Updated rapidjson w/ fix for empty structured arrays & pandas data frames

Author: langmm

rapidjson

@@ -125,6 +125,15 @@ def test_structured_array(dumps, loads):
     np.testing.assert_equal(loaded, value)
 
 
+def test_structured_array_empty(dumps, loads):
+    value = np.array([], dtype=[('name', 'U0'), ('age', 'i4'),
+                                ('weight', 'f4'), ('color', 'S0')])
+    dumped = dumps(value)
+    loaded = loads(dumped)
+    assert type(loaded) is type(value) and loaded.dtype == value.dtype
+    np.testing.assert_equal(loaded, value)
+
+
 def test_pandas(dumps, loads):
     try:
         import pandas as pd
@@ -147,20 +156,13 @@ def test_pandas_empty(dumps, loads):
         pytest.skip("requires pandas")
     value = np.array([], dtype=[('name', 'U0'), ('age', 'i4'),
                                 ('weight', 'f4'), ('color', 'S0')])
+    # Because pandas stores both unicode & bytes as object type, the
+    # type defaults to bytes for empty arrays where the elements cannot
+    # be inspected for their type
+    value_def = np.array([], dtype=[('name', 'S0'), ('age', 'i4'),
+                                    ('weight', 'f4'), ('color', 'S0')])
     value_pd = pd.DataFrame(value)
     dumped = dumps(value_pd)
     loaded = loads(dumped)
-    loaded_value = loads(dumps(value))
-    print(loaded, loaded_value)
-    print(type(loaded), type(loaded_value))
-    # import pdb; pdb.set_trace()
-    assert (type(loaded) is type(loaded_value))
-    if isinstance(loaded, list):
-        assert len(loaded) == len(loaded_value)
-        np.testing.assert_equal(loaded, loaded_value)
-        # for a, b in zip(loaded, loaded_value):
-        #     np.testing.assert_equal(a, b)
-    else:
-        assert loaded.dtype == loaded_value.dtype
-        np.testing.assert_equal(loaded, value)
-        np.testing.assert_equal(loaded, loaded_value)
+    assert type(loaded) is type(value) and loaded.dtype == value_def.dtype
+    np.testing.assert_equal(loaded, value_def)