Bootstrapping Compiler Generators from Partial Evaluators

Abstract

Abstract. This paper shows that bootstrapping of compiler generators from program specializers is a viable alternative to the third Futamura projection. To practically validate the technique, a novel partial evaluation-based compiler generator was designed and implemented for a recursive flowchart language. Three-step bootstrapping was found to be faster and to produce the same compiler generator that Gomard and Jones produced two decades ago by double self-application. Compilergenerator bootstrapping has distinct properties that are not present in the classic three Futamura projections, such as the ability to turn a specializer into a compiler generator in one step without self-application. Up to now, the approach of hand-writing compiler generators has only been used to avoid difficulties when specializing strongly-typed languages, not as a first step towards compiler-generator bootstrapping.