/
queries_rewrite.h
1148 lines (923 loc) · 58.9 KB
/
queries_rewrite.h
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
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
#ifndef _HOME_SYSTEM_DEVELOPMENT_PROJECTS_TRINITY_QUERIES_REWRITE_H
#define _HOME_SYSTEM_DEVELOPMENT_PROJECTS_TRINITY_QUERIES_REWRITE_H
// Query rewrites is as important, maybe even more so, than a good ranking (precision) function.
// However, it's also important to consider the final query complexity. A simple query with say, 5-8 terms, can easily blow up to
// be extremely complex, many 100s of nodes in size, so you need to compromise and accept tradeoffs.
#pragma once
#include "queries.h"
#include <unordered_set>
namespace Trinity {
enum class RewriteFlags : uint8_t {
NoShallowCopy = 1
};
struct flow;
struct flow_ent final {
enum class Type : uint8_t {
Node,
Flow
} type;
union {
ast_node *n;
flow * f;
};
flow_ent(ast_node *const node)
: type{Type::Node}, n{node} {
}
flow_ent(flow *const _flow)
: type{Type::Flow}, f{_flow} {
}
inline ast_node *materialize(simple_allocator &a) const;
flow_ent()
: n{nullptr} {
}
flow_ent(const flow_ent &o) {
type = o.type;
n = o.n;
}
flow_ent(flow_ent &&o) {
type = o.type;
n = o.n;
}
auto &operator=(const flow_ent &o) {
type = o.type;
n = o.n;
return *this;
}
auto &operator=(flow_ent &&o) {
type = o.type;
n = o.n;
return *this;
}
};
struct gen_ctx;
struct flow final {
range32_t range;
flow * parent{nullptr};
Operator op{Operator::OR};
std::vector<flow_ent> ents;
void replace_child_flow(flow *from, flow *to) {
for (auto &it : ents) {
if (it.type == flow_ent::Type::Flow && it.f == from) {
it.f = to;
to->parent = this;
}
}
}
flow(flow &&o)
: range{o.range}, parent{o.parent}, op{o.op} {
ents = std::move(o.ents);
}
auto &operator=(flow &&o) {
range = o.range;
parent = o.parent;
op = o.op;
ents = std::move(o.ents);
return *this;
}
// with no contructor (e.g not even flow() = default;) it will break in so many ways
flow() = default;
inline bool overlaps(const range32_t range) const noexcept;
void push_back_flow(flow *const f) {
f->parent = this;
ents.emplace_back(f);
}
bool replace_self(flow *with) {
if (auto p = parent) {
for (size_t i{0}; i != p->ents.size(); ++i) {
if (p->ents[i].type == flow_ent::Type::Flow && p->ents[i].f == this) {
with->range.Set(UINT32_MAX, 0);
with->parent = parent;
p->ents[i].f = with;
for (auto &it : ents) {
if (it.type == flow_ent::Type::Flow && it.f->parent == this)
it.f->parent = with;
}
return true;
}
}
}
return false;
}
ast_node *materialize(simple_allocator &a) const {
[[maybe_unused]] static ast_node dummy{.type = ast_node::Type::Dummy};
if (const auto n = ents.size()) {
ast_node *lhs{nullptr};
for (size_t i{0}; i != n;) {
ast_node *node;
if (ents[i].type == flow_ent::Type::Flow) {
// sequences of flows in ents[] need to be combined with OR
auto localLHS = ents[i].materialize(a);
const auto op = ents[i].f->op;
for (++i; i != n && ents[i].type == flow_ent::Type::Flow && ents[i].f->op == op; ++i) {
auto b = ast_node::make_binop(a);
b->binop.op = op;
b->binop.lhs = localLHS;
b->binop.rhs = ents[i].materialize(a);
localLHS = b;
}
node = localLHS;
} else {
