pglite-adapter

@pgpmjs/pglite-adapter

A PGlite driver for pgpm. It registers an in-process PGlite instance (WASM Postgres) as the pg-cache pool factory, so the unmodified pgpm engine deploys, verifies, and reverts migrations against PGlite with no Postgres server and no socket.

All @electric-sql/pglite* dependencies live here — @pgpmjs/core, pg-cache, and pgsql-test never import them.

How it works

pgpm's engine only needs a node-pg-shaped pool (query / connect / end). Step 1 added a driver seam in pg-cache (registerPgPoolFactory). This package implements that seam against PGlite's JS API:

• parameterized queries → db.query (extended protocol)
• parameterless / multi-statement SQL (pgpm's bootstrap) → db.exec

Because PGlite is a single in-process session, both pool.query and the client returned by connect() share one session — so pgpm's transactional deploy works without the socket PoC's useTransaction: false workaround.

Usage

import { PgpmMigrate } from '@pgpmjs/core';
import { registerPglite } from '@pgpmjs/pglite-adapter';

const { db, close } = await registerPglite();

const migrate = new PgpmMigrate({ database: 'postgres' /* ...pg config */ });
await migrate.deploy({ modulePath: './my-module' });
await migrate.verify({ modulePath: './my-module' });

await close(); // restores the previous factory and closes PGlite

Extensions

pgpm's cleanSql strips CREATE EXTENSION from migrations, so extensions are provisioned out-of-band — register the WASM extension at construction and run CREATE EXTENSION at bootstrap:

import { vector } from '@electric-sql/pglite-pgvector';

const { db, close } = await registerPglite({
  extensions: { vector },
  extensionSql: ['CREATE EXTENSION IF NOT EXISTS vector;']
});