We address the problem of refining and completing a partially specified high-level design model and a partially-defined mapping from source code to design model. This is related but not identical to tasks that have been automated with a variety of reverse engineering tools to support software modification tasks. We posited that set-based flow analysis algorithms would provide a convenient and powerful basis for refining an initial rough model and partial mapping, and in particular that the ability to compute fixed points of set equations would be useful in propagating constraints on the relations among the model, the mapping, and facts extracted from the implementation. Here we report our experience applying this approach to a modest but realistic example problem. We were successful in expressing a variety of useful transformations very succinctly as flow equations, and the propagation of recursively-defined constraints was indeed useful in refining the mapping from implementation to model. On the other hand, our experience highlights remaining challenges to make this an attractive approach for general use. Special measures are required to identify and remove inconsistent constraints before they propagate through a system. Also, while the required flow equations are succinct, they are also rather opaque; it is not obvious how their expressive power might be preserved in a more accessible notation.

The past three decades have seen the creation of several tools that extract, visualize, and manipulate graph-structured representations of program information. To facilitate interconnection and exchange of information between these tools, and to support the prototyping and development of new tools, it is desirable to have some generic support for the specification of graph transformations and exchanges between them.GenSet is a generic programmable tool for transformation of graph-structured data. The implementation of the GenSet system and the programming paradigm of its language are both based on the view of a directed graph as a binary relation. Rather than use traditional relational algebra to specify transformations, however, we opt instead for the more expressive class of flow equations. Flow equations---or, more generally, systems of simultaneous fixpoint equations---have seen fruitful applications in several areas, including data and control flow analysis, formal verification, and logic programming. In GenSet, they provide the fundamental construct for the programmer to use in defining new transformations.