Title: An Automated Approach to Program Repair with Semantic Code Search
Date/Time: November 4th, 2015 @ 8:30 AM
Place: 213 Atanasoff Hall
Faculty Advisor: Professor Kathryn Stolee
Every year software companies dedicate numerous developer hours to debugging and fixing defects. Automated program repair has the potential to greatly decrease the costs of debugging. Existing automated repair techniques, such as Genprog, TSPRepair, and AE, show great promise but are not able to repair all bugs. We propose a new automated program repair technique, SearchRepair, which is a complementary program repair technique. We take advantage of existing open source code to find potential fixes based on the assumption that there are correct implementations in open source project code for some defects. The key challenges lie in efficiently finding code semantically similar (but not identical) to defective code and then appropriately integrating that code into the buggy program. The technique we present, SearchRepair, addresses these challenges by (1) encoding a large database of human-written code fragments as SMT constraints on input-output behavior, (2) localizing a given defect to likely-buggy program fragments, (3) dynamically analyzing those buggy fragments to derive input-output pairs that describe likely buggy behavior and that can be encoded as SMT constraints, (4) using state-of-the-art constraint solvers to find fragments in the code database that satisfy those constraints, and (5) validating patches that repair the bug against program test suites.