19:49 <korvo> Okay, I've finished investigating my claim that packrat-style PEG parsers are kind of like partial evaluation. It's not worth a blog post, but I did end up creating a language that I'll probably poke at more.

19:50 <korvo> Short version: yes, technically, a packrat parser can evaluate a rule many times, with different runtime parameters, over a fixed (section of) input text. Each evaluation can generate a fresh semantic action, so it could generate multiple versions of a rule which are themselves specializable.

19:51 <korvo> But *all* of the semantic state can be captured in a sufficiently-rich tree algebra, and we can just ask a rule to emit trees which are specialized by later tree transformations once the parser is finished running.

19:54 <korvo> Also, although I haven't built it, a packrat-memoized Earley parser should be capable of specializing multiple rules in parallel. Really, Earley with lookahead assertions is just parallel PEG.

19:55 <korvo> Anyway I forked META II into a language that compiles to RPython. The language lets users describe a PEG grammar as well as multiple tree algebras and functors from one algebra to another. It compiles itself in 250 lines using three algebras: https://github.com/MostAwesomeDude/zaddy/blob/meta/zaddy.zaddy

19:57 <korvo> The old partial-evaluation annotations are removed because the tree algebras generate much tighter code which doesn't need to be optimized. Still, I documented them here: https://esolangs.org/wiki/Zaddy