Towards Efficient Evaluation of Methods by Reduction

Basic Idea For DOOD, especially when using bottom-up evaluation, optimization is essential. However, for advanced languages like F-logic 2, 4] the general syntactical techniques well{known from Datalog are not suucient; e.g. due to the rich structure of the language (comprising multiple inheritance, variables on class or method position, and rule{deened class hierarchy) magic-set transformation is applicable only for a restricted subset of F-logic. Hence, attention should also be payed to specialized optimization strategies which take advantage from semantical knowledge about certain parts of a program. As a simple example consider railway connections: There are objects belonging to a class station which have a method next giving the next station depending on one parameter, the line. We deene on lines a method dist which, for any two stations, gives the number of stations between them. With F-logic rules, the deenition is as follows: A pure bottom{up evaluation will deduce all distances between arbitrary stations and store it in the computed model, even if the user queries only one certain distance. If magic{set optimization is used, due to the recursive deeni-tion, still the distances of stations \along the way" are deduced, claiming space and iterations of the consequence operator. To overcome this problem we propose to complement the general bottom{up xpoint evaluation by functional reduction as a kind of subordinate computation taking place whenever a call to a so called computed method occurs which has been deened appropriately. Contributions The idea is similar to the aim of functional{logic programming 3] but there are crucial diierences: Instead of an equality relation over the universe, in F-logic the declarative semantics of a method deenition is a set of additional facts extending a given Herbrand{interpretation. As operational semantics functional{logic programming extends the top{down principle for the A full paper may be obtained as technical report Nr. 82 under the URL: