aifollow.news
Back Rohan Paul
Rohan Paul· @rohanpaul_ai · X· · Original publication time AI score54

Google paper describes FlowAgent for fixing failing tests before code mergesMachine translation

AI introduction

FlowAgent edits code, builds it and reruns tests after a continuous integration failure. Verified fixes appear in code review with a brief root-cause note. Experts judged 67.18% of its fixes correct across 195 real failures, and developers applied 28,554 fixes in production, though many suggestions still arrived too late.

Original post · AI translation

@rohanpaul_ai
Translation · refer to the original

谷歌的一篇论文表明,一个 AI 智能体能够修复谷歌各处未通过的测试,而修复速度决定开发者是否采用修复方案。

在合并前,持续集成(CI)中未通过的测试会打断开发者的工作节奏,而现有的修复智能体大多在合并后离线运行。

他们构建了 FlowAgent:它会捕获每次测试失败,并运行一个智能体,让其编辑代码、构建并重新运行测试,直到测试通过。经过验证的修复方案会出现在代码审查中,并附有一段简短的根本原因说明。

专家对 195 次真实失败的评估认为,其修复方案中有 67.18% 是正确的;开发者在生产环境中应用了 28,554 个修复方案。其运行时间中位数为 9.85 分钟,而开发者开始编辑代码前的等待时间中位数为 27.48 分钟,但许多修复方案仍来得太晚。

Source: Rohan Paul · x.com

Research
Found an error? Send a correction