CRStore

Conflict-free replicated store.

WARNING: Still in development! Expect breaking changes!

BREAKING (v0.20.0): Added support for React & Solid. For Svelte import database from crstore/svelte. Renamed store to replicated.

BREAKING (v0.19.0): Updated cr-sqlite from v13 to v16. See changelog

BREAKING (v0.18.0): If you want to support older browsers consider adding navigator.locks polyfill to your project. CRStore does not ship it since 0.18.0!

• ✨ Elegance of Svelte / SolidJS / React
• 💪 Power of SQLite
• 🛡️ Safety with Kysely
• ⚡ CRDTs powered by cr-sqlite
• 🔮 Automagical schema using superstruct
• 🤝 First class tRPC support
• 🐇 Supports bun:sqlite (experimental)

Install crstore and superstruct (for automatic schema):

npm install crstore superstruct

Using CRStore

To start using CRStore first you need to define a schema for your database. This is like a Kysely schema, but defined with superstruct, so we can have a runtime access to it.

import { crr, primary } from "crstore";

// Struct that represents the table
const todos = object({
  id: string(),
  title: string(),
  text: string(),
  completed: boolean(),
});
crr(todos); // Register table with conflict-free replicated relations
primary(todos, "id"); // Define a primary key (can be multi-column)

const schema = object({ todos });

Now you can establish a database connection with your schema:

import { database } from "crstore/svelte";

const { replicated } = database(schema);

Note, that this example uses Svelte version (replicated). For React database function will return useReplica and createReplica for SolidJS. Learn more how to use CRStore with these frameworks here.

With the replicated function we can create arbitrary views to our database which are valid svelte stores. For example let's create a store that will have our entire todos table:

const todos = replicated((db) => db.selectFrom("todos").selectAll());

To mutate the data we can either call .update on the store or add built-in actions upon creation:

const todos = replicated((db) => db.selectFrom("todos").selectAll(), {
  // Define actions for your store
  toggle(db, id: string) {
    return db
      .updateTable("todos")
      .set({ completed: sql`NOT(completed)` })
      .where("id", "=", id)
      .execute();
  },
  remove(db, id: string) {
    return db.deleteFrom("todos").where("id", "=", id).execute();
  },
});

// Call an update manually
todos.update((db) => db.insertInto("todos").values({ ... }).execute());
// Call an action
todos.toggle("id");

We can simple iterate the store to render the results:

Note that the database loads asynchronously, so the store will contain an empty array util it loads.

{#each $todos as todo}
  <h2>{todo.title}</h2>
  <p>{todo.text}</p>
{/each}

This we dynamically react to all the changes in our database even if we make them from a different store. Each store we create reacts only to changes in tables we have selected from.

Connecting with tRPC

You can provide custom handlers for your network layer upon initialization. push method is called when you make changes locally that need to be synchronized. pull is called when crstore wants to subscribe to any changes coming from the network. Let's say you have a push tRPC mutation and a pull tRPC subscription then you can use them like so when connection to a database:

const { replicated } = database(schema, {
  push: trpc.push.mutate,
  pull: trpc.pull.subscribe,
});

Then your server implementation would look something like this:

import { database } from "crstore";

const { subscribe, merge } = database(schema);
const { router, procedure } = initTRPC.create();

const app = router({
  push: procedure.input(any()).mutation(({ input }) => merge(input)),
  pull: procedure
    .input(object({ version: number(), client: string() }))
    .subscription(({ input }) =>
      observable(({ next }) => subscribe(["*"], next, input))
    ),
});

If you are using vite-node to run your server, you should add define: { "import.meta.env.SSR": false } to your vite config file.

Advanced Usage

Depend on other stores

When creating a crstore you might want it to subscribe to some other stores. For example you can have a writable query store and a search crstore. Where search updates every time query updates. To do so you can use .with(...stores) syntax when creating a store. All the resolved dependencies will be passed to your SELECT callback.

import { database } from "crstore/svelte";
import { writable } from "svelte/store";

const { replicated } = database(schema);

const query = writable("hey");
const search = replicated.with(query)((db, query) => 
  db.selectFrom("todos").where("text", "=", query).selectAll()
);

Specify custom paths

If needed you can specify custom paths to better-sqlite3 binding, crsqlite extension and crsqlite-wasm binary. To do so, provide path option upon database initialization:

import { database } from "crstore/svelte";

const { replicated } = database(schema, {
  // These are the default values:
  paths: {
    wasm: "/sqlite.wasm",
    extension: "node_modules/@vlcn.io/crsqlite/build/Release/crsqlite.node",
    binding: undefined,
  }
});

