Generalization in Hierarchies of Online Program Specialization Systems

In recent work, we proposed a simple functional language S-graph-n to study metaprogramming aspects of self-applicable online program specialization. The primitives of the language provide support for multiple encodings of programs. An important component of online program specialization is the termination strategy. In this paper we show that such a representation has the great advantage of simplifying generalization of multiply encoded data. After developing and formalizing the basic metaprogramming concepts, we extend two basic methods to multiply encoded data: most specific generalization and the homeomorphic embedding relation. Examples and experiments with the initial design of an online specializer illustrate their use in hierarchies of online program specializers.