A domain-specific language for dynamic white-box evaluation of Java assignments

Abstract

Programming exercises involving algorithms typically involve time and spatial constraints. Automated assessments for such implementations are often carried out in a black-box manner or through static analysis of the code, without considering the internal execution properties, which could lead to falsely positive evaluations of students' solutions. We present Witter, a domain-specific language for defining white-box test cases for the Java language. We evaluated programming assignment submissions from a Data Structures and Algorithms course against Witter’s test cases to determine if our approach could offer additional insight regarding incomplete algorithmic behaviour requirements. We found that a significant amount of student solutions fail to meet the desired algorithmic behavior (approx. 21%), despite passing black-box tests. Hence, we conclude that white-box tests are useful to achieve a thorough automated evaluation of this kind of exercises.