Abstract

 A large software system contains millions of lines of source code, and the development often involves many developers over a long period. How to understand and analyze its complex code dependencies is challenging but meaningful to developers for program comprehension. In this paper, we propose a novel visual analytics system to explore code dependencies between files for visually understanding software architecture and interactively analyzing bad dependencies. The dependencies between files are abstracted as a directed graph (i.e., the dependency graph) at different levels via the file hierarchy to show the modularization design of source code. Node embeddings of the dependency graph are learned to identify the files with specific dependencies and analyze the similarity between bad dependencies. Finally, we evaluate the usability of our system by two case studies in different JavaScript libraries as well as a user study on software architecture understanding and bad dependency analysis.

Graphic abstract

中文翻译：

---

通过基于嵌入的相似性对源代码中的依赖图进行可视化探索

摘要

 大型软件系统包含数以百万计的源代码行，并且开发过程通常需要很长一段时间才能参与。如何理解和分析其复杂的代码依赖性是具有挑战性的，但对于开发人员来说对于程序理解是有意义的。在本文中，我们提出了一种新颖的视觉分析系统，以探索文件之间的代码依赖性，以直观地了解软件体系结构并交互式分析不良依赖性。文件之间的依赖关系通过文件层次结构在不同级别抽象为有向图（即依赖关系图），以显示源代码的模块化设计。学习依赖图的节点嵌入，以识别具有特定依赖项的文件并分析不良依赖项之间的相似性。最后，

图形摘要