ICFP 2006. Joint work with James Leifer.
In most programming languages, type abstraction is guaranteed by syntactic scoping in a single program, but is not preserved by marshalling during distributed communication. A solution is to generate hash types at compile time that consist of a fingerprint of the source code implementing the data type. These hash types can be tupled with a marshalled value and compared efficiently at unmarshall time to guarantee abstraction safety. In this paper, we extend a core calculus of ML-like modules, functions, distributed communication, and hash types, to integrate structural subtyping, user- declared subtyping between abstract types, and bounded existential types. Our semantics makes two contributions: (1) the explicit tracking of the interaction between abstraction boundaries and subtyping; (2) support for user-declared module upgrades with propagation of the resulting subhashing relation throughout the network during communication. We prove type preservation, progress, determinacy, and erasure for our system.
This short version is quite similar to the one that appeared in the proceedings [pdf].