Automatic Repair of Timestamp Comparisons

Giovanni Liva, Muhammad Taimoor Khan, Martin Pinzger, Francesco Spegni, Luca Spalazzi

IEEE Transactions on Software Engineering, 2019

Automated program repair has the potential to reduce the developers' effort to fix errors in their code. In particular, modern programming languages, such as Java, C, and C\#, represent time as integer variables that suffer from integer overflow, introducing subtle errors that are hard to discover and repair. Recent researches on automated program repair rely on test cases to discover failures to correct, making them suitable only for regression errors. We propose a new strategy to automatically repair programs that suffer from timestamp overflows that are manifested in comparison expressions. It unifies the benefits of static analysis and automatic program repair avoiding dependency on testing to identify and correct defected code. Our approach performs an abstract analysis over the time domain of a program using a Time Type System to identify the problematic comparison expressions. The repairing strategy rewrites the timestamp comparisons exploiting the binary representation of machine numbers to correct the code. We have validated the applicability of our approach with 20 open source Java projects. The results show that it is able to correctly repair all 246 identified errors. To further validate the reliability of our approach, we have proved the soundness of both, type system and repairing strategy. Furthermore, several patches for three open source projects have been acknowledged and accepted by their developers.