Associations with generic types

[moulins]

I have the following two tables (with the corresponding schema!s):

#[derive(Debug, Queryable, Insertable, Identifiable)]
#[table_name = "foos"]
pub struct Foo<'a> {
    id: i32,
    stuff: Cow<'a, String>
}

#[derive(Debug, Queryable, Insertable, Identifiable, Associations)]
#[belongs_to(Foo)]
#[table_name = "bars"]
pub struct Bar {
    id: i32,
    foo_id: i32
}

The #[derive(Associations)] fail with expected lifetime parameter, because there is no way to refer to the generic lifetime 'a of Foo in the #[belongs_to]. Did I overlook something which could allow me to use the derive?

---

Of course, I can implement BelongsTo manually, and everything seems to work:

impl<'a> BelongsTo<Foo<'a>> for Bar {
    type ForeignKey = i32;
    type ForeignKeyColumn = schema::bars::foo_id;
    
    fn foreign_key(&self) -> Option<&Self::ForeignKey> {
        Some(&self.foo_id)
    }

    fn foreign_key_column() -> Self::ForeignKeyColumn {
        schema::bars::foo_id
    }
}

But it would be nicer if we could specify these kinds of generic lifetimes, maybe with something like: #[belongs_to(type="for<'a> Foo<'a>")]?

[sgrif]

This is something we should support, but I'd like to see more discussion around what the API should be. We 100% need the lifetime declared in belongs_to, but I'm wondering if just belongs_to("Foo<'a>") makes more sense

[moulins]

We 100% need the lifetime declared in belongs_to, but I'm wondering if just belongs_to("Foo<'a>") makes more sense

Well, I was thinking "What if you want to specify constraints on your lifetimes?"; but when thinking more about it, all diesel objects you'll get back via belongs_to should be 100% owned, so all lifetime parameters can be unconstrained.

Using belongs_to("Foo<'a>") still seems wierd to me, because the 'a isn't defined anywhere, and I'm not sure what belongs_to("Foo<'a, 'a>") should mean:

• impl<'a> ... for Foo<'a, 'a>, or
• impl<'a, 'b> ... for Foo<'a, 'b>?

Instead, I would propose belongs_to("Foo<'_>"): the elided lifetime makes it clear that any lifetime is acceptable.

[sgrif]

I think allowing elided lifetimes as the syntax makes perfect sense.

[moulins]

Nice, thanks for the work!