rusty-ecma / RESS

Rusty EcmaScript Scanner
MIT License
93 stars 10 forks source link

RESS

Rusty EcmaScript Scanner

Github Actions crates.io last commit master

A scanner/tokenizer for JS written in Rust

Usage

The primary way to interact with ress is through the Scanner struct which implements Iterator over the Item struct. Item has three fields token for the Token found, span which represents the start and end of the byte position in the original string and location which represents start and end character position with a line and column. It's definition looks like this.

Item {
    token: Token::Punct(Punct::Bang),
    span: Span {
        start: 0,
        end: 1,
    },
    location: SourceLocation {
        start: Position {
            line: 1,
            column: 1,
        },
        end: Position {
            line: 1,
            column: 2,
        }
    }
}

Note: the EcmaScript spec allows for 4 new line characters, only two of which are normally rendered by modern text editors the location line numbers will count these un-rendered lines.

Here is an example that will check some JS text for the existence of a semicolon and panics if one is found.

use ress::Scanner;

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

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

By far the most important part of Item is the Token enum, which will represent the 11 different types of token's supported by the ECMAScript specification.

In Javascript it is hard to know if a forward slash means divide or is the start of a regular expression. The above Scanner will detect RegEx automatically by keeping track of the previously parsed tokens, this makes things very convenient, however if you are parsing Javascript into an AST, you likely already need to keep track of the same information. In that case, you may not want to pay the performance cost of that automatic RegEx detection, you would want to reach for the ManualScanner. Instead of exposing the basic Iterator interface, it exposes two primary methods for driving the scanner next_token and next_regex. The first of those will always return a / or /= when encountering a regular expression, the latter will fail if the next token isn't a regular expression.

use ress::{ManualScanner, prelude::*};

fn main() {
    let mut s = ManualScanner::new("let x = /[a-z]+/g");
    while let Some(Ok(item)) = s.next_token() {
        if item.token.matches_punct(Punct::ForwardSlash)
        || item.token.matches_punct(Punct::ForwardSlashEqual) {
            // it could be a 1 or 2 length prefix
            let regex = s.next_regex(1).unwrap().unwrap();
            println!("{:?}", regex);
        } else {
            println!("{:?}", item);
        }
    }
}

ES Tokens

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, 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. As of the writing of this readme ress supports all tokens in the Stage 2 and Stage 3 ECMAScript Proposals with the exception of the #! comments and number seperators.

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.

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 can be converted into a Peekable 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().unwrap().token, Token::Keyword(Keyword::Function));
assert_eq!(s.next().unwrap().unwrap().token, Token::Punct(Punct::OpenParen));
assert_eq!(s.next().unwrap().unwrap().token, Token::Punct(Punct::CloseParen));
s.set_state(start);
assert_eq!(s.next().unwrap().unwrap().token, Token::Keyword(Keyword::Function));

Why?

Wouldn't it be nice to write new JS development tools in Rust? The clear-comments 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