Type-safe compilation of covariant specialization: A practical case

Abstract

Despite its lack of type safety, some typed object-oriented languages use covariant specialization for methods. In this work, we show how one may modify the semantics of languages that use covariant specialization in order to improve their type safety. We demonstrate our technique using O₂, a strongly and statically typed object-oriented database programming language which uses covariant specialization. We propose a modification to the O₂ compiler that adds code to correct previously ill-typed computations that arise from the use of covariant specialization. The modification we propose does not affect the semantics of those computations without type errors. Furthermore, the new semantics of the previously ill-typed computations is defined in a very “natural” way and ensures the type safety (w.r.t. covariance) of the program. Since the solution consists of a conservative backward-compatible modification of the compiler, it does not require, unlike other solutions, any modification of existing O₂ code. Our solution is based solely on a type-theoretic analysis and thus is general. Therefore, although this paper applies it to a particular programming language, the same ideas could easily be applied to other languages that use covariant specialization.

The work of John Boyland was supported in part by Advanced Research Projects Agency grant MDA972-92-J-1028, and by NSF Infrastructure grant CDA-9401156. The content of this paper does not necessarily reflect the position or the policy of the U. S. Government, and no official endorsement should be inferred.