REF: pass dtype explicitly to _from_sequence (pandas-dev#59774)

Author: jbrockmendel

pandas/core/arrays/arrow/array.py

@@ -2300,7 +2300,13 @@ def _groupby_op(
         )
         if isinstance(result, np.ndarray):
             return result
-        return type(self)._from_sequence(result, copy=False)
+        elif isinstance(result, BaseMaskedArray):
+            pa_result = result.__arrow_array__()
+            return type(self)(pa_result)
+        else:
+            # DatetimeArray, TimedeltaArray
+            pa_result = pa.array(result, from_pandas=True)
+            return type(self)(pa_result)
 
     def _apply_elementwise(self, func: Callable) -> list[list[Any]]:
         """Apply a callable to each element while maintaining the chunking structure."""

pandas/core/arrays/datetimelike.py

@@ -1393,7 +1393,7 @@ def __add__(self, other):
         if isinstance(result, np.ndarray) and lib.is_np_dtype(result.dtype, "m"):
             from pandas.core.arrays import TimedeltaArray
 
-            return TimedeltaArray._from_sequence(result)
+            return TimedeltaArray._from_sequence(result, dtype=result.dtype)
         return result
 
     def __radd__(self, other):
@@ -1453,7 +1453,7 @@ def __sub__(self, other):
         if isinstance(result, np.ndarray) and lib.is_np_dtype(result.dtype, "m"):
             from pandas.core.arrays import TimedeltaArray
 
-            return TimedeltaArray._from_sequence(result)
+            return TimedeltaArray._from_sequence(result, dtype=result.dtype)
         return result
 
     def __rsub__(self, other):
@@ -1472,7 +1472,7 @@ def __rsub__(self, other):
                 # Avoid down-casting DatetimeIndex
                 from pandas.core.arrays import DatetimeArray
 
-                other = DatetimeArray._from_sequence(other)
+                other = DatetimeArray._from_sequence(other, dtype=other.dtype)
             return other - self
         elif self.dtype.kind == "M" and hasattr(other, "dtype") and not other_is_dt64:
             # GH#19959 datetime - datetime is well-defined as timedelta,

pandas/core/arrays/datetimes.py

@@ -818,11 +818,7 @@ def _add_offset(self, offset: BaseOffset) -> Self:
                     stacklevel=find_stack_level(),
                 )
             res_values = self.astype("O") + offset
-            # TODO(GH#55564): as_unit will be unnecessary
-            result = type(self)._from_sequence(res_values).as_unit(self.unit)
-            if not len(self):
-                # GH#30336 _from_sequence won't be able to infer self.tz
-                return result.tz_localize(self.tz)
+            result = type(self)._from_sequence(res_values, dtype=self.dtype)
 
         else:
             result = type(self)._simple_new(res_values, dtype=res_values.dtype)

pandas/core/arrays/period.py

@@ -812,7 +812,7 @@ def to_timestamp(self, freq=None, how: str = "start") -> DatetimeArray:
         new_parr = self.asfreq(freq, how=how)
 
         new_data = libperiod.periodarr_to_dt64arr(new_parr.asi8, base)
-        dta = DatetimeArray._from_sequence(new_data)
+        dta = DatetimeArray._from_sequence(new_data, dtype=np.dtype("M8[ns]"))
 
         if self.freq.name == "B":
             # See if we can retain BDay instead of Day in cases where

pandas/tests/arrays/datetimes/test_constructors.py

@@ -28,10 +28,12 @@ def test_mixing_naive_tzaware_raises(self, meth):
         # GH#24569
         arr = np.array([pd.Timestamp("2000"), pd.Timestamp("2000", tz="CET")])
 
-        msg = (
-            "Cannot mix tz-aware with tz-naive values|"
-            "Tz-aware datetime.datetime cannot be converted "
-            "to datetime64 unless utc=True"
+        msg = "|".join(
+            [
+                "Cannot mix tz-aware with tz-naive values",
+                "Tz-aware datetime.datetime cannot be converted "
+                "to datetime64 unless utc=True",
+            ]
         )
 
         for obj in [arr, arr[::-1]]:
@@ -63,10 +65,10 @@ def test_bool_dtype_raises(self):
 
     def test_copy(self):
         data = np.array([1, 2, 3], dtype="M8[ns]")
-        arr = DatetimeArray._from_sequence(data, copy=False)
+        arr = DatetimeArray._from_sequence(data, dtype=data.dtype, copy=False)
         assert arr._ndarray is data
 
