Lambda the Ultimate

Wow. That really makes the idea crystal clear--it's all just partial evaluation in principle. It's a deep principle. Plus a good quote: "More generally, we can think of many, many code optimization problems as moving binding times around, where the same logic applies."

Too bad there's not good language support for this kind of thing in most languages. It's a shame too, since writing one yourself is tantamount to writing an interpreter for the language. That's a lot of duplicated effort.

It seems like so often the newer static languages get this close to letting you do some really neat metaprogramming. Like the language designers either just didn't see the potential, or consciously decided to seal off what would otherwise be a natural part of the language.

It seems like so often the newer static languages get this close to letting you do some really neat metaprogramming.

You may then be interested in what I did with the Heron design. Heron is a simple object-oriented language, implemented using a two-phase interpreter. There is a meta-phase, where the AST is provided to an entry-point function called "Main" (think of it like a compile-time "Main()"). During this phase any part of the program can be executed, and any transformation can be performed on the AST.

When it comes time to execute "Main", the modified AST is executed.

So in theory compilation can be done by the interpreter from within any program by including a Heron library. (It's only theoretical, because I haven't completed the compiling library yet, I have done pretty printers and various AST manipulations).

Thanks for the reference. But the longer journal version is probably a better bet than the condensed APLAS version. And you'll probably want to look at the accompanying code to play around with the ideas.