Piston-Meta

A DSL parsing library for human readable text documents

Introduction

Piston-Meta makes it easy to write parsers for human readable text documents. It can be used for language design, custom formats and data driven development.

Meta parsing is a development technique that goes back to the first modern computer. The idea is to turn pieces of a computer program into a programmable pipeline, thereby accelerating development. An important, but surprisingly reusable part across projects, is the concept of generating structured data from text, since text is easy to modify and reason about.

Most programs that work with text use the following pipeline:

f : text -> data

The problem with this approach is that f changes from project to project, and the task of transforming text into a data structure can get very complex. For example, to create a parser for the syntax of a programming language, one might need several thousands lines of code. This slows down development and increases the chance of making errors.

Meta parsing is a technique where f gets split into two steps:

f <=> f2 . f1
f1 : text -> meta data
f2 : meta data -> data

The first step f1 takes text and converts it into meta data. A DSL (Domain Specific Language) is used to describe how this transformation happens. The second step f2 converts meta data into data, and this is often written as code.

Rules

The meta language is used to describe how to read other documents. First you define some strings to reuse, then some node rules. The last node is used to read the entire document.

20 document = [.l(string:"string") .l(node:"node") .w?]

Strings start with an underscore and can be reused among the rules:

_opt: "optional"

Nodes start with a number that gets multiplied with 1000 and used as debug id. If you get an error #4003, then it was caused by a rule in the node starting with 4.

"Hello world" in Piston-Meta

extern crate piston_meta;

use piston_meta::*;

fn main() {
    let text = r#"hi James!"#;
    let rules = r#"
        1 say_hi = ["hi" .w? {"James":"james" "Peter":"peter"} "!"]
        2 document = say_hi
    "#;
    // Parse rules with meta language and convert to rules for parsing text.
    let rules = match syntax_errstr(rules) {
        Err(err) => {
            println!("{}", err);
            return;
        }
        Ok(rules) => rules
    };
    let mut data = vec![];
    match parse_errstr(&rules, text, &mut data) {
        Err(err) => {
            println!("{}", err);
            return;
        }
        Ok(()) => {}
    };
    json::print(&data);
}

Bootstrapping

When the meta language changes, bootstrapping is used to hoist the old meta syntax into the new meta syntax. Here is how it works:

1. Piston-Meta contains composable rules that can parse many human readable text formats.
2. Piston-Meta knows how to parse and convert to its own rules, known as "bootstrapping".
3. Therefore, you can tell Piston-Meta how to parse other text formats using a meta language!
4. Including the text format describing how to parse its own syntax, which generates equivalent rules to the ones hard coded in Rust.
5. New versions of the meta language can describe older versions to keep backwards compatibility, by changing the self syntax slightly, so it can read an older version of itself.