Automatic Repair of Infinite Loops

Abstract

Research on automatic software repair is concerned with the development of systems that automatically detect and repair bugs. One well-known class of bugs is the infinite loop. Every computer programmer or user has, at least once, experienced this type of bug. We state the problem of repairing infinite loops in the context of test-suite based software repair: given a test suite with at least one failing test, generate a patch that makes all test cases pass. Consequently, repairing infinites loop means having at least one test case that hangs by triggering the infinite loop. Our system to automatically repair infinite loops is called $Infinitel$. We develop a technique to manipulate loops so that one can dynamically analyze the number of iterations of loops; decide to interrupt the loop execution; and dynamically examine the state of the loop on a per-iteration basis. Then, in order to synthesize a new loop condition, we encode this set of program states as a code synthesis problem using a technique based on Satisfiability Modulo Theory (SMT). We evaluate our technique on seven seeded-bugs and on seven real-bugs. $Infinitel$ is able to repair all of them, within seconds up to one hour on a standard laptop configuration.