evaluator: should close() distribute across elements of a disjunction?

[myitcv]

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

$ cue version
cue version v0.0.0-20220905100514-9e9786e93b63

       -compiler gc
     CGO_ENABLED 1
          GOARCH arm64
            GOOS linux
             vcs git
    vcs.revision 9e9786e93b63af979d9f105a12c8519a6216e56f
        vcs.time 2022-09-05T10:05:14Z
    vcs.modified false

Does this issue reproduce with the latest release?

Yes

What did you do?

exec cue eval x.cue
cmp stdout stdout.golden

-- x.cue --
x: close({name: string}) | close({age: int})
x: name: "test"

y: close({name: string} | {age: int})
y: name: "test"
-- stdout.golden --
x: {
    name: "test"
}
y: {
    name: "test"
}

Playground link

What did you expect to see?

Passing test.

What did you see instead?

> exec cue eval x.cue
[stdout]
x: {
    name: "test"
}
y: close({
    name: string
} | {
    age: int
}) & {
    name: "test"
}
> cmp stdout stdout.golden
--- stdout
+++ stdout.golden
@@ -1,10 +1,6 @@
 x: {
     name: "test"
 }
-y: close({
-    name: string
-} | {
-    age: int
-}) & {
+y: {
     name: "test"
 }

FAIL: /tmp/testscript3032515904/repro.txtar/script.txtar:2: stdout and stdout.golden differ

[myitcv]

cc @wonderflow @FogDong @Somefive based on our conversation earlier.

[myitcv]

Noting in passing that numexists() as proposed in #943 addresses one of the core use-cases of using close() in this way, namely one-of semantics.

[myitcv]

Closing per #1878 (comment). The use case behind this question is better solved with oneof semantics. So closing this issue in the absence of something concrete to motivate changing the existing semantics.