Allow recursive untagged unions

[spinda]

The current implementation of RFC #1444 disallows the following:

#![feature(untagged_unions)]

union Foo {
    f0: u8,
    f1: Foo,
}

error[E0072]: recursive type `Foo` has infinite size
 --> <anon>:3:1
  |
3 | union Foo {
  | ^ recursive type has infinite size
  |
  = help: insert indirection (e.g., a `Box`, `Rc`, or `&`) at some point to make `Foo` representable

Unlike structs, since all fields of a union exist in the same space, a union whose definition involves infinite recursion can be represented in finite space. In this case, Foo can be represented as union { f0: u8 }, with &foo.f1 equivalent to &foo.

[ranma42]

What is the use case? As you mentioned in your comment, it looks like it would only provide a different (and more verbose) name to the same data.
An alternative could involve improving the error so that instead of infinite size it complains about the almost-useless self-referential field.

[spinda]

My use case is generic unions generated by a macro, of the form:

union Foo<A, B> {
    a: A,
    b: B,
}

Foo<A, Foo<B, C>> is a type whose size is the maximum of the sizes of A, B, and C. Sometimes this ends up producing a recursive union definition, and instead of quietly accepting it, rustc raises an unnecessary error.

[oli-obk]

related: #1151 (recursive enums)

[joshtriplett]

@spinda It seems like the generic types there should eliminate the circularity. A Foo can't unconditionally contain a Foo, but it can contain a T and a user can map that T to a different Foo type; that doesn't create a loop. I wonder what makes the compiler think it does?

Can you post a concrete code example with a macro invocation that produces the error?

[spinda]

@joshtriplett Hrm... In the months since I submitted this, I changed my approach to the project that needed it, and now I can't recall a specific example. I'd be okay with closing this.

[vcfxb]

@spinda I'm trying to get around this right now, how did you change your approach?

[Centril]

Triaging this issue: @spinda Are you interested in this still?

[spinda]

@Centril I'm late here, but no, not really. Moved on to other things.

[Centril]

Alright. Closing this then.