Relating Two Semantics of Locally Scoped Names

Abstract

The operational semantics of programming constructs involving locally
scoped names typically makes use of stateful "dynamic allocation": a
set of currently-used names forms part of the state and upon entering
a scope the set is augmented by a new name bound to the scoped
identifier. More abstractly, one can see this as a transformation of
local scopes by expanding them outward to an implicit top-level. By
contrast, in a neglected paper from 1994, Odersky gave a stateless
lambda calculus with locally scoped names whose dynamics contracts
scopes inward. The properties of "Odersky-style" local names are quite
different from dynamically allocated ones and it has not been clear,
until now, what is the expressive power of Odersky's notion. We show
that in fact it provides a direct semantics of locally scoped names
from which the more familiar dynamic allocation semantics can be
obtained by continuation-passing style (CPS) translation. More
precisely, we show that there is a CPS translation of typed lambda
calculus with dynamically allocated names (the Pitts-Stark
nu-calculus) into Odersky's lambda-nu-calculus which is
computationally adequate with respect to observational equivalence in
the two calculi.