node = ents[i++].materialize(a);
}
if (!lhs)
lhs = node;
else {
auto bn = ast_node::make_binop(a);
bn->binop.op = Operator::AND;
bn->binop.lhs = lhs;
bn->binop.rhs = node;
lhs = bn;
}
}
return lhs ?: &dummy;
//return lhs ?: ast_node::make(a, ast_node::Type::Dummy);
}
return &dummy;
//return ast_node::make(a, ast_node::Type::Dummy);
}
void validate() const noexcept {
#if 0
std::size_t n{0};
for (auto it = this->parent; it; it = it->parent)
{
if (++n == 100)
std::abort();
}
#endif
}
inline void push_back_node(const std::pair<range32_t, ast_node *> p, gen_ctx &, std::vector<flow *> &, uint32_t &, const Operator op = Operator::AND);
};
inline flow *flow_for_node(const std::pair<range32_t, ast_node *> p, gen_ctx &, std::vector<flow *> &, uint32_t &);
struct gen_ctx final {
uint32_t logicalIndex;
uint32_t K;
simple_allocator *qa;
simple_allocator flowsAllocator{4096};
// Could have used just one container insted of two, and a map or multimap for tracking flows by range
// TODO(markp): we can just reuse allocatedFlows (reusable = std::move(allocatedFlows))
// (we shoudn't use a single std::vector<flow *> for all allocated flows and for tracking because
// we really only need to track 'useful' flows)
std::vector<flow *> allocatedFlows, flows, flows_1, flows_2;
std::unordered_set<flow *> S[2];
std::vector<void *> large_allocations;
~gen_ctx() {
while (!allocatedFlows.empty()) {
allocatedFlows.back()->~flow();
allocatedFlows.pop_back();
}
for (auto p : large_allocations)
std::free(p);
}
void prepare_run_capture() {
flows.clear();
while (!allocatedFlows.empty()) {
allocatedFlows.back()->~flow();
allocatedFlows.pop_back();
}
flowsAllocator.reuse();
}
template <typename... T>
auto new_flow(T &&... o) {
auto res = flowsAllocator.construct<flow>(std::forward<T>(o)...);
allocatedFlows.push_back(res);
return res;
}
void clear(const uint8_t _k) {
while (!allocatedFlows.empty()) {
allocatedFlows.back()->~flow();
allocatedFlows.pop_back();
}
flowsAllocator.reuse();
flows.clear();
flows_1.clear();
flows_2.clear();
K = _k;
}
};
// generate a list of expressions(ast_nodes) from a run, starting at (i), upto (i + maxSpan)
template <typename L>
static auto run_next(std::size_t &budget, query &q, const std::vector<ast_node *> &run, const uint32_t i, const uint8_t maxSpan, L &&l, gen_ctx &genCtx) {
static constexpr bool trace{false};
require(i < run.size());
const auto token = run[i]->p->terms[0].token;
static thread_local std::vector<std::pair<std::pair<str32_t, query_term_flags_t>, uint8_t>> vTLS;
auto & v{vTLS};
static thread_local std::vector<std::pair<str32_t, query_term_flags_t>> altsTLS;
auto & alts{altsTLS};
static thread_local simple_allocator altAllocatorInstanceTLS;
auto & altAllocator{altAllocatorInstanceTLS};
const auto normalizedMaxSpan = std::min<uint8_t>(maxSpan, genCtx.K);
strwlen8_t tokens[normalizedMaxSpan];
std::vector<std::pair<Trinity::ast_node *, uint8_t>> expressions;
auto tokensParser = q.tokensParser;
auto & allocator = q.allocator;
uint16_t base_app_phrase_id = run[i]->p->app_phrase_id;
const char * base_token_data = base_app_phrase_id ? token.data() : nullptr;
// TODO: maybe this needs to be configurable?
