forked from google-deepmind/deepmind-research
-
Notifications
You must be signed in to change notification settings - Fork 0
/
saturation.rkt
970 lines (864 loc) · 40.1 KB
/
saturation.rkt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
#lang racket/base
;**************************************************************************************;
;**** Saturation Algorithm ****;
;**************************************************************************************;
(require bazaar/cond-else
bazaar/date
bazaar/debug
bazaar/dict
bazaar/mutation
data/heap/unsafe
data/queue
define2
global
racket/block
racket/dict
racket/format
racket/list
racket/math
racket/pretty
racket/string
satore/Clause
satore/clause-format
satore/clause
satore/interact
satore/json-output
satore/misc
satore/rewrite-tree
satore/tptp
satore/unification-tree
satore/unification)
(provide (all-defined-out))
;===============;
;=== Globals ===;
;===============;
(define-global:boolean *quiet-json?* #false
'("JSON output format and silent mode. Deactivates some verbose options."))
(define-global *memory-limit* 4096
"Memory limit in MB, including the Racket VM."
exact-positive-integer?
string->number)
(define-global:boolean *find-unit-rules-in-candidates?* #false
'("Search for unit rewrite rules in the condidate set?"
"This may speed up the search significantly, or slow it down significantly,"
"depending on how many unit rules can be generated."))
(define-global:boolean *backward-rw?* #true
'("Use binary-clause rewriting? (aka θ-equivalences)."
"The iterative mode does not use backward rewrites."))
(define-global:boolean *dynamic-rules?* #false
"Use dynamic rules? Experimental.")
(define-global:boolean *proof?* #false
"Display the proof when found?")
(define-global *age:cost* '(1 9)
"Age:cost ratio. Format: <age-freq>:<cost-freq>, e.g., '1:3'"
(λ (p) (and (list? p) (= 2 (length p)) (andmap exact-nonnegative-integer? p)))
(λ (s) (map string->number (string-split s ":"))))
(define-global:category *cost-type* 'weight-fair
'(weight-fair weight)
"Cost type.")
(define-global *cost-depth-factor* 1.5
"Cost = weight + cost-depth-factor * depth"
(λ (x) #true)
string->number)
(define-global *cost-noise* 0.
"Noise factor to add to costs"
(λ (x) (and (real? x) (>= x 0)))
string->number)
(define-global:boolean *parent-discard?* #false
"Discard clauses when at least one parent has been discarded?")
(define-global:boolean *discover-online?* #true
'("Use rewrite rules as soon as they are discovered?"
"The rules will not be made confluence until the next restart"))
(define-global:boolean *negative-literal-selection?* #true
"When resolving, if a clause has negative literals, only select one of them.")
(define-global *cpu-limit* +inf.0
"CPU limit in seconds per problem."
exact-positive-integer?
string->number)
(define-global *cpu-first-limit* +inf.0
"CPU limit in seconds per problem."
exact-positive-integer?
string->number)
(define-global *cpu-limit-factor* 3
'("Increase of cpu limit of cpu-first-limit once an iteration has failed."
"Set to 0 to avoid restarting."
"Assumes --cpu-limit.")
(λ (x) (and (real? x) (>= x 0)))
string->number)
(define-global *input-rules* #false
'("File to read rewrite rules from.")
file-exists?
values)
(define-global *output-rules* #false
'("File to write rewrite rules to. If 'auto', a unique name is chosen automatically.")
(λ (x) #true)
(λ (str)
(if (string=? str "auto")
(build-path "rules" (string-append "rules-" (date-iso-file) ".txt"))
str)))
;======================;
;=== Rule discovery ===;
;======================;
;; Finds new binary equivalences between `C` and the clauses of `utree`,
;; and adds and returns the set of resulting new rules that can be added to `rwtree`.
;;
;; rwtree-out : rewrite-tree?
;; C : Clause?
;; utree : unification-tree?
(define (discover-new-rules! rwtree-out C utree)
(cond/else
[(not rwtree-out) '()]
;; FIND UNIT REWRITE RULES
;; Unit rewrite using the binary rewrite tree
[(unit-Clause? C)
(when-debug>= steps (displayln "Found unit clause"))
;; Add the new unit clause to the set of unit rewrite rules
(rewrite-tree-add-unit-Clause! rwtree-out C #:rewrite? #false)]
[(not (binary-Clause? C))
'()]
;; FIND BINARY REWRITE RULES
#:else
(when-debug>= steps (displayln "Found binary clause"))
;; We search for the converse implication in the active set.
;; This takes care of clauses that have converse clauses but not the converse.
;; Ex: C1 = p(X) | q(X) ; C2 = ~p(a) | ~q(a)
;; C2 has a converse clause, but not C1. Hence C2 can be added as a binary rewrite rule
;; but not C1.
