Jo is a Lisp dialect that compiles directly to Go source code.

A Lisp dialect which compiles directly to Go source code

This is a fizzbuzz solution in Jo:

(package main)

(import "fmt")

(func main ()
    (for (define i 0) (< i 100) (inc i)
        (if (= 0 (% i 15))
            (fmt.Println "fizzbuzz")
            (if (= 0 (% i 3))
                (fmt.Println "fizz")
                (if (= 0 (% i 5))
                    (fmt.Println "buzz")
                    (fmt.Println i))))))

And the corresponding Go code (playground link):

package main

import "fmt"

func main() {
	for i := 0; i < 100; i++ {
		if 0 == i%15 {
		} else {
			if 0 == i%3 {
			} else {
				if 0 == i%5 {
				} else {

Building Jo

The Jo compiler is really just a parser that takes Jo source code as input and produces a Go AST, which can be converted to Go source code using the standard library's go/printer or go/format packages so nothing more has to be done.

I implemented the parser using parser combinators, starting by adapting the article Learning Parser Combinators With Rust into Go.

After moving on to parsing Jo code, I had an “Aha!” moment when I realised how all the parsers I was writing could be combined into a single parser which directly returned the entire AST. Before that, I had actually mixed up lexical analysis and parsing, due to a preconception that compilers always had separate lexical analysis and parsing steps.

From that point on, it was a relatively mechanical task of referring to the Go Programming Language Specification and building parsers for each syntax element, then checking the output against actual ASTs parsed from Go programs. The tool astextract was invaluable while doing this.

The biggest challenge was determining how a given Go language construct should be written in Jo. Certain decisions were easy to make, like writing package main as (package main) or using (f a b c) for function call syntax (because Lisp). But decisions like using (define x 0) for short variable declaration x := 0 and (assign x 0) for normal assignment x = 0 were mostly stylistic, and decisions like sometimes requiring an additional set of parentheses around expression lists were just to make parsing easier.


From the start Jo was just a learning project, there's absolutely no reason to use it other than for fun, unlike some other hosted Lisps like Clojure on the JVM or Hy in Python.

Nevertheless, I still want to continue improving it, and apart from supporting more Go features (key features like slices, struct literals and goroutines are all not implemented yet), there's one thing I really want to add: macros!

What is a Lisp without metaprogramming? At one point I tried to add a feature to Jo that wasn't present in Go in the form of an assert statement, but in the process of doing so, I realised that rather baking it into the compiler, it was just one of many possible language extensions with a macro system.

In fact, code generation via external tools is something that Go has clearly embraced, and Jo itself is a Go code-generating tool, so I think adding a macro system makes sense.

At least that's the plan—I'm not an experienced Lisp programmer nor have I done much metaprogramming.

Another feature I've thought about is type inference. Go already has some form of type inference in short variable declarations, and it follows that Jo does as well. But if Jo had a more advanced type system, it could become more than the Go AST mapped to S-expressions. For example, maybe function literals could be written in a more compact form but compile to appropriately-typed Go function literals.

I've got no idea how it would turn out though. Such a feature would definitely increase Jo's complexity exponentially since it would no longer be just a dumb syntax transformation.

Finally, I'm also interested in further exploring parsing techniques. Apparently parser combinators are a form of recursive descent parser which is one of the more basic strategies, for reasons related to terms I don't understand right now like “LL”, “LR” and “left recursion”. Specifically, I'd like to see how I could use ANTLR, which is something called a parser generator, to implement Jo instead.