Treffer: Program Repair by Automated Generation of Hints

Programming has become an important skill in today's technology-driven world. It is a complex activity because of which programmers make mistakes in their software. Student programmers make mistakes in their programs due to lack of programming experience and also due to lack of understanding of the algorithmic concepts being used. An experienced programmer in the industry, on the other hand, makes mistakes in the software he develops because of the mere complexity of an industry-grade software. Debugging one's programs, i.e.finding and fixing faults is one of the challenging activities for a programmer. This necessitates the need to develop automated techniques that are effective in assisting programmers repair faults in their programs. In this thesis, we present new computer-assisted approaches to help programmers repair their programs. Finding repair for a fault in the program is essentially a search over the space of all possible repairs. Traditional repair techniques attempt to output a complete repaired program that eliminates the fault in the original program. This ambitious goal of finding a complete repair is often unachievable because the search space becomes too large to navigate efficiently. The key insight of our work is that programmers only need guidelines that help them understand the faults they made and suggestions on how to x them, as opposed to a complete repair. Therefore, this thesis proposes a novel perspective of solving the program repair problem, where we generate textual hints that direct the programmer on how to x the fault. A hint specifies which part of the program needs modification as well as suggests how to modify it. The programmer uses them as guidance to arrive at a complete repair. For student programmers, our hints help them reflect on the mistakes they made and improve their understanding of the problem being solved in the program. For experienced programmers, ou hints help them easily identify the cause of fault and guide them to a repair. Our approaches use novel ...