We are heavily inspired by ActiveRecord, Eloquent and TypeORM.
- Intuitive and ergonomic
API should state the intention clearly. Provide syntax sugar for common things.
- Fast(er) compilation
Balance between compile-time checking and compilation speed.
- Avoid 'symbol soup'
Avoid macros with DSL, use derive macros where appropriate. Be friendly with IDE tools.
After some bitterness we realized it is not possible to capture everything compile time. But we don't want to encounter problems at run time either. The solution is to perform checking at 'test time' to uncover problems. These checks will be removed at production so there will be no run time penalty.
Consider the following method:
fn left_join<E>(self) -> Self
where
E: EntityTrait,
{
// ...
}
which has to be invoked like:
.left_join::<fruit::Entity>()
If we instead do:
fn left_join<E>(self, _: E) -> Self
where
E: EntityTrait,
{
// ...
}
then the Turbofish can be omitted:
.left_join(fruit::Entity)
provided that fruit::Entity
is a unit struct.
Instead of:
fn has_many(entity: Entity, from: Column, to: Column);
has_many(cake::Entity, cake::Column::Id, fruit::Column::CakeId)
we'd prefer having a builder and stating the params explicitly:
has_many(cake::Entity).from(cake::Column::Id).to(fruit::Column::CakeId)
Consider the following two methods, which accept the same parameter but in different forms:
fn method_with_model(m: Model) { ... }
fn method_with_active_model(a: ActiveModel) { ... }
We would define a trait
pub trait IntoActiveModel {
fn into_active_model(self) -> ActiveModel;
}
Such that Model
and ActiveModel
both impl this trait.
In this way, we can overload the two methods:
pub fn method<A>(a: A)
where
A: IntoActiveModel,
{
let a: ActiveModel = a.into_active_model();
...
}