-        arr = DatetimeArray._from_sequence(data, copy=True)
+        arr = DatetimeArray._from_sequence(data, dtype=data.dtype, copy=True)
         assert arr._ndarray is not data
 
     def test_numpy_datetime_unit(self, unit):
@@ -163,7 +165,9 @@ def test_from_arrow_from_empty(unit, tz):
     dtype = DatetimeTZDtype(unit=unit, tz=tz)
 
     result = dtype.__from_arrow__(arr)
-    expected = DatetimeArray._from_sequence(np.array(data, dtype=f"datetime64[{unit}]"))
+    expected = DatetimeArray._from_sequence(
+        np.array(data, dtype=f"datetime64[{unit}]"), dtype=np.dtype(f"M8[{unit}]")
+    )
     expected = expected.tz_localize(tz=tz)
     tm.assert_extension_array_equal(result, expected)
 
@@ -179,7 +183,9 @@ def test_from_arrow_from_integers():
     dtype = DatetimeTZDtype(unit="ns", tz="UTC")
 
     result = dtype.__from_arrow__(arr)
-    expected = DatetimeArray._from_sequence(np.array(data, dtype="datetime64[ns]"))
+    expected = DatetimeArray._from_sequence(
+        np.array(data, dtype="datetime64[ns]"), dtype=np.dtype("M8[ns]")
+    )
     expected = expected.tz_localize("UTC")
     tm.assert_extension_array_equal(result, expected)
 

pandas/tests/arrays/test_array.py

@@ -370,11 +370,15 @@ def test_array_copy():
         ),
         (
             np.array([1, 2], dtype="m8[ns]"),
-            TimedeltaArray._from_sequence(np.array([1, 2], dtype="m8[ns]")),
+            TimedeltaArray._from_sequence(
+                np.array([1, 2], dtype="m8[ns]"), dtype=np.dtype("m8[ns]")
+            ),
         ),
         (
             np.array([1, 2], dtype="m8[us]"),
-            TimedeltaArray._from_sequence(np.array([1, 2], dtype="m8[us]")),
+            TimedeltaArray._from_sequence(
+                np.array([1, 2], dtype="m8[us]"), dtype=np.dtype("m8[us]")
+            ),
         ),
         # integer
         ([1, 2], IntegerArray._from_sequence([1, 2], dtype="Int64")),

pandas/tests/arrays/test_datetimelike.py

@@ -257,7 +257,8 @@ def test_fillna_method_doesnt_change_orig(self, method):
         if self.array_cls is PeriodArray:
             arr = self.array_cls(data, dtype="period[D]")
         else:
-            arr = self.array_cls._from_sequence(data)
+            dtype = "M8[ns]" if self.array_cls is DatetimeArray else "m8[ns]"
+            arr = self.array_cls._from_sequence(data, dtype=np.dtype(dtype))
         arr[4] = NaT
 
         fill_value = arr[3] if method == "pad" else arr[5]
@@ -273,7 +274,8 @@ def test_searchsorted(self):
         if self.array_cls is PeriodArray:
             arr = self.array_cls(data, dtype="period[D]")
         else:
-            arr = self.array_cls._from_sequence(data)
+            dtype = "M8[ns]" if self.array_cls is DatetimeArray else "m8[ns]"
+            arr = self.array_cls._from_sequence(data, dtype=np.dtype(dtype))
 
         # scalar
         result = arr.searchsorted(arr[1])
@@ -739,10 +741,10 @@ def test_array_i8_dtype(self, arr1d):
     def test_from_array_keeps_base(self):
         # Ensure that DatetimeArray._ndarray.base isn't lost.
         arr = np.array(["2000-01-01", "2000-01-02"], dtype="M8[ns]")
-        dta = DatetimeArray._from_sequence(arr)
+        dta = DatetimeArray._from_sequence(arr, dtype=arr.dtype)
 
         assert dta._ndarray is arr
-        dta = DatetimeArray._from_sequence(arr[:0])
+        dta = DatetimeArray._from_sequence(arr[:0], dtype=arr.dtype)
         assert dta._ndarray.base is arr
 
     def test_from_dti(self, arr1d):

pandas/tests/arrays/test_datetimes.py

@@ -499,7 +499,7 @@ def test_value_counts_preserves_tz(self):
     @pytest.mark.parametrize("method", ["pad", "backfill"])
     def test_fillna_preserves_tz(self, method):
         dti = pd.date_range("2000-01-01", periods=5, freq="D", tz="US/Central")
-        arr = DatetimeArray._from_sequence(dti, copy=True)
+        arr = DatetimeArray._from_sequence(dti, dtype=dti.dtype, copy=True)
         arr[2] = pd.NaT
 