(define Conv (make-converse-Clause C))
(define self-conv? (Clause-subsumes C Conv))
#:cond
[self-conv?
; Self-converse Clause, so only one Clause to add (but will lead to 2 rules).
(when-debug>= steps (displayln "Self-converse"))
(rewrite-tree-add-binary-Clause! rwtree-out C C #:rewrite? #false)]
#:else
(define Subs (utree-find/any utree Conv (λ (a Conv) (and (binary-Clause? a)
(Clause-subsumes a Conv)))))
#:cond
[Subs
; We found a converse Clause in the active set, or a Clause that subsumes the
; converse Clause, so we can add selected-Clause as a rule and also its
; converse Clause.
; We can't add Subs, since Subs may not itself have a converse
; clause in the active set. Ex:
; selected-Clause = p(a) | q(a), so Conv = ~p(a) | ~q(a) and Subs = ~p(X) | ~q(X)
; Subs cannot be added as a rule because (supposedly) we haven't found a converse
; clause yet, but we can still add Conv as a rule.
; If Subs is aleardy a rule, Conv will be rewritten to a tautology and discarded.
(when-debug>= steps (printf "Found converse subsuming clause: ~a\n"
(Clause->string Subs)))
; Since C has already been rewritten, there is no need to do it again.
(rewrite-tree-add-binary-Clause! rwtree-out C Subs #:rewrite? #false)]
#:else
; Asymmetric rules.
; Even when selected-Clause is not a rule (yet?), it may still enable
; other more specific Clauses of the active set to be added as rules too.
; Eg, selected-Clause = p(X) | q(X) and some other active clause is ~p(a) | ~q(a).
(define Subsd (utree-find/all utree Conv (λ (a Conv) (and (binary-Clause? a)
(Clause-subsumes Conv a)))))
#:cond
[(not (empty? Subsd))
(when-debug>= steps
(displayln "Found converse subsumed clauses:")
(print-Clauses Subsd))
; rewrite=#true is ok here because Clause->rules already considers both directions
; of the implication. Hence if a potential rule is rewritten to a tautology,
; we know it's already redundant anyway.
; TODO: Rename this to add-asymmetric-rules ?
(rewrite-tree-add-binary-Clauses! rwtree-out Subsd C #:rewrite? #true)]
#:else '()))
;====================;
;=== Clause costs ===;
;====================;
;; Clause comparison for the cost queue.
;;
;; Clause? Clause? -> boolean?
(define (Candidate<= C1 C2)
(<= (Clause-cost C1) (Clause-cost C2)))
;; Sets the cost of a list of Clauses that all have the same parent `parent`.
;; Some noise can be added to the cost via *cost-noise*.
;;
;; Cs : (listof Clause?)
;; cost-type: symbol?
;; parent : Clause?
;; cost-depth-factor : number?
;; -> void?
(define (Clauses-calculate-cost! Cs
cost-type
parent
#:! cost-depth-factor)
(case cost-type
;; Very simple cost function that uses the weight of the Clause. May not be fair.
[(weight)
(for ([C (in-list Cs)])
(set-Clause-cost! C
(if (empty? (Clause-clause C))
-inf.0 ; empty clause should always be top of the list
(Clause-size C))))]
;; Very simple cost function that is fair
[(weight-fair)
(for ([C (in-list Cs)])
(set-Clause-cost! C
(if (empty? (Clause-clause C))
-inf.0 ; empty clause should always be top of the list
(+ (* (Clause-depth C) cost-depth-factor)
(Clause-size C)))))])
;; Add noise to the cost so as to potentially solve more later.
;; To combine with a slowly increasing step-limit-factor in iterative saturation
(unless (zero? (*cost-noise*))
(define ε (*cost-noise*))
(for ([C (in-list Cs)])
(set-Clause-cost! C
(* (+ (- 1. ε) (* ε (random)))
(Clause-cost C))))))
;==================;
;=== Saturation ===;
;==================;
;; List of possible status values. Used to prevent mistakes.
(define statuses '(refuted saturated time memory steps running))
;; Returns whether the result dictionary `res` has the status `status`.
;;
;; res : dict?
;; status : symbol?
(define (check-status res status)
(define res-status (dict-ref res 'status #false))
; To avoid silent typo bugs.
(assert (memq status statuses) status)
(assert (memq res-status statuses) res-status)
(eq? status res-status))
;; The main algorithm. Saturates the formula given by the input clauses
;; by adding new clauses (either resolutions or factors) until either
;; the empty clause is produced, or a resource limit is reached (steps, time, memory).
;;
;; input-clauses : (listof clause?)
;; step-limit : number?
;; memory-limit : number?
;; cpu-limit : number?
