Evolution of partial evaluators: Removing inherited limits

Abstract

Abstract. We show the evolution of partial evaluators over the past ten years from a particular perspective: the attempt to remove limits on the structure of residual programs that are inherited from structural bounds in the original programs. It will often be the case that a language allows an unbounded number or size of a particular features, but each program being nite will only have a nite number or size of these features. If the residual programs cannot overcome the bounds given in the original program, that can be seen as a weakness in the partial evaluator, as it potentially limits the e ectiveness of residual programs. The inherited limits are best observed through specializing a self-interpreter and examining the object programs produced by specialisation of this. We show how historical developments in partial evaluators gradually remove inherited limits, and suggest how this principle can be used as a guideline for further development.

Jones optimality, binding-time improvements, and the strength of program specializers

“…The results also support the observation [25] that a specializer has a weakness if it cannot overcome inherited limits and that they are best observed through specializing a self-interpreter (which amounts to testing whether a specializer is Jones-optimal).…”

“…In particular, when programs p and p are identical, it is safe to conclude that this goal is achieved. Also, as explained in [25], the limits on the structure of residual programs that are inherited from structural bounds in the source programs are best observed by specializing a self-interpreter (e.g., the arity of function definitions in a residual program).…”

“…For instance, for a partial evaluator for a first-order functional language with algebraic data types [25], a combination of several transformation methods is necessary (constant folding, unfolding, polyvariant specialization, partially static values, constructor specialization, type specialization).…”

“…6. More realistic Jones-optimal specializers, for instance [26,23,25,28], also satisfy the property of static expression reduction.…”

“…These works focus on the problem of self-application; none of them considers the role of Jones-optimal specialization for binding-time improvements. Also, it was said [25] that a specializer is 'weak' if it cannot overcome inherited limits and that this is best observed by specializing a selfinterpreter. Our results underline this argument from the perspective of binding-time improvements.…”