iterators for core eve library, how should that look.

[DenisYaroshevskiy]

eve core library has some things that I'd classify as algorithms:

load/store aggregated wide
compress_store (should generate the individual components to store).

Based on this I see the following operations required:

unalign - compress store need to advance by arbitrary runtime n.
reduce_alignment - load aggregated needs to reduce alignment twice
+= or advance

I would also say that the following operations should work
previous_partially_aligned(it, eve::fixed)
next__partially_aligned(it, eve::fixed)
equality and ordering
distance
read if input
write if output
load for arbitrary cardinal if input
store for arbitrary cardinal if output

The list of extra operations implies that kumi::tuple<pointer ...> cannot be made a valid iterator - because equality/ordering are done by all components. While they should only be done via one component.

This leads me to think that we need a separate zip_pointer/ tuple_pointer/soa_pointer type.

The problem is, we kinda have one in algo::views: algo::views::zip.

However, that one I find a little problematic to bring into core - it'd drag a lot of algo with it.

So I suggest maybe having an soa_pointer in eve. Or zip_pointer.
It will be a duplication with algo and in certain cases we'll have two types that mean the same thing, like:

soa_pointer<int*, int*> and views::zip_iterator<int*, int*>

But I don't see a better option.

Why not bring eve::algo::views::zip in eve?

Mostly because zip can be applied to ranges.
Also kinda seems akward, some views are in eve, some in views.