Improving Search-Based Automatic Program Repair With Neural Machine Translation

The challenge of automatically repairing bugs in programs to reduce debugging expenses and increase program quality is known as automated repair. To overcome this issue, test-suite-based repair techniques use a specified test suite an oracle alter the input faulty pass full suite. GenProg well-known example kind repair, which genetic programming used reorder statements already present program. However, recent practical experiments suggest that GenProg’s performance, notably for Java, not sufficient. Improved dependability necessitates automatic techniques. Template-based have recently been combined with search-based solve issues automatically. Although intriguing, it has two fundamental drawbacks: Its search space often lacks correct solution, technique disregards expertise, such precise code language. Compared template-based approach, existing neural-machine-translation-based approaches are limited by these constraints due their ability learn generate new solutions. We propose approach combines approach. More specifically, we both redundancy assumption sequence-to-sequence learning patches source potential fix feed into multiobjective evolutionary algorithm find test-suite-adequate patches. In work, novel framework called ARJANMT introduced Java programs. Two sets controlled conducted on 410 from benchmarks investigate repairability correctness our proposed framework. A comparison between state-of-the-art frameworks made. experimental results indicate combining those types (search-based neural-machine-translation-based) produces better or fixes they previously were unable individually.