[SEG-staff] Talk announcement: "Efficient Exploration and Analysis of Program Repair Search Spaces" by Khashayar Someoliayi, KTH Stockholm (December 4)

Dear all, Another talk announcement for Dec. 4, scheduled in a way that avoids conflicts with Adi Seredinschi's talk ;-) *Date and Location* 4. December 2024, 10:00-11:00 (s.t.) Engehalde 8, Seminarraum 109 *Speaker* Dr. Khashayar Someoliayi *Title* Efficient Exploration and Analysis of Program Repair Search Spaces *Abstract* In the presented work, we aim at improving the efficiency of three main components involved in the exploration and analysis of APR search spaces: patch generation, automated patch assessment, and manual patch assessment. For this purpose, we present the three following contributions. To make patch generation efficient, we introduce Sorald, a template-based APR approach for fixing SonarJava static warnings. Sorald employs accurately designed templates to generate exactly one patch that is highly likely to fix the bug. The lightweight patch generation technique and the small search space that needs little analysis makes Sorald an efficient APR approach. For making automated patch assessment efficient, we propose LighteR, a lightweight tool for estimating the potential of fix templates used by template-based APR approaches. LighteR compares fix templates against developer-made bug-fixes to assess if the templates follow the modification patterns used by experts. The result of this assessment is used to rank the patches based on the potential of templates used for their generation. This ranking is used to prioritize patches for manual assessment and thus, finding the correct patch with minimal manual analysis. Finally, we introduce Collector-Sahab, which aims at helping code reviewers better understand behavioral changes caused by patches. Given two versions of a program P & Q, Collector-Sahab collects the execution trace of both P & Q. It next compares the traces and identifies runtime differences at variable and return value level. Finally, it augments the code diff between P & Q with a concise selection of extracted runtime differences. This code augmentation helps code reviewers to better understand the behavior of APR patches and thus, reduces human effort needed for manual patch assessment. To sum up, in this study we aim at making APR useful in practice by improving its efficiency. For this purpose, we propose novel methods to make patch generation, automated patch assessment, and manual patch assessment efficient. *Short Bio* Khashayar Etemadi received his PhD in software engineering from KTH Royal Institute of Technology, Sweden. His research interests are automated program repair, software testing, and software evolution Happy to see you there! Best, Timo