Replace SparseArray Option<T> with T::MAX to cut down on branching

[alice-i-cecile]

would enable cheaper get_unchecked() operations

From Future Work of #1525.

[bjorn3]

Cranelift has PackedOption. This preserves an Option like api while using a reserved value of your choice as None.

[cart]

Ooh I dig that!

[NathanSWard]

Is it worth bringing in cranelift-entity for PackedOption or should we roll our own.
I've been messing around with PackedOption, however it's quite limiting and missing quite a few features including as_ref for example.

Also inorder to use PackedOption with builtin-in type such a usize we have to create an extra level of indirection
e.g.

pub struct SparseArray<I, V = I>
where
    V: ReservedValue,
{
    values: Vec<PackedOption<V>>,
    marker: PhantomData<I>,
}

struct Wrapper<T>(T);

// ReservedValue is requires for PackedOptiont<T>
impl ReservedValue for Wrapper<usize> { /* impl */ }

pub struct ComponentSparseSet {
    // ..fields
    sparse: SparseArray<Entity, Wrapper<usize>>,
}

[cart]

The code is small enough that I think rolling our own is probably the move for maximum flexibility / keeping dep count low. Afaik Wrapper would get compiled out, so thats largely just a question of ergonomic right?

[NathanSWard]

The code is small enough that I think rolling our own is probably the move for maximum flexibility / keeping dep count low.

Cool, this is the path I was planning on.

Wrapper would get compiled out, so thats largely just a question of ergonomic right?

Yep, performance shouldn't be an issue. It was more so a code-cleanliness/ergonomic issue ahah :)

[james7132]

This was implemented for non-component SparseSets for all ECS metadata stores (i.e. Resources, Tables, etc.) in #12083. ComponentSparseSet not supporting the full index range of entities will likely be a correctness error. Closing this out as completed.