evalv3: "conflicting values" error regression when a disjunction involves missing fields

[mvdan]

# With the old evaluator.
env CUE_EXPERIMENT=evalv3=0
exec cue vet -c=false

# With the new evaluator.
env CUE_EXPERIMENT=evalv3=1
exec cue vet -c=false

-- input.cue --
package p

#PatchOp: {
	op:    "replace"
	value: _
} | {
	op:   "remove"
} | {
	op:   "copy"
	from: string
}
#Patch: {
	input:  #PatchOp
	output: input
}

params: {}
patched: (#Patch & {
	input: {
		op:    "replace"
		value: "\(params.missing)"
	}
}).output

As of c29fd02:

# With the old evaluator. (0.010s)
> env CUE_EXPERIMENT=evalv3=0
> exec cue vet -c=false
# With the new evaluator. (0.028s)
> env CUE_EXPERIMENT=evalv3=1
> exec cue vet -c=false
[stderr]
output.op: conflicting values "copy" and "replace":
    ./input.cue:9:8
    ./input.cue:20:10
output.op: conflicting values "remove" and "replace":
    ./input.cue:7:8
    ./input.cue:20:10

The config is incomplete, as params.missing is not a field that exists, but still - I think v2 was correct in not reporting the "conflicting values" errors which do not matter here.

[mvdan]

Slightly smaller testscript:

# With the old evaluator.
env CUE_EXPERIMENT=evalv3=0
exec cue vet -c=false

# With the new evaluator.
env CUE_EXPERIMENT=evalv3=1
exec cue vet -c=false

-- input.cue --
package p

#PatchOp: {
	op:    "replace"
	value: _
} | {
	op:   "remove"
} | {
	op:   "copy"
	from: string
}

params: {}
patched: #PatchOp & {
	op:    "replace"
	value: "\(params.missing)"
}

[mpvl]

This issue is fixed by https://cuelang.org/cl/1206284.

However, the above code still outputs a somewhat misleading value for cue eval, where the value of patched.value is input. Similarly, for cue def this suffers from the let at wrong scope issue.

I'll keep this open to address these issue.

[mpvl]

I can no longer reproduce the aforementioned issues. They do occur in another Issue, so at least they are still reproducible elsewhere.

Note that the original issue was already fixed.