Vetting an object that can be a subset of another object

[MikkelHJuul]

When vetting an object the vet fails validating an object that is a subset of another object in an or-list of possible structures.

if #A is a subset of #B then
obj: #A | #B fails validating an object describing #A
with message: incomplete value {...#A<exploded - values replaced>} | {...#B<exploded - values replaced>}

Is this the correct behavior, or a missing feature?

this sprang from the discussion: https://cuelang.slack.com/archives/CLT3ULF6C/p1626443556244700

[myitcv]

@MikkelHJuul please can you provide an example of what is failing?

[MikkelHJuul]

It is unclear if the proposal would fix this, maybe something like:

obj: {
    numexist(=1, a) |  numexist(=2, a, b)
    a: string
    b: string
}

which seems a bit like a hack (if it is even valid, I'm also not sure how the bulitin numexist work)

[myitcv]

all fields of both objects are required, but #B subsumes #A

If it is the case that all fields in both objects are required, then #B does not subsume #A AFAIU. Because for that to be the case, a value v: #A would also have to satisfy #B. And that cannot be the case under the required fields proposal (unless #B and #A are identical). Because any required field that #B defines beyond #A would not be permitted by #A because of closedness, and could not be omitted in a check against #B because it is required.

A very basic example of what I understand from your situation would be as follows:

#A: {
	name!: string
}

#B: {
	#A
	age!: int
}

Both name and age are marked as required, so a value {name: "cueckoo"} is unambiguously of type #A, and {name: "gopher", age: 42} is unambiguously of type #B.

[MikkelHJuul]

exactly - now it almost sounds like I am supposed to be able to do this, but I am not. I always get the error: incomplete value {...#A<exploded - values replaced>} | {...#B<exploded - values replaced>}

fx

#A: a: string
#B: a: string
#B: b: string

a: #A|#B

> echo '{"a":"asd"}' >> t.json
> cue vet t.json test.cue -d a
incomplete value {a:"asd"} | {a:"asd",b:string}

or

close({
   a: string
}) | close({
   a: string
   b: string
})

> cue vet t.json test.cue
incomplete value {a:"asd"} | {a:"asd",b:string}

or am I doing something incorrect?

[myitcv]

or am I doing something incorrect?

You are not doing anything incorrect at all. It's just that you need #822 to be implemented in order to achieve what you want 😄

Notice in my example above, the use of ! in the declaration of the two fields. That is the proposed syntax for specifying that a field is required. And it is those required semantics that allow such a disjunction to be resolved unambiguously. Without those semantics, i.e. today, you cannot do it.

[MikkelHJuul]

okay, great! thanks

[verdverm]

This is expected behavior when the fields of one struct are a strict subset of another struct in a disjunction. There is no way for CUE to decide which is "correct." Each option to a disjunction needs to be uniquely differentiated. When you find that one is a subset of another, there may be a different way to organize things depending am what you are trying to achieve.