BUG: Series construction with mismatched dt64 data vs td64 dtype (pandas-dev#38764)

Author: jbrockmendel

doc/source/whatsnew/v1.3.0.rst

@@ -189,7 +189,7 @@ Datetimelike
 ^^^^^^^^^^^^
 - Bug in :class:`DataFrame` and :class:`Series` constructors sometimes dropping nanoseconds from :class:`Timestamp` (resp. :class:`Timedelta`) ``data``, with ``dtype=datetime64[ns]`` (resp. ``timedelta64[ns]``) (:issue:`38032`)
 - Bug in :meth:`DataFrame.first` and :meth:`Series.first` returning two months for offset one month when first day is last calendar day (:issue:`29623`)
-- Bug in constructing a :class:`DataFrame` or :class:`Series` with mismatched ``datetime64`` data and ``timedelta64`` dtype, or vice-versa, failing to raise ``TypeError`` (:issue:`38575`)
+- Bug in constructing a :class:`DataFrame` or :class:`Series` with mismatched ``datetime64`` data and ``timedelta64`` dtype, or vice-versa, failing to raise ``TypeError`` (:issue:`38575`, :issue:`38764`)
 - Bug in :meth:`DatetimeIndex.intersection`, :meth:`DatetimeIndex.symmetric_difference`, :meth:`PeriodIndex.intersection`, :meth:`PeriodIndex.symmetric_difference` always returning object-dtype when operating with :class:`CategoricalIndex` (:issue:`38741`)
 - Bug in :meth:`Series.where` incorrectly casting ``datetime64`` values to ``int64`` (:issue:`37682`)
 -

pandas/core/dtypes/cast.py

@@ -166,14 +166,23 @@ def maybe_unbox_datetimelike(value: Scalar, dtype: DtypeObj) -> Scalar:
     elif isinstance(value, Timedelta):
         value = value.to_timedelta64()
 
-    if (isinstance(value, np.timedelta64) and dtype.kind == "M") or (
-        isinstance(value, np.datetime64) and dtype.kind == "m"
+    _disallow_mismatched_datetimelike(value, dtype)
+    return value
+
+
+def _disallow_mismatched_datetimelike(value: DtypeObj, dtype: DtypeObj):
+    """
+    numpy allows np.array(dt64values, dtype="timedelta64[ns]") and
+    vice-versa, but we do not want to allow this, so we need to
+    check explicitly
+    """
+    vdtype = getattr(value, "dtype", None)
+    if vdtype is None:
+        return
+    elif (vdtype.kind == "m" and dtype.kind == "M") or (
+        vdtype.kind == "M" and dtype.kind == "m"
     ):
-        # numpy allows np.array(dt64values, dtype="timedelta64[ns]") and
-        #  vice-versa, but we do not want to allow this, so we need to
-        #  check explicitly
         raise TypeError(f"Cannot cast {repr(value)} to {dtype}")
-    return value
 
 
 def maybe_downcast_to_dtype(result, dtype: Union[str, np.dtype]):
@@ -1715,6 +1724,8 @@ def construct_1d_ndarray_preserving_na(
     if dtype is not None and dtype.kind == "U":
         subarr = lib.ensure_string_array(values, convert_na_value=False, copy=copy)
     else:
+        if dtype is not None:
+            _disallow_mismatched_datetimelike(values, dtype)
         subarr = np.array(values, dtype=dtype, copy=copy)
 
     return subarr

pandas/tests/frame/test_constructors.py

@@ -2925,12 +2925,31 @@ def get1(obj):
 
 
 class TestFromScalar:
-    @pytest.fixture
-    def constructor(self, frame_or_series):
-        if frame_or_series is Series:
-            return functools.partial(Series, index=range(2))
+    @pytest.fixture(params=[list, dict, None])
+    def constructor(self, request, frame_or_series):
+        box = request.param
+
+        extra = {"index": range(2)}
+        if frame_or_series is DataFrame:
+            extra["columns"] = ["A"]
+
+        if box is None:
+            return functools.partial(frame_or_series, **extra)
+
+        elif box is dict:
+            if frame_or_series is Series:
+                return lambda x, **kwargs: frame_or_series(
+                    {0: x, 1: x}, **extra, **kwargs
+                )
+            else:
+                return lambda x, **kwargs: frame_or_series({"A": x}, **extra, **kwargs)
         else:
-            return functools.partial(DataFrame, index=range(2), columns=range(2))
+            if frame_or_series is Series:
+                return lambda x, **kwargs: frame_or_series([x, x], **extra, **kwargs)
+            else:
+                return lambda x, **kwargs: frame_or_series(
+                    {"A": [x, x]}, **extra, **kwargs
+                )
 
     @pytest.mark.parametrize("dtype", ["M8[ns]", "m8[ns]"])
     def test_from_nat_scalar(self, dtype, constructor):
@@ -2951,7 +2970,8 @@ def test_from_timestamp_scalar_preserves_nanos(self, constructor):
         assert get1(obj) == ts
 
     def test_from_timedelta64_scalar_object(self, constructor, request):
-        if constructor.func is DataFrame and _np_version_under1p20:
+        if getattr(constructor, "func", None) is DataFrame and _np_version_under1p20:
+            # getattr check means we only xfail when box is None
             mark = pytest.mark.xfail(
                 reason="np.array(td64, dtype=object) converts to int"
             )
@@ -2964,7 +2984,15 @@ def test_from_timedelta64_scalar_object(self, constructor, request):
         assert isinstance(get1(obj), np.timedelta64)
 
     @pytest.mark.parametrize("cls", [np.datetime64, np.timedelta64])
-    def test_from_scalar_datetimelike_mismatched(self, constructor, cls):
+    def test_from_scalar_datetimelike_mismatched(self, constructor, cls, request):
+        node = request.node
+        params = node.callspec.params
+        if params["frame_or_series"] is DataFrame and params["constructor"] is not None:
+            mark = pytest.mark.xfail(
+                reason="DataFrame incorrectly allows mismatched datetimelike"
+            )
+            node.add_marker(mark)
+
         scalar = cls("NaT", "ns")
         dtype = {np.datetime64: "m8[ns]", np.timedelta64: "M8[ns]"}[cls]