DOC/TST: DataFrame constructor with a list of DataFrames

doc/source/whatsnew/v1.1.0.rst

@@ -695,6 +695,7 @@ Other API changes
 - :func: `merge` now checks ``suffixes`` parameter type to be ``tuple`` and raises ``TypeError``, whereas before a ``list`` or ``set`` were accepted and that the ``set`` could produce unexpected results (:issue:`33740`)
 - :class:`Period` no longer accepts tuples for the ``freq`` argument (:issue:`34658`)
 - :meth:`Series.interpolate` and :meth:`DataFrame.interpolate` now raises ValueError if ``limit_direction`` is 'forward' or 'both' and ``method`` is 'backfill' or 'bfill' or ``limit_direction`` is 'backward' or 'both' and ``method`` is 'pad' or 'ffill' (:issue:`34746`)
+- The :class:`DataFrame` constructor no longer accepts a list of ``DataFrame`` objects. Because of changes to NumPy, ``DataFrame`` objects are now consistently treated as 2D objects, so a list of ``DataFrames`` is considered 3D, and no longer acceptible for the ``DataFrame`` constructor (:issue:`32289`).
 
 
 Increased minimum versions for dependencies

pandas/core/internals/construction.py

@@ -321,7 +321,7 @@ def convert(v):
     if values.ndim == 1:
         values = values.reshape((values.shape[0], 1))
     elif values.ndim != 2:
-        raise ValueError("Must pass 2-d input")
+        raise ValueError(f"Must pass 2-d input. shape={values.shape}")
 
     return values
 

pandas/tests/frame/test_constructors.py

@@ -11,7 +11,7 @@
 import pytz
 
 from pandas.compat import PY37, is_platform_little_endian
-from pandas.compat.numpy import _is_numpy_dev
+from pandas.compat.numpy import _np_version_under1p19
 
 from pandas.core.dtypes.common import is_integer_dtype
 
@@ -147,14 +147,20 @@ def test_constructor_dtype_list_data(self):
         assert df.loc[1, 0] is None
         assert df.loc[0, 1] == "2"
 
-    @pytest.mark.xfail(_is_numpy_dev, reason="Interprets list of frame as 3D")
-    def test_constructor_list_frames(self):
-        # see gh-3243
-        result = DataFrame([DataFrame()])
-        assert result.shape == (1, 0)
+    @pytest.mark.skipif(_np_version_under1p19, reason="NumPy change.")
+    def test_constructor_list_of_2d_raises(self):
+        # https://github.com/pandas-dev/pandas/issues/32289
+        a = pd.DataFrame()
+        b = np.empty((0, 0))
+        with pytest.raises(ValueError, match=r"shape=\(1, 0, 0\)"):
+            pd.DataFrame([a])
 
-        result = DataFrame([DataFrame(dict(A=np.arange(5)))])
-        assert isinstance(result.iloc[0, 0], DataFrame)
+        with pytest.raises(ValueError, match=r"shape=\(1, 0, 0\)"):
+            pd.DataFrame([b])
+
+        a = pd.DataFrame({"A": [1, 2]})
+        with pytest.raises(ValueError, match=r"shape=\(2, 2, 1\)"):
+            pd.DataFrame([a, a])
 
     def test_constructor_mixed_dtypes(self):
         def _make_mixed_dtypes_df(typ, ad=None):
@@ -507,22 +513,6 @@ def test_constructor_error_msgs(self):
         with pytest.raises(ValueError, match=msg):
             DataFrame({"a": False, "b": True})
 
-    @pytest.mark.xfail(_is_numpy_dev, reason="Interprets embedded frame as 3D")
-    def test_constructor_with_embedded_frames(self):
-
-        # embedded data frames
-        df1 = DataFrame({"a": [1, 2, 3], "b": [3, 4, 5]})
-        df2 = DataFrame([df1, df1 + 10])
-
-        df2.dtypes
-        str(df2)
-
-        result = df2.loc[0, 0]
-        tm.assert_frame_equal(result, df1)
-
-        result = df2.loc[1, 0]
-        tm.assert_frame_equal(result, df1 + 10)
-
     def test_constructor_subclass_dict(self, float_frame, dict_subclass):
         # Test for passing dict subclass to constructor
         data = {