A Transformation-Based Foundation for Semantics-Directed Code Generation

Interpreters and compilers are two different ways of implementing programming languages. An interpreter directly executes its program input. It is a concise definition of the semantics of a programming language and is easily implemented. A compiler translates its program input into another language. It is more difficult to construct, but the code that it generates runs faster than interpreted code. In this dissertation, we propose a transformation-based foundation for deriving compilers from semantic specifications in the form of four rules. These rules give apriori advice for staging, and allow explicit compiler derivation that would be less succinct with partial evaluation. When applied, these rules turn an interpreter that directly executes its program input into a compiler that emits the code that the interpreter would have executed. We formalize the language syntax and semantics to be used for the interpreter and the compiler, and also specify a notion of equality. It is then possible to precisely state the transformation rules and to prove both local and global correctness theorems. And although the transformation rules were developed so as to apply to an interpreter written in a denotational style, we consider how to modify non-denotational interpreters so that the rules apply. Finally, we illustrate these ideas by considering a larger example: a Prolog implementation