evaluator: a disjunction of a struct with an interpolation error and a struct with an empty string remains not concrete

[bttk]

What version of CUE are you using (cue version)?

Playground v0.4.0

Does this issue reproduce with the latest release?

yes

What did you do?

let fn = {
    in: {
        content: string
        ch:      string | *"-"
    }
    out: "\(in.ch)\(in.content)\(in.ch)"
} | {
    in: content: null
    out: ""
}

let outer = {inner: string | *null }

let x = (fn & {in: content: outer.inner}).out
ox: "\(x)"
oy: "\((fn & {in: content: null}).out)"
oz: "\((fn & {in: content: "b"}).out)"

What did you expect to see?

In ox, the interpolation in the first part of fn is not possible, so I'd expect it to result in a concrete value.

ox: ""
oy: ""
oz: "-b-"

What did you see instead?

invalid interpolation: invalid interpolation: unresolved disjunction

let fn = {
	in: {
		content: string
		ch:      string | *"-"
	}
	out: "\(in.ch)\(in.content)\(in.ch)"
} | {
	in: {
		content: null
	}
	out: ""
}
let outer = {
	inner: string | *null
}
let x = (fn & {
	in: {
		content: outer.inner
	}
}).out
ox: "\(x)"
oy: ""
oz: "-b-"

[mpvl]

Analysis: CUE normally eliminates erroneous disjuncts. It somehow does not seem to do that for interpolations (probably as it gobbles the error).

Regarding oy, the spec also needs clarification on how to handle null in interpolation.

[myitcv]

Drive-by observation: the playground (where this behaviour was observed) by default behaves more like cue eval:

exec cue eval x.cue
-- x.cue --
let fn = {
    in: {
        content: string
        ch:      string | *"-"
    }
    out: "\(in.ch)\(in.content)\(in.ch)"
} | {
    in: content: null
    out: ""
}

let outer = {inner: string | *null }

let x = (fn & {in: content: outer.inner}).out
ox: "\(x)"
oy: "\((fn & {in: content: null}).out)"
oz: "\((fn & {in: content: "b"}).out)"

Hence why we are likely not surfacing an error in this situation (if, that is, there should be one).

Drive-by question:

CUE normally eliminates erroneous disjuncts

Is this an erroneous disjunct though? Or just one that is incomplete? Given what we're looking at here is cue eval like behaviour, I don't think this can be an erroneous disjunct. But under #822 and the changes discussed there regarding cue export I can see a world in which this would export.