static constexpr bool inherit_app_phrase_id{true};
EXPECT(tokensParser);
if constexpr (trace)
SLog(ansifmt::bold, ansifmt::color_green, "AT ", i, " ", token, ansifmt::reset, " (maxSpan = ", maxSpan, "(", normalizedMaxSpan, "), budget = ", budget, ", base_app_phrase_id = ", base_app_phrase_id, ")\n");
alts.clear();
v.clear();
if (run[i]->p->rep > 1 || run[i]->p->flags || 0 == budget) {
// special care for reps
auto n = allocator.Alloc<ast_node>();
if constexpr (trace)
SLog("Special case, rep = ", run[i]->p->rep, " ", run[i]->p->flags, "\n");
n->type = ast_node::Type::Token;
n->p = run[i]->p;
expressions.emplace_back(n, 1);
if (budget && budget != std::numeric_limits<std::size_t>::max())
--budget;
return expressions;
}
// This may be handy. e.g for possibly generating composite terms
//WAS: const std::pair<uint32_t, uint32_t> runCtx(i, run.size());
const std::pair<uint32_t, const std::vector<ast_node *> *> runCtx(i, &run);
v.clear();
v.push_back({{{token.data(), uint32_t(token.size())}, 0}, 1});
altAllocator.reuse();
// Looks like we should do this in reverse
// i.e consider longer sequences before shorter sequences
// This facilitates considering longer sequences before shorter sequences in our callback
const std::size_t upto = std::min<std::size_t>(run.size(), i + normalizedMaxSpan);
std::size_t n{0};
for (auto end{i}; end != upto && run[end]->p->rep == 1; ++end)
tokens[n++] = run[end]->p->terms[0].token;
while (n) {
alts.clear();
l(runCtx, tokens, n, altAllocator, &alts);
if constexpr (trace)
SLog("FOR n = ", n, " => ", alts.size(), "\n");
for (const auto &it : alts)
v.push_back({{it.first, it.second}, n});
--n;
}
// Now with the new algo, we don't really need to group
// all alts in an OR group; we can just parse them and emit them directly one after the other into
// expressions[]. This could be potentially beneficial as well.
// TODO(markp): consider this implementation.
//
// This is important so that we will disregard dupes, and also, if e.g
// (united, states, of, america) => [usa]
// and (united, states) => [usa]
// we will ignore the second (united, states) rule because we already matched it earlier (we process by match span descending)
std::sort(v.begin(), v.end(), [](const auto &a, const auto &b) noexcept {
return b.second < a.second; // sort by span
});
expressions.clear();
alts.clear();
for (const auto *p = v.data(), *const e = p + v.size(); p != e;) {
const auto span = p->second;
ast_node * node;
const auto saved = alts.size();
if constexpr (trace)
SLog("For SPAN ", span, "\n");
do {
uint32_t k{0};
const auto s = p->first.first;
const auto F = p->first.second;
while (k != alts.size() && !(alts[k].second == F && alts[k].first == s))
++k;
if (k == alts.size()) {
// ignore duplicates or already seen
if constexpr (trace)
SLog("Accepting Alt [", s, "] ", p->first.second, "\n");
alts.emplace_back(s, p->first.second);
} else if constexpr (trace)
SLog("Ignoring Alt [", s, "] ", p->first.second, ": seen already\n");
} while (++p != e && p->second == span);
const auto n = alts.size() - saved;
if (0 == n)
continue;
static constexpr uint32_t parser_flags = unsigned(Trinity::ast_parser::Flags::ParseConstTrueExpr);
if (n > 1) {
auto lhs = ast_parser(alts[saved].first, allocator, tokensParser, parser_flags).parse();
if (unlikely(nullptr == lhs))
throw Switch::data_error("Failed to parse [", alts[saved].first, "]");
if (inherit_app_phrase_id) {
if (base_app_phrase_id) {
lhs->set_app_phrase_id(base_app_phrase_id);
}
} else if (alts[saved].first.data() == base_token_data) {
if (trace)
SLog("BASE token [", alts[saved].first, "] => ", base_app_phrase_id, "\n");
lhs->set_app_phrase_id(base_app_phrase_id);
}
if (lhs->type == ast_node::Type::Token && span > 1) {
// source range span > 1 and alt is a token
lhs->p->rewrite_ctx.srcSeqSize = span;
}
lhs->set_rewrite_translation_coeff(span);
