README.md

Benchmarks

Here you will find some benchmarks and rough performance comparison with similar libraries.

How to run locally

Like a piece of cake.

Note

Make sure to run npm i at the root of the repository as well before running benchmarks, because benchmarks depend on the @nrsk/sigma sources outside this package.

npm i
npm run bench

Notes

Unfortunately, it's difficult to come up with sensible benchmarks, given that how a parser written with sigma depends entirely on what you are parsing, how you structure your parser(s), which patterns the parser attempts to match first, what is involved in constructing your AST (if any), and so on.

All that said, here are some rough numbers from the JSON parsing benchmark (running on my MacBook Pro 16" 2019 with Intel i9-9880H @ 2.30GHz and Node@18).

Running "JSON :: sigma vs parjs" suite...

  sigma: 790 ops/s, ±0.71%   | fastest
  parjs: 146 ops/s, ±0.56%   | slowest, 81.52% slower

I have included results from Sigma and Parjs (another parser combinator library). I wanted to also add Arcsecond, because I like its API with functional flavor, but somehow their JSON example is atrociously slow (like, orders of magnitude, 250-500 times slower).

The JSON sample being parsed is a typical JSON data, which has 923 lines. This translates to ~700-730k lines of JSON per second, and that is actually on par with some Rust parser combinator crates like pom.