BUG: frame[object].astype(M8[unsupported]) not raising (#50015)

Author: jbrockmendel

pandas/core/dtypes/astype.py

@@ -135,16 +135,15 @@ def astype_nansafe(
     elif is_object_dtype(arr.dtype):
 
         # if we have a datetime/timedelta array of objects
-        # then coerce to a proper dtype and recall astype_nansafe
+        # then coerce to datetime64[ns] and use DatetimeArray.astype
 
         if is_datetime64_dtype(dtype):
             from pandas import to_datetime
 
-            return astype_nansafe(
-                to_datetime(arr.ravel()).values.reshape(arr.shape),
-                dtype,
-                copy=copy,
-            )
+            dti = to_datetime(arr.ravel())
+            dta = dti._data.reshape(arr.shape)
+            return dta.astype(dtype, copy=False)._ndarray
+
         elif is_timedelta64_dtype(dtype):
             # bc we know arr.dtype == object, this is equivalent to
             #  `np.asarray(to_timedelta(arr))`, but using a lower-level API that

pandas/tests/dtypes/test_common.py

@@ -757,7 +757,13 @@ def test_astype_datetime64_bad_dtype_raises(from_type, to_type):
 
     to_type = np.dtype(to_type)
 
-    with pytest.raises(TypeError, match="cannot astype"):
+    msg = "|".join(
+        [
+            "cannot astype a timedelta",
+            "cannot astype a datetimelike",
+        ]
+    )
+    with pytest.raises(TypeError, match=msg):
         astype_nansafe(arr, dtype=to_type)
 
 

pandas/tests/frame/methods/test_astype.py

@@ -377,6 +377,16 @@ def test_astype_column_metadata(self, dtype):
         df = df.astype(dtype)
         tm.assert_index_equal(df.columns, columns)
 
+    @pytest.mark.parametrize("unit", ["Y", "M", "W", "D", "h", "m"])
+    def test_astype_from_object_to_datetime_unit(self, unit):
+        vals = [
+            ["2015-01-01", "2015-01-02", "2015-01-03"],
+            ["2017-01-01", "2017-01-02", "2017-02-03"],
+        ]
+        df = DataFrame(vals, dtype=object)
+        with pytest.raises(TypeError, match="Cannot cast"):
+            df.astype(f"M8[{unit}]")
+
     @pytest.mark.parametrize("dtype", ["M8", "m8"])
     @pytest.mark.parametrize("unit", ["ns", "us", "ms", "s", "h", "m", "D"])
     def test_astype_from_datetimelike_to_object(self, dtype, unit):

pandas/tests/frame/test_constructors.py

@@ -1955,19 +1955,11 @@ def test_constructor_datetimes_with_nulls(self, arr):
 
     @pytest.mark.parametrize("order", ["K", "A", "C", "F"])
     @pytest.mark.parametrize(
-        "dtype",
-        [
-            "datetime64[M]",
-            "datetime64[D]",
-            "datetime64[h]",
-            "datetime64[m]",
-            "datetime64[s]",
-            "datetime64[ms]",
-            "datetime64[us]",
-            "datetime64[ns]",
-        ],
+        "unit",
+        ["M", "D", "h", "m", "s", "ms", "us", "ns"],
     )
-    def test_constructor_datetimes_non_ns(self, order, dtype):
+    def test_constructor_datetimes_non_ns(self, order, unit):
+        dtype = f"datetime64[{unit}]"
         na = np.array(
             [
                 ["2015-01-01", "2015-01-02", "2015-01-03"],
@@ -1977,13 +1969,16 @@ def test_constructor_datetimes_non_ns(self, order, dtype):
             order=order,
         )
         df = DataFrame(na)
-        expected = DataFrame(
-            [
-                ["2015-01-01", "2015-01-02", "2015-01-03"],
-                ["2017-01-01", "2017-01-02", "2017-02-03"],
-            ]
-        )
-        expected = expected.astype(dtype=dtype)
+        expected = DataFrame(na.astype("M8[ns]"))
+        if unit in ["M", "D", "h", "m"]:
+            with pytest.raises(TypeError, match="Cannot cast"):
+                expected.astype(dtype)
+
+            # instead the constructor casts to the closest supported reso, i.e. "s"
+            expected = expected.astype("datetime64[s]")
+        else:
+            expected = expected.astype(dtype=dtype)
+
         tm.assert_frame_equal(df, expected)
 
     @pytest.mark.parametrize("order", ["K", "A", "C", "F"])

pandas/tests/io/xml/test_xml_dtypes.py

@@ -128,14 +128,14 @@ def test_dtypes_with_names(parser):
     df_result = read_xml(
         xml_dates,
         names=["Col1", "Col2", "Col3", "Col4"],
-        dtype={"Col2": "string", "Col3": "Int64", "Col4": "datetime64"},
+        dtype={"Col2": "string", "Col3": "Int64", "Col4": "datetime64[ns]"},
         parser=parser,
     )
     df_iter = read_xml_iterparse(
         xml_dates,
         parser=parser,
         names=["Col1", "Col2", "Col3", "Col4"],
-        dtype={"Col2": "string", "Col3": "Int64", "Col4": "datetime64"},
+        dtype={"Col2": "string", "Col3": "Int64", "Col4": "datetime64[ns]"},
         iterparse={"row": ["shape", "degrees", "sides", "date"]},
     )
 

pandas/tests/reshape/merge/test_merge.py

@@ -730,13 +730,13 @@ def test_other_datetime_unit(self, unit):
         ser = Series([None, None], index=[101, 102], name="days")
 
         dtype = f"datetime64[{unit}]"
-        df2 = ser.astype(dtype).to_frame("days")
 
         if unit in ["D", "h", "m"]:
             # not supported so we cast to the nearest supported unit, seconds
             exp_dtype = "datetime64[s]"
         else:
             exp_dtype = dtype
+        df2 = ser.astype(exp_dtype).to_frame("days")
         assert df2["days"].dtype == exp_dtype
 
         result = df1.merge(df2, left_on="entity_id", right_index=True)