;; rwtree : (or/c #false rewrite-tree?)
;; rwtree-out : (or/c #false rewrite-tree?)
;; backward-rewrite? : boolean?
;; parent-discard? : boolean?
;; age:cost : (list/c exact-nonnegative-integer? exact-nonnegative-integer?)
;; cost-type : symbol?
;; disp-proof? : boolean?
;; L-resolvent-pruning : boolean?
;; find-unit-rules-in-candidates? : boolean?
;; negative-literal-selection? : boolean?
;; -> dict?
(define (saturation input-clauses
#:? [step-limit +inf.0]
#:? [memory-limit (*memory-limit*)] ; in MB
#:? [cpu-limit +inf.0] ; in seconds
; The rewrite tree holding the binary rules.
; Set it to #false to deactivate it.
#:? [rwtree (make-rewrite-tree #:atom<=> (get-atom<=>)
#:dynamic-ok? (*dynamic-rules?*)
#:rules-file (*input-rules*))]
; rewrite-tree where new rules found during saturation are stored.
; If rwtree-out is different from rwtree, new rules are not used but only stored,
; and backward rewriting is deactivated.
#:? [rwtree-out rwtree] ; or #false if don't care
; Only effective if (eq? rwtree rwtree-out)
#:? [backward-rewrite? (*backward-rw?*)]
#:? [parent-discard? (*parent-discard?*)]
#:? [age:cost (*age:cost*)] ; chooses age if (< (modulo t (+ age cost)) age)
#:? [cost-type (*cost-type*)]
#:? [disp-proof? (*proof?*)]
#:? [L-resolvent-pruning? (*L-resolvent-pruning?*)]
#:? [find-unit-rules-in-candidates? (*find-unit-rules-in-candidates?*)]
#:? [negative-literal-selection? (*negative-literal-selection?*)])
;; Do NOT reset the clause-index, in particular if rwtree is kept over several calls to saturation.
#;(reset-clause-index!)
;; Tree containing the active Clauses
(define utree (make-unification-tree))
;; Clauses are pulled from priority first, unless empty.
;; INVARIANT: active-Clauses U priority is (should be!) equisatisfiable with input-Clauses.
;; In other words, if priority is empty, then the set of active-Clauses
;; is equisatisfiable with the input-Clauses.
;; Some active clauses may be removed from utree and pushed back into priority for further
;; processing like 'backward' rewriting. In that case, the active Clauses (utree) is not
;; 'complete'.
(define priority (make-queue))
;; Both heaps contain the candidate clauses (there may be duplicates between the two,
;; but this is checked when extracting Clauses from either heap).
(define candidates (make-heap Candidate<=))
(define age-queue (make-heap Clause-age>=))
;; Frequency of extracting Clauses from either heap.
(define age-freq (first age:cost))
(define cost-freq (second age:cost))
(define age+cost-freq (+ age-freq cost-freq))
;; Add the Clauses Cs to the priority queue for priority processing.
(define (add-priority-Clauses! Cs)
(for ([C (in-list Cs)])
; Need to set candidate? to #true otherwise it may be skipped.
; (or maybe we should not skip clauses loaded from `priority`?)
(set-Clause-candidate?! C #true)
(enqueue! priority C)))
(define cost-depth-factor (*cost-depth-factor*)) ; immutable value
(define (add-candidates! parent Cs)
;; Calculate costs and add to candidate heap.
(unless (empty? Cs)
(Clauses-calculate-cost! Cs cost-type parent #:cost-depth-factor cost-depth-factor)
(for ([C (in-list Cs)])
(set-Clause-candidate?! C #true))
(unless (= 0 cost-freq) (heap-add-all! candidates Cs))
(unless (= 0 age-freq) (heap-add-all! age-queue Cs))
(when-debug>= steps
(printf "#new candidates: ~a #candidates: ~a\n"
(length Cs)
(heap-count candidates)))))
(define input-Clauses
(map (λ (c) (make-Clause c '() #:type 'in)) input-clauses))
;; This maintains the invariant: If priority is empty, then the set of active-Clauses
;; is equisatisfiable with the input-Clauses.
;; In other words active-Clauses U priority is (should be!) equisatisfiable with input-Clauses.
(add-priority-Clauses! input-Clauses)
;; We add the Clauses of the binary rules as candidates, so as to not cluter the active set
;; in case there are many rules.
;; Another option is to add them to the priority queue because they can be seen as possibly
;; useful lemmas.
(when rwtree
; A mock root clause parent of all input rules
(define C0rules (make-Clause (list ltrue) '() #:type 'rules-root))
; rewrite=#false should not be necessary (since rewriting checks if a clause is from the original
(add-candidates! C0rules (rewrite-tree-original-Clauses rwtree)))
(define step 0)
(reset-n-tautologies!)
