cue vet could be more helpful for incomplete comprehension conditions

[rogpeppe]

Consider the following example:

! exec cue vet -c=false x.cue

-- x.cue --
#Y: {
	cond!: bool
	x!: int
}

y: #Y & {
	cond!: bool
	if cond {
		x: "hello"
	}
	if !cond {
		x: "blah"
	}
}

In this example, y represents a template where the condition is not known yet.
Whatever the condition, the result will always be a conflict, because in all cases, the CUE is setting x to a string where it's defined in the schema as an integer.

If this pattern occurs in real-world CUE, it would almost certainly be a mistake that we'd like to catch early.
However cue vet currently succeeds on this example.

I think it would be useful if cue vet was to fail if there is a conflict of this kind - theoretically we know that y is unified with #Y and that the field #Y.x is an int, so any occurrence of x that doesn't unify with int is a likely mistake. That way we could catch more failures early, even in the presence of incomplete results.

[mvdan]

I believe the previous issue that @myitcv was talking about is #1730. Should we close as a duplicate? @rogpeppe you can always add your example as a comment there.

[rogpeppe]

Closing as dup of #1730.