Characterizing Python Method Evolution with PyMevol: An Essential Step Towards Enabling Reliable Software Systems

Abstract

Understanding the evolution of library methods is essential for maintaining high-quality and reliable software systems as those libraries often evolve rapidly in order to meet new requirements such as adding new features, improving performance, or fixing vulnerabilities. Failing to incorporate this evolution may result in compatibility issues that may manifest themselves as runtime crashes, leading to a poor user experience. This is not uncommon for the most popular programming language, Python, for which our community has developed over 380,000 libraries. To help developers better understand their used libraries, we propose to the community a prototype tool called PyMevol to model Python libraries' APIs and their evolution. Specifically, given a Python library, PyMevol statically examines its code to extract APIs (including aliases introduced by Python's import-flow mechanism) from all its released versions to build a history-sensitive alias-aware API explorer tree, a tree structure that allows users to explore the biography of each API so as to quickly locate where and when a given API is introduced, changed, or removed. Our experimental results over five popular real-world Python libraries show that our approach is reliable in achieving its purpose (i.e., over 90 % of accuracy) and helpful in supporting further API-relevant analyses.