npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2024 – Pkg Stats / Ryan Hefner

nodehun-native

v3.0.6

Published

The Hunspell binding for nodejs that exposes as much of hunspell as possible and also adds new features. Hunspell is a first class spellcheck library used by Google, Apple, and Mozilla.

Downloads

273

Readme

Nodehun-Native

👋 Notice - November 2023

This is a fork of ONLYOFFICE's fork of Nodehun, that allows Nodehun to link against the actual hunspell source code. Without their great work, this wouldn't have worked. I take no credit for this, I am only maintaining the package so it will work with the latest version of Hunspell.

The primary difference between this fork and Nodehun is that the .dic and .aff files must be loaded from a file on disk, instead of via a Buffer.

Since their repo had become stale and lacked a npm package, Nodehun-Native provides feature parity with the latest Hunspell (at time of writing v1.7.2), easily installable via NPM. I've also updated the documentation and examples.

Introduction

Nodehun aims to expose as much of hunspell's functionality as possible in an easy to understand and maintainable way, while also maintaining the performance characteristics expected of a responsible node module.

Features

  • Native performance.
  • Exposes all of hunspell's functionality:
    • Spell checking,
    • suggestions,
    • personal dictionaries and word management,
    • stems/roots of words,
    • morphological generation, and,
    • word analysis.
  • TypeScript declaration file.
  • Synchronous + promise-based async API.
  • Extensive unit testing.
  • Completely re-written using N-API (thus, stability in future v8 versions)

Installation

npm install nodehun-native

If you run into any build errors, make sure you satisfy the requirements for node-gyp.

Quick Start

import { Nodehun } from 'nodehun'

const affix       = 'path/to/*.aff'
const dictionary  = 'path/to/*.dic'

const nodehun     = new Nodehun(affix, dictionary)

// Promise example
nodehun.suggest('colour')
		   .then(suggestions => { })

// async/await example
async function example() {
	const suggestions = await nodehun.suggest('colour')
}

// sync example
const suggestions = nodehun.suggestSync('colour')

Table of Contents

  1. Examples
    • Spell checking
    • Spelling suggestions
    • Adding a dictionary
    • Add a word
    • Add a word (with affix)
    • Remove a word
    • Word stem
    • Word analysis
    • Word generation
  2. Notes
    • Improving performance
    • A Warning on Synchronous Methods
    • A Note About Open Office Dictionaries
    • A Note About Creating Dictionaries
    • Where To Get Dictionaries
  3. Development and Contribution
    • Building
    • Scripts
    • Notes
    • Mentions

Examples

The following section includes short examples of various exposed operations. For complete examples, see the /examples directory.

Checking for Correctness

Nodehun offers a method that returns true or false if the passed word exists in the dictionary, i.e. is "correct".

await nodehun.spell('color') // => true
await nodehun.spell('colour') // => false, assuming en_US dictionary

Spelling Suggestions

Nodehun also offers a method that returns an array of words that could possibly match a misspelled word, ordered by most likely to be correct.

await nodehun.suggest('color')
// => null (since it's correctly spelled)

await nodehun.suggest('calor')
// => ['carol','valor','color','cal or','cal-or','caloric','calorie']

Add Dictionary

Nodehun also can add another dictionary on top of an existing dictionary object at runtime (this means it is not permanent) in order to merge two dictionaries.

const en_CA = './path/to/en_CA.dic';

await nodehun.suggest('colour') // => [ ...suggestions... ]
// because "colour" is not a defined word in the US English dictionary
await nodehun.addDictionary(en_CA)
await nodehun.suggest('colour') // => null
// (since the word is considered correctly spelled now)

Add Word

Nodehun can also add a single word to a dictionary at runtime (this means it is not permanent) in order to have a custom runtime dictionary. If you know anything about Hunspell you can also add flags to the word.

await nodehun.suggest('colour') // => [ ...suggestions...]
// because "colour" is not a defined word in the US English dictionary
await nodehun.add('colour')
await nodehun.suggest('colour') // => null
// (since 'colour' is correct now)

Note: colouring will still be considered incorrect. See the the addWithAffix example below.

