Understanding and Detecting Usage of Deprecated Remote APIs

Abstract

Context: Modern software systems reuse features from other software modules or services via Application Programming Interfaces (APIs). While deprecation mechanisms and automated tool support are well-established for local APIs (e.g., Java libraries), a significant gap exists for remote APIs (e.g., HTTP communication). Currently, there is no tool support to alert client developers when a remote API they call becomes deprecated.

Objective: The primary objective of this dissertation is to improve the detection of deprecated remote API usage by designing and implementing a language-independent framework. This framework integrates multiple data sources to enable client developers to react effectively to evolving software ecosystems.

Methods: Following a design science research approach, the dissertation includes a systematic mapping study of research on API deprecation to identify research gaps, an empirical quantitative and qualitative analysis of open-source Java repositories to characterize the usage of HTTP clients, an analysis of OpenAPI descriptions to evaluate how deprecation is documented in practice, and the construction, implementation and evaluation of a hybrid detection framework combining static and dynamic approaches.

Results: The systematic study confirms that research on remote API deprecation is missing. Furthermore, the empirical analyses of HTTP clients in Java and deprecation practices in OpenAPI descriptions shows that any relevant solution needs to cover a variety of implementation options for client code and deprecation documentation. The hybrid detection framework achieves this widespread applicability by using auto-instrumentation for clients and combining multiple sources of deprecation documentation. Finally, the implemented prototype achieves a precision of 1.00 and recall values between 0.93 and 1.00 in controlled evaluations.

Conclusions: Automated, multi-source detection for usage of deprecated remote APIs is technically feasible and accurate. By combining dynamic runtime observation with static documentation analysis, the proposed hybrid approach provides a foundation for practical tools in evolving distributed systems.