UTFix - Change Aware Unit Test Repairing using LLM

Explore a conference presentation introducing UTFix, a novel approach for automatically repairing unit tests when their corresponding focal methods undergo changes in software development. Learn how this Large Language Model-based solution addresses the critical problem where 14%-22% of software failures stem from outdated tests that fail to reflect codebase changes, as revealed by Meta's research. Discover how UTFix tackles two key issues: assertion failures and reduced code coverage caused by changes in focal methods by leveraging contextual information including static code slices, dynamic code slices, and failure messages. Examine the evaluation results showing UTFix's impressive performance on synthetic benchmarks (Tool-Bench) where it successfully repaired 89.2% of assertion failures and achieved 100% code coverage for 96 out of 369 tests, plus its effectiveness on real-world benchmarks repairing 60% of assertion failures while achieving 100% code coverage for 19 out of 30 unit tests. Understand the significance of this first comprehensive study focused on unit test repair in evolving Python projects, including the development of UTFix, creation of Tool-Bench and real-world benchmarks, and demonstration of LLM-based methods' effectiveness in addressing unit test failures due to software evolution.