for errors out on solvable cycles

[cueckoo]

Originally opened by @Ereski in cuelang/cue#779

Current master (2c86835)

Similar to the previous issues I've opened, this time it's for that is having trouble with cycles:

X: Y.message

STATE: {
    for k, v in Y {
        if k != "message" {
            "\(k)": v
        }
    }
}

Y: STATE & {
    message: X
}

This is a simple stateful bidirectional projection expressed in CUE (STATE is saved and restored as needed). In one way, Y is forked into X and STATE. In the other way, X and STATE are joined into Y. Two usage examples are:

X: "test"

STATE: {
    code: 101
}

and

Y: {
    message: "test"
    code: 101
}

Both of which errors out with a cycle error. Even defining X, STATE, and Y all at the same time causes a cycle error.

[cueckoo]

Original reply by @mpvl in cuelang/cue#779 (comment)

Another interesting one. :)

The cycle here is detected here because the for comprehension iterators over the the very struct in which it embeds.

Arguably, of course, that should be okay as long as it does not add new fields that then should have been used during evaluation.

FTR, in v0.2 the cycle check did not work correctly and this construct would produce inaccurate results.

[mpvl]

@Ereski We could probably break this cycle if we allow comprehensions that insert fields in structs that are used as a source that do not introduce any new fields.