Lifting and Flattening of Typed ASTs
Ed Amsden
Abstract
Strongly-typed functional languages such as Haskell have several advantages as languages for language implementation. In particular, Algebraic Datatypes with pattern-matching present a natural way to write much of the functionality of the compiler or interpreter in an inductive way. Generalized Algebraic Datatypes allow the typing features of the host language to be employed to ensure that transformations of the AST preserve some notion of type soundness on the AST.
A useful pattern in language implementation is “flattening”, where nested expressions in an AST are converted to a spine of let-bindings of AST primitives, whose sub-expressions are variable references brought into scope by the let bindings. I show how to implement a flattening transformation from a typed AST to a new datatype which preserves the type-soundness of the AST while guaranteeing by its structure that no nested expressions are present. This permits the user to write total functions over the AST which are checked for the preservation of AST type-soundness and assume a flattened representation.