         fill_val = dti[1] if method == "pad" else dti[3]
@@ -665,7 +665,9 @@ def test_shift_fill_value(self):
         dti = pd.date_range("2016-01-01", periods=3)
 
         dta = dti._data
-        expected = DatetimeArray._from_sequence(np.roll(dta._ndarray, 1))
+        expected = DatetimeArray._from_sequence(
+            np.roll(dta._ndarray, 1), dtype=dti.dtype
+        )
 
         fv = dta[-1]
         for fill_value in [fv, fv.to_pydatetime(), fv.to_datetime64()]:
@@ -731,7 +733,11 @@ def test_iter_zoneinfo_fold(self, tz):
         )
         utc_vals *= 1_000_000_000
 
-        dta = DatetimeArray._from_sequence(utc_vals).tz_localize("UTC").tz_convert(tz)
+        dta = (
+            DatetimeArray._from_sequence(utc_vals, dtype=np.dtype("M8[ns]"))
+            .tz_localize("UTC")
+            .tz_convert(tz)
+        )
 
         left = dta[2]
         right = list(dta)[2]

pandas/tests/arrays/test_timedeltas.py

@@ -263,10 +263,10 @@ def test_searchsorted_invalid_types(self, other, index):
 class TestUnaryOps:
     def test_abs(self):
         vals = np.array([-3600 * 10**9, "NaT", 7200 * 10**9], dtype="m8[ns]")
-        arr = TimedeltaArray._from_sequence(vals)
+        arr = TimedeltaArray._from_sequence(vals, dtype=vals.dtype)
 
         evals = np.array([3600 * 10**9, "NaT", 7200 * 10**9], dtype="m8[ns]")
-        expected = TimedeltaArray._from_sequence(evals)
+        expected = TimedeltaArray._from_sequence(evals, dtype=evals.dtype)
 
         result = abs(arr)
         tm.assert_timedelta_array_equal(result, expected)
@@ -276,7 +276,7 @@ def test_abs(self):
 
     def test_pos(self):
         vals = np.array([-3600 * 10**9, "NaT", 7200 * 10**9], dtype="m8[ns]")
-        arr = TimedeltaArray._from_sequence(vals)
+        arr = TimedeltaArray._from_sequence(vals, dtype=vals.dtype)
 
         result = +arr
         tm.assert_timedelta_array_equal(result, arr)
@@ -288,7 +288,7 @@ def test_pos(self):
 
     def test_neg(self):
         vals = np.array([-3600 * 10**9, "NaT", 7200 * 10**9], dtype="m8[ns]")
-        arr = TimedeltaArray._from_sequence(vals)
+        arr = TimedeltaArray._from_sequence(vals, dtype=vals.dtype)
 
         evals = np.array([3600 * 10**9, "NaT", -7200 * 10**9], dtype="m8[ns]")
         expected = TimedeltaArray._from_sequence(evals)

pandas/tests/base/test_conversion.py

@@ -333,7 +333,8 @@ def test_array_multiindex_raises():
         # Timedelta
         (
             TimedeltaArray._from_sequence(
-                np.array([0, 3600000000000], dtype="i8").view("m8[ns]")
+                np.array([0, 3600000000000], dtype="i8").view("m8[ns]"),
+                dtype=np.dtype("m8[ns]"),
             ),
             np.array([0, 3600000000000], dtype="m8[ns]"),
         ),

pandas/tests/dtypes/test_generic.py

@@ -20,8 +20,8 @@ class TestABCClasses:
     df = pd.DataFrame({"names": ["a", "b", "c"]}, index=multi_index)
     sparse_array = pd.arrays.SparseArray(np.random.default_rng(2).standard_normal(10))
 
-    datetime_array = pd.core.arrays.DatetimeArray._from_sequence(datetime_index)
-    timedelta_array = pd.core.arrays.TimedeltaArray._from_sequence(timedelta_index)
+    datetime_array = datetime_index.array
+    timedelta_array = timedelta_index.array
 
     abc_pairs = [
         ("ABCMultiIndex", multi_index),

pandas/tests/series/accessors/test_dt_accessor.py

@@ -790,7 +790,8 @@ def test_end_time_timevalues(self, input_vals):
         # GH#17157
         # Check that the time part of the Period is adjusted by end_time
         # when using the dt accessor on a Series
-        input_vals = PeriodArray._from_sequence(np.asarray(input_vals))
+        dtype = pd.PeriodDtype(input_vals[0].freq)
+        input_vals = PeriodArray._from_sequence(np.asarray(input_vals), dtype=dtype)
 
         ser = Series(input_vals)
         result = ser.dt.end_time