Self-Admitted Library Migrations in Java, JavaScript, and Python Packaging Ecosystems: A Comparative Study

Reusing open-source software libraries has become the norm in modern software development, but libraries can fail due to various reasons, e.g., security vulnerabilities, lacking features, and end of maintenance. In some cases, developers need to replace a library with another competent library with similar functionalities, i.e., library migration. Previous studies have leveraged library migrations as a unique lens of observation to reveal insights into library selection and dependency management in general. However, they are heavily biased toward Java while the generalizability of their findings remains unknown.In this paper, we present a comparative study on self-admitted library migrations (SALMs) from three packaging ecosystems: Java/Maven, JavaScript/npm, and Python/PyPI. For this study, we design a set of semi-automatic methods that accurately locate SALMs, their domains, and their rationales from git repositories. We reveal that SALMs are prevalent and highly unidirectional in all three ecosystems, and the underlying rationales can be well covered by a previous theoretical framework. Also, SALMs in these ecosystems present domain similarity (testing frameworks, web frameworks, HTTP clients, and serialization). However, we observe differences in the longitudinal trends, the distributions of rationales, the ecosystem-specific domains, and the levels of unidirectionality, all of which indicate that Python/PyPI sees increasingly intense competition between libraries and deserves more research on library recommendation and migration.