Butane

An experimental ORM for Rust with a focus on simplicity and on writing Rust, not SQL

Butane takes an object-oriented approach to database operations. It may be thought of as much as an object-persistence system as an ORM -- the fact that it is backed by a SQL database is mostly an implementation detail to the API consumer.

Features

• Relational queries using Rust-like syntax (via proc-macros)
• Automatic migrations without writing SQL (although the generated SQL may be hand-tuned if necessary)
• Ability to embed migrations in Rust code (so that a library may easily bundle its migrations)
• SQLite and PostgreSQL backends
• Write entirely or nearly entirely the same code regardless of database backend

Getting Started

Models, declared with struct attributes define the database schema. For example the Post model for a blog might look like this:

#[model]
#[derive(Default)]
struct Post {
    id: AutoPk<i32>,
    title: String,
    body: String,
    published: bool,
    likes: i32,
    tags: Many<Tag>,
    blog: ForeignKey<Blog>,
    byline: Option<String>,
}

An object is an instance of a model. An object is created like a normal struct instance, but must be saved in order to be persisted.

let mut post = Post::new(blog, title, body);
post.save(conn)?;

Changes to the instance are only applied to the database when saved:

post.published = true;
post.save(conn)?;

Queries are performed ergonomically with the query! macro.

let posts = query!(Post, published == true).limit(5).load(&conn)?;

For a detailed tutorial, see the Getting Started Guide.

Cargo Features

Butane exposes several features to Cargo. By default, no backends are enabled: you will want to enable sqlite and/or pg:

• default: Turns on datetime, json and uuid.
• async: Turns on async support. This is automatically enabled for the pg backend, which is implemented on the tokio-postgres crate.
• async-adapter: Enables the use of async with the sqlite backend, which is not natively async.
• debug: Used in developing Butane, not expected to be enabled by consumers.
• deadpool: Connection pooling using deadpool.
• datetime: Support for timestamps (using chrono crate).
• fake: Support for the fake crate's generation of fake data.
• json: Support for storing structs as JSON, including using postgres' JSONB field type.
• log: Log certain warnings to the log crate facade (target "butane").
• pg: Support for PostgreSQL using postgres crate.
• r2d2: Connection pooling using r2d2. (See butane::db::ConnectionManager).
• sqlite: Support for SQLite using rusqlite crate.
• sqlite-bundled: Bundles sqlite instead of using the system version.
• tls: Support for TLS when using PostgreSQL, using postgres-native-tls crate.
• uuid: Support for UUIDs (using the uuid crate).

Limitations

• Butane, and its migration system especially, expects to own the database. It can be used with an existing database accessed also by other consumers, but it is not a design goal and there is no facility to infer butane models from an existing database schema.
• API ergonomics are prioritized above performance. This does not mean Butane is slow, but that when given a choice between a simple, straightforward API and eking out the smallest possible overhead, the API will win.

Migration of Breaking Changes

0.8 (not yet released)

Async

This is a major release which adds Async support. Effort has been made to keep the sync experience as unchanged as possible. Async versions of many types have been added, but the sync ones generally retain their previous names.

In order to allow sync and async code to look as similar as possible for types and traits which do not otherwise need separate sync and async variants, several "Ops" traits have been introduced which contain methods split off from prior types and traits.

For example, if obj is an instance of DataObject, then you may call obj.save(conn) (sync) or obj.save(conn).await (async). The save method no longer lives on DataObject. Instead, you must use either butane::DataObjectOpsSync or butane::DataObjectOpsAsync. Which trait is in scope will determine whether the save method is sync or async.

The Ops traits are:

• DataObjectOpsSync / DataObjectOpsAsync (for use with DataObject)
• QueryOpsSync / QueryOpsSync (for use with Query, less commonly needed directly if you use the query or filter macros)
• ForeignKeyOpsSync / ForeignKeyOpsAsync (for use with ForeignKey)
• ManyOpsSync / ManyOpsAsync (for use with Many)

ConnectionManager

The ConnectionManager struct has moved from butane::db::r2 to butane::db. It no longer implements ConnectionMethods as this was unnecessary due to Deref. The butane::db::r2 module is no longer public.

0.7

AutoPk

Replace model fields like

#[auto]
pub id: i64

with

pub id: AutoPk<i64>

ObjectState is removed

Remove any references to ObjectState or to the (previously automatically generated) state field on models.

Roadmap

Butane is young. The following features are currently missing, but planned

• Foreign key constraint cascade setting
• Incremental object save
• Back-references for ForeignKey and Many.
• Field/column rename support in migrations
• Prepared/reusable queries
• Benchmarking and performance tuning
• Support for other databases such as MySQL or SQL Server are not explicitly planned, but contributions are welcome.

Comparison to Diesel

Butane is inspired by Diesel and by Django's ORM. If you're looking for a mature, performant, and flexible ORM, go use Diesel. Butane doesn't aim to be better than Diesel, but makes some different decisions, including:

1. It is more object-oriented, at the cost of flexibility.

2. Automatic migrations are prioritized.

3. Rust code is the source of truth. The schema is understood from the definition of Models in Rust code, rather than inferred from the database.

4. Queries are constructed using a DSL inside a proc-macro invocation rather than by importing DSL methods/names to use into the current scope. For Diesel, you might write

   use diesel_demo::schema::posts::dsl::*;
   let posts = posts.filter(published.eq(true))
        .limit(5)
        .load::<Post>(&conn)?

   whereas for Butane, you would instead write

   let posts = query!(Post, published == true).limit(5).load(&conn)?;

   Which form is preferable is primarily an aesthetic judgement.

5. Differences between database backends are largely hidden.

6. Diesel is overall significantly more mature and full-featured.

For a detailed tutorial, see the getting started guide.

License

Butane is licensed under either of the MIT license or the Apache License, Version 2.0 at your option.

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in Butane by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.