Nominal unification inspired by alphaKanren

[namin]

Review by @swannodette.

[swannodette]

This is truly, truly fantastic. However official Clojure projects can't just merge pull requests. I need a ticket+patch in JIRA - http://dev.clojure.org/jira/browse/LOGIC. If it's not too much bother could you please submit a Clojure Contributor Agreement (CA) - http://clojure.org/contributing. If you're willing to become a contributor, I'm more than happy to go ahead and apply the patch ASAP while the CA slowly finds its way to North Carolina via snail mail :)

Again, thank you very much for this addition.

[namin]

Sure. I'll do the ticket / patch on JIRA as well as sign the agreement tomorrow.

Also please let me know if you have suggestions.

Thanks. :-)

[swannodette]

I'm curious about this, why do we need to check for nil?

[namin]

In reifyc for the nom-hash constraint, I just reify the constraint if it's relevant to the end result. This is consistent with alphaKanren.

For example,

(run* [q]
  (nom-fresh [c d]
     (typo [] ['lam (nom-tie c ['lam (nom-tie d ['var c])])] q)))

would otherwise return

(([-> _0 [-> _1 _0]] :- lvar:x5037#lvar:xc5049 nom:b5036#lvar:gc5025 nom:b5036#_0 nom:b5036#[nom:b5026 lvar:trand5024]))

instead of just

((-> _0 (-> _1 _0)))

Because the constraints are on vars and noms that are not in the final result, they're just noisy and not interesting to the end user.

Simpler examples. Contrast:

(is (= (run* [q] (nom-fresh [a b] (nom-hash a q) (nom-hash b q))) '(_0)))
(is (= (run* [q] (fresh [x]  (nom-fresh [a b] (nom-hash a x) (nom-hash b x) (== [x a b] q))))
      '(([_0 a_1 a_2] :- a_2#_0 a_1#_0))))

For the first case, we don't want:

((_0 :- <nom:b5130>#_0 <nom:a5129>#_0))

[swannodette]

Ok, makes sense.

[swannodette]

Just making sure that you actually want to test that q unifies with a list that contains 2 numbers and 1 symbol. Clojure doesn't have pairs like Scheme. lcons and the supporting protocol fns lfirst lnext exist so that you can construct sequence like things with an improper tail as you can in Scheme via cons - but in the case of core.logic only logic vars are truly allowed. Though we don't do any explicit validation around lcons now - we may do so in the future. A closer translation of this test would be:

(= (run* [q] 
     (nom/fresh [a b] 
       (== q (lcons 1 1))
       (nom/hash a q)))
   (lcons 1 1))

[namin]

Good catch. Done.

[namin]

I am just not sure how to deeply apply swaps without having suspensions. Let's take it on chat sometimes.

[swannodette]

Yes would also need to add a constraint to each var that gets a swap. This constraint would apply the swap list when the var becomes ground. Related I think we can eliminate the Suspension type altogether. We can just leverage the constraint solving infrastructure to do that work.

[swannodette]

Starting to look more closely at the nominal implementation, I believe removing Suspension and leveraging the constraint solving framework is the key.

[swannodette]

Check out master, we have deep constraints now in the form of treec :)

[namin]

Cool, @swannodette. I could use treec for nom/hash. I added some hacks for special handling and reifyc (but just saw that you anticipated that as well).

[namin]

@swannodette : I've been working on an implementation of core.logic.nominal which doesn't use suspensions. I am almost there, but the quine example gets generated with extra nils at the end of lists. Debugging that.

In addition, with the latest core.logic master, I'm having trouble printing to the repl for debugging. Have you been experiencing that?

[namin]

FYI, the no-susp branch is here: https://github.com/namin/core.logic/compare/nominal-no-susp

All tests pass, but there's still a problem with quines related to extra nils.

[swannodette]

Wow, cool! Now that run expressions are lazy you need to wrap them in a doall when debugging. I'm happy to add a debug atom that we can set to make this less tedious. Let me know.

[swannodette]

Code is looking pretty! :) :) :)

[swannodette]

Please feel free to attach a patch in JIRA. As soon as your CA is in more than happy to give you commit rights to core.logic :)

[namin]

Neat, thanks :)

The quines now run with the no-susp branch. I'll attach a patch on JIRA soon. Just want to iron out a few more quirks.

[namin]

Patch attached to http://dev.clojure.org/jira/browse/LOGIC-78

Closing this pull request.