Add Word (with affix)

Like the method above, except it also applies the example word's affix definition to the new word.

await nodehun.suggest('colouring') // => [ ...suggestions...]
// because "colour" is not a defined word in the US English dictionary
await nodehun.addWithAffix('colour', 'color')
await nodehun.suggest('colouring') // => null
// (since 'colouring' is correct now)

Remove Word

Nodehun can also remove a single word from a dictionary at runtime (this means it is not permanent) in order to have a custom runtime dictionary. If you know anything about Hunspell this method will ignore flags and just strip words that match.

await nodehun.suggest('color') // => null (since the word is correctly spelled)
await nodehun.remove('color')
await nodehun.suggest('color') // => ['colon', 'dolor', ...etc ]

Word Stems

Nodehun exposes the Hunspell stem function which analyzes the roots of words. Consult the Hunspell documentation for further understanding.

await nodehun.stem('telling') // => [telling, tell]

Word Analysis

Nodehun exposes the Hunspell analyze function which analyzes a word and return a morphological analysis. Consult the Hunspell documentation for further understanding.

await nodehun.analyze('telling') 
// with the appropriate dictionaries files, it will return:
// => [' st:telling ts:0', ' st:tell ts:0 al:told is:Vg']

Word Generation

Nodehun exposes the Hunspell generate function which generates a variation of a word by matching the morphological structure of another word. Consult the Hunspell documentation for further understanding.

await nodehun.generate('telling', 'ran') // => [ 'told' ]
await nodehun.generate('told', 'run') // => [ 'tell' ]

Notes

Improving Performance

If the native performance isn't fast enough for your workload, you can try using an LRU cache for your operations. The idea is to cache the results of the operation and only repeat the operations on cache misses.

const LRUCache = require('lru-native2')

var cache = new LRUCache({ maxElements: 1000 })

async function suggestCached() {
  let cachedResult = cache.get(word)
  if (cachedResult) {
    // cache hit
    return cachedResult
  } else {
    // cache miss
    let result = await nodehun.suggest(word)
    cache.set(word, result)
    return result
  }
}

// ... example usage:

const suggestions = await suggestCached('Wintre')
// now 'wintre' results are cached

// ... some time later...

const suggestions = await suggestCached('Wintre')
               // => this is fetched from the cache

Here are two LRU implementations you can consider:

A Warning on Synchronous Methods

There are synchronous versions of all the methods listed above, but they are not documented as they are only present for people who really know and understand what they are doing. I highly recommend looking at the C++ source code if you are going to use these methods in a production environment as the locks involved with them can create some counterintuitive situations. For example, if you were to remove a word synchronously while many different suggestion threads were working in the background the remove word method could take seconds to complete while it waits to take control of the read-write lock. This is obviously disastrous in a situation where you would be servicing many requests.

A Note About Open Office Dictionaries

All files must be UTF-8 to work! When you download open office dictionaries don't assume that the file is UTF-8 just because it is being served as a UTF-8 file. You may have to convert the file using the iconv unix utility (easy enough to do) to UTF-8 in order for the files to work.

A Note About Creating Dictionaries

If you want to create a new Hunspell dictionary you will need a base affix file. I recommend simply using one of the base affix files from the open office dictionaries for the language you are creating a dictionary for. Once you get around to creating a dictionary read the hunspell documentation to learn how to properly flag the words. However, my guess is that the vast majority of people creating dictionaries out there will be creating a dictionary of proper nouns. Proper nouns simply require the "M" flag. This is what a dictionary of proper nouns might look like:

Aachen/M
aardvark/SM
Aaren/M
Aarhus/M
Aarika/M
Aaron/M

Notice that the "S" flag denotes a proper noun that isn't capitalized, otherwise look in the docs.

Where To Get Dictionaries

The included dictionaries were extracted from Libre Office. The Libre Office versions have a modified aff file that makes generate() and analyze() much more useful. However, any MySpell style dictionary will work. Here are a few sources:

Also, check out @wooorm's UTF-8 dictionary collection here.

Let the community know if you've found other dictionary repositories!

Development and Contribution

Building

To build nodehun-native, take the following steps:

# Clone the repo and enter directory
gh repo clone bjarkebech/nodehun-native
cd nodehun-native

# Install npm dependencies
npm i

# Fetch the latest changes to our submodule
git submodule update --recursive --remote

# Then, check out the latest release version of hunspell (in our case, 1.7.2).
cd src/hunspell
git checkout v1.7.2

# Go back to the root of our repository
cd ../../

# Build and run our tests!
npm run start

Scripts

The following is a a list of commands and their descriptions which may help in development.

npm start: to jumpstart the development server. This will automatically recompile the c++ source when changes are made and run the tests once more.

npm run start-test: if you don't want to continuously compile the c++ source, but do want the tests to re-run when changes are made to the test files.

npm run build: to compile the addon once.

npm test: to run the tests once.

npm run performance-test: to run the performance tests and output updated graphs. (see test/performance)

Notes

Make node-gyp build faster by increasing the number of cores it uses:

export JOBS=max
npm run build # super fast now!

Mentions

Special thanks to @nathanjsweet for his grass roots efforts with this project, including the hunspell-distributed package upon which this library relies to provide buffer-based Hunspell initialization.

Huge props to @Wulf and @ONLYOFFICE for their great work on Nodehun and the fork.