Reduce API Debugging Overhead via Knowledge Prepositioning (2304.06692v1)

Abstract: OpenAPI indicates a behavior where producers offer Application Programming Interfaces (APIs) to help end-users access their data, resources, and services. Generally, API has many parameters that need to be entered. However, it is challenging for users to understand and document these parameters correctly. This paper develops an API workbench to help users learn and debug APIs. Based on this workbench, much exploratory work has been proposed to reduce the overhead of learning and debugging APIs. We explore the knowledge, such as parameter characteristics (e.g., enumerability) and constraints (e.g., maximum/minimum value), from the massive API call logs to narrow the range of parameter values. Then, we propose a fine-grained approach to enrich the API documentation by extracting dependency knowledge between APIs. Finally, we present a learning-based prediction method to predict API execution results before the API is called, significantly reducing user debugging cycles. The experiments evaluated on the online system show that this work's approach substantially improves the user experience of debugging OpenAPIs.