Avaliando o merge semi-estruturado: um estudo comparativo de corretude e desempenho da ferramenta SESAME

Abstract

The integration of concurrent modifications is a critical challenge in collaborative software development. While unstructured (textual) merge tools often generate spurious conflicts, structured approaches, although more accurate, can introduce silent semantic errors and have high computational costs. In this context, the SESAME tool proposes an innovative semi-structured approach, using syntactic separators to infer the code structure in a lightweight manner. This work presents an empirical evaluation of SESAME through a differentiated replication of the study by Schesch et al. (2024). Adopting a rigorous experimental protocol, the tool was evaluated not only by counting conflicts, but also by semantic correctness verified through the execution of automated test suites on a vast set of Java projects. SESAME’s performance was compared with state-of-the-art tools (Git Merge, Spork, and IntelliMerge) using a robust dataset with over 5,000 merge scenarios, with metrics for Correctness, Execution Time, and Effort Reduction. The results reveal that, although purely structured tools maximize conflict resolution, they introduce high error rates (11% in Spork) and impose prohibitive latency. In contrast, SESAME has established itself as a sweet spot: it maintained security close to the textual standard (only 3% error) with an execution time and order of magnitude lower than structured tools, validating itself as a pragmatic and efficient alternative for Continuous Integration pipelines.