if (budget && budget != std::numeric_limits<std::size_t>::max()) {
if (const auto n = lhs->nodes_count(); budget >= n)
budget -= n;
else
budget = 0;
}
if (const auto flags = alts[saved].second) {
if constexpr (trace)
SLog("FLAGS:", flags, "\n");
lhs->set_alltokens_flags(flags);
}
for (size_t i{1}; budget && i != n; ++i) {
auto n = ast_node::make_binop(allocator);
if (budget && budget != std::numeric_limits<std::size_t>::max())
--budget;
n->binop.op = Operator::OR;
n->binop.lhs = lhs;
n->binop.rhs = ast_parser(alts[i + saved].first, allocator, tokensParser, parser_flags).parse();
lhs = n;
if constexpr (trace)
SLog("[", alts[i + saved].first, "] to [", *n->binop.rhs, "]\n");
if (unlikely(nullptr == n->binop.rhs))
throw Switch::data_error("Failed to parse [", alts[i + saved].first, "]");
if (inherit_app_phrase_id) {
if (base_token_data) {
n->binop.rhs->set_app_phrase_id(base_app_phrase_id);
}
} else if (alts[i + saved].first.data() == base_token_data) {
if (trace) {
SLog("BASE token [", alts[i + saved].first, "] => ", base_app_phrase_id, "\n");
}
n->binop.rhs->set_app_phrase_id(base_app_phrase_id);
}
if (n->binop.rhs->type == ast_node::Type::Token && span > 1) {
n->binop.rhs->p->rewrite_ctx.srcSeqSize = span;
}
n->binop.rhs->set_rewrite_translation_coeff(span);
if (budget && budget != std::numeric_limits<std::size_t>::max()) {
if (const auto _n = n->binop.rhs->nodes_count(); budget >= _n)
budget -= _n;
else
budget = 0;
}
if (const auto flags = alts[i + saved].second) {
if constexpr (trace)
SLog("FLAGS:", flags, " for [", *n->binop.rhs, "]\n");
n->binop.rhs->set_alltokens_flags(flags);
}
if constexpr (trace)
SLog("CREATED:", *n, "\n");
}
node = lhs;
} else {
node = Trinity::ast_parser(alts[saved].first, allocator, tokensParser, parser_flags).parse();
if (unlikely(nullptr == node))
throw Switch::data_error("Failed to parse [", alts[saved].first, "], parser flags ", parser_flags);
if (inherit_app_phrase_id) {
if (base_app_phrase_id) {
node->set_app_phrase_id(base_app_phrase_id);
}
} else if (alts[saved].first.data() == base_token_data) {
if (trace)
SLog("BASE token [", alts[saved].first, "] => ", base_app_phrase_id, "\n");
node->set_app_phrase_id(base_app_phrase_id);
}
if (node->type == ast_node::Type::Token && span > 1) {
// source range span > 1 and alt is a token
node->p->rewrite_ctx.srcSeqSize = span;
}
node->set_rewrite_translation_coeff(span);
if constexpr (trace)
SLog("Parsed [", alts[saved], "] ", *node, "\n");
if (budget && budget != std::numeric_limits<std::size_t>::max()) {
if (const auto n = node->nodes_count(); budget >= n)
budget -= n;
else
budget = 0;
}
if (const auto flags = alts[saved].second) {
if constexpr (trace)
SLog("FLAGS:", flags, "\n");
node->set_alltokens_flags(flags);
}
}
expressions.emplace_back(node, span);
if constexpr (trace)
SLog(ansifmt::color_brown, "<<<<<< [", *node, "] ", span, ansifmt::reset, "\n");
}
if constexpr (trace) {
SLog(ansifmt::bold, ansifmt::color_blue, "CAPTURED EXPRESSIONS", ansifmt::reset, "\n");
for (const auto &it : expressions)
SLog(it.second, ":", *it.first, "\n");
}
return expressions;
}
// TODO(markp): all kinds of IMPLEMENT_ME(), also
// not sure the use of replace_self() makes sense. Need to reproduce the problem though
template <typename L>
static std::pair<ast_node *, uint8_t> run_capture(const uint32_t rewriteFlags,
std::size_t & budget,
query & q,
const std::vector<ast_node *> &run,
const uint32_t i,
L && l,
const uint8_t maxSpan,
gen_ctx & genCtx) {
static constexpr bool trace{false};
[[maybe_unused]] auto & allocator = q.allocator;
static thread_local std::vector<std::pair<range32_t, ast_node *>> list_tl;
auto & list{list_tl};
const auto baseIndex{i};
genCtx.prepare_run_capture();
// New faster, simpler, more efficient, and likely correct scheme
// Beats all past attempts and alternative schemes that required all kind of heuristics and recursion.
