Apex: Automatic Programming Assignment Error Explanation
Dohyeong Kim, Yongwhi Kwon, Peng Liu, I Luk Kim, David Mitchel Perry, Xiangyu Zhang, and Gustavo Rodriguez-Rivera: "Apex: Automatic Programming Assignment Error Explanation", OOPSLA 2016.
This paper presents Apex, a system that can automatically generate explanations for programming assignment bugs, regarding where the bugs are and how the root causes led to the runtime failures. It works by comparing the passing execution of a correct implementation (provided by the instructor) and the failing execution of the buggy implementation (submitted by the student). The technique overcomes a number of technical challenges caused by syntactic and semantic differences of the two implementations. It collects the symbolic traces of the executions and matches assignment statements in the two execution traces by reasoning about symbolic equivalence. It then matches predicates by aligning the control dependences of the matched assignment statements, avoiding direct matching of path conditions which are usually quite different. Our evaluation shows that Apex is every effective for 205 buggy real world student submissions of 4 programming assignments, and a set of 15 programming assignment type of buggy programs collected from stackoverflow.com, precisely pinpointing the root causes and capturing the causality for 94.5% of them. The evaluation on a standard benchmark set with over 700 student bugs shows similar results. A user study in the classroom shows that Apex has substantially improved student productivity.
In this paper, the authors combine several sophisticated techniques to diagnose the root causes of errors in novice programming assignments even when the novice's code is significantly different from the correct implementation provided by the instructor. They do this by collecting traces of the programs' execution and reasoning about them symbolically to figure out why the final states of variables differ, regardless of the paths taken to create those variables.
This may seem like magic, but the methods used are well established. What they aren't is part of the usual undergraduate curriculum. I suspect this is due to the mathematical sophistication they require, but in truth, there is nothing here more complicated than the partial differential equations commonly taught in the junior or senior year of most engineering programs.
What really made this paper for me, though, was that the authors didn't stop once they had built a complex tool: they assessed its real-world utility by examining some programs found on the web, and then did a small controlled experiment to see whether the feedback provided by the tool helped novice programmers finish assignments faster. All together, this is a wonderful exemplar of how theory can guide the implementation of new things, which can then be assessed empirically—in short, of how engineering ought to be done.