Vado-Vı́rseda. Algorithmic debugging of missing answers in constraint functional-logic programming

Abstract. We present a declarative method for diagnosingmissing com-puted answers in CFLP (D), a generic scheme for lazy Constraint Func-tional-Logic Programming which can be instantiated by any constraint domain D given as parameter. As far as we know, declarative diagnosis of missing answers in such an expressive framework has not been tackled be-fore. Our approach combines and extends previous work done separately for constraint logic programming and lazy functional programming lan-guages. Diagnosis can be started whenever a user finds that the set of computed answers for a given goal with finite search space misses some expected solution w.r.t. an intended interpretation of the program, that provides a declarative descriptions of its expected behavior. Diagnosis proceeds by exploring a proof tree, that provides a declarative view of the answer-collection process performed by the computation, and it ends up with the detection of some function definition in the program that is incomplete w.r.t. the intended interpretation. We can prove the logi-cal correctness of the diagnosis method under the assumption that the recollection of computed answers performed by the goal solving system can be represented as a proof tree. We argue the plausibility of this assumption, and we describe the prototype of a tool which implements the diagnosis method.