// 1. Collect all (range, tokens list)s in this run
list.clear();
for (uint32_t it{i}, cnt = run.size(); it != cnt; ++it) {
static constexpr bool assignRanges{true};
auto e = run_next(budget, q, run, it, maxSpan, l, genCtx);
const auto normalized = it - baseIndex;
for (const auto &eit : e) {
auto node{eit.first};
if (assignRanges) {
const range_base<uint16_t, uint8_t> rewriteRange{uint16_t(genCtx.logicalIndex + it), uint8_t(eit.second)};
node->set_rewrite_range(rewriteRange);
}
list.push_back({{normalized, eit.second}, node}); // range [offset, end) => ast_node
}
}
// This is what makes everything work.
// Sorting the list _properly_
// Figuring out the right order is extremely important; it's what make this scheme design possible
std::sort(list.begin(), list.end(), [](const auto &a, const auto &b) noexcept {
return a.first.offset < b.first.offset || (a.first.offset == b.first.offset && a.first.stop() < b.first.stop()); // SORT METHOD #4 (start asc, stop asc)
});
if (trace) {
for (size_t i{0}; i != list.size(); ++i) {
const auto &it = list[i];
SLog(ansifmt::bold, ansifmt::color_brown, "LIST: ", i, ": ", it.first, ":", ansifmt::reset, *it.second, "\n");
}
}
auto & flows = genCtx.flows;
uint32_t maxStop{0};
auto & S = genCtx.S;
const auto find_flows_by_range = [&](const auto range, auto *const v1, auto *const v2) {
v1->clear();
v2->clear();
if (trace) {
SLog("Finding flows for range ", range, " among ", flows.size(), "\n");
for (const auto f : flows)
SLog("FLOW:", f->range, "\n");
}
for (const auto f : flows) {
f->validate();
if (f->range.offset == range.offset)
v1->push_back(f);
else if (f->range.stop() == range.offset)
v2->push_back(f);
}
};
[[maybe_unused]] const auto consider_stop = [&](const uint32_t stop) noexcept {
maxStop = std::max<uint32_t>(stop, maxStop);
};
// Fast-simple way to identify the common ancestor among a list of flows[]
// TODO(markp): we can probably avoid using std::unordered_set<> here which would make sense, but let's consider alternatives later
[[maybe_unused]] const auto common_anchestor = [&](flow **const flows, const size_t cnt, const bool oneFlowUseParent = false) -> flow * {
if (cnt == 0)
return nullptr;
else if (cnt == 1) {
if (oneFlowUseParent)
return flows[0]->parent ?: flows[0];
else
return flows[0];
}
uint16_t active{0};
const auto penultimate{cnt - 1};
flows[0]->validate();
S[0].clear();
S[1].clear();
for (auto it{flows[0]}; it; it = it->parent)
S[0].insert(it);
for (size_t i{1}; i != penultimate; ++i) {
flows[i]->validate();
for (auto it{flows[i]}; it; it = it->parent) {
if (S[active].count(it))
S[1 - active].insert(it);
}
S[active].clear();
active = 1 - active;
}
for (auto it{flows[penultimate]}; it; it = it->parent) {
if (S[active].count(it))
return it;
}
return nullptr;
};
flows.clear();
// Create a root here
// This is not strictly required(we could check if there is a root, and if not, assign the first flow to the root)
// but this is a good idea, because we won't need to do that, and because we can rely on it to be the common ancestor for most flows/paths
auto root = genCtx.new_flow();
auto &atOffset = genCtx.flows_1;
