12 unstable releases (3 breaking)

0.7.1 Oct 25, 2023
0.7.0 Oct 16, 2023
0.6.3 Oct 16, 2023
0.5.7 Oct 12, 2023
0.0.0 Oct 9, 2023

#1270 in Procedural macros

36 downloads per month
Used in exemplar

MIT/Apache

17KB
433 lines

Exemplar

A boilerplate eliminator for rusqlite.

Getting Started

A taste of what you can do:

#[derive(Debug, PartialEq, Model)]
#[table("users")]
#[check("../tests/schema.sql")]
struct User {
   username: String,
   #[bind(bind_path)]
   #[extr(extr_path)]
   home_dir: PathBuf,
   #[column("pwd")]
   password: Vec<u8>,
}
 
fn main() -> Result<()> {
    let conn = Connection::open_in_memory()?;
 
    conn.execute_batch(
        include_str!("../tests/schema.sql")
    )?;
 
    let alice = User {
        username: "Alice".to_owned(),
        home_dir: "/var/home/alice".into(),
        password: b"hunter2".to_vec()
    };
 
    let bob = User {
        username: "Bob".to_owned(),
        home_dir: "/var/home/robert".into(),
        password: b"password".to_vec()
    };
 
    alice.insert(&conn)?;
    bob.insert(&conn)?;
 
    let mut stmt = conn.prepare("
        SELECT * FROM users ORDER BY username ASC
    ")?;
    
    let mut iter = stmt.query_and_then([], User::from_row)?;
 
    assert_eq!(alice, iter.next().unwrap()?);
    assert_eq!(bob, iter.next().unwrap()?);
 
    Ok(())
}

Exemplar is based around the Model trait, which has its own derive macro.

  • See the aformentioned macro's documentation to get started.
  • For handling enums in models, check out the sql_enum macro.
  • For working with "anonymous" record types, look at the record macro.

Features

  • Works with raw SQL, not against it.
  • Thin, zero-cost API.
    • Most of Exemplar revolves around the Model trait, which gets inlined and monomorphized away before runtime. The resulting code is roughly what you'd write by hand when using pure rusqlite.
    • Designed to be drop-in; reuses rusqlite's existing types where possible, including its Result type alias.
    • Supports any type that Deref's to rusqlite::Connection, such as transactions or pooled connections.
  • Optional test derivation for guarding against drift between your database schema and Rust model types.
  • Macros for working with SQL-compatible enums and "anonymous" record types that map to ad-hoc queries.
  • Some ability to reflect on/work with dyn Models at runtime.

If you just need to CRUD some Rust data with sqlite and don't want a whole ORM or enterprise-grade DBMS, then Exemplar is for you!

FAQ

"What does Exemplar not do?"

A few key things:

  • Schema generation and management. Exemplar is explicitly not an ORM, and it's difficult to represent concepts like foreign keys and migrations without falling into ORM territory.
    • If this is a "must" for you, check out diesel or sqlx/seaorm, which both support SQLite.
  • Query generation (excluding INSERT.)
  • Interface portability. Only rusqlite is supported.

"Is it blazing fast?"

Yes. On my machine (according to these benchmarks) Exemplar can:

  • Insert a non-trivial model type in ~600 nanoseconds (1.6 million rows/sec)
  • Query and reconstruct the same type in ~9 microseconds (111,000 rows/sec, using SELECT * LIMIT 1)

Obviously the credit for this speed goes to the SQLite and rusqlite developers, but I can confidently say that I didn't slow things down!

"How does this compare to serde-rusqlite?"

serde_rusqlite is a clever hack, but it still involved too much contorting and boilerplate for my taste - that's why I created Exemplar.

The pain points I tried to fix were:

  • Needing to allocate and juggle a slice of String column names to efficiently deserialize rows - probably due to serde limitations?
    • Exemplar statically knows what columns to expect, so from_row requires no extra inputs and makes no superfluous allocations.
  • Odd design choices for field-less enums - they are inefficiently serialized as TEXT instead of INTEGER. This was nice for debugging, but I figured the faster option should be Exemplar's default.
  • to_params_named(&row1).unwrap().to_slice().as_slice() doesn't quite roll off the tongue (although this is likely serde weirdness showing up again.)
    • Equivalent to row1.insert(&conn) or row1.insert_with(&stmt) in Exemplar.
  • General serde overhead popping up, both at compile and runtime.
    • Benchmarking shows that serde_rusqlite is ~25% slower on insert operations compared to Exemplar.
    • Retrieval operations are equally fast, likely because the final conversion step is nothing compared to query calculation and I/O.

Acknowledgements

  • rusqlite, for providing the foundation on which this library is built.
  • David Tolnay, for his various proc macro incantations crates.

Dependencies

~310–770KB
~18K SLoC