Extending Existing Inference Tools to Mine Dynamic APIs

APIs often feature dynamic relations between client and service provider, such as registering for notifications or establishing a connection to a service. Dynamic specification mining techniques attempt to fill gaps in missing or decaying documentation, but current miners are blind to relations established dynamically. Because they cannot recover properties involving these dynamic structures, they may produce incomplete or misleading specifications. We have devised an extension to current dynamic specification mining techniques that ameliorates this shortcoming. The key insight is to monitor not only values dynamically, but also properties to track dynamic data structures that establish new relations between client and service provider. We have implemented this approach as an extension to the instrumentation component of Daikon, the leading example of dynamic invariant mining in the research literature. We evaluated our tool by applying it to selected modules of widely used software systems published on GitHub.