Dynamic allocation of space for attribute instances in multi-pass evaluators of attribute grammars

Abstract

In this paper we address the second problem: how to store and address the values of attribute instances. In a one-pass implementation a stack is a natural data structure to be used for attribute values. Saarinen [24] shows how stacks can be used in treewalk © 1979-ACM 0-89791-002-8/79/0800-0026 $00.75 see i i 2 6 evaluators. The same principles apply to multi-pass evaluators as well [7]. However, because pass-oriented evaluators use a predetermined traversal order, an even better method of space allocation is possible. We present an algorithm which is based on finding equivalence classes of attribute instances. Intuitively, two attribute instances belong to the same equivalence class if the value of one instance is transmitted to the value of the other by copy rules. It is possible to use the same storage area for all the instances in an equivalence class. • The main advantage of our method lies in its ability to use the same area for several equivalence classes. This is achieved by analyzing each equivalence class to find out the last instance which uses the value of an instance in the equivalence class. The instances in the same class with this last user can now use the area which has already served its purpose for the previous class. This is important, for instance, in connection with symbol tables, which are usually built up little by little, adding information on new identifiers as they are declared. Since symbol tables are large structures, it is not tolerable to store a separate version of each intermediate symbol table. Our method makes it possible to use the same space repeatedly throughout the handling of the declarations. 2. BASIC CONCEPTS Our definition of attribute grammars follows the approach of [21] and [24].