Type-changing rewriting and semantics-preserving transformation

Abstract

We have identified a class of regular, whole-program transformations that cannot be safely performed with typical transformation techniques because transformation requires changing the types of terms. In these transformations, we want to change typically large parts of a program from using one type to using another type while simultaneously preserving the original program semantics after transformation.In this paper, we present type-and-transform systems, an automated approach to the whole-program transformation of terms involving type A to terms involving the isomorphic type B using typechanging rewrite rules. Type-and-transform systems establish typing and semantics relations between all source and target subprograms such that a complete transformation can guarantee the equivalent semantics of a whole program. We describe the type-andtransform system for the lambda calculus with let-polymorphism and general recursion, including several examples from the literature and properties of the system.

Late data layout

“…In the area of formal descriptions, two papers on typedirected coercion insertion stand out as very closely related to this paper [41,69]. The work of Swami et al [69] focuses on automatically composing several coercions together in order to bridge the gap between different types.…”

“…This work resembles the mechanisms used to introduce implicit conversions in Scala, although the rules provide more flexibility and are proven not to diverge. On the other hand, Leather et al [41] describe a coercion insertion mechanism which deliberately produces ambiguous rewritings from which heuristics can pick the best. More importantly, the formalism presented in [41] is also capable of consistently changing types in the rewrite rules, making the transformation very versatile.…”

“…On the other hand, Leather et al [41] describe a coercion insertion mechanism which deliberately produces ambiguous rewritings from which heuristics can pick the best. More importantly, the formalism presented in [41] is also capable of consistently changing types in the rewrite rules, making the transformation very versatile. Unfortunately, the two formalisms do not handle backward propagation, object orientation and subtyping, all of which are crucial to performing optimal data representation transformations in Scala.…”

Towards Trustworthy Refactoring in Erlang

“…Executing the "fun2value" refactoring on the fun expression checks if the value "apple" is side-effect free, and then it removes the unnecessary abstraction and application. With this scheme, one might implement API-adaptation, or type-changing refactorings [11] as well. Nevertheless, for such refactorings, data references should be gathered based on behavioural dependencies rather than just ordinary dataflow.…”

Type-changing rewriting and semantics-preserving transformation

“…Our ultimate goal is to extend the theory to Haskell, implement full support for the language, and investigate the realworld effectiveness of type-and-transform systems. This article is an update of another paper published with the same name [22]. We have made a number of changes since that first paper.…”