(define n-parent-discard 0)
(define n-forward-subsumed 0)
(define n-backward-subsumed 0)
(reset-n-binary-rewrites!)
(reset-n-rule-added!)
(reset-subsumes-stats!)
(reset-n-L-resolvent-pruning!)
(define start-time (current-milliseconds))
;; TODO: Some calls are very slow...
(define (make-return-dict status [other '()])
(assert (memq status statuses) status)
(define stop-time (current-milliseconds))
`((status . ,status)
(steps . ,step)
(generated . ,clause-index) ; includes all input clauses and rules and intermediate steps
(actives . ,(length (unification-tree-Clauses utree)))
(candidates . ,(heap-count candidates))
(priority-remaining . ,(queue-length priority))
(tautologies . ,n-tautologies) ; counted in generated, (mostly) not in candidates
,@(rewrite-tree-stats rwtree)
(binary-rewrites . ,n-binary-rewrites)
(forward-subsumed . ,n-forward-subsumed)
(backward-subsumed . ,n-backward-subsumed)
(subsumes-checks . ,n-subsumes-checks)
(subsumes-steps . ,n-subsumes-steps)
(subsumes-breaks . ,n-subsumes-breaks)
(parent-discard . ,n-parent-discard)
(L-resolvent-pruning . ,n-L-resolvent-pruning)
(memory . ,(current-memory-use)) ; doesn't account for GC---this would take too much time
(time . ,(- stop-time start-time))
. ,other))
(define (make-refuted-dict C)
(define proof (Clause-ancestor-graph C)) ; no duplicates
(define flat-proof (flatten proof))
(define type-occs (occurrences flat-proof #:key Clause-type))
(when disp-proof?
(displayln "#| begin-proof")
(display-Clause-ancestor-graph C #:tab " ")
(displayln "end-proof |#"))
(make-return-dict 'refuted
`((proof-length . ,(length flat-proof)) ; doesn't account for compound rewrites
(proof-steps . ,(for/sum ([C2 (in-list flat-proof)])
(define n (length (Clause-parents C2)))
(if (< n 2) n (- n 1))))
(proof-inferences . ,(count (λ (C2) (not (empty? (Clause-parents C2))))
flat-proof))
,@(for/list ([(t o) (in-dict type-occs)])
(cons (string->symbol (format "proof-type:~a" t)) o)))))
;:::::::::::::::::::::;
;:: Saturation Loop ::;
;:::::::::::::::::::::;
(define result
(let loop ()
(++ step)
(define time-passed (- (current-milliseconds) start-time)) ; this is fast
(define mem (current-memory-use-MB)) ; mflatt says it's fast
(when-debug>= steps
(printf "\nstep: ~a generated: ~a processed/s: ~a generated/s: ~a\n"
step
clause-index
(quotient (* 1000 step) (+ 1 time-passed))
(quotient (* 1000 clause-index) (+ 1 time-passed))))
(cond/else
[(and (= 0 (heap-count candidates))
(= 0 (heap-count age-queue))
(= 0 (queue-length priority)))
(when-debug>= steps (displayln "Saturated"))
(make-return-dict 'saturated)]
[(> step step-limit) (make-return-dict 'steps)]
[(> time-passed (* 1000 cpu-limit)) (make-return-dict 'time)]
[(and (> mem memory-limit)
(block
(define pre (current-milliseconds))
;; Memory is full, but try to collect garbage first.
(unless (*quiet-json?*)
(printf "; before GC: memory-limit: ~a memory-use: ~a\n" memory-limit mem))
(collect-garbage)
(collect-garbage)
(define mem2 (current-memory-use-MB))
(define post (current-milliseconds))
(unless (*quiet-json?*)
(printf "; after GC: memory-limit: ~a memory-use: ~a gc-time: ~a\n"
memory-limit mem2 (* 0.001 (- post pre))))
(> mem2 memory-limit)))
; mem is full even after GC, so exit
(make-return-dict 'memory)]
#:else
;; Choose a queue/heap to extract the selected-Clause from.
(define queue
(cond [(> (queue-length priority) 0)
; Always has priority.
priority]
[(or (= 0 (heap-count candidates))
(and (> (heap-count age-queue) 0)
(< (modulo step age+cost-freq) age-freq)))
; TODO: This is somewhat defeated by the `priority` queue.
age-queue]
[else candidates]))
(when-debug>= steps
(printf "Selected queue: ~a\n" (cond [(eq? queue priority) "priority"]
[(eq? queue candidates) "candidates"]
[else "age queue"])))
(define selected-Clause
(if (heap? queue)
(begin0 (heap-min queue)
(heap-remove-min! queue))
(dequeue! queue)))
#:cond
;; ALREADY PROCESSED
[(not (Clause-candidate? selected-Clause))
(when-debug>= steps (displayln "Clause already processed. Skipping."))
