sql-gen

Crates.iosql-gen
lib.rssql-gen
version0.1.5
sourcesrc
created_at2023-07-12 06:58:14.678903
updated_at2024-03-03 12:01:08.525044
descriptionA CLI tool for generating models based on a SQL Database using SQLx
homepage
repository
max_upload_size
id914329
size131,266
James H. (jayy-lmao)

documentation

README

SQL-Gen - Rust CLI Tool for PostgreSQL Database Operations Generation extending SQLX

codegen_example

PR's and Issues welcome! This is still early stages of devopment, though hopefully useful to some!

SQL-Gen is a command-line tool written in Rust that helps you generate Rust structs, queries, and SQL migrations based on an existing PostgreSQL database schema. It utilizes the sqlx and clap libraries to provide a user-friendly and efficient experience for working with your PostgreSQL database.

This project draws inspiration from rsdbgen (GitHub: brianhv/rsdbgen) and we appreciate their contribution to the Rust database tooling ecosystem. Originally this was going to be extending rsdbgen but at a certain point of changes it seemed to have diverged.

Use Cases

  • Generate Rust structs and queries for PostgreSQL database tables or from migrations via a Postgres test container
  • Generate SQL migrations based on changes in the structs vs a database or migration via a Postgres test container
  • Handle nullable/optional fields in the generated code.
  • Optional --force flag to overwrite existing files.
  • Use environment variables instead of clap flags (optional).

Installation

To use SQL-Gen, make sure you have Rust and Cargo installed on your system. You can install them by following the instructions at https://www.rust-lang.org/.

Once you have Rust and Cargo installed, you can build SQL-Gen by running the following command:

cargo install sql-gen

Or for the latest github

cargo install --git https://github.com/jayy-lmao/sql-gen --branch main

Usage

sql-gen [SUBCOMMAND] [OPTIONS]

Subcommands:

generate - Generate structs and queries for tables in your db

https://github.com/jayy-lmao/sql-gen/assets/32926722/55f3391f-47e1-42dd-b24e-6903f96971d5

migrate - Generate SQL migrations based on struct differences for structs that have a database table matching them

https://github.com/jayy-lmao/sql-gen/assets/32926722/ea3b9739-be8f-43e5-b48d-83d130ffd1c5

Options:

  • generate subcommand options:

    • -o, --models <SQLGEN_MODEL_OUTPUT_FOLDER> - Sets the output folder for generated structs (required)
    • -d, --database <DATABASE_URL> - Sets the database connection URL (required). Can be set as docker to spin up a testcontainer instance for applying migrations.
    • -c, --context <SQLGEN_CONTEXT_NAME> - The name of the context for calling functions. Defaults to DB name
    • -f, --force - Overwrites existing files sharing names in that folder
    • -m, --migrations <SQLGEN_MIGRATION_INPUT> - Sets the input folder for migrations (only if using docker as the database)
  • migrate subcommand options:

    • -o, --models <SQLGEN_MODEL_FOLDER> - Sets the folder containing existing struct files (required)
    • -m, --migrations <SQLGEN_MIGRATION_OUTPUT> - Sets the output folder for migrations (required)
    • -d, --database <DATABASE_URL> - Sets the database connection URL (required). Can be set as docker to spin up a testcontainer instance for applying migrations.

Example .env file

Create a .env file in the project root directory with the following content:

DATABASE_URL=postgres://username:password@localhost/mydatabase
SQLGEN_MODEL_OUTPUT_FOLDER=./src/models/
SQLGEN_MODEL_FOLDER=./src/models/
SQLGEN_MIGRATION_OUTPUT=./migrations
SQLGEN_MIGRATION_INPUT=./migrations

Make sure to replace the values with your actual database connection URL and desired folder paths for generated structs and migrations.

Generate Structs and Queries

To generate Rust structs and queries for a PostgreSQL database, use the generate command:

sql-gen generate --output db --database <DATABASE_URL>

Replace <DATABASE_URL> with the URL of your PostgreSQL database. The generated code will be saved in the db folder.

Example:

Assuming we have the following input database schema:

CREATE TABLE customer (
    id SERIAL PRIMARY KEY,
    created_at TIMESTAMPTZ DEFAULT NOW()
    email VARCHAR(255) UNIQUE,
);

Running SQLGen with the generate command:

sql-gen generate --output db --database postgresql://postgres:password@localhost/mydatabase

This will generate the following Rust structs and queries (based on primary key, foreign keys, and unique fields):

// in db/customer.rs

#[derive(sqlx::FromRow, Debug)]
struct Customer {
    pub id: i32,
    pub created_at: Option<chrono::DateTime<chrono::Utc>>,
    pub email: Option<String>,
}

// in db/customer_db_set.rs
use sqlx::{query, query_as, PgExecutor, Result};
use super::Customer;

pub struct CustomerSet;

impl CustomerSet {
    pub async fn all<'e, E: PgExecutor<'e>>(&self, executor: E) -> Result<Vec<Customer>> {
        query_as::<_, Customer>(r#"SELECT * FROM "customer""#)
            .fetch_all(executor)
            .await
    }

    pub async fn by_id<'e, E: PgExecutor<'e>>(&self, executor: E, id: i64) -> Result<Customer> {
        query_as::<_, Customer>(r#"SELECT * FROM "customer" WHERE "id" = $1"#)
            .bind(id)
            .fetch_one(executor)
            .await
    }

    pub async fn by_id_optional<'e, E: PgExecutor<'e>>(&self, executor: E, id: i64) -> Result<Option<Customer>> {
        query_as::<_, Customer>(r#"SELECT * FROM "customer" WHERE "id" = $1"#)
            .bind(id)
            .fetch_optional(executor)
            .await
    }

