Haskell Is Not Not ML

We present a typed calculus IL (“intermediate language”) which supports the embedding of ML-like (strict, eager) and Haskell-like (non-strict, lazy) languages, without favoring either. IL’s type system includes negation (continuations), but not implication (function arrow). Within IL we find that lifted sums and products can be represented as the double negation of their unlifted counterparts. We exhibit a compilation function from IL to AM—an abstract von Neumann machine—which maps values of ordinary and doubly negated types to heap structures resembling those found in practical implementations of languages in the ML and Haskell families. Finally, we show that a small variation in the design of AM allows us to treat any ML value as a Haskell value at runtime without cost, and project a Haskell value onto an ML type with only the cost of a Haskell deepSeq. This suggests that IL and AM may be useful as a compilation and execution model for a new language which combines the best features of strict and non-strict functional programming.