hypermerge

Hypermerge

Warning Hypermerge is deprecated. This library is no longer maintained and uses an ancient and slow version of Automerge. We strongly recommend you adopt https://github.com/automerge/automerge-repo instead.

Hypermerge is a Node.js library for building p2p collaborative applications without any server infrastructure. It combines Automerge, a CRDT, with hypercore, a distributed append-only log.

This project provides a way to have apps data sets that are conflict free and offline first (thanks to CRDT's) and serverless (thanks to hypercore/DAT).

While the DAT community has done a lot of work to secure their tool set, zero effort has been made with hypermerge itself to deal with security and privacy concerns. Due to the secure nature of the tools its built upon a properly audited and secure version of this library would be possible in the future.

How it works

Hypermerge stores an automerge document as a set of separate Hypercores, one per actor ID. Each actor ID in the document is the discovery key of a Hypercore, which allows recursive lookup.

Critique

This strategy works for a small number of documents, but is very slow to synchronize due to the vast number of Hypercores required for a large collection. It also doesn't work well with the new Automerge file format which consolidates sequential changes during storage: each change has its own Hypercore entry with all the attendant metadata. To put this in context, a single keystroke in a textfield will result in hundreds of bytes of data if you use Hypermerge.

Examples

There are several example repos in the /examples directory, including a very simple two-repo code demo and a simple CLI-based chat application.

The best demonstration of Hypermerge is PushPin, which shows Hypermerge in "full flight", including taking advantage of splitting the fast, simple front-end from the more expensive, slower back-end.

Concepts

The base object you make with hypermerge is a Repo. A repo is responsible for managing documents and replicating to peers.

Basic Setup (Serverless or with a Server)

import { Repo } from 'hypermerge'
import Hyperswarm from 'hyperswarm'

const path = '.data'

const repo = new Repo({ path })

repo.setSwarm(Hyperswarm())

Create / Edit / View / Delete a document

…

NOTE: If you're familiar with Automerge: the change function in Hypermerge is asynchronous, while the Automerge.change function is synchronous. What this means is that although Automerge.change returns an object representing the new state of your document, repo.change (or handle.change) does NOT. So:

// ok in Automerge!
doc1 = Automerge.change(doc1, 'msg', (doc) => {
  doc.foo = 'bar'
})

// NOT ok in Hypermerge!
doc1 = repo.change(url1, (doc) => {
  doc.foo = 'bar'
})

Instead, you should expect to get document state updates via repo.watch (or handle.subscribe) as shown in the example above.

Two repos on different machines

…

Accessing Files

Hypermerge supports a special kind of core called a hyperfile. Hyperfiles are unchanging, written-once hypercores that store binary data.

Here's a simple example of reading and writing files.

// Write an hyperfile
const fileStream = fs.createReadStream('image.png')
const { url } = await repo.files.write(fileStream, 'image/png')

// Read an hyperfile
const fileStream = fs.createWriteStream('image.png')
const hyperfileStream = await repo.files.read(url)

hyperfileStream.pipe(fileStream)

Note that hyperfiles are conveniently well-suited to treatment as a native protocol for Electron applications. This allows you to refer to them throughout your application directly as though they were regular files for images and other assets without any special treatment. Here's how to register that:

protocol.registerStreamProtocol(
  'hyperfile',
  (request, callback) => {
    try {
      const stream = await repo.files.read(request.url)
      callback(stream)
    } catch (e) {
      log(e)
    }
  },
  (error) => {
    if (error) {
      log('Failed to register protocol')
    }
  }
)

Splitting the Front-end and Back-end

Both Hypermerge and Automerge supports separating the front-end (where materialized documents live and changes are made) from the backend (where CRDT computations are handled as well as networking and compression.) This is useful for maintaining application performance by moving expensive computation operations off of the render thread to another location where they don't block user input.

The communication between front-end and back-end is all done via simple Javascript objects and can be serialized/deserialized through JSON if required.

…

Note: each back-end only supports a single front-end today.

Related libraries