Fair Termination with Cruel Schedulers

Abstract

The problem of transforming nondeterministic or concurrent programs involving fairness assumptions into fairness-free schedulers is considered. The availability of such transformations provides an indirect way of proving fair termination. Building upon a general tree-transformation result of Harel, and scheduler-yielding transformations of devised by several researchers, we provide a general way of finding such transformations for a very broad class of fairness criteria. In passing, we explicitly construct a scheduler for Pnueli’s rather elusive notion of extreme fairness. Our schedulers are really nondeterministic, and may be termed “cruel” since they leave choices to the program, but immediately thereafter cause it to abort if it happens to make the wrong choice.