SubDatetime(datetime):