API Beauty is in the eye of the Clients: 2.2 Million Maven Dependencies reveal the Spectrum of Client-API Usages
This study provides insights for API maintainers and developers in software engineering to prioritize efforts and manage public APIs effectively, though it is incremental as it extends prior empirical observations.
The paper investigates the paradox between Hyrum's law and research suggesting limited API usage by analyzing 2.2 million Maven dependencies, finding that while most API types are used at least once with enough clients, a small set of types is used by the vast majority of clients.
Hyrum's law states a common observation in the software industry: "With a sufficient number of users of an API, it does not matter what you promise in the contract: all observable behaviors of your system will be depended on by somebody". Meanwhile, recent research results seem to contradict this observation when they state that "for most APIs, there is a small number of features that are actually used". We investigate this seeming paradox between the observations in industry and the research literature, with a large scale empirical study of client API relationships in one single ecosystem: Maven Central. We study the 94 most popular libraries in Maven Central, as well as the 829,410 client artifacts that declare a dependency to these libraries and that are available in Maven Central, summing up to 2.2M dependencies. Our analysis indicates the existence of a wide spectrum of API usages, with enough clients, most API types end up being used at least once. Our second key observation is that, for all libraries, there is a small set of API types that are used by the vast majority of its clients. The practical consequences of this study are two-fold: (i) it is possible for API maintainers to find an essential part of their API on which they can focus their efforts; (ii) API developers should limit the public API elements to the set of features for which they are ready to have users.