Improving binding times without explicit CPS-conversion

  • Authors:

  • Anders Bondorf

  • Affiliations:

  • -

  • Venue:
  • LFP '92 Proceedings of the 1992 ACM conference on LISP and functional programming
  • Year:
  • 1992

Quantified Score

Hi-index 0.00

Visualization

Abstract

A major obstacle in partial evaluation (program specialization) is the need for binding time improvements. By reorganizing a source program, the residual programs obtained by specializing the source program may be improved: more computations can be done statically, that is, at specialization time.One well-known effective reorganization is (manual or automatic) conversion into continuation passing style (cps). This conversion allows data consumers to be propagated through frozen expressions to the data producers. In this paper we show how such improvements can be obtained without affecting the source program: by writing the program specializer itself in cps; traditionally, specialization has been formulated in direct style.The advantages of avoiding cps-converting source programs are: (1) no cps-conversion phase is needed; (2) the generated residual programs are not in cps; (3) since no source level continuations are added, there is no overhead of manipulating closure representations in the generating extensions (e.g. compilers) obtained by self-application; (4) manual “binding time debugging” is easier since binding time analysis is done on a non-converted program.We have implemented a cps-based program specializer; it is integrated in the partial evaluator Similix 4.0. Using a cps-specializer, partially static data structures can be handled safely in a straightforward way. The difficulty is to ensure automatically that residual expressions that become part of a partially static data structure are neither duplicated nor discarded. This is achieved by binding such residual expressions in automatically inserted frozen let-expressions; cps is needed to propagate operations on the partially static data structure through these frozen let-expressions. Based on this idea, we have implemented an extension of Similix 4.0 that handles partially static data structures.