(2022) Automated Program Repair Using Formal Verification Techniques.
Text
FGRS22.pdf - Accepted Version Download (449kB) |
Abstract
We focus on two different approaches to automatic program repair, based on formal verification methods. Both repair techniques consider infinite-state C-like programs, and consist of a generate-validate loop, in which potentially repaired programs are repeatedly generated and verified. Both approaches are incremental – partial information gathered in previous verification attempts is used in the next steps. However, the settings of both approaches, including their techniques for finding repairs, are quite distinct. The first approach uses syntactic mutations to repair sequential programs with respect to assertions in the code. It is based on a reduction to the problem of finding unsatisfiable sets of constraints, which is addressed using an interplay between SAT and SMT solvers. A novel notion of must-fault-localization enables efficient pruning of the search space, without losing any potential repair. The second approach uses an Assume-Guarantee (AG) style reasoning in order to verify large programs, composed of two concurrent components. The AG reasoning is based on automata-learning techniques. When verification fails, the procedure repeatedly repairs one of the components, until a correct repair is found. Several different repair methods are considered, trading off precision and convergence to a correct repair.
Item Type: | Other |
---|---|
Divisions: | Bernd Finkbeiner (Reactive Systems Group, RSG) |
Depositing User: | Hadar Frenkel |
Date Deposited: | 08 Sep 2023 13:44 |
Last Modified: | 08 Sep 2023 13:44 |
Primary Research Area: | NRA2: Reliable Security Guarantees |
URI: | https://publications.cispa.saarland/id/eprint/4027 |
Actions
Actions (login required)
View Item |