Skip to content
master
Switch branches/tags
Code

Files

Permalink
Failed to load latest commit information.

lib-ruby-parser

test unsafe forbidden Crates.io codecov MIT Licence dependency status Docs

lib-ruby-parser is a Ruby parser written in Rust.

Basic usage:

use lib_ruby_parser::{Parser, ParserOptions};

fn main() -> Result<(), Box<dyn std::error::Error>> {
    let options = ParserOptions {
        buffer_name: "(eval)".to_string(),
        ..Default::default()
    };
    let mut parser = Parser::new(b"2 + 2".to_vec(), options);

    println!("{:#?}", parser.do_parse());

    Ok(())
}

Full documentation

Features

TLDR; it's fast, it's precise, and it has a beautiful interface.

Comparison with Ripper/RubyVM::AST:

  1. It's based on MRI's parse.y, and so it returns exactly the same sequence of tokens.
  2. It's been tested on top 300 gems (by total downlads, that's about 3M LOC), rubyspec and ruby/ruby repos and there's no difference with Ripper.lex.
  3. It's ~3 times faster than Ripper (with jemalloc), Ripper parses 3.9M LOC in ~16s, lib-ruby-parser does it in ~6.5s. That's ~600K LOC/s. You can find some benchmarks in the bench/ directory, they don't include IO and GC.
  4. It has a much, much better interface. AST is strongly typed and well documented.
  5. It doesn't throw away information about tokens. All nodes have information about their source locations.

Comparison with whitequark/parser:

  1. It's much faster (the same corpus of 3M LOC can be parsed in 180s on the same machine)
  2. It has a very similar interface (both in terms of AST structure and errors reporting)
  3. However, AST is strongly typed, and so if something is nullable it's explicitly defined and documented.
  4. What's important, it doesn't depend on Ruby

Grammar versioning

lib-ruby-parser follows MRI/master. There are no plans to support multiple versions like it's done in whitequark/parser.

Library versioning

Ruby version lib-ruby-parser version
3.0.0 3.0.0+

Encodings

By default lib-ruby-parser can only parse source files encoded in UTF-8 or ASCII-8BIT/BINARY.

It's possible to pass a decoder function in ParserOptions that takes a recognized (by the library) encoding and a byte array. It must return a UTF-8 encoded byte array or an error:

use lib_ruby_parser::source::{InputError, Decoder, DecoderResult};
use lib_ruby_parser::{Parser, ParserOptions, ParserResult};

fn decode(encoding: String, input: Vec<u8>) -> DecoderResult {
    if "US-ASCII" == encoding.to_uppercase() {
        // reencode and return Ok(result)
        return DecoderResult::Ok(b"# encoding: us-ascii\ndecoded".to_vec());
    }
    DecoderResult::Err(InputError::DecodingError(
        "only us-ascii is supported".to_string(),
    ))
}

let options = ParserOptions {
    decoder: Some(Decoder::new(Box::new(decode))),
    ..Default::default()
};
let mut parser = Parser::new(b"# encoding: us-ascii\n3 + 3".to_vec(), options);
let ParserResult { ast, input, .. } = parser.do_parse();

assert_eq!(ast.unwrap().expression().source(&input).unwrap(), "decoded".to_string())

Invalid string values

Ruby doesn't require string literals to be valid in their encodings. This is why the following code is valid:

# encoding: utf-8

"\xFF"

Byte sequence 255 is invalid in UTF-8, but MRI ignores it.

But not all languages support it, and this is why string and symbol nodes encapsulate a custom StringValue instead of a plain String.

If your language supports invalid strings you can use raw .bytes of this StringValue. For example, a Ruby wrapper for this library could do that.

If your language doesn't support it, better call .to_string_lossy() that replaces all unsupported chars with a special U+FFFD REPLACEMENT CHARACTER (�).

Regexes

Ruby constructs regexes from literals during parsing to:

  1. validate them
  2. declare local variables if regex is used for matching AND it contains named captures

To mirror this behavior lib-ruby-parser uses Onigurama to compile, validate and parse regex literals.

This feature is disabled by default, but you can add it by enabling "onig" feature.

Bison

The grammar of lib-ruby-parser is built using a custom bison skeleton that was written for this project.

For development you need the latest version of Bison installed locally. Of course, it's not necessary for release builds from crates.io (because compiled parser.rs is included into release build AND build.rs that converts it is excluded).

If you use it from GitHub directly you also need Bison (because parser.rs is under gitignore)

Bindings for other languages

Profiling

You can use parse example:

$ cargo run --features run-examples --example parse -- --print=N --run-profiler=1 "<pattern>"

Benchmarking

A codebase of 3.9M LOCs can be generated using a download.rb script:

$ ruby gems/download.rb

Then, run a script that compares Ripper and lib-ruby-parser (attached results are from Feb 2021):

$ ./bench/compare.sh
    Finished release [optimized] target(s) in 0.08s
Running lib-ruby-parser
Run 1:
Time taken: 6.6232788220 (total files: 18018)
Run 2:
Time taken: 6.6498335800 (total files: 18018)
Run 3:
Time taken: 7.0684415810 (total files: 18018)
Run 4:
Time taken: 6.7987308510 (total files: 18018)
Run 5:
Time taken: 6.6954798760 (total files: 18018)
--------
Running MRI/ripper
Run 1:
Time taken: 22.92822499992326 (total files: 18017)
Run 2:
Time taken: 21.8613000002224 (total files: 18017)
Run 3:
Time taken: 21.96083900006488 (total files: 18017)
Run 4:
Time taken: 21.44488099985756 (total files: 18017)
Run 5:
Time taken: 21.738944000098854 (total files: 18017)

Profile-guided optimization

# Build recording executable
RUSTFLAGS="-Cprofile-generate=$(PWD)/target/pgo/pgo.profraw" cargo build --release --example parse

# Record raw profiling data
target/release/examples/parse --no-output "gems/repos/**/*.rb"

# Merge profiled data
llvm-profdata merge -o target/pgo/pgo.profraw/merged.profdata target/pgo/pgo.profraw

# Build optimized executable
RUSTFLAGS="-Cprofile-use=$(PWD)/target/pgo/pgo.profraw/merged.profdata" cargo build --release --example parse

PGO, No LTO:

$ repeat 5 time target/release/examples/parse --no-output "gems/repos/**/*.rb"
9.46s user 1.27s system 80% cpu 13.371 total
8.51s user 0.66s system 99% cpu 9.171 total
8.52s user 0.68s system 99% cpu 9.208 total
9.63s user 0.74s system 99% cpu 10.381 total
9.70s user 0.73s system 99% cpu 10.443 total

No PGO, LTO=fat:

$ repeat 5 time target/release/examples/parse --no-output "gems/repos/**/*.rb"
9.90s user 1.29s system 80% cpu 13.917 total
9.42s user 0.71s system 99% cpu 10.138 total
10.24s user 0.76s system 99% cpu 11.004 total
10.21s user 0.75s system 99% cpu 10.962 total
10.22s user 0.74s system 99% cpu 10.966 total

The diff seems to be too small to use this feature.

When both PGO and LTO are enabled building a parse example gives a bunch of LLVM errors about wrong types of functions (like expected a Function or null).

If you know how to fix them, please open an issue.

Fuzz testing

First, make sure to switch to nightly:

$ rustup default nightly

Then install cargo-fuzz:

$ cargo install cargo-fuzz

And run the fuzzer (change the number of --jobs as you need or remove it to run only 1 parallel process):

$ RUST_BACKTRACE=1 cargo fuzz run parse --jobs=8 -- -max_len=50