Static Memory Management for Logic Programming Languages

Introduction. Logic programming (LP) languages aim to free programmers from procedural details such as memory management tasks. One classical, automatic memory management technique in logic programming is to use a heap memory for all the structured terms and rely on backtracking and on a runtime garbage collector to reclaim memory. While efficient implementation of garbage collectors for LP languages can reuse more than 90% heap space, they introduce performance penalties to the execution of a program because the collectors need to temporarily stop the main program to do their job. Background Literature. To remedy this shortcoming there has been a lot of research on compile-time memory management techniques, which automatically enhance programs with instructions to reuse memory. This static method generally follows two approaches: region-based memory management (RBMM) and compile-time garbage collection (CTGC). The basic idea of RBMM is to divide the heap memory into different regions. The dynamically created terms and their subterms have to be distributed over the regions in such a way that at a certain point in the execution of the program all terms in a region are dead and the region can be removed. CTGC detects when allocated memory cells are no longer used and instructs the program to reuse those cells to construct new terms, reducing memory consumption and in some cases achieving faster code. RBMM has long been a topic of intensive research for functional programming languages [7, 1, 4] and more recently also for procedural languages [3, 2]. For LP languages, there has been only one attempt to make RBMM work for Prolog [5]. The idea of CTGC has been used to reuse memory cells locally in the procedures of the LP language Mercury [6]. Goal of the Research. RBMM achieves competitive memory consumption for many programs due to timely removal of dead regions and memory management operations are time bound. CTGC can exploit many reuse opportunities. Taking those advantages to have a system in which reusable memory can be reused and non-reusable dead memory can be deallocated timely is the motivation for our current research. The ultimate research goal is to investigate the possibility and practicality of a hybrid static memory management technique, which combines RBMM and CTGC. Current Status. Developing an algorithm that combines the nontrivial program analyses of CTGC in [6] and the type-based region inference in [5] has not been straightforward. Its main difficulty comes from the fact …