Semantic‐aware two‐phase test case prioritization for continuous integration

Summary Continuous integration (CI) is a widely applied development practice to allow frequent of software changes, detecting early faults. However, extremely builds consume amounts time and resources in such scenario. It quite challenging for existing test case prioritization (TCP) address this issue due the time‐consuming information collection (e.g. coverage) or inaccurately modelling code semantics result unsatisfied prioritization. In paper, we propose semantic‐aware two‐phase TCP framework, named SatTCP, which combines coarse‐grained filtering fine‐grained perform precise with low costs CI. consists three parts: (1) representation, parsing programme changes cases obtain change representations; (2) filtering, conducting preliminary ranking based on retrieval; (3) prioritization, training pretrained Siamese language model filtered set further sort via semantic similarity. We evaluate SatTCP large‐scale, real‐world dataset cross‐project validation from fault detection efficiency compare it five baselines. The results show that outperforms all baselines by 6.3%–45.6% mean average percentage detected per cost (APFDc), representing an obvious upward trend as project scale increases. Meanwhile, can reduce real CI testing 71.4%, outperforming best baseline 17.2% average. Furthermore, discuss impact different configurations, flaky tests hybrid techniques performance respectively.