(I am using CHICKEN Scheme's miniKanren, but I would appreciate portable answers!)
Suppose I have some relation:
(define (rel° x)
(conde
((pred?° x))
((== x (my-awesome-imperative-thing)))))
The intention is that this relation can allow x to be either ground or variable. When it is ground, it satisfies some relationally-computed predicate. When it's variable, it satisfies the predicate and also is unified with some imperative value-generating procedure's output.
There's a relatively obvious problem, though; conde clauses need to be idempotent. Suppose my imperative thing isn't idempotent, but I can guarantee that we'll only call rel° at most once per logic variable; if nothing else grounds that variable, then rel° will be the "writer" of the variable and all other relations will be the "readers", and it's safe/correct-ish. In this special case, is this pattern possible with miniKanren? And if you want to use more words, what are the costs and tradeoffs?
My motivation is to explore a possible solution for an existing question of mine. In that case, the relation is something like:
(define (gensym° x)
(conde
((symbol° x))
((== x (gensym)))))
I've tried a few different things, but nothing works yet.