    // Doesn't exist in this example, but foreign keys will functions like this, assuming customer has a fk field called category
    // pub async fn all_by_categories_id<'e, E: PgExecutor<'e>>(executor: E, categories_id: i64) -> Result<Vec<Customer>> {
    //     query_as::<_, Customer>(r#"SELECT * FROM "customer" WHERE category = $1"#)
    //         .bind(categories_id)
    //         .fetch_all(executor)
    //         .await
    // }

    pub async fn by_email<'e, E: PgExecutor<'e>>(&self, executor: E, email: String) -> Result<Customer> {
        query_as::<_, Customer>(r#"SELECT * FROM "customer" WHERE "email" = $1"#)
            .bind(email)
            .fetch_one(executor)
            .await
    }

    pub async fn many_by_email_list<'e, E: PgExecutor<'e>>(&self, executor: E, email_list: Vec<String>) -> Result<Vec<Customer>> {
        query_as::<_, Customer>(r#"SELECT * FROM "customer" WHERE "email" = ANY($1)"#)
            .bind(email_list)
            .fetch_all(executor)
            .await
    }

    pub async fn by_email_optional<'e, E: PgExecutor<'e>>(&self, executor: E, email: String) -> Result<Option<Customer>> {
        query_as::<_, Customer>(r#"SELECT * FROM "customer" WHERE "email" = $1"#)
            .bind(email)
            .fetch_optional(executor)
            .await
    }


    pub async fn insert<'e, E: PgExecutor<'e>>(&self, executor: E, products: Customer) -> Result<Customer> {
        query_as::<_, Customer>(r#"INSERT INTO "customer" ("id", "created_at", "email", "category") VALUES ($1, $2, $3, $4) RETURNING *;"#)
            .bind(products.id)
            .bind(products.created_at)
            .bind(products.email)
            .fetch_one(executor)
            .await
    }

    pub async fn update<'e, E: PgExecutor<'e>>(&self, executor: E, products: Customer) -> Result<Customer> {
        query_as::<_, Customer>(r#"UPDATE "customer" SET "created_at" = $2, "email" = $3 WHERE "id" = 1 RETURNING *;"#)
            .bind(products.id)
            .bind(products.created_at)
            .bind(products.email)
            .fetch_one(executor)
            .await
    }

    pub async fn delete<'e, E: PgExecutor<'e>>(&self, executor: E) -> Result<()> {
        query(r#"DELETE FROM "customer" WHERE "id" = 1"#)
            .execute(executor)
            .await
            .map(|_| ())
    }

}


// in db/mod.rs
pub mod customer;
pub use customer::Customer;
pub mod customer_db_set;
pub use customer_db_set::CustomerSet;

pub struct PostgresContext;

impl PostgresContext {
  pub fn customer(&self) -> CustomerSet { CustomerSet }

}

The name of the context will default to the name of your database, or can be set with the '--context' flag. These queries may need modifying or changing, but they can serve as a good start. You should be able to run commands like:

let customers = PostgresContext.customer().all(&pool).await?;

The suggested way to add customer queries etc would be to add them somewhere like db/customer_custom_queries.rs so that they are not overwritten by codgen. If you impl CustomerSet and add functions it should extend it.

Generate Migrations

To generate SQL migrations based on changes in the structs, use the migrate generate command:

sql-gen migrate generate --database <DATABASE_URL> --include <FOLDER_PATH> --output migrations

Replace <DATABASE_URL> with the URL of your PostgreSQL database, <FOLDER_PATH> with the folder containing the generated structs (db in the previous example), and migrations with the output folder for the SQL migrations.

Example:

Running SQLGen with the migrate generate command:

sql-gen migrate generate --database postgresql://postgres:password@localhost/mydatabase --include db --output migrations

This will perform a dry run of the previous database generation, compare it with the existing structs in the db folder, and generate SQL migrations for any detected changes. The migrations will be saved in the migrations folder.

Example Migration:

Assuming a change is made to the Customer struct, adding a new field:

pub struct Customer {
    pub id: i32,
    pub created_at: Option<chrono::DateTime<chrono::Utc>>,
    pub email: Option<String>,
    pub address: Option<String>, // New field
}

Running SQL-Gen with the migrate generate command will generate the following migration:

-- Migration generated for struct: Customer
ALTER TABLE customer ADD COLUMN address TEXT;

For a complete list of available commands and options, you can use the --help flag:

sql-gen --help

Roadmap

SQL-Gen is under active development, and future enhancements are planned. Here are some items on the roadmap:

Types

  • Arrays / Vectors
  • Enums
  • Composite types
  • Decimal

Other

  • Singularise/Pluralise correctly with inflection crate (e.g. your table is called customers, maybe you want the struct to still be Customer)
  • A way to customise generated derives
  • Clearer input types that allow for leaving out default fields
  • Tests
  • Linting on output (rustfmt has proven tricky with derive macros)
  • Support for comments and annotations
  • Support for more data types in struct and query generation.
  • Integration with other database systems (MySQL, SQLite, etc.).
  • Advanced migration scenarios (renaming columns, table-level changes, etc.).
  • Dry run mode for generating migration code without writing to files.

Your contributions and feedback are highly appreciated! If you encounter any issues or have suggestions

Commit count: 0

cargo fmt