mk: phase 5A — conda, soft-cut without onceo
Some checks failed
Test, Build, and Deploy / test-build-deploy (push) Failing after 54s

conda-try mirrors condu-try but on the chosen clause it (mk-bind
(head-goal s) (rest-conj)) — all head answers flow through. condu by
contrast applies rest-conj to (first peek), keeping only one head
answer.

7 new tests covering: first-non-failing-wins, skip-failing-head, all-fail,
no-clauses, the conda-vs-condu divergence (`(1 2)` vs `(1)`), rest-goals
running on every head answer, and the soft-cut no-fallthrough property.

169/169 cumulative.
This commit is contained in:
2026-05-07 21:51:52 +00:00
parent f4ab7f2534
commit 240ed90b20
3 changed files with 124 additions and 2 deletions

42
lib/minikanren/conda.sx Normal file
View File

@@ -0,0 +1,42 @@
;; lib/minikanren/conda.sx — Phase 5 piece A: `conda`, the soft-cut.
;;
;; (conda (g0 g ...) (h0 h ...) ...)
;; — first clause whose head g0 produces ANY answer wins; ALL of g0's
;; answers are then conj'd with the rest of that clause; later
;; clauses are NOT tried.
;; — differs from condu only in not wrapping g0 in onceo: condu
;; commits to the SINGLE first answer, conda lets the head's full
;; answer-set flow into the rest of the clause.
;; (Reasoned Schemer chapter 10; Byrd 5.3.)
(define
conda-try
(fn
(clauses s)
(cond
((empty? clauses) mzero)
(:else
(let
((cl (first clauses)))
(let
((head-goal (first cl)) (rest-goals (rest cl)))
(let
((peek (stream-take 1 (head-goal s))))
(if
(empty? peek)
(conda-try (rest clauses) s)
(mk-bind (head-goal s) (mk-conj-list rest-goals))))))))))
(defmacro
conda
(&rest clauses)
(quasiquote
(fn
(s)
(conda-try
(list
(splice-unquote
(map
(fn (cl) (quasiquote (list (splice-unquote cl))))
clauses)))
s))))

View File

@@ -0,0 +1,75 @@
;; lib/minikanren/tests/conda.sx — Phase 5 piece A tests for `conda`.
;; --- conda commits to first non-failing head, keeps ALL its answers ---
(mk-test
"conda-first-clause-keeps-all"
(run*
q
(conda
((mk-disj (== q 1) (== q 2)))
((== q 100))))
(list 1 2))
(mk-test
"conda-skips-failing-head"
(run*
q
(conda
((== 1 2))
((mk-disj (== q 10) (== q 20)))))
(list 10 20))
(mk-test
"conda-all-fail"
(run*
q
(conda ((== 1 2)) ((== 3 4))))
(list))
(mk-test "conda-no-clauses" (run* q (conda)) (list))
;; --- conda DIFFERS from condu: conda keeps all head answers ---
(mk-test
"conda-vs-condu-divergence"
(list
(run*
q
(conda
((mk-disj (== q 1) (== q 2)))
((== q 100))))
(run*
q
(condu
((mk-disj (== q 1) (== q 2)))
((== q 100)))))
(list (list 1 2) (list 1)))
;; --- conda head's rest-goals run on every head answer ---
(mk-test
"conda-rest-goals-run-on-all-answers"
(run*
q
(fresh
(x r)
(conda
((mk-disj (== x 1) (== x 2))
(== r (list :tag x))))
(== q r)))
(list (list :tag 1) (list :tag 2)))
;; --- if rest-goals fail on a head answer, that head answer is filtered;
;; the clause does not fall through to next clauses (per soft-cut). ---
(mk-test
"conda-rest-fails-no-fallthrough"
(run*
q
(conda
((mk-disj (== q 1) (== q 2)) (== q 99))
((== q 200))))
(list))
(mk-tests-run!)

View File

@@ -124,8 +124,9 @@ Key semantic mappings:
escapes to ground values for arithmetic or string ops
- [ ] `matche` — pattern matching over logic terms (extension from core.logic)
`(matche l ((head . tail) goal) (() goal))`
- [ ] `conda` — soft-cut disjunction (like Prolog `->`)
- [ ] `condu` — committed choice (already in phase 2; refine semantics here)
- [x] `conda` — soft-cut: first non-failing head wins; ALL of head's answers
flow through rest-goals; later clauses not tried (`Phase 5 piece A`)
- [x] `condu` — committed choice (Phase 2)
- [ ] `nafc` — negation as finite failure with constraint
- [ ] Tests: Zebra puzzle, N-queens, Sudoku via `project`, family relations via `matche`
@@ -155,6 +156,10 @@ _(none yet)_
_Newest first._
- **2026-05-07** — **Phase 5 piece A — conda**: soft-cut. Mirrors `condu` minus
the `onceo` on the head: all head answers are conjuncted through the rest of
the chosen clause. 7 new tests including the conda-vs-condu divergence test.
169/169 cumulative.
- **2026-05-07** — **Phase 4 piece B — reverseo + lengtho**: reverseo runs forward
cleanly and `run 1`-cleanly backward; lengtho uses Peano-encoded lengths so it
works as a true relation in both directions (tests use the encoding directly).