Parser combinators have a reputation of poor performance—worst-case\nexponential time—and deemed good for prototyping but not for production. Or so\nthe story goes. While techniques like Packrat parsing or static analysis offer\nlinear-time guarantees by addressing naive backtracking, they come with their\nown trade-offs.<\/p>\n
In this post, I want to explore an optimization route that complements the\nothers: avoiding the performance penalty of abstraction.<\/p>\n
By leaning into Scala 3’s metaprogramming capabilities, our combinators combine\ncode fragments<\/em>. By moving to a staged<\/em> continuation-passing style (CPS)\nencoding, we make the control flow explicit and continuations are fully\nevaluated at compile time. Results show that a staged parser runs ~50% faster\nthan a handwritten recursive descent parser and over 25x faster than a\nnon-staged version. Interestingly, this approach is deeply rooted in partial\nevaluation history, closely mirroring the First Futamura\nProjection<\/a>.<\/p>"}}}