Why Authors and Maintainers Link (or Don't Link) Their PyPI Libraries to Code Repositories and Donation Platforms
Alexandros Tsakpinis, Nicolas Raube, Alexander Pretschner
TL;DR
This study investigates why PyPI libraries link (or fail to link) to code repositories and donation platforms, using two large-scale surveys of authors and maintainers and an LLM-based topic-modeling pipeline to analyze thousands of short-text responses. The approach demonstrates robust topic extraction across 30 runs and provides a structured, human-evaluated view of motivations, barriers, and cross-cutting issues in PyPI metadata practices. Key findings show repository links are common and valued for transparency and collaboration, while donation links are infrequent and hampered by usability, awareness, and organizational constraints. The work offers practical recommendations for PyPI to improve metadata guidance and tooling, and validates an effective LLM-based method for analyzing short-text survey data in software engineering research.
Abstract
Metadata of libraries on the Python Package Index (PyPI)-including links to source code repositories and donation platforms-plays a critical role in supporting the transparency, trust, and sustainability of open-source libraries. Yet, many packages lack such metadata, and little is known about the underlying reasons. This paper presents a large-scale empirical study combining two targeted surveys sent to 50,000 PyPI authors and maintainers. We analyze more than 1,400 responses using large language model (LLM)-based topic modeling to uncover key motivations and barriers related to linking repositories and donation platforms. While repository URLs are often linked to foster collaboration, increase transparency, and enable issue tracking, some maintainers omit them due to oversight, laziness, or the perceived irrelevance to their project. Donation platform links are reported to support open source work or receive financial contributions, but are hindered by skepticism, technical friction, and organizational constraints. Cross-cutting challenges-such as outdated links, lack of awareness, and unclear guidance-affect both types of metadata. We further assess the robustness of our topic modeling pipeline across 30 runs (84% lexical and 89% semantic similarity) and validate topic quality with 23 expert raters (Randolph's kappa = 0.55). The study contributes empirical insights into PyPI's metadata practices and provides recommendations for improving them, while also demonstrating the effectiveness of our topic modeling approach for analyzing short-text survey responses.