cue: references to ancestor nodes should be direct and not through outer scope

[mpvl]

If a struct is referred to (and copied), any reference within this copy to the root of the struct should resolve to the root of the copied struct.

Both these cases below should work.

a: {
  b: int
  c: a.b  // workaround: use b, intstead of a.b
}
d: a & { b: 2 } // a.c == 2

Foo: {
   foo?: Foo
   bar:  string
}
foo: Foo & {
	foo: {
		bar: "a"
	}
	bar: "b"
}

See also Issue #145

[mpvl]

Or maybe the way it is now is actually correct. Depending on how a references is interpreted, a reference to root should bind to the copy or not, but it should probably be consistent.

[rudolph9]

Or maybe the way it is now is actually correct. Depending on how a references is interpreted, a reference to root should bind to the copy or not, but it should probably be consistent.

@mpvl Agreed, either case is probably valid but needs to be consistent.

In either case, it should probably factor into the discussion around the @ notation on the Proposal: Querying extension.

[rudolph9]

Relevant comment to this this issue here

[mpvl]

I'm pretty convinced the current behavior is correct now. It is just unfortunate there is this dichotomy. Field aliases mitigate it a bit, as one can alias a possibly shadowed variable.

I think it will be possible to detect when users make a mistake and then warn them in a vet command, similar to what Go does.

[mpvl]

Closing for now.

[cueckoo]

This issue has been migrated to cue-lang/cue#152.

For more details about CUE's migration to a new home, please see cue-lang/cue#1078.