#abstract #binding #tree

nightly rabbot-plugin

Compiler plugin for Rabbot

1 unstable release

Uses old Rust 2015

0.1.0 Aug 26, 2016

#19 in #abstract

Apache-2.0/MIT

2KB

rabbot

Rabbot generates interfaces to abstract binding trees (ABTs). It is a port of SML's Abbot to Rust.

What is an ABT?

An abstract binding tree is a data structure similar to an abstract syntax tree, except that it lets you pleasantly and correctly handle the declaration and binding of symbols. To motivate this, let's consider a syntax tree for a simple pseudo-language. For the expression 2 + 1, this would produce a syntax tree Plus(Number(2), Number(1)). For the expression let x = 1 in x + 2, this would produce Let(Var("x"), Number(1), Plus(Var("x"), Number(2)). For these examples, using a normal abstract syntax tree works fine.

However, we run into trouble if we introduce multiple bindings with the same name. For example, in the expression let x = 1 in (let x = 2 in x) + x, we now need some notion of scope analysis. Which x does each instance refer to? ABTs eliminate this problem by providing constructs to handle the binding and substitution of variables. The prior expression translates roughly to Let(1, x . (Let(2, x . x) + x)) where the dot . denotes a binding site. See Bob Harper's Practical Foundations for Programming Languages, chapter 1 for more information.

What does Rabbot do?

The usual way to create an abstract syntax is tree is to define an enum with branches for each part of your syntax, e.g.

enum Expression {
    Number(i32),
    Plus(Box<Expression>, Box<Expression>),
    ...
}

However, ABTs work differently. At its core, an ABT is a fixed tree data structure that is parameterized by a set of operators, where each operator describes functionality like Number or Plus. See rabbot/src/abt.rs for the definition of the data type. This lets you easily re-use the ABT structure for different kinds of programs, but it requires more boilerplate and more code in practice to use.

To alleviate the verbosity of using ABTs, Rabbot takes a description of an abstract binding tree and generates a more concise interface for the programmer to use. For example, one can implement lambda calculus with Peano numerals and its interpreter in a few lines:

#![feature(plugin, box_syntax)]
#![plugin(rabbot_plugin)]

#[macro_use] extern crate rabbot;

rabbot! {
    sort Term {
        Z,
        S(Term),
        Lam(Binding<Term> . Term),
        App((Term, Term))
    }
}

use rabbot::var::Var;
use term::*;
fn interpret(t: Term) -> Term {
    match out(t) {
        v @ View::Z | v @ View::Lam(_) => into(v),
        View::S(t) => {
            bind!(View::S{v} = out(interpret(t)));
            into(View::S(v))
        },
        View::App((fun, arg)) => {
            bind!(View::Lam{(var, body)} = out(interpret(fun)));
            subst(arg, var, body)
        },
        View::Var(_) => unreachable!()
    }
}

#[test]
fn test() {
    let x = Var::new("x".to_string());
    let term = into(View::App((
        into(View::Lam((
            x.clone(), into(View::S(into(View::Var(x.clone()))))))),
        into(View::Z))));

    println!("{:?}", interpret(term)); // prints S(Z)
}

Caveats

This is still in heavy development, so talk to me (email: wcrichto@stanford.edu) if you want to use it. The code generator works as a compiler plugin, so it only works on nightly, although that's not a hard requirement.

Dependencies