core::mem::align_of requires Self: Sized, even though it only needs to know the type's alignment. Notably, slices have an alignment known at compile time, even though they are not Sized. Should Rust have an Aligned trait (as a supertrait of Sized), so this bound can be loosened?
It makes sense to me, but do you have a practical example where you would have used this?
1 Like
I'm trying to write a Vec-like type that can store unsized values (slices) as elements. I would prefer for my Vec to start out with an aligned dangling pointer, so that I don't need to change said pointer until I necessary. It's a minor annoyance, which I can easily work around, but still nice to have.
Is there an example of !Aligned type, other than extern types?
dyn Trait
So if it's essentially Vec<[T]>, can't you use align_of::<T>() for that pointer? You could also initialize that pointing to [T; 0], not even dangling.
1 Like
I could, hence why I stated it's "easy to work around". Though this becomes less fun if you also want to support structs with a slice as their last field.
Yeah, custom slice-based DSTs are more interesting.
Language-wise, ?Sized constraints are special, but I suppose that would also have to imply ?Aligned for backward compatibility. And vice versa, ?Aligned must imply ?Sized for the supertrait. You would explicitly combine T: Aligned + ?Sized for dealing with the middle ground.
2 Likes
This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.