New Patch Validation Tool and New Results on a Patch Generator

[Yingfei Xiong]

Dear colleagues,

I would like to catch your attention to our two newest advances, which may be useful in future program repair research.

  * ExpressAPR: ExpressAPR (published at TSE 2024, https://doi.org/10.1109/TSE.2024.3359969) is an efficient general-purpose patch validator that takes a set of patches and produces the test execution results of the patches. It integrates five acceleration techniques and is 137.1X faster than directly using defects4j commands and 8.8X faster than UniAPR. ExpressAPR works out-of-the-box for Maven projects and Defects4J projects, and could be easily configured for other projects.

  * Tare: Tare is an effective patch generator. Tare is built on top of Recoder by guiding the neural network to learn type inference. Tare was published at ICSE 2023 (https://doi.org/10.1109/ICSE48619.2023.00126) and we have recently updated its repair result with perfect localization on Defects4J 2.0 (https://github.com/pkuzqh/ICSE23Repair). Tare in total repairs 132 bugs on Defects4J 2.0 with perfect fault localization by considering only Top-1 plausible patch, outperforming multiple recent approaches using large pre-training models (The neural model of Tare has only 35m parameters).

We welcome you to try our implementations and report any issues identified.

  * ExpressAPR: https://github.com/ExpressAPR/ExpressAPR/

  * Tare: https://github.com/pkuzqh/ICSE23Repair

The tool that combines the two approaches, ET, has won the first place in the Java functional-bug track in APR-COMP 2024.

Best Regards,
Yingfei Xiong

-- 
Xiong, Yingfei / 熊英飞 / 熊英飛

Associate Professor
Institute of Software
School of Computer Science 
Peking University
China

Web: https://xiongyingfei.github.io/