Technical Debt Tools: a Survey and an Empirical Evaluation

Background: The life cycle of a technical debt from its identification to its payment is long and may include several activities, such as identification and management. There is a lot of research in the literature to address different sets of these activities by different means. Specifically, several tools have already tackled such technical debt identification problem. However, only a few studies empirically assessed those tools. Method: In this article, we carried a multi- method research. We first surveyed the literature for the technical debt tools available an then we evaluated two of them, which aim at identification of self-admitted technical debt. Thy are named eXcomment e DebtHunter. Results: We found 92 tools employing different approaches to support technical debt life cycle management. Most of them (59%) address only the high level task of management, instead of actually identify and pay the debt. Additionally, as for our empirical evaluation of tools, our results show that DebtHunter found only 7% of debt identified by eXcomment. In the other way around, eXcomment found 19.9% the debt find by DebtHunter. Besides, both tools have low levels of precision and recall. Conclusion: It is hard to find technical debt through comments. Both tools can find indicators of debt items, however they struggle on the precision and recall. In fact, although eXcomment and DebtHunter diverge on the amount of debt identified, they seem to converge with regard to the type o debt present in the system under evaluation.背景：技术债务（Technical Debt）从识别到偿付的全生命周期漫长，涵盖识别、管理等多项活动。现有大量文献研究通过不同手段针对该生命周期的不同环节展开探索。目前已有多款工具致力于解决技术债务识别问题，但仅有少数研究对这些工具开展了实证评估。 方法：本文采用多方法研究范式。首先，我们对现有公开的技术债务工具开展文献调研；随后针对两款面向自我承认的技术债务（Self-admitted Technical Debt）识别的工具展开评估，二者分别名为eXcomment与DebtHunter。 结果：本次调研共筛选出92款采用不同方法支撑技术债务全生命周期管理的工具，其中59%仅聚焦于管理层面的高层任务，并未实际开展债务识别与偿付工作。此外，针对两款工具的实证评估结果显示：DebtHunter仅能识别出eXcomment所发现债务的7%；反之，eXcomment仅能覆盖DebtHunter识别债务的19.9%。同时，两款工具的精确率（Precision）与召回率（Recall）均处于较低水平。 结论：通过代码注释识别技术债务存在较大难度。两款工具虽可捕捉债务项的相关指标，但在精确率与召回率上表现欠佳。尽管eXcomment与DebtHunter在识别的债务规模上存在显著分歧，但二者在所评估系统中检测到的债务类型上表现出一致性。