auto &atStop = genCtx.flows_2;
root->range.Set(UINT32_MAX, 0);
flows.push_back(root);
for (const auto p : list) {
atOffset.clear();
atStop.clear();
find_flows_by_range(p.first, &atOffset, &atStop);
if (trace) {
SLog("\n\n", ansifmt::bold, ansifmt::color_green, "Processing ", p.first, ansifmt::reset, " ", *p.second, " => maxStop = ", maxStop, " (atOffset.size = ", atOffset.size(), ", atStop.size = ", atStop.size(), ") (", flows.size(), " flows)\n");
SLog("root:", *root->materialize(*genCtx.qa), ": ", *root, "\n");
#if 0
for (const auto f : flows)
SLog("Registered ", f->range, ":(", f->op, ", parent:", ptr_repr(f->parent), ", self:", ptr_repr(f), ") ", *f->materialize(*genCtx.qa), "\n");
#endif
}
if (atOffset.empty()) {
if (atStop.empty()) {
[[maybe_unused]] auto nf = flow_for_node(p, genCtx, flows, maxStop);
root->push_back_flow(nf);
if (trace) {
SLog("Registered first\n");
}
} else {
if (atStop.size() == 1) {
auto front = atStop.front();
if (trace) {
SLog("front:", *front, "\n");
for (auto p = front->parent; p; p = p->parent)
SLog("up:", *p, "\n");
}
[[maybe_unused]] auto nf = flow_for_node(p, genCtx, flows, maxStop);
if (trace) {
SLog("Will just append to a flow ", ptr_repr(atStop.front()), "\n");
SLog("Before:", *atStop.front(), "\n");
}
nf->op = Operator::AND;
atStop.front()->push_back_flow(nf);
if (trace) {
SLog("NOW:", *atStop.front(), "\n");
SLog("root:", *root->materialize(*genCtx.qa), "\n");
SLog("hierarchy after appended\n");
for (auto p = nf->parent; p; p = p->parent)
SLog("up:", *p, "\n");
}
} else {
// atOffset.empty == true && atStop.size() > 1
if (trace) {
SLog("Candidates\n");
for (auto f : atStop) {
SLog(*f->materialize(*genCtx.qa), " ", *f, "\n");
for (auto p = f->parent; p; p = p->parent)
SLog("up:", *p, "\n");
}
}
// this needs to be intelligent enoguh
// i.e it should accept the common ancestor from: (MACBOOKPRO OR (MAC BOOK OR MACBOOK)) -- for PRO
// but not from: (WORLD ((OF (WAR OR WARCRAFT)) OR OFWAR)) -- for CRAFT
// If we can figure this out, we are probably fine.
// Maybe we should check common ancestor, if there is another flow with range.stop() >= p.stop()
// XXX: breaks down for mac book pro lap top
// need to consider overlap for each candidate
if (auto ac = common_anchestor(atStop.data(), atStop.size(), true); ac && false == ac->overlaps(p.first)) {
if (trace) {
SLog("common:", *ac->materialize(*genCtx.qa), "\n");
}
[[maybe_unused]] auto nf = flow_for_node(p, genCtx, flows, maxStop);
nf->op = Operator::AND;
ac->push_back_flow(nf);
if (trace) {
SLog("AC now:", *ac->materialize(*genCtx.qa), "\n");
}
} else {
if (trace) {
if (ac) {
SLog("Have common ancestor[", *ac->materialize(*genCtx.qa), "] but overlaps\n");
Print("\n\n");
} else
SLog("No common ancestor\n");
}
for (auto f : atStop) {
// XXX: we are making shallow copies here, which mean that the token's phrase may be shared among
// 2+ nodes, which can cause problems, if you are e.g stemming all tokens after you have rewritten the query because
// you will first stem the token and then because the phrase is shared with another node, you will end up attemtping to stem
// the now already stemmed token.
