ress 0.7.0-beta.5

A sanner/tokenizer for JS files
Documentation

RESS

Rusty EcmaScript Scanner

travis appveyor crates.io last commit master

A scanner/tokenizer for JS written in Rust

Usage

There are two main interfaces for using ress in your Rust code.

The first is the very simple function tokenize, this takes in a String and outputs a Vec<Token>.

extern crate ress;

use ress::tokenize;

static JS: &str = include_str!("index.js");

fn main() {
    let tokens = tokenize(JS);
    it !tokens.iter().any(|t| t.matches_punct_str(";")) {
        panic!("No semi-colon!? You nave!");
    } else {
        println!("At least you are sane at one point");
    }
}

The other option is to create a Scanner, an iterator over the Item struct. Item has two fields token for the RefToken found and Span for the position in the string.

extern crate ress;

use ress::{Scanner};

static JS: &str = include_str!("index.js");

fn main() {
    let s = Scanner::new(JS);
    for token in s {
        if token.matches_punct_str(";") {
            panic!("A semi-colon!? Heathen!");
        }
    }
    println!("Good show! Why use something that's optional?")
}

In either method the major construct that you would be dealing with is a RefToken enum. This enum represents the 10 different tokens defined in the ECMAScript specification.

ES Tokens

  • Boolean Literal
  • End of File
  • Identifier
  • Keyword
  • Null Literal
  • Numeric Literal
  • Punctuation
  • String Literal
  • Regular Expression Literal
  • Comment

Keep in mind that keywords have been moving around a lot in JS between ES3 through ES2019 so you might find some items parsed as keywords in the ES2019 context that are not in the ES3 context and since my goal is keep this scanner context free this should be dealt with at a higher level. A good example of this is yield which is sometimes a keyword and sometimes an identifier, this package will always parse this as a Keyword. Please also note that any pending specifications at tc39 may not be included

For each of the token cases there is either a struct or enum to provide additional information with the exception of NullLiteral and EoF which should be self explanatory. The more complicated items do implement ToString which should get you back to the original js text for that token. The Token enum also provides a number of helper functions for building that picture without pulling the inner data our of the enum. Using the Punct case as an example the helper functions look like this

fn is_punct(&self) -> bool;
fn matches_punct(&self, p: Punct) -> bool;
fn matches_punct_str(&self, s: &str) -> bool;

A similar set of functions are available for each case. Be aware that some _str implementations panic if the wrong string is provided meaning these would also panic.

let p = Token::Punct(Keyword::This);
if p.matches_keyword_str("junk") {
    // panic!
}
if p.matches_keyword(Keyword::This) {
    // Don't panic!
}

Like all Iterators the Scanner has a next method, It also has a look_ahead method that will allow you to parse the next value without advancing. Using this method can be a convenient way to get the next token without performing a mutable borrow, however you will be incurring the cost of parsing that token twice. All Iterators implement Peekable that will convert them into a new iterator with a peek method, this will allow you to look ahead while only paying the cost once however peek performs a mutable borrow which means it needs to be in a different scope than a call to next.

// look_ahead
let js = "function() { return; }";
let mut s = Scanner::new(js);
let current = s.next();
let next = s.look_ahead();
let new_current = s.next();
assert_eq!(next, new_current);
// peekable (fails to compile)
let p = Scanner::new(js).peekable();
let current = s.next(); // <-- first mutable borrow
let next = p.peek(); // <-- second mutable borrow

For more intense lookahead scenarios Scanner makes available the get_state and set_state methods. These methods will allow you to capture a snapshot of the current position and any context, and then later reset to that position and context.

let js = "function() {
    return 0;
};";
let mut s = Scanner::new(js);
let start = s.get_state();
assert_eq!(s.next().unwrap().token, RefToken::Keyword(Keyword::Function));
assert_eq!(s.next().unwrap().token, RefToken::Punct(Punct::OpenParen));
assert_eq!(s.next().unwrap().token, RefToken::Punct(Punct::CloseParen));
s.set_state(start);
assert_eq!(s.next().unwrap().token, RefToken::Keyword(Keyword::Function));

In addition to the RefToken enum this crate also provides an OwnedToken enum, this will semantically work the same as a RefToken (as they both impl Token) however it will require a string be allocated to construct. This can be useful if you need to work outside of the lifetime bounds of a RefToken.

Why?

Wouldn't it be nice to write new JS development tools in Rust? The (clear-comments)[https://github.com/FreeMasen/RESS/blob/master/examples/clear-comments/src/main.rs] example is a proof of concept on how you might use this crate to do just that. This example will take in a JS file and output a version with all of the comments removed. An example of how you might see it in action is below (assuming you have a file called in.js in the project root).

$ cargo run --example clear-comments -- ./in.js ./out.js

Performance

The below stats are from running cargo +nightly bench on a MBP (2.9 GHz i9-8850H & 16bg RAM).

Lib Size Time +/-
Angular 1.5 1.16mb 18.991 ms 4.393 ms
jquery 271.75kb 7.218 ms 577.236 μs
React 59.09kb 1.976 ms 116.139 μs
React-dom 641.51kb 16.880 ms 3.614 ms
Vue 289.30kb 9.675 ms 1.402 ms

If you are interested in getting an idea about performance without waiting for cargo bench to complete you can run the following command.

$ cargo run --example major_libs

Contributing

see contributing.md