fix #19335 Disallow void in type section

[SirOlaf]

Fixes #19335

Edit: Other option would be having the field disappear like with generics but that seems confusing

[Araq]

No, what we should do instead is to map void to some Unit-like type and optimize it away in the backends.

[SirOlaf]

Would that actually address the issue with using a field typed as void and is there a good reason to allow typing a field as void in the first place

[Araq]

In the example it makes no sense but for

type G[T] = object
  x: T

G[void]

[SirOlaf]

Yeah that is the case that makes fields disappear, and it continues to work

[SirOlaf]

Actually this presents the question why void can't be modeled as an empty object instead of a special case

type VoidType = object

Edit: Rust seems to do it that way and it allows cool stuff like saying Set<Key> = Map<Key, ()> (https://doc.rust-lang.org/nomicon/exotic-sizes.html#zero-sized-types-zsts), but getting that to work seems a lot more involved

[arnetheduck]

typeof(G[void]().x) should still work in the above example - it's useful in generic code to not have to introduce a special case for void, often

[SirOlaf]

That never worked because as I said it makes the fields disappear completely https://play.nim-lang.org/#ix=4dGQ

Edit: Specifically here https://github.com/nim-lang/Nim/blob/devel/compiler/semtypinst.nim#L208

[Araq]

Edit: Rust seems to do it that way and it allows cool stuff like saying Set = Map<Key, ()>

Nim had that before Rust, see e.g. CritbitTree[void].

[arnetheduck]

That never worked

too bad - makes for a good feature request :)