// For now, just deal with it in your application because shallow copies are so much cheaper -- and dealing with it in your app. is trivial
// TODO(markp): expose a flag that controls the semantics(shallow or full copies)
// UPDATE: now controlled via flags
auto *const clone = (rewriteFlags & unsigned(RewriteFlags::NoShallowCopy))
? p.second->copy(genCtx.qa, &genCtx.large_allocations)
: p.second->shallow_copy(genCtx.qa, &genCtx.large_allocations);
f->push_back_node({p.first, clone}, genCtx, flows, maxStop, Operator::AND);
}
}
if (trace) {
SLog("root:", *root->materialize(*genCtx.qa), "\n");
}
}
}
} else {
if (atStop.empty()) // atOffset.empty() == false && atStop.empty() == true
{
[[maybe_unused]] auto nf = flow_for_node(p, genCtx, flows, maxStop);
auto ca = common_anchestor(atOffset.data(), atOffset.size(), true);
if (ca) {
if (trace) {
for (const auto f : atOffset)
SLog("Candidate:", *f->materialize(*genCtx.qa), "\n");
SLog("Will merge, common ancestor:", ptr_repr(ca), " ", *ca->materialize(*genCtx.qa), " ", ca->ents.size(), ": ", *ca, "\n");
}
} else {
if (trace) {
SLog("no common ancestor\n");
for (const auto f : atOffset)
SLog(*f->materialize(*genCtx.qa), "\n");
}
IMPLEMENT_ME();
}
if (atOffset.size() == 1) {
auto pg = genCtx.new_flow();
auto g = genCtx.new_flow();
auto first = atOffset.front();
pg->push_back_flow(g);
if (first->replace_self(pg)) {
}
first->op = Operator::OR;
nf->op = Operator::OR;
g->push_back_flow(first);
g->push_back_flow(nf);
if (trace) {
SLog("Created container:", *pg->materialize(*genCtx.qa), "\n");
SLog("g = ", ptr_repr(g), ", pg = ", ptr_repr(pg), "\n");
for (auto f : atOffset) {
SLog("flow of ", ptr_repr(f), " ", *f, "\n");
for (auto p = f->parent; p; p = p->parent)
SLog(ptr_repr(p), ":", *p, "\n");
}
}
if (trace) {
auto ac = common_anchestor(atOffset.data(), atOffset.size(), true);
SLog("ac = ", ptr_repr(ac), "\n");
require(ac == g);
}
} else {
if (trace) {
for (auto f : atOffset) {
SLog("flow ", ptr_repr(f), ": ", *f, "\n");
for (auto p = f->parent; p; p = p->parent)
SLog(ptr_repr(p), ":", *p, "\n");
}
Print("\n\n");
}
auto g = genCtx.new_flow();
if (auto p = ca->parent)
p->replace_child_flow(ca, g);
g->op = ca->op;
g->push_back_flow(ca);
g->push_back_flow(nf);
ca->op = nf->op = Operator::OR;
if (trace) {
for (auto f : atOffset) {
SLog("flow ", ptr_repr(f), " ", *f, "\n");
for (auto p = f->parent; p; p = p->parent)
SLog(ptr_repr(p), ":", *p, "\n");
}
auto ac = common_anchestor(atOffset.data(), atOffset.size(), true);
SLog("ac = ", ptr_repr(ac), ":", *ac, "\n");
}
}
} else // false == atOffset.empty() && false == atStop.empty()
{
// FIXME:
// if (atOffset.size() > 1)
// this breaks for e.g [play station 4 video games] and [key board micro soft]
[[maybe_unused]] auto nf = flow_for_node(p, genCtx, flows, maxStop);
if (trace)
SLog("atOffset.size = ", atOffset.size(), "\n");
if (trace) {
for (auto f : atOffset) {
SLog("atOffset:", *f->materialize(*genCtx.qa), "\n");
for (auto p = f->parent; p; p = p->parent)
SLog("up:", *p, "\n");
}
for (auto f : atStop) {
SLog("atStop:", *f->materialize(*genCtx.qa), "\n");
for (auto p = f->parent; p; p = p->parent)
SLog("up:", *p, "\n");
}
}
if (atOffset.size() > 1) {
// This is really tricky
// for now, we 'll just append, and this works fine
