eval: regression in closedness handling with disjunctions

[myitcv]

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

$ cue version
cue version v0.0.0-20250128173939-870bfc2bb495

go version go1.23.2
      -buildmode exe
       -compiler gc
  DefaultGODEBUG asynctimerchan=1,gotypesalias=0,httpservecontentkeepheaders=1,tls3des=1,tlskyber=0,x509keypairleaf=0,x509negativeserial=1
     CGO_ENABLED 1
          GOARCH arm64
            GOOS linux
         GOARM64 v8.0
             vcs git
    vcs.revision 870bfc2bb495f05faed7c65ec51109b9eed1762c
        vcs.time 2025-01-28T17:39:39Z
    vcs.modified false
cue.lang.version v0.12.0

Does this issue reproduce with the latest release?

Yes

What did you do?

# evalv2
env CUE_EXPERIMENT=evalv3=0
exec cue vet -c -d '#site' schema.cue site.json

# evalv3
env CUE_EXPERIMENT=evalv3=1
exec cue vet -c -d '#site' schema.cue site.json

-- schema.cue --
package preprocessor

#site: {
	_kind: {
		kind!: string
	}

	#sanitiser:  _kind & (#patternSanitiser | #ellipsisSanitiser)
	#comparator: _kind & (#patternComparator | #unstableLineOrderComparator)

	#patternSanitiser: {
		kind:     "patternSanitiser"
		pattern?: string
	}

	#patternComparator: {
		kind:     "patternComparator"
		pattern?: string
	}

	#ellipsisSanitiser: {
		kind: "ellipsisSanitiser"
	}

	#unstableLineOrderComparator: {
		kind: "unstableLineOrderComparator"
	}

	sanitiser?:  #sanitiser
	comparator?: #comparator
}
-- site.json --
{
  "comparator": {
    "kind": "patternComparator",
    "pattern": "expr"
  },
  "sanitiser": {
    "kind": "patternSanitiser",
    "pattern": "expr"
  }
}

What did you expect to see?

Passing test.

What did you see instead?

# evalv2 (0.453s)
# evalv3 (0.451s)
> env CUE_EXPERIMENT=evalv3=1
> exec go run cuelang.org/go/cmd/cue vet -c -d '#site' schema.cue site.json
[stderr]
sanitiser: 2 errors in empty disjunction:
sanitiser.kind: conflicting values "ellipsisSanitiser" and "patternSanitiser":
    ./schema.cue:22:9
    ./site.json:7:13
sanitiser.pattern: field not allowed:
    ./site.json:8:5
exit status 1
[exit status 1]
FAIL: /tmp/testscript2499769875/repro.txtar/script.txtar:10: unexpected command failure

This bisects to 45e3256

[mvdan]

Here is the same testscript but slightly simplified:

# With the old evaluator.
env CUE_EXPERIMENT=evalv3=0
exec cue vet -c -d '#site' input.cue data.json

# With the new evaluator.
env CUE_EXPERIMENT=evalv3=1
exec cue vet -c -d '#site' input.cue data.json

-- input.cue --
package preprocessor

#site: {
	_kind: kind!: string

	#foo: _kind & (#foo1 | #foo2)
	#bar: _kind & (#bar1 | #bar2)

	#foo1: {
		kind:       "foo1"
		foo1field?: string
	}
	#foo2: {
		kind: "foo2"
	}

	#bar1: {
		kind:       "bar1"
		bar1field?: string
	}
	#bar2: {
		kind: "bar2"
	}

	foo?: #foo
	bar?: #bar
}
-- data.json --
{
  "foo": {
    "kind": "foo1",
    "foo1field": "expr"
  },
  "bar": {
    "kind": "bar1",
    "bar1field": "expr"
  }
}

As of 870bfc2:

# With the old evaluator. (0.011s)
> env CUE_EXPERIMENT=evalv3=0
> exec cue vet -c -d '#site' input.cue data.json
# With the new evaluator. (0.039s)
> env CUE_EXPERIMENT=evalv3=1
> exec cue vet -c -d '#site' input.cue data.json
[stderr]
bar: 2 errors in empty disjunction:
bar.bar1field: field not allowed:
    ./data.json:8:5
bar.kind: conflicting values "bar2" and "bar1":
    ./data.json:7:13
    ./input.cue:22:9
bar.bar1field: field not allowed:
    ./input.cue:7:17
    ./data.json:8:5
[exit status 1]

And here's a slight alteration which, as far as I can tell, should be equivalent. But, surprisingly enough, it fixes evalv3... and breaks evalv2!

# With the old evaluator.
env CUE_EXPERIMENT=evalv3=0
exec cue vet -c -d '#site' input.cue data.json

# With the new evaluator.
env CUE_EXPERIMENT=evalv3=1
exec cue vet -c -d '#site' input.cue data.json

-- input.cue --
package preprocessor

#site: {
	#kind: {
		kind!: string
		...
	}

	foo?: #kind & (#foo1 | #foo2)
	bar?: #kind & (#bar1 | #bar2)

	#foo1: {
		kind:       "foo1"
		foo1field?: string
	}
	#foo2: kind: "foo2"

	#bar1: {
		kind:       "bar1"
		bar1field?: string
	}
	#bar2: kind: "bar2"
}
-- data.json --
{
  "foo": {
    "kind": "foo1",
    "foo1field": "expr"
  },
  "bar": {
    "kind": "bar1",
    "pattern2": "expr"
  }
}

# With the old evaluator. (0.010s)
> env CUE_EXPERIMENT=evalv3=0
> exec cue vet -c -d '#site' input.cue data.json
[stderr]
bar: 2 errors in empty disjunction:
bar.kind: conflicting values "bar2" and "bar1":
    ./data.json:7:13
    ./input.cue:10:25
    ./input.cue:22:15
bar.pattern2: field not allowed:
    ./data.json:8:5
    ./input.cue:3:8
    ./input.cue:6:3
    ./input.cue:10:8
    ./input.cue:10:17
    ./input.cue:18:9
[exit status 1]
FAIL: repro-evalv3.txtar:3: unexpected command failure
# With the new evaluator. (0.036s)
> env CUE_EXPERIMENT=evalv3=1
> exec cue vet -c -d '#site' input.cue data.json

[rogpeppe]

Here's a simpler repo. Looks like the reference to the inner #site._kind field is closing it up (adding a ... fixes the issue).

# With the old evaluator.
env CUE_EXPERIMENT=evalv3=0
exec cue vet -c -d '#site' input.cue data.json

# With the new evaluator.
env CUE_EXPERIMENT=evalv3=1
exec cue vet -c -d '#site' input.cue data.json

-- input.cue --
package preprocessor

#site: {
	_kind: kind!: string

	bar?: _kind & {
		kind!:       "bar1"
		bar1field?: string
	}
}
-- data.json --
{
  "bar": {
    "kind": "bar1",
    "bar1field": "expr"
  }
}

Output:

# With the old evaluator. (0.013s)
# With the new evaluator. (0.048s)
> env CUE_EXPERIMENT=evalv3=1
> exec cue vet -c -d '#site' input.cue data.json
[stderr]
bar.bar1field: field not allowed:
    ./data.json:4:5
    ./input.cue:8:3
[exit status 1]
FAIL: /tmp/x.txtar:7: unexpected command failure