(-- step) ; don't count this as a step
(loop)]
;; ONE PARENT DISCARDED
[(and parent-discard? (ormap Clause-discarded? (Clause-parents selected-Clause)))
(when-debug>= steps (displayln "At least one parent has been discarded. Discard too."))
(discard-Clause! selected-Clause)
(++ n-parent-discard)
(loop)]
#:else
(set-Clause-candidate?! selected-Clause #false)
;; FORWARD REWRITE
;; BINARY CLAUSE REWRITE OF SELECTED
;; NOTICE: We do binary rewrites first because if we did unit then binary
;; we would need to attempt a second unit-rewrite after that.
;; (This may lead to unnecessary binary rewrites, but it's cleaner this way.)
(define selected-Clause-brw
(if rwtree
(binary-rewrite-Clause rwtree selected-Clause)
selected-Clause))
(when-debug>= steps
(printf "|\nstep ~a: selected: ~a\n"
step (Clause->string/alone selected-Clause 'all))
(define binary-rewritten? (not (eq? selected-Clause-brw selected-Clause)))
(when binary-rewritten?
(displayln "Binary rewritten:")
(display-Clause-ancestor-graph selected-Clause-brw #:depth 1))
(unless (eq? selected-Clause-brw selected-Clause-brw)
(displayln "Unit rewritten:")
(display-Clause-ancestor-graph selected-Clause-brw #:depth 1))
(when-debug>= interact
(interact-saturation
(priority utree rwtree selected-Clause make-return-dict)
selected-Clause-brw selected-Clause-brw)))
(set! selected-Clause selected-Clause-brw)
;;; From now on, only selected-Clause should be used
(define selected-clause (Clause-clause selected-Clause))
#:cond
;; REFUTED?
[(empty-clause? selected-clause)
(make-refuted-dict selected-Clause)]
;; TAUTOLOGY?
[(clause-tautology? selected-clause)
(when-debug>= steps (displayln "Tautology."))
(discard-Clause! selected-Clause)
(loop)] ; skip clause
;; FORWARD SUBSUMPTION
[(utree-find/any utree selected-Clause Clause-subsumes)
;; TODO: Tests
=>
(λ (C2)
(++ n-forward-subsumed)
(when-debug>= steps (printf "Subsumed by ~a\n" (Clause->string C2 'all)))
(discard-Clause! selected-Clause)
(loop))] ; skip clause
#:else
;; Clause is being processed.
;; BACKWARD SUBSUMPTION
(define removed (utree-inverse-find/remove! utree selected-Clause Clause-subsumes))
(for-each discard-Clause! removed)
(+= n-backward-subsumed (length removed))
(when-debug>= steps
(define n-removed (length removed))
(when (> n-removed 0)
(printf "#backward subsumed: ~a\n" n-removed)
(when-debug>= interact
(print-Clauses removed 'all))))
;; FIND NEW REWRITE RULES
(define clause-index-before-discover clause-index)
(define new-rules (discover-new-rules! rwtree-out selected-Clause utree))
(define new-rule-Clauses (rules-original-Clauses new-rules))
;; NOTICE: We MUST add Clauses that are newly generated to the set of active rules
;; (via priority) otherwise we may miss some resolutions.
;; Only the Clauses that have been created during the discovery process need to be added.
;; Notice: To prevent the clauses from which the rules have originated to be rewritten to
;; tautologies, a test is performed in binary-rewrite-literal.
;; But this applies *only* to the `eq?`-Clause of the rule, hence beware of copies or
;; rewrites.
(add-priority-Clauses!
(filter (λ (C) (> (Clause-idx C) clause-index-before-discover))
new-rule-Clauses))
;; BACKWARD BINARY REWRITING
;; We don't need to backward rewrite if the new rules are not stored in rwtree,
;; as this means the set of used rules does not change during the whole saturation.
(when (and backward-rewrite?
rwtree
(eq? rwtree rwtree-out) ; not storing new rules in a different rwtree
(not (empty? new-rules)))
; Remove active Clauses that can be rewritten, and push them into priority.
; We must check whether the clauses we remove will be rewritten,
; otherwise we might add all the same candidates again when the removed Clause
; is popped from priority.
(define removed-active-Clauses
;; TODO: This is inefficient. We should modify utree-inverse-find/remove!
;; TODO: to handle multiple rule-C so as to take advantage of its hash/cache.
(remove-duplicates
(flatten
(for/list ([rule-C (in-list new-rule-Clauses)])
(utree-inverse-find/remove! utree rule-C
(λ (_rule-C C2)
(binary-rewrite-Clause? rwtree C2)))))
eq?))
