Fix empty string bug on dataframe regression

CHANGELOG.rst

@@ -2,6 +2,7 @@
 ------------------
 
 * `#54 <https://github.com/ESSS/pytest-regressions/pull/54>`__: New ``--with-test-class-names`` command-line flag to consider test class names when composing the expected and obtained data filenames. Needed when the same module contains different classes with the same method names.
+* `#74 <https://github.com/ESSS/pytest-regressions/pull/74>`__: Fix ``empty string bug`` on dataframe regression.
 
 2.2.0 (2020-01-27)
 ------------------

src/pytest_regressions/dataframe_regression.py

@@ -97,8 +97,8 @@ def _check_fn(self, obtained_filename, expected_filename):
 
         __tracebackhide__ = True
 
-        obtained_data = pd.read_csv(str(obtained_filename))
-        expected_data = pd.read_csv(str(expected_filename))
+        obtained_data = pd.read_csv(str(obtained_filename), keep_default_na=False)
+        expected_data = pd.read_csv(str(expected_filename), keep_default_na=False)
 
         comparison_tables_dict = {}
         for k in obtained_data.keys():

tests/test_dataframe_regression.py

@@ -252,3 +252,14 @@ def test_non_pandas_dataframe(dataframe_regression):
         " *Object with type '%s' was given." % (str(type(data)),),
     ):
         dataframe_regression.check(data)
+
+
+def test_frame(dataframe_regression):
+    df = pd.DataFrame.from_records(
+        [
+            {"a": "a", "b": "b"},
+            {"a": "a1", "b": ""},
+        ]
+    )
+
+    dataframe_regression.check(df)

tests/test_dataframe_regression/test_frame.csv

@@ -0,0 +1,3 @@
+,a,b
+0,a,b
+1,a1,