Crate diesel_derive_newtype [] [src]

#[derive(DieselNewType)]

This crate exposes a single custom-derive macro DieselNewType which implements ToSql, FromSql, FromSqlRow, Queryable, AsExpression and QueryId for the single-field tuple struct (NewType) it is applied to.

The goal of this project is that:

Example

This implementation:

#[macro_use]
extern crate diesel_derive_newtype;

#[derive(DieselNewType)] // Doesn't need to be on its own line
#[derive(Debug, Hash, PartialEq, Eq)] // required by diesel
struct MyId(i64);

Allows you to use the MyId struct inside your entities as though they were the underlying type:

table! {
    my_items {
        id -> Integer,
        val -> Integer,
    }
}

#[derive(Debug, PartialEq, Identifiable, Queryable, Associations)]
struct MyItem {
    id: MyId,
    val: u8,
}

Oooohhh. Ahhhh.

See the tests for a more complete example.

Limitations

The DieselNewtype derive does not create new database types, or Diesel serialization types. That is, if you have a MyId(i64), this will use Diesel's underlying BigInt type, which means that even though your newtypes can be used anywhere the underlying type can be used, the underlying types, or any other newtypes of the same underlying type, can be used as well.

At a certain point everything does become bits on the wire, so if we didn't do it this way then Diesel would have to do it somewhere else, and this is reasonable default behavior (it's pretty debuggable), but I'm investigating auto-generating new proxy types as well to make it impossible to construct an insert statement using a tuple or a mis-typed struct.

Here's an example of that this type-hole looks like:

Be careful when using this code, it's not being tested!
#[derive(Debug, Hash, PartialEq, Eq, DieselNewType)]
struct OneId(i64);

#[derive(Debug, Hash, PartialEq, Eq, DieselNewType)]
struct OtherId(i64);

#[derive(Debug, Clone, PartialEq, Identifiable, Insertable, Queryable)]
#[table_name="my_entities"]
pub struct MyEntity {
    id: OneId,
    val: i32,
}

fn darn(conn: &Connection) {
    // shouldn't allow constructing the wrong type, but does
    let OtherId: Vec<OtherId> = my_entities
        .select(id)
        .filter(id.eq(OtherId(1)))  // shouldn't allow filtering by wrong type
        .execute(conn).unwrap();
}

See tests/should-not-compile.rs for the things I think should fail to compile.

I believe that the root cause of this is that Diesel implements the various expression methods for types that implement AsExpression, based on the SQL type, not caring about self and other's Rust type matching. That seems like a pretty good decision in general, but it is a bit unfortunate here.

I hope to find a solution that doesn't involve implementing every *Expression trait manually with an extra bound, but for now you have to keep in mind that the Diesel methods basically auto-transmute your data into the underlying SQL type.