(unless (empty? removed-active-Clauses)
(when-debug>= steps
(displayln "Some active Clauses can be backward binary rewritten:")
(print-Clauses removed-active-Clauses))
(add-priority-Clauses! removed-active-Clauses)))
;; Note that backward-rewritable Clauses are not yet discarded. They may be discarded
;; when they are pulled from priority and deemed discardable.
;;; Even if the selected Clause is a unit/binary rewrite rule, we must continue processing it
;;; and generate resolutions (because rewriting is only left-unification, not full unification)
;;; NEW CANDIDATES
(define L-resolvent-pruning-allowed?
(and L-resolvent-pruning?
; As per the invariant, if no Clause is in the priority queue,
; then the set of active Clauses of utree is equisatisfiable with the input clauses.
(= 0 (queue-length priority))))
(define new-Candidates
(if negative-literal-selection?
(utree-resolve+unsafe-factors/select utree selected-Clause
#:rewriter (λ (C) (binary-rewrite-Clause rwtree C)))
(utree-resolve+unsafe-factors utree selected-Clause
#:rewriter (λ (C) (binary-rewrite-Clause rwtree C))
#:L-resolvent-pruning? L-resolvent-pruning-allowed?)))
(when-debug>= interact
(displayln "New candidates:")
(print-Clauses new-Candidates))
;; If a clause has no resolvent with the active set (when complete)
;; then it will never resolve with anything and can thus be discarded.
#:cond
[(and L-resolvent-pruning-allowed?
(empty? new-Candidates))
(discard-Clause! selected-Clause)
(when-debug>= steps
(printf "No resolvent (L-resolvent-pruning?=~a). Clause discarded. \n"
L-resolvent-pruning?))
(loop)]
#:else
;; ADD CLAUSE TO ACTIVES
;; Rewrite the candidates with unit and binary rules, filter out tautologies,
;; calculate their costs and add them to the queues.
(add-candidates! selected-Clause new-Candidates)
;; UNIT RULE DISCOVERY IN CANDIDATES
;; Look for unit rewrite rules in the candidate set.
;; (Looking for binary rules would be too costly here)
(when find-unit-rules-in-candidates?
(when rwtree-out
(for ([C (in-list new-Candidates)])
(when (unit-Clause? C)
;; TODO: Should be calling/merged with discover-rules! to avoid inconsistencies
(rewrite-tree-add-unit-Clause! rwtree-out C #:rewrite? #false)))))
(add-Clause! utree selected-Clause)
(when-debug>= steps
(displayln "Adding clause.")
(print-active-Clauses utree #false))
(loop))))
(when-debug>= interact
(displayln "Saturation loop finished.")
(pretty-print result)
(define selected-Clause #false) ; mock up
(interact-saturation
(priority utree rwtree selected-Clause make-return-dict)))
result)
;========================;
;=== User interaction ===;
;========================;
;; Some commands to use with '--debug interact'. Type 'help' for information.
(define interact-commands '())
(define-namespace-anchor ns-anchor)
(define-syntax-rule (interact-saturation
(priority utree rwtree selected-Clause make-return-dict)
more ...)
(begin
(define what '(idx parents clause-pretty))
(interact
#:command (and (not (empty? interact-commands))
(begin0 (first interact-commands)
(rest! interact-commands)))
#:variables (priority utree rwtree what more ...)
#:namespace-anchor ns-anchor
#:readline? #true
[(list 'steps (? number? n))
"skips n steps"
(when (> n 0)
(cons! "" interact-commands)
(cons! (format "steps ~a" (- n 1)) interact-commands))]
[(list (or 'ancestors 'ancestor-graph 'graph))
"display the ancestor graph of the selected Clause."
(display-Clause-ancestor-graph selected-Clause)]
[(list (or 'ancestors 'ancestor-graph 'graph) (? number? depth))
"display the ancestor graph of the selected Clause down to the given depth."
(display-Clause-ancestor-graph selected-Clause #:depth depth)]
[(list 'what-fields)
(string-append
"Prints which fields are available for 'what,\n"
"which is used for printing clause information.")
(displayln Clause->string-all-fields)]
[(list 'selected)
"Selected clause"
(print-Clauses (list selected-Clause) what)]
[(list 'active)
"Active clauses"
(print-active-Clauses utree #true what)]
[(list (or 'binary 'rules))
"Found binary rules"
(print-binary-rules rwtree #true)]
[(list 'stats)
"Return-dictionary-like stats"
(pretty-print (make-return-dict 'running))]
[(list 'save-rules)
"Save the binary rules from the default rules-file"
(save-rules! rwtree #:rules-file (*output-rules*))])))
;; Prints the set of active Clauses (held in utree).
;;
;; utree : unification-tree?
