Lightweight source model extraction

Reverse engineers depend on the automatic extrac-t ion of information from source code. Some use-ful kinds of information+ource models—are well-known: call graphs, file dependence, etc. Predict-ing every kind of source model that a reverse engi-neer may need is impossible. We have developed a lightweight approach for generating flexible and tol-erant source model extractors from lexical specifica-t ions. The approach is lightweight in that the speci-fications are relatively small and easy to write. It is flexible in that there are few constraints on the in-formation in the source that can be extracted (e.g., we can extract from macros, comments, etc.). It is tolerant in that information can be extracted from source that cannot necessarily be compiled or linked. In essence, we scan for source constructs that con-tribute to the specified source model while ignor-ing constructs that do not contribute to that source model. We have developed tools to support this ap-proach and applied the tools to the extraction of a number of different source models (file dependence, event interactions, call graphs) from systems imple-mented in a variety of programming languages (C, C++, CLOS, Eiffel). We discuss our approach and describe its application to extract source models not available using existing systems; for example, we com-pute the invoked-by relation over Field tools. We compare and contrast our approach to the conven-tional approach of generating source models from a program database