2 files changed, 69 insertions, 4 deletions
diff --git a/yaksh/evaluator_tests/test_python_evaluation.py b/yaksh/evaluator_tests/test_python_evaluation.py
index 1fba73e..82cf4c3 100644
--- a/yaksh/evaluator_tests/test_python_evaluation.py
+++ b/yaksh/evaluator_tests/test_python_evaluation.py
@@ -400,7 +400,7 @@ class PythonAssertionEvaluationTestCases(EvaluatorBaseTest):
         result = grader.evaluate(kwargs)
         error_as_str = ''.join(result.get("error"))
         err = error_as_str.splitlines()
-        
+
         # Then
         self.assertFalse(result.get("success"))
         self.assertEqual(5, len(err))
@@ -578,7 +578,10 @@ class PythonStdIOEvaluationTestCases(EvaluatorBaseTest):
 
         # Then
         self.assertFalse(result.get('success'))
-        self.assert_correct_output("Incorrect answer", result.get('error'))
+        self.assert_correct_output(
+            "ERROR:\nExpected:\n3\nGiven:\n-1\n\nError in line 1 of output.",
+            result.get('error')
+        )
 
     def test_file_based_answer(self):
         # Given
@@ -728,7 +731,7 @@ class PythonHookEvaluationTestCases(EvaluatorBaseTest):
         # Then
         self.assertFalse(result.get('success'))
         self.assert_correct_output('Incorrect Answer', result.get('error'))
-    
+
     def test_assert_with_hook(self):
         # Given
         user_answer = "def add(a,b):\n\treturn a + b"
@@ -815,7 +818,7 @@ class PythonHookEvaluationTestCases(EvaluatorBaseTest):
         # Then
         self.assertTrue(result.get('success'))
         self.assertEqual(result.get("weight"), 1.5)
-        
+
     def test_infinite_loop(self):
         # Given
         user_answer = "def add(a, b):\n\twhile True:\n\t\tpass"
diff --git a/yaksh/evaluator_tests/test_python_stdio_evaluator.py b/yaksh/evaluator_tests/test_python_stdio_evaluator.py
new file mode 100644
index 0000000..db5028a
--- /dev/null
+++ b/yaksh/evaluator_tests/test_python_stdio_evaluator.py
@@ -0,0 +1,62 @@
+from textwrap import dedent
+
+from yaksh.python_stdio_evaluator import compare_outputs
+
+
+def test_compare_outputs():
+    exp = "5\n5\n"
+    given = "5\n5\n"
+    success, msg = compare_outputs(given, exp)
+    assert success
+
+    exp = "5\n5\n"
+    given = "5\n5"
+    success, msg = compare_outputs(given, exp)
+    assert success
+
+    exp = "5\r5"
+    given = "5\n5"
+    success, msg = compare_outputs(given, exp)
+    assert success
+
+    exp = " 5 \r 5 "
+    given = "  5  \n  5  "
+    success, msg = compare_outputs(given, exp)
+    assert success
+
+    exp = "5\n5\n"
+    given = "5 5"
+    success, msg = compare_outputs(given, exp)
+    assert not success
+    m = dedent("""\
+    ERROR: Got 1 lines in output, we expected 2.
+    Expected:
+    5
+    5
+
+    Given:
+    5 5
+    """)
+    assert m == msg
+
+    exp = "5\n5\n"
+    given = "5\n6"
+    success, msg = compare_outputs(given, exp)
+    assert not success
+    m = dedent("""\
+    ERROR:
+    Expected:
+    5
+    5
+
+    Given:
+    5
+    6
+
+    Error in line 2 of output.
+    Expected line 2:
+    5
+    Given line 2:
+    6
+    """)
+    assert m == msg