;; long? : boolean?
;; what : (or/c 'all (listof symbol?))
;; -> void?
(define (print-active-Clauses utree long? [what 'all])
(define actives (unification-tree-Clauses utree))
(printf "#active clauses: ~a\n" (length actives))
(when long?
(displayln "Active clauses:")
(print-Clauses (sort actives < #:key Clause-idx) what)))
;; Prints the set of binary rules.
;;
;; rewrite-tree? boolean? -> void?
(define (print-binary-rules rwtree long?)
(define rules (rewrite-tree-rules rwtree))
(printf "#binary rules: ~a #original clauses: ~a\n"
(length rules)
(length (remove-duplicates (map rule-Clause rules) eq?)))
(when long?
(display-rules (rewrite-tree-rules rwtree))))
;============================;
;=== Iterative saturation ===;
;============================;
;; A struct holding information about a given input formula.
(struct problem (file name clauses [time-used #:mutable] [last-time #:mutable]))
;; file? (or/ #false string?) (listof clause?) -> problem?
(define (make-problem file name clauses)
(problem file name clauses 0 0))
;; Returns the same values as body ... but evaluates time-body ... before returning.
;; The result of time-body ... is discarded.
(define-syntax-rule (with-time-result [(cpu real gc) time-body ...] body ...)
(let-values ([(res cpu real gc) (time-apply (λ () body ...) '())])
time-body ...
(apply values res)))
;; Calls saturation for a set of problems in a loop.
;; Tries again each unsolved problem after multiplying the step-limit by step-limit-factor
;; and so on untill all problems are solved.
;; Loading time from files is *not* taken into account.
;;
;; problems : (listof problem?)
;; saturate : procedure?
;; memory-limit : number?
;; cpu-limit : number?
;; cpu-first-limit : number?
;; cpu-limit-factor? : number?
;; -> void?
(define (iterative-saturation/problem-set problems
saturate
#:! memory-limit
#:! cpu-limit ; in second
#:! cpu-first-limit ; in seconds
#:! cpu-limit-factor) ; in seconds
(define n-problems (length problems))
(define n-attempted 0)
(define n-solved 0)
(with-time-result [(cpu real gc)
(unless (*quiet-json?*)
(printf "; Total time: cpu: ~a real: ~a gc: ~a\n" cpu real gc))]
(let loop ([problems problems] [iter 0])
(define n-unsolved (length problems))
(define n-solved-iter 0)
(define n-attempted-iter 0)
(define new-unsolved
(for/fold ([unsolved '()]
[cumu-time 0]
#:result (reverse unsolved))
([prob (in-list problems)])
(define input-clauses (problem-clauses prob))
(when-debug>= init (for-each (compose displayln clause->string) input-clauses))
;; Collecting garbage can take time even when there's nothing to collect,
;; and can take a significant proportion of the time when solving is fast,
;; hence it's better to trigger GC only if needed.
(when (>= (current-memory-use-MB) (* 0.8 memory-limit))
(collect-garbage)
(collect-garbage))
;; Main call
(define cpu-limit-problem (min (max cpu-first-limit
(* cpu-limit-factor (problem-last-time prob)))
(- cpu-limit (problem-time-used prob))))
(define res (saturate input-clauses cpu-limit-problem))
(set! res (append `((name . ,(problem-name prob))
(file . ,(problem-file prob)))
res))
(++ n-attempted-iter)
(when (= 0 iter) (++ n-attempted))
(define solved? (or (check-status res 'refuted)
(check-status res 'saturated)))
(when solved?
(++ n-solved-iter)
(++ n-solved))
(define last-time (* 0.001 (dict-ref res 'time)))
(set-problem-last-time! prob last-time)
(set-problem-time-used! prob (+ (problem-time-used prob) last-time))
(set! res (dict-set res 'cumulative-time (exact-ceiling (* 1000 (problem-time-used prob)))))
(define remove-problem?
(or solved?
(check-status res 'memory) ; more time won't help if status=memory
(>= (problem-time-used prob) cpu-limit))) ; cpu exhausted for this problem
; Don't pretty-print to keep it on a single line which is simpler for parsing.
; Only print the last iteration of a problem for borg.
(cond
[(*quiet-json?*)
(when remove-problem? (displayln (saturation-result->json res)))]
[else
(pretty-write res)
(printf "; ~a/~a solved (iter: ~a/~a/~a success: ~a% avg-time: ~as ETA: ~as)\n"
n-solved n-attempted
n-solved-iter n-attempted-iter n-unsolved
(~r (* 100 (/ n-solved-iter n-attempted-iter)) #:precision '(= 1))
(~r (/ cumu-time n-attempted-iter 1000.) #:precision '(= 3))
(~r (* (/ cumu-time n-attempted-iter 1000.)
