Definition recursion and struct disjunction

[b4nst]

So first sorry for the title, I does not really know where the important part is in the following question

So I have this structure:

-- invalid.yaml --
components:
  one:
    components:
      two:
        name: foo
-- main.cue --
#Node: {
	id?:   string
	name?: string
	components: [string]: #Leaf | #Node
}

#Leaf: {
	id: string
}
-- valid.yaml --
components:
  one:
    components:
      two:
        id: foo

I was expecting that cue vet invalid.yaml main.cue -d '#Node' would fail and that cue vet valid.yaml main.cue -d '#Node' would succeed when it's actually exactly the opposite.

Any idea what I'm missing there?

[seh]

#Node.components is not a required field, per #822, so "components.one.components. two" satisfies #Node, which tolerates a "name" field.

Unifying with #Node shows that the "components" field winds up in there, but is not required to be there from cue vet's perspective, at least for today.

[b4nst]

Oh wow, for some reason I thought #Node.components was required but now it makes total sense. Thanks for the enlightening!

[seh]

That's how I've read it too. You can see by comments starting at #822 (comment) that I shared your interpretation. I'm coming around to believe that #822 is necessary, but the present state of CUE is confusing. It seems to me that there are two different lenses or perspectives through which one can read the same CUE code—say, export versus vet, or generation versus validation—by which the same code means something different in each context.

[myitcv]

Yep, the discussion in #822 is relevant here.

But just to note, do you even need #Leaf here? Is a #Leaf not simply a #Node without a components field defined?

The builtins discussed in #943 will also be of interest (they are strongly linked to #822)

To also note that the disjunction #Leaf | #Node is problematic because of the reasons described in #1163

[b4nst]

#Leaf is not exactly a #Node without a components field. I wanted to describe the constraint of having an id set on leafs, while it's optional on a node.

[myitcv]

Ah I see, ok. You definitely either need #822 or some other mechanism to disambiguate #Node and #Leaf