Must Fault Localization for Program Repair

Abstract

This work is concerned with fault localization for automated program repair. We define a novel concept of a must location set. Intuitively, such a set includes at least one program location from every repair for a bug. Thus, it is impossible to fix the bug without changing at least one location from this set. A fault localization technique is considered a must algorithm if it returns a must location set for every buggy program and every bug in the program. We show that some traditional fault localization techniques are not must. We observe that the notion of must fault localization depends on the chosen repair scheme, which identifies the changes that can be applied to program statements as part of a repair. We develop a new algorithm for fault localization and prove that it is must with respect to commonly used schemes in automated program repair. We incorporate the new fault localization technique into an existing mutation-based program repair algorithm. We exploit it in order to prune the search space when a buggy mutated program has been generated. Our experiments show that must fault localization is able to significantly speed-up the repair process, without losing any of the potential repairs.

Automated Program Repair Using Formal Verification Techniques

“…This work is concerned with automated program repair. It focuses on two specific approaches, presented in [50,48] and [22,21], that demonstrate many of the guiding principles in program repair, when it is based on formal methods. While the two approaches have much in common, they are also quite distinct, due to their different settings, including their type of programs, specifications and repair mechanisms.…”

“…The syntax refers to the program code, which might be updated for the purpose of repair. In [50] a predefined set of mutations is used for syntactic update, where [22] uses abduction to derive constraints that are added to the program code. In both cases, SMT solvers are used to answer semantic questions that arise during verification.…”

“…In [50] the search space consists of all mutated programs and the goal is to return all good repairs that are minimal. A notion of must-fault localization is developed in order to guarantee that similarly failed programs will not be considered in the future.…”

“…As stated, the differences between the two approaches are quite significant. [50] exploits a predefined set of mutations for repair. Its goal is to return all minimal repairs and its focus is on the notion of must-fault localization, which achieves efficiency and completeness.…”

Software Fault Localization through Aggregation-Based Neural Ranking for Static and Dynamic Features Selection

“…Program constraints are assertions introduced to a program that must be true to function correctly throughout its execution. These constraints are generally specified in the program as a conditional statement, which causes the program to stop execution if any constraint is not satisfied [11]. Program log [12] inserts are often included with the code on a hit-or-miss basis to check variables and other program information updating processes that utilize stored log files or printed run-time information to identify failure when they see strange program behavior.…”