Erasable coercions:
a unified approach to type systems

Defense informations

The defense took place on the 30th of January, 2014 at 14:30 in room 227C of Paris 7 University:

Hall aux Farines
16 rue Françoise Dolto
75013 Paris
The slides are available here.

Abstract

Functional programming languages, like OCaml or Haskell, rely on the Lambda Calculus for their core language. Although they have different reduction strategies and type system features, their proof of soundness and normalization (in the absence of recursion) should be factorizable. This thesis does such a factorization for theoretical type systems featuring recursive types, subtyping, bounded polymorphism, and constraint polymorphism. Interestingly, soundness and normalization for strong reduction imply soundness and normalization for all usual strategies. Our observation is that a generalization of existing coercions permits to describe all type system features stated above in an erasable and composable way. We illustrate this by proposing two concrete type systems: first, an explicit type system with a restricted form of coercion abstraction to express subtyping and bounded polymorphism; and an implicit type system with unrestricted coercion abstraction that generalizes the explicit type system with recursive types and constraint polymorphism—but without the subject reduction property. A side technical result is an adaptation of the step-indexed proof technique for type-soundness to calculi equipped with a strong notion of reduction.

PhD informations

The manuscript and formalization (coq files and documentation) can be found at http://phd.ia0.fr/ .