Optimizing Pattern Matching Compilation by Program Transformation

Abstract

Motivated by the promotion of rewriting techniques and their use in major industrial applications, we have designed Tom: a pattern matching layer on top of conventional programming languages. The main originality is to support pattern matching against native data-structures like objects or records.

While crucial to the efficient implementation of functional languages
as well as rewrite rule based languages, in our case, this combination of algebraic constructs with arbitrary native data-structures makes the pattern matching algorithm more difficult to compile. In particular, well-known many-to-one automaton-based techniques cannot be used. We present a two-stages approach which first compiles pattern matching constructs in a naive way, and then optimize the resulting code by program transformation using rewriting. As a benefit, the compilation algorithm is simpler, easier to extend, and the resulting pattern matching code is almost as efficient as best known implementations.