@jballanc: Yes, in the code, b could be anything as long as 1 and 3 are member of the vector q and a is 1 or c is 3. So with the 8th result, there's no way to complete it to a solution, no matter what you insert for b.

Does someone have a minimal example showing the difference between conda and condu? As far as I understand,

(run* [x y]
  (conda  ;; or condu
   [(== x 1) (membero 2 y)]
   [(== x 2) (== y 3)]))

Well, with condu it might also return ([1 #{2}]) or ([1 [2]]) since the collection type isn't fixed, and of course the collection doesn't need to be minimal although that would make sense.

Or with a much simpler example:

(defn y-or-n [x]
  (conde
   [(== x :y)]
   [(== x :n)]))

(run* [x y]
  (condu
   [(== x 1) (y-or-n y)]
   [(== x 2) (== y 3)]))
;;=> ([1 :y] [1 :n])

Hm, it works when I swap the two goals in the first clause. So it seems that only the head goal has the once-semantics.

tsdh: I implemented FizzBuzz with core.logic (don’t ask), and made use of condu (in the form of defnu) - https://github.com/jballanc/logicbuzz/blob/master/src/logicbuzz/core.clj#L17

essentially, as I understand it, you’re building up a tree of solutions: first make the first condition pass, then cycle back and fill in any remaining fresh vars, then make the second condition pass, then cycle back and fill remaining fresh, etc.

with condu, if you can make the first condition pass, then you fill the remaining fresh vars and stop

if you can’t, then you try the second, and so on

@jballanc: I think with your fizzbuzzo you can use defna and defnu. Both commit to a clause as soon as the head goal succeeds.

@jballanc: And in your case, x is always ground.

it’s been a while, but I think I tried defna and didn’t have success...

@jballanc: The difference with conda/`defna` versus condu/`defnu` only shows up when the head goal of a clause may produce multiple solutions. With the former, all of them are considered, with the latter, only the first one is considered. Thus, some programs which diverge with conda may terminate with condu.

hmm…ok

At least that's what I've read in W. Byrd's thesis and validated with experimentation here.

ok, yeah, stated that way I think it matches with what I’ve found in the past

(that FizzBuzz solution was a lot of trial and error to get everything done with only using core.logic)