Transforming programs into networks : call-graph caching, applications, and examples

There are computer programs that use the same flow of control when run on different inputs. This redundancy in their program execution traces can be exploited by preserving suitably abstracted callgraphs for subsequent reuse. We introduce a new programming transformation Call-Graph Caching (CGC) which partially evaluates the control flow of sets of such programs into a network formed from their call-graphs. CGC can automatically generate efficient state-saving structure-sharing incremental algorithms from simple program specifications. As an example, we show how a straightforward, inefficient LISP program for conjunctive match is automatically transformed into the RETE network algorithm. Simple and understandable changes to elegant functional (and other) programs are automatically translated by CGC into new efficient incremental network algorithms; this abstraction mechanism is shown for a class of conjunctive matching algorithms. We establish criteria for the appropriate application of CGC to other Al methods, such as chart parsing, consistency maintenance, and analogical reasoning. Detailed pedagogical examples are also presented. publication was supported in part by grant R29 LM 04707 from the National Library of Medicine. Table of