A Garbage Collection Algorithm for Tricia

We present a new algorithm for garbage collection of the term stack in David H. D. Warren's abstract Prolog machine (WAM). The algorithm exploits the possibilities of using a large address space, with special attention paid to virtual memory behaviour. The time required for the algorithm depends linearly on the size of the areas in which garbage is collected. Pseudo-code is given for the whole algorithm. It has been fully implemented in Tricia, a Prolog developed at UPMAIL for the DEC-2060 computer. 1. PROBLEM In languages with dynamic memory allocation it often happens that an instance of a data structure is not needed for the rest of the computation. Assuming that virtual memory is a limited resource (although it may be very large) it is desirable to be able to re-use such garbage structures. Placing the responsibility of returning garbage on the programmer is a too heavy burden for non-trivial programs so it should be mechanized and invisible to the programmer. In modern abstract Prolog machines, such as WAM 19] or ZIP 8] some garbage collection is built into the machine itself. On backtracking, allocated memory in the environment and term stacks is immediately reclaimed. In addition, environment stack memory is reclaimed during deterministic execution and tail-recursive calls. These and other peculiarities of Prolog machines have to be considered when writing a garbage collector, or it is likely to be ineecient in time and/or space. A garbage collection algorithm for WAM or a similar machine must have the following properties: 1. The garbage collected term stack must be compactiied. 2. The order of objects in the term stack must be preserved, at least within each term stack segment. (A term stack segment is deened as the part of the term stack allocated between two subsequent choice points. Environment stack segment and trail segment are deenied analogously.) This criterion rules out many LISP garbage collectors. 3. Non-atomic objects may be of arbitrary size. 4. There may be pointers with diierent tags to the same address. The standard example is a list whose head is an unbound variable. Both variable and list-tagged pointers to the rst word of the list object may appear. This can be avoided, at a price, by always moving unbound variables out of lists and other compound terms. There may be other reasons than garbage collection to do this 7]. The ideal garbage collection would progress in parallel …