Implement HistoryBufferView on top of #486

[sosthene-nitrokey]

OldestOrdered was already part of the public API.

When merging we will need to add its removal to #494 so that there is only OldestOrederedView (which can be renamed to just OldestOrdered)

[Dirbaio]

OldestOrdered was already part of the public API.

oof 🥲

maybe it's "less bad" to make OldestOrdered the "new" way already, and add a type alias at the root for compat that just ignores the const generic. This way in 0.9 we can just remove the type alias.

mod histbuf {
    pub struct OldestOrdered<'a, T> {
        inner: core::iter::Chain<core::slice::Iter<'a, T>, core::slice::Iter<'a, T>>,
    }
}

#[deprecated = "use `heapless::histbuf::OldestOrdered` instead."]
pub type OldestOrdered<'a, T, const N: usize> = OldestOrderedInner<'a, T>;

i'm not sure if there's any problem in "ignoring" N that way, it seems to work in quick testing.

[sosthene-nitrokey]

i'm not sure if there's any problem in "ignoring" N that way, it seems to work in quick testing.

I did not know that was possible!

[sosthene-nitrokey]

i'm not sure if there's any problem in "ignoring" N that way, it seems to work in quick testing.

I think I found a possible issue, there will be some cases where the compiler won't be able to infer N. For exemple the following compiles now, but wouldn't compile with the new alias approach.

fn testing<const N: usize>(iter: OldestOrdered<'_, u32, N> {}

fn main() {
   let a = HistoryBuffer::<u32, 3>::new();
   testing(a.oldest_ordered());
}

[Dirbaio]

aw, that's a shame :(. Nice find. Let's do the PhantomData then.