BUG: fix parquet roundtrip with unsigned integer dtypes (#31918)

Author: jorisvandenbossche

doc/source/whatsnew/v1.0.2.rst

@@ -35,6 +35,8 @@ Bug fixes
 **I/O**
 
 - Using ``pd.NA`` with :meth:`DataFrame.to_json` now correctly outputs a null value instead of an empty object (:issue:`31615`)
+- Fixed bug in parquet roundtrip with nullable unsigned integer dtypes (:issue:`31896`).
+
 
 **Experimental dtypes**
 

pandas/core/arrays/integer.py

@@ -103,6 +103,10 @@ def __from_arrow__(
         import pyarrow  # noqa: F811
         from pandas.core.arrays._arrow_utils import pyarrow_array_to_numpy_and_mask
 
+        pyarrow_type = pyarrow.from_numpy_dtype(self.type)
+        if not array.type.equals(pyarrow_type):
+            array = array.cast(pyarrow_type)
+
         if isinstance(array, pyarrow.Array):
             chunks = [array]
         else:

pandas/tests/arrays/test_integer.py

@@ -1036,9 +1036,9 @@ def test_arrow_array(data):
     assert arr.equals(expected)
 
 
-@td.skip_if_no("pyarrow", min_version="0.15.1.dev")
+@td.skip_if_no("pyarrow", min_version="0.16.0")
 def test_arrow_roundtrip(data):
-    # roundtrip possible from arrow 1.0.0
+    # roundtrip possible from arrow 0.16.0
     import pyarrow as pa
 
     df = pd.DataFrame({"a": data})
@@ -1048,6 +1048,19 @@ def test_arrow_roundtrip(data):
     tm.assert_frame_equal(result, df)
 
 
+@td.skip_if_no("pyarrow", min_version="0.16.0")
+def test_arrow_from_arrow_uint():
+    # https://github.com/pandas-dev/pandas/issues/31896
+    # possible mismatch in types
+    import pyarrow as pa
+
+    dtype = pd.UInt32Dtype()
+    result = dtype.__from_arrow__(pa.array([1, 2, 3, 4, None], type="int64"))
+    expected = pd.array([1, 2, 3, 4, None], dtype="UInt32")
+
+    tm.assert_extension_array_equal(result, expected)
+
+
 @pytest.mark.parametrize(
     "pandasmethname, kwargs",
     [

pandas/tests/io/test_parquet.py

@@ -533,25 +533,28 @@ def test_additional_extension_arrays(self, pa):
         df = pd.DataFrame(
             {
                 "a": pd.Series([1, 2, 3], dtype="Int64"),
-                "b": pd.Series(["a", None, "c"], dtype="string"),
+                "b": pd.Series([1, 2, 3], dtype="UInt32"),
+                "c": pd.Series(["a", None, "c"], dtype="string"),
             }
         )
-        if LooseVersion(pyarrow.__version__) >= LooseVersion("0.15.1.dev"):
+        if LooseVersion(pyarrow.__version__) >= LooseVersion("0.16.0"):
             expected = df
         else:
             # de-serialized as plain int / object
-            expected = df.assign(a=df.a.astype("int64"), b=df.b.astype("object"))
+            expected = df.assign(
+                a=df.a.astype("int64"), b=df.b.astype("int64"), c=df.c.astype("object")
+            )
         check_round_trip(df, pa, expected=expected)
 
         df = pd.DataFrame({"a": pd.Series([1, 2, 3, None], dtype="Int64")})
-        if LooseVersion(pyarrow.__version__) >= LooseVersion("0.15.1.dev"):
+        if LooseVersion(pyarrow.__version__) >= LooseVersion("0.16.0"):
             expected = df
         else:
             # if missing values in integer, currently de-serialized as float
             expected = df.assign(a=df.a.astype("float64"))
         check_round_trip(df, pa, expected=expected)
 
-    @td.skip_if_no("pyarrow", min_version="0.15.1.dev")
+    @td.skip_if_no("pyarrow", min_version="0.16.0")
     def test_additional_extension_types(self, pa):
         # test additional ExtensionArrays that are supported through the
         # __arrow_array__ protocol + by defining a custom ExtensionType