A Case Study in Extracting the Runtime Architecture of an Object-Oriented System

Many legacy systems benefit from having an up-to-date documented runtime architecture. But it is hard to statically extract a sound hierarchal runtime architecture from a legacy object-oriented system that is written in a general-purpose programming language and that follows common design idioms, so many have preferred dynamic analyses. Abi-Antoun and Aldrich recently developed a static analysis to extract runtime architectures from legacy object-oriented programs written in a mainstream language. The approach relies on adding annotations to the code, and statically extracts a hierarchical object graph from the annotated code. The annotations implement a type system, so a typechecker can check that the annotations are consistent with each other and with the code. We present a case study to evaluate the method and the tools on a 36-KLOC system that is in wide use and is actively evolving. We developed a tool to reduce the annotation burden by generating the initial set of annotations. Then, we used a typechecker to check the annotations, and manually fixed the annotation warnings. We then extracted a hierarchical object graph, which we refined on our own, before showing it to the system maintainer. The maintainer was able to match the extracted object graph to his mental model of the system and gave us suggestions for further refinement. We incorporated the maintainer’s feedback by slightly adjusting the annotations and using additional features in the extraction tool, to make the extracted diagram convey the maintainer’s design intent.