BUG: Coerce to object for mixed concat with extension arrays (#20799)

Author: TomAugspurger

pandas/core/dtypes/concat.py

@@ -8,6 +8,7 @@
 from pandas.core.dtypes.common import (
     is_categorical_dtype,
     is_sparse,
+    is_extension_array_dtype,
     is_datetimetz,
     is_datetime64_dtype,
     is_timedelta64_dtype,
@@ -173,6 +174,10 @@ def is_nonempty(x):
     elif 'sparse' in typs:
         return _concat_sparse(to_concat, axis=axis, typs=typs)
 
+    extensions = [is_extension_array_dtype(x) for x in to_concat]
+    if any(extensions):
+        to_concat = [np.atleast_2d(x.astype('object')) for x in to_concat]
+
     if not nonempty:
         # we have all empties, but may need to coerce the result dtype to
         # object if we have non-numeric type operands (numpy would otherwise
@@ -210,7 +215,7 @@ def _concat_categorical(to_concat, axis=0):
 
     def _concat_asobject(to_concat):
         to_concat = [x.get_values() if is_categorical_dtype(x.dtype)
-                     else x.ravel() for x in to_concat]
+                     else np.asarray(x).ravel() for x in to_concat]
         res = _concat_compat(to_concat)
         if axis == 1:
             return res.reshape(1, len(res))
@@ -548,6 +553,8 @@ def convert_sparse(x, axis):
         # coerce to native type
         if isinstance(x, SparseArray):
             x = x.get_values()
+        else:
+            x = np.asarray(x)
         x = x.ravel()
         if axis > 0:
             x = np.atleast_2d(x)

pandas/tests/extension/base/reshaping.py

@@ -41,6 +41,29 @@ def test_concat_all_na_block(self, data_missing, in_frame):
             expected = pd.Series(data_missing.take([1, 1, 0, 0]))
             self.assert_series_equal(result, expected)
 
+    def test_concat_mixed_dtypes(self, data):
+        # https://github.com/pandas-dev/pandas/issues/20762
+        df1 = pd.DataFrame({'A': data[:3]})
+        df2 = pd.DataFrame({"A": [1, 2, 3]})
+        df3 = pd.DataFrame({"A": ['a', 'b', 'c']}).astype('category')
+        df4 = pd.DataFrame({"A": pd.SparseArray([1, 2, 3])})
+        dfs = [df1, df2, df3, df4]
+
+        # dataframes
+        result = pd.concat(dfs)
+        expected = pd.concat([x.astype(object) for x in dfs])
+        self.assert_frame_equal(result, expected)
+
+        # series
+        result = pd.concat([x['A'] for x in dfs])
+        expected = pd.concat([x['A'].astype(object) for x in dfs])
+        self.assert_series_equal(result, expected)
+
+        # simple test for just EA and one other
+        result = pd.concat([df1, df2])
+        expected = pd.concat([df1.astype('object'), df2.astype('object')])
+        self.assert_frame_equal(result, expected)
+
     def test_align(self, data, na_value):
         a = data[:3]
         b = data[2:5]