Nesting objects using const generics requires non-local bounds

[TomDeRybel]

Hello,
First of, thanks for all the work on const generics. As an embedded user, this is really useful and makes my life significantly more pleasant.

For a big project I'm working on, I've run into an issue where nesting objects using const generics expressions requires repeating bounds in the top-level object that are specified in the lower-level (embedded) object. This makes using this feature difficult, as naming clashes are rife, and it requires implementation details of lower-level objects to be carried-through the hierarchy.

I'm not quite confident I'm explaining this right. To clarify the issue, I've attached a minimal example. UwU embeds OwO. Both have const generics expressions. The compiler gives the following output;

error: unconstrained generic constant
  --> src/uwu.rs:12:11
   |
12 |     woof: OwO<N>,
   |           ^^^^^^
   |
   = help: try adding a `where` bound using this expression: `where [(); Q + OWO_CONST - 1]:`
note: required by a bound in `OwO`
  --> src/owo.rs:5:10
   |
3  | pub struct OwO<const Q: usize>
   |            --- required by a bound in this
4  | where
5  |     [(); Q + OWO_CONST - 1]: Sized,
   |          ^^^^^^^^^^^^^^^^^ required by this bound in `OwO`

As I understand it, UwU should not need additional bounds due to those imposed in OwO. But, I could definitely be wrong. In any case, all insights are greatly appreciated.

const_generic_locality_issue.zip

[lcnr]

this is related to #41 and we're not yet sure whether we should require these bounds.

If these bounds are required, your issue seems unavoidable though. The bound of OwO is evaluatable(N + 3) while your only other bound is evaluatable(N + U). Without adding the N + 3 bound, UwU<usize::MAX, 0> would have a field which isn't well formed.

[coreylowman]

What will help determine whether the bounds should be required? As a user, copy-pasting bounds everywhere feels really bad because its verbose and its hard to understand why the compiler needs that.

[TomDeRybel]

Thank you for the feed-back. What surprised me most is that internal concerns of one object (how a memory size is computed from const generic parameters) must be propagated upward, thus breaking the "locality of concerns". I'm curious to see how all of this will evolve over time.