Specify database name

If you need to manage multiple databases you can specify name database option. This will be used as a filename on a server or a VFS path on a client.

import { database } from "crstore/svelte";

const { replicated } = database(schema, {
  name: "data/example.db"
});

Specify a custom online checker

push and pull capabilities rely on checking current online status. When available navigator.onLine is used by default. You have an option to override it by providing a custom online function.

import { database } from "crstore/svelte";

const { replicated } = database(schema, {
  online: () => true // Always online
});

Note that this is only really needed if you use pull and push helpers. If your server implementation uses subscribe and merge methods instead, the online checker is unnecessary (defaults to false).

Apply updates without creating a store

Use can apply any updates right after you have initialized your database connection by using the update function. If there are any stores initialized, they will also be updated if you change any tables they depend on.

import { database } from "crstore";

const { update } = database(schema);
update((db) => db.insertInto("todos").values({ ... }));

Access raw database connection

Use can access the raw database connection. This can sometime be useful for debugging. Note that any mutations you do directly from the connection will not trigger any reactive updates! To mutate data safely please use the update function instead.

import { database } from "crstore";

const { connection } = database(schema);
const db = await connection;

const data = await db.selectFrom("todos").selectAll().execute()
console.log(data);

Nested JSON queries

crstore provides support for nested JSON queries via it's own JSON Kysely plugin. You can see how it's used in practice be looking at the library demo.

import { groupJSON } from "crstore";

const grouped = replicated((db) =>
  db
    .selectFrom("tracks")
    .leftJoin("artists", "tracks.artist", "artists.id")
    .leftJoin("albums", "tracks.album", "albums.id")
    .select([
      "albums.title as album",
      (qb) =>
        // Here we aggregate all the tracks for the album
        groupJSON(qb, {
          id: "tracks.id",
          title: "tracks.title",
          artist: "artists.title",
          album: "albums.title",
        }).as("tracks"),
    ])
    // `groupBy` is essential for the aggregation to work
    .groupBy("album")
);

$grouped[0] // ↓ The type is inferred from `json`
// {
//   album: string | null;
//   tracks: {
//     id: string;
//     title: string;
//     artist: string | null;
//     album: string | null;
//   }[]
// }

Specify indexes in the schema

You can specify one or more indexes for your tables.

import { index } from "crstore";

const todos = object({
  id: string(),
  title: string(),
  text: string(),
  completed: boolean(),
});
index(todos, "title");
index(todos, "text", "completed"); // Multi-column index

Define a fractional index for a table

cr-sqlite supports conflict free fractional indexing. To use them in CRStore first you should define table as ordered in your schema:

import { ordered } from "crstore";

const todos = object({
  id: string(),
  text: string(),
  completed: boolean(),
  collection: string(),
  order: string()
});
// Sort by 'order' column in each 'collection'
ordered(todos, "order", "collection");

Then you can append or prepend items by putting the exported constants as your order value.

import { APPEND, PREPEND } from "crstore";

db.insertInto("todos")
  .values({
    id: "4321",
    text: "Hello",
    completed: false,
    collection: "1234",
    order: APPEND,
  })
  .execute();

To move an item you should update the {you_table}_fractindex virtual table with the after_id value.

db
  .updateTable("todos_fractindex" as any)
  .set({ after_id: "2345" })
  .where("id", "=", "4321")
  .execute();

Check out the sortable example for more details.

Setup server side rendering

When defining your database set ssr option to true:

const { replicated, merge, subscribe } = database(schema, {
  ssr: true,
});

Add +page.server.ts file to preload your data with SvelteKit. You can call .then on a store to get a promise with its latest state (the await keyword would achieve the same effect). Pass down the value of your store to your page like this:

import type { PageServerLoad } from "./$types";
import { items } from "./stores";

export const load: PageServerLoad = async () => ({ ssr: await items });

In your +page.svelte you render the server-side data until client database is ready.

<script lang="ts">
  import type { PageData } from "./$types";
  import { items } from "./stores";
  import { ready } from "$lib";

  export let data: PageData;
</script>

{#each ready($items) ? $items : data.ssr as item}
  <li>{item.data}</li>
{/each}

Check out the ssr example for complete implementation.

Error handling

You can add an error handler to your database connection.

const { replicated } = database(schema, {
  error: (reason) => console.log(reason),
});

It will handle all the errors that happen during subscriber callbacks.