shared-only fields

[nikomatsakis]

Proposal

The current RFC proposes only "mutable" fields. That is, if you have an &mut T, and T: Trait, you can mutate the fields declared in Trait. However, it'd be nice to be able to have "shared-only" fields, where the fields act shared no matter what. So even with a &mut T, you can only get a shared borrow of those fields.

Among other things, this is needed if we ever wish to support defining traits with fields that can be implemented by &T types. Example:

trait Foo {
    value: u32
}

impl Foo for &u32 {
    value = *self
}

Right now, though, that is blocked by #4 in any case.

Shared-only fields are also interesting because they would have looser overlap requirements (i.e., two shared-only fields could point at the same data).

Interactions:

• mapping to non-interior fields #4 -- mapping to non-interior fields

[glaebhoerl]

(What does it mean when a P-essential is blocked on a P-nice-to-have?)

Anyway, this seems like a nice-to-have to me, but awkardly, syntax-wise the most obvious (and likely overall best) way to signify this would be by the absence of a mut qualifier - that is, "normal" mutable fields would be let mut or whatever. So we probably want to decide one way or another on the syntax aspect at least before stabilizing the core feature.

(Edit: I see you mentioned the same thing in #3.)

[burdges]

I think "blocked" means #4 requires adding mut here so that smart pointer Derefs can do their thing.

[nikomatsakis]

@glaebhoerl

(What does it mean when a P-essential is blocked on a P-nice-to-have?)

Are you referring to #4? This particular issue is not blocked on #4, I don't think, it just "interacts" with it. (In particular, combining the two enables fields that are reachable through an &-ref.)