for (auto it : atOffset) {
auto g = genCtx.new_flow();
auto nf = flow_for_node(p, genCtx, flows, maxStop);
if (auto p = it->parent)
p->replace_child_flow(it, g);
g->op = it->op;
g->push_back_flow(it);
g->push_back_flow(nf);
it->op = nf->op = Operator::OR;
}
if (trace) {
SLog("root now:", *root->materialize(*genCtx.qa), "\n");
}
} else {
auto ca = common_anchestor(atOffset.data(), atOffset.size(), false);
auto g = genCtx.new_flow();
if (trace) {
if (ca) {
SLog("CA = ", *ca->materialize(*genCtx.qa), " ", *ca, ", overlaps:", ca->overlaps(p.first), "\n");
} else {
SLog("ca = nullptr\n");
exit(1);
}
} else if (!ca) {
IMPLEMENT_ME();
}
if (auto p = ca->parent)
p->replace_child_flow(ca, g);
g->op = ca->op;
g->push_back_flow(ca);
g->push_back_flow(nf);
ca->op = nf->op = Operator::OR;
if (trace) {
for (auto f : atOffset) {
SLog("atOffset:", *f->materialize(*genCtx.qa), "\n");
for (auto p = f->parent; p; p = p->parent)
SLog("up:", *p, "\n");
}
for (auto f : atStop) {
SLog("atStop:", *f->materialize(*genCtx.qa), "\n");
for (auto p = f->parent; p; p = p->parent)
SLog("up:", *p, "\n");
}
SLog("root ", *root->materialize(*genCtx.qa), "\n");
}
}
}
}
}
auto res = root->materialize(*genCtx.qa);
#if 0
SLog("ROOT:", *root, "\n");
SLog(ansifmt::bold, ansifmt::color_green, "FINAL:", ansifmt::reset, *res, "\n"); exit(0);
#endif
return {res, run.size()}; // we process the whole run
}
// Very handy utility function that faciliaties query rewrites
// It generates optimal structures, and it is optimised for performance.
// `K` is the span of tokens you want to consider for each run. The lambda
// `l` will be passed upto that many tokens.
//
// Example:
/*
*
Trinity::rewrite_query(inputQuery, 3, [](const auto runCtx, const auto tokens, const auto cnt, auto &allocator, auto out)
{
if (cnt == 1)
{
if (tokens[0].Eq(_S("PS4"))
out->push_back({"PLAYSTATION AND 4", 1});
}
else if (cnt == 2)
{
if (tokens[0].Eq(_S("MAC")) && tokens[1].Eq(_S("BOOK")))
out->push_back({"MACBOOK", 1});
}
else if (cnt == 3)
{
if (tokens[0].Eq(_S("WORLD")) && tokens[1].Eq(_S("OF")) || tokens[2].Eq(_S("WARCRAFT")))
out->push_back({"WOW", 1});
}}
}
You probably want another runs pass in order to e.g convert all stop word nodes to
<stopword> so that <the> becomes optional.
Currently, the budget is an approximate threshold which controls our run sequence=>expressions extrapolation.
If it's set to std::numeric_limits<std::size_t>::max(), it's not taken into account, otherwise, for every expression we 'll compute the number
of nodes it is comprised of, and deduct that from the budget; once the budget has reached 0, we will not attempt to match a sequence to multiple alternatives.
TODO: In a future revision, budget semantics should be implemented in run_capture() where this makes a lot more sense, because we are bound to rewrite, extend, duplicate etc there
and that would be more appropriate, and accurate, to adjust the budget there.
*
*/
static inline void dummy_rcb(const std::vector<ast_node *> &) {
}
template <typename L, typename RCB = void (*)(const std::vector<ast_node *> &)>
void rewrite_query(const uint32_t rewriteFlags, Trinity::query &q,
std::size_t budget,
const uint8_t K,
L && l,
RCB && rcb = dummy_rcb) {
static constexpr bool trace{false};
if (!q) {