Merging graph-based and rule-based computation : the language G-Log

In this paper we discuss the merging of two different computation paradigms: the fixpoint computation for deductive databases and the pattern-matching computation for graph-based languages. We show how these paradigms can be combined on the example of the declarative, graph-based, database query language G-Log. A naive algorithm to compute G-Log programs turns out to be very inefficient. However, we also present a backtracking fixpoint algorithm for Generative G-Log, a syntactical sublanguage of G-Log that, like G-Log, is non-deterministic complete. This algorithm is considerably more efficient, and reduces to the standard fixpoint computation for a sublanguage of Generative G-Log that is a graphical equivalent of Datalog. The paper also studies some interesting properties like satisfiability and triviality, that are undecidable for full G-Log and turn out to be decidable for sufficiently general classes of Generative G-Log programs.