(- n-unsolved n-attempted-iter)) #:precision '(= 2)))])
(flush-output)
(values
(if remove-problem?
unsolved
(cons prob unsolved))
(+ cumu-time last-time))))
(unless (or (empty? new-unsolved)
(= cpu-limit-factor 0))
(loop new-unsolved (+ iter 1))))))
;; Calls saturate on a single set of clauses, first with a time limit of cpu-first-limit,
;; then restarts and doubles it until the cumulative time reaches cpu-limit.
;; Loading time is taken into account.
;; During a call to saturate, the new rewrite rules are saved in a separate tree,
;; which means that no new rule is introduced until the next restart—and thus the first
;; call to saturate uses no rewrite rule.
;;
;; NOTICE: In this mode the unit rewrites are gathered only for the next round, but this is
;; likely not necessary!
;;
;; saturate : procedure?
;; tptp-program : string?
;; rwtree-in : rewrite-tree?
;; discover-online? : boolean?
;; cpu-limit : number?
;; cpu-first-limit : number?
;; cpu-limit-factor? : number?
;; -> void?
(define (iterative-saturation saturate
#:! tptp-program
#:! rwtree-in
#:? [discover-online? (*discover-online?*)]
#:? [cpu-limit (*cpu-limit*)]
#:? [cpu-first-limit (*cpu-first-limit*)]
#:? [cpu-limit-factor (*cpu-limit-factor*)])
(define cpu-start (current-inexact-seconds))
; Don't make new Clauses here, they need to be created at each `saturation` call.
(define clauses (tptp-prog->clauses tptp-program))
(define quiet? (*quiet-json?*))
(define n-rules-init (rewrite-tree-count rwtree-in))
(let loop ([iter 1] [uncapped-current-cpu-limit cpu-first-limit] [rwtree-in rwtree-in])
(define remaining-cpu (- cpu-limit (- (current-inexact-seconds) cpu-start)))
(define current-cpu-limit (min remaining-cpu uncapped-current-cpu-limit))
(unless quiet?
(printf "; iter: ~a remaining-cpu: ~a current-cpu-limit: ~a\n"
iter
remaining-cpu
current-cpu-limit))
; Simplify the set of rules (only once)
(unless (and (= 1 iter)
(= 0 n-rules-init)) ; don't do this if no restarting
; Note that these steps destroy the Clause ancestry, and proofs will be incomplete.
(unless quiet?
(printf "; Rules stats: ~v\n" (rewrite-tree-stats rwtree-in))
(displayln "; Simplifying the rules via re-add-rules!"))
;; Rewrite lhs and rhs of rules, remove subsumed and tautologies.
(re-add-rules! rwtree-in)
(unless quiet?
(printf "; Rules stats: ~v\n" (rewrite-tree-stats rwtree-in))
(printf "; Confluence! bounded? = ~a\n" (*bounded-confluence?*)))
;; Unify rhs of rules to produce new rules.
(rewrite-tree-confluence! rwtree-in)
(unless quiet?
(printf "; Rules stats: ~v\n" (rewrite-tree-stats rwtree-in))
(displayln "; Simplifying the rules via re-add-rules! (again)"))
;; Rewrite and simplify again.
(re-add-rules! rwtree-in)
(unless quiet? (printf "; Rules stats: ~v\n" (rewrite-tree-stats rwtree-in))))
(flush-output)
(define rwtree-out (if discover-online? rwtree-in (rewrite-tree-shallow-copy rwtree-in)))
(define res (saturate #:clauses clauses
#:cpu-limit current-cpu-limit
#:rwtree-in rwtree-in
#:rwtree-out rwtree-out))
(define new-cumulative-cpu (- (current-inexact-seconds) cpu-start))
(set! res (dict-set res 'cumulative-time (exact-ceiling (* 1000. new-cumulative-cpu)))) ; ms
(set! res (dict-set res 'saturation-iter iter))
(define solved? (or (check-status res 'refuted)
(check-status res 'saturated)))
;; We exit also if memory limit has been reached, but we could instead restart
;; if new rules have been found.
(define finished? (or solved?
(check-status res 'memory)
(> new-cumulative-cpu cpu-limit)))
(cond
[(*quiet-json?*)
(when finished? (displayln (saturation-result->json res)))]
[else
(pretty-write res)])
(flush-output)
(cond
[finished?
(when (*output-rules*)
(unless quiet?
(printf "Saving rules to ~a\n"
(if (string? (*output-rules*))
(*output-rules*)
(path->string (*output-rules*)))))
(save-rules! rwtree-out #:rules-file (*output-rules*)))]
[else
(loop (+ iter 1)
(* uncapped-current-cpu-limit cpu-limit-factor)
rwtree-out)])))