evalv3: disjunct of definitions that each reference an identical optional field results in incomplete evaluation

[marxarelli]

I am trying out the v3 evaluator with an existing project and ran into some strange behavior with regards to incomplete (or unsimplified?) evaluation. After completely paring down my schema, I was able to repro the issue with the minimal configuration below.

Note that the golden examples below are from the v2 evaluator.

exec cue version
env CUE_EXPERIMENT=evalv3
exec cue eval -e foo foo.cue
cmp stdout stdout.golden

-- foo.cue --
#Run: {
	run!: string
	options?: #Option | [#Option, ...]
	if options != _|_ {
		optionsValue: options
	}
}

#Copy: {
	copy!: string
	options?: #Option | [#Option, ...]
	if options != _|_ {
		optionsValue: options
	}
}

#Option: {}

foo: #Run | #Copy
foo: run: "make"

-- stdout.golden --
run: "make"

If I remove the conditional, the entire evaluation result is empty.

env CUE_EXPERIMENT=evalv3
exec cue eval -e foo foo.cue
cmp stdout stdout.golden

-- foo.cue --
#Run: {
	run!: string
	options?: #Option | [#Option, ...]
	optionsValue: options
}

#Copy: {
	copy!: string
	options?: #Option | [#Option, ...]
	optionsValue: options
}

#Option: {}

foo: #Run | #Copy
foo: run: "make"

-- stdout.golden --
run: "make"

(BTW, thanks for all the hard work on the new evaluator. I'm very much looking forward to it when it's ready.)