Lock-free atom garbage collection for multithreaded Prolog - ERRATUM

Abstract

Algorithm 4 on page 960 of the above named article (Wielemaker and Harris 2016) is flawed. The issue is illustrated by algorithm 1 (supplementary figure 1). If a thread A detects the condition table too full is false it proceeds adding its atom to the table. If thread B detects the table is (now) too full it starts a resize. The resize allocates a new table and copies the atoms from the old to the new table. If thread A adds the new atom after the copy loop passes its location and before thread B activates the new table the insertion is considered successful, but the new atom is only in the deactivated old table.