Should we be allowing Const nodes with void types?

[doug-q]

By void type I mean sums with 0 variants.

This type is very useful and I am not advocating touching it.

We should be able to infer from a wire of void type that it's source never terminates. For example, you can return a void type from panic no problem. But what does LoadConstant a Const of void type mean?

[cqc-alec]

Since there can be no value of void type, it should be impossible to construct a Const of void type, right?

[doug-q]

Hmm, yes for Value::Sum but you can make a CustomConst that reports it as it's type. Maybe that's ok? Maybe the semantics of LoadConstanting your constant is that it runs a web server forever

[cqc-alec]

Whether or not we formally allow it in a HUGR I think anything downstream trying to compile or interpret it would be well within its rights to throw up its hands in despair.

[acl-cqc]

Hmmm. Why did we not force CustomConsts to have CustomTypes? That would prevent this...I mean, it doesn't make sense for a CustomConst to have even a type like Sum( () | () ) (i.e. bool), does it?

[acl-cqc]

Also giving Sum private fields might be an idea, but we at least validate that a Sum's contents (including tag) are correct (...right?), which would rule out Sums of void type