ARJA: Automated Repair of Java Programs via Multi-Objective Genetic Programming

摘要

Recent empirical studies show that the performance of GenProg is notsatisfactory, particularly for Java. In this paper, we propose ARJA, a new GPbased repair approach for automated repair of Java programs. To be specific, wepresent a novel lower-granularity patch representation that properly decouplesthe search subspaces of likely-buggy locations, operation types and potentialfix ingredients, enabling GP to explore the search space more effectively.Based on this new representation, we formulate automated program repair as amulti-objective search problem and use NSGA-II to look for simpler repairs. Toreduce the computational effort and search space, we introduce a test filteringprocedure that can speed up the fitness evaluation of GP and three types ofrules that can be applied to avoid unnecessary manipulations of the code.Moreover, we also propose a type matching strategy that can create newpotential fix ingredients by exploiting the syntactic patterns of the existingstatements. We conduct a large-scale empirical evaluation of ARJA along withits variants on both seeded bugs and real-world bugs in comparison with severalstate-of-the-art repair approaches. Our results verify the effectiveness andefficiency of the search mechanisms employed in ARJA and also show itssuperiority over the other approaches. In particular, compared to jGenProg (animplementation of GenProg for Java), an ARJA version fully following theredundancy assumption can generate a test-suite adequate patch for more thantwice the number of bugs (from 27 to 59), and a correct patch for nearly fourtimes of the number (from 5 to 18), on 224 real-world bugs considered inDefects4J. Furthermore, ARJA is able to correctly fix several realmulti-location bugs that are hard to be repaired by most of the existing repairapproaches.