Using Execution Paths to Evolve Software Patches

摘要

We present an evolutionary approach using genetic programming (GP) to automatically create software repairs. By concentrating the modifications on regions related to where the bug occurs, we effectively minimize the search space complexity and hence increase the performance of the GP process. To preserve the core functionalities of the program, we evolve programs only from code in the original program. Early experimental results show our GP approach is able to fix various program defects in reasonable time.