Leveraging Change Types and Contexts to Guide LLMs for Automated Test Code Updating
Abstract
In the continuous iteration of software, synchronizing the updates of test code and production code is crucial for ensuring project quality. However, in actual development, test code maintenance often lags behind the evolution of production code, which can lead to issues such as invalid functionality verification. Existing test code repair methods based on pre-trained models or Large Language Models (LLMs) show less effective or low generalization due to their lack of precise contextual awareness or strong limitations in use scenarios.
Publication details
- Journal / source
- Proceedings of the 2026 34th IEEE/ACM International Conference on Program Comprehension
- Publisher
- ACM
- Pages
- 270–281
- Publication date
- 12 April 2026