Skip to main content

TypedSQL

Getting started with TypedSQL​

To start using TypedSQL in your Prisma project, follow these steps:

  1. Ensure you have @prisma/client and prisma installed and updated to at least version 5.19.0.

    npm install @prisma/client@latest
    npm install -D prisma@latest

  2. Add the typedSql preview feature flag to your schema.prisma file:

    generator client {
      provider = "prisma-client-js"
      previewFeatures = ["typedSql"]
    }

  3. Create a sql directory inside your prisma directory. This is where you'll write your SQL queries.

    mkdir -p prisma/sql

  4. Create a new .sql file in your prisma/sql directory. For example, getUsersWithPosts.sql. Note that the file name must be a valid JS identifier and cannot start with a $.

  5. Write your SQL queries in your new .sql file. For example:

    prisma/sql/getUsersWithPosts.sql
    SELECT u.id, u.name, COUNT(p.id) as "postCount"
    FROM "User" u
    LEFT JOIN "Post" p ON u.id = p."authorId"
    GROUP BY u.id, u.name

  6. Generate Prisma Client with the sql flag to ensure TypeScript functions and types for your SQL queries are created:

    warning

    Make sure that any pending migrations are applied before generating the client with the sql flag.

    prisma generate --sql

    If you don't want to regenerate the client after every change, this command also works with the existing --watch flag:

    prisma generate --sql --watch

  7. Now you can import and use your SQL queries in your TypeScript code:

    /src/index.ts
    import { PrismaClient } from '@prisma/client'
    import { getUsersWithPosts } from '@prisma/client/sql'

    const prisma = new PrismaClient()

    const usersWithPostCounts = await prisma.$queryRawTyped(getUsersWithPosts())
    console.log(usersWithPostCounts)

Passing Arguments to TypedSQL Queries​

To pass arguments to your TypedSQL queries, you can use parameterized queries. This allows you to write flexible and reusable SQL statements while maintaining type safety. Here's how to do it:

  1. In your SQL file, use placeholders for the parameters you want to pass. The syntax for placeholders depends on your database engine:

    For PostgreSQL, use the positional placeholders $1, $2, etc.:

    prisma/sql/getUsersByAge.sql
    SELECT id, name, age
    FROM users
    WHERE age > $1 AND age < $2
    note

    See below for information on how to define argument types in your SQL files.

  2. When using the generated function in your TypeScript code, pass the arguments as additional parameters to $queryRawTyped:

    /src/index.ts
    import { PrismaClient } from '@prisma/client'
    import { getUsersByAge } from '@prisma/client/sql'

    const prisma = new PrismaClient()

    const minAge = 18
    const maxAge = 30
    const users = await prisma.$queryRawTyped(getUsersByAge(minAge, maxAge))
    console.log(users)

By using parameterized queries, you ensure type safety and protect against SQL injection vulnerabilities. The TypedSQL generator will create the appropriate TypeScript types for the parameters based on your SQL query, providing full type checking for both the query results and the input parameters.

Passing array arguments to TypedSQL​

TypedSQL supports passing arrays as arguments for PostgreSQL. Use PostgreSQL's ANY operator with an array parameter.

prisma/sql/getUsersByIds.sql
SELECT id, name, email
FROM users
WHERE id = ANY($1)
/src/index.ts
import { PrismaClient } from '@prisma/client'
import { getUsersByIds } from '@prisma/client/sql'

const prisma = new PrismaClient()

const userIds = [1, 2, 3]
const users = await prisma.$queryRawTyped(getUsersByIds(userIds))
console.log(users)

TypedSQL will generate the appropriate TypeScript types for the array parameter, ensuring type safety for both the input and the query results.

note

When passing array arguments, be mindful of the maximum number of placeholders your database supports in a single query. For very large arrays, you may need to split the query into multiple smaller queries.

Defining argument types in your SQL files​

Argument typing in TypedSQL is accomplished via specific comments in your SQL files. These comments are of the form:

-- @param {Type} $N:alias optional description

Where Type is a valid database type, N is the position of the argument in the query, and alias is an optional alias for the argument that is used in the TypeScript type.

As an example, if you needed to type a single string argument with the alias name and the description "The name of the user", you would add the following comment to your SQL file:

-- @param {String} $1:name The name of the user

Currently accepted types are Int, BigInt, Float, Boolean, String, DateTime, Json, Bytes, and Decimal.

Taking the example from above, the SQL file would look like this:

-- @param {Int} $1:minAge
-- @param {Int} $2:maxAge
SELECT id, name, age
FROM users
WHERE age > $1 AND age < $2

The format of argument type definitions is the same regardless of the database engine.

note

Manually argument type definitions are not supported for array arguments. For these arguments, you will need to rely on the type inference provided by TypedSQL.

Examples​

For practical examples of how to use TypedSQL in various scenarios, please refer to the Prisma Examples repo. This repo contains a collection of ready-to-run Prisma example projects that demonstrate best practices and common use cases, including TypedSQL implementations.

Limitations of TypedSQL​

Supported Databases​

TypedSQL supports modern versions of MySQL and PostgreSQL without any further configuration. For MySQL versions older than 8.0 and all SQLite versions, you will need to manually describe argument types in your SQL files. The types of inputs are inferred in all supported versions of PostgreSQL and MySQL 8.0 and later.

TypedSQL does not work with MongoDB, as it is specifically designed for SQL databases.

Active Database Connection Required​

TypedSQL requires an active database connection to function properly. This means you need to have a running database instance that Prisma can connect to when generating the client with the --sql flag. If a directUrl is provided in your Prisma configuration, TypedSQL will use that for the connection.

Dynamic SQL Queries with Dynamic Columns​

TypedSQL does not natively support constructing SQL queries with dynamically added columns. When you need to create a query where the columns are determined at runtime, you must use the $queryRaw and $executeRaw methods. These methods allow for the execution of raw SQL, which can include dynamic column selections.

Example of a query using dynamic column selection:

const columns = 'name, email, age'; // Columns determined at runtime
const result = await prisma.$queryRawUnsafe(
  `SELECT ${columns} FROM Users WHERE active = true`
);

In this example, the columns to be selected are defined dynamically and included in the SQL query. While this approach provides flexibility, it requires careful attention to security, particularly to avoid SQL injection vulnerabilities. Additionally, using raw SQL queries means foregoing the type-safety and DX of TypedSQL.

Acknowledgements​

This feature was heavily inspired by PgTyped and SQLx. Additionally, SQLite parsing is handled by SQLx.