File: DBIHacks.pm

package info (click to toggle)
libdbix-class-perl 0.082844-1
  • links: PTS, VCS
  • area: main
  • in suites: forky, sid, trixie
  • size: 5,320 kB
  • sloc: perl: 27,215; sql: 322; sh: 29; makefile: 16
file content (1489 lines) | stat: -rw-r--r-- 48,398 bytes parent folder | download | duplicates (2)
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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
package   #hide from PAUSE
  DBIx::Class::Storage::DBIHacks;

#
# This module contains code supporting a battery of special cases and tests for
# many corner cases pushing the envelope of what DBIC can do. When work on
# these utilities began in mid 2009 (51a296b402c) it wasn't immediately obvious
# that these pieces, despite their misleading on-first-sighe-flakiness, will
# become part of the generic query rewriting machinery of DBIC, allowing it to
# both generate and process queries representing incredibly complex sets with
# reasonable efficiency.
#
# Now (end of 2019), more than 10 years later the routines in this class have
# stabilized enough, and are meticulously covered with tests, to a point where
# an effort to formalize them into user-facing APIs might be worthwhile.
#
# An implementor working on publicizing and/or replacing the routines with a
# more modern SQL generation framework should keep in mind that pretty much all
# existing tests are constructed on the basis of real-world code used in
# production somewhere.
#
# Please hack on this responsibly ;)
#

use strict;
use warnings;

use base 'DBIx::Class::Storage';
use mro 'c3';

use Scalar::Util 'blessed';
use DBIx::Class::_Util qw(UNRESOLVABLE_CONDITION serialize);
use SQL::Abstract::Util qw(is_plain_value is_literal_value);
use DBIx::Class::Carp;
use namespace::clean;

#
# This code will remove non-selecting/non-restricting joins from
# {from} specs, aiding the RDBMS query optimizer
#
sub _prune_unused_joins {
  my ($self, $attrs) = @_;

  # only standard {from} specs are supported, and we could be disabled in general
  return ($attrs->{from}, {})  unless (
    ref $attrs->{from} eq 'ARRAY'
      and
    @{$attrs->{from}} > 1
      and
    ref $attrs->{from}[0] eq 'HASH'
      and
    ref $attrs->{from}[1] eq 'ARRAY'
      and
    $self->_use_join_optimizer
  );

  my $orig_aliastypes = $self->_resolve_aliastypes_from_select_args($attrs);

  my $new_aliastypes = { %$orig_aliastypes };

  # we will be recreating this entirely
  my @reclassify = 'joining';

  # a grouped set will not be affected by amount of rows. Thus any
  # purely multiplicator classifications can go
  # (will be reintroduced below if needed by something else)
  push @reclassify, qw(multiplying premultiplied)
    if $attrs->{_force_prune_multiplying_joins} or $attrs->{group_by};

  # nuke what will be recalculated
  delete @{$new_aliastypes}{@reclassify};

  my @newfrom = $attrs->{from}[0]; # FROM head is always present

  # recalculate what we need once the multipliers are potentially gone
  # ignore premultiplies, since they do not add any value to anything
  my %need_joins;
  for ( @{$new_aliastypes}{grep { $_ ne 'premultiplied' } keys %$new_aliastypes }) {
    # add all requested aliases
    $need_joins{$_} = 1 for keys %$_;

    # add all their parents (as per joinpath which is an AoH { table => alias })
    $need_joins{$_} = 1 for map { values %$_ } map { @{$_->{-parents}} } values %$_;
  }

  for my $j (@{$attrs->{from}}[1..$#{$attrs->{from}}]) {
    push @newfrom, $j if (
      (! defined $j->[0]{-alias}) # legacy crap
        ||
      $need_joins{$j->[0]{-alias}}
    );
  }

  # we have a new set of joiners - for everything we nuked pull the classification
  # off the original stack
  for my $ctype (@reclassify) {
    $new_aliastypes->{$ctype} = { map
      { $need_joins{$_} ? ( $_ => $orig_aliastypes->{$ctype}{$_} ) : () }
      keys %{$orig_aliastypes->{$ctype}}
    }
  }

  return ( \@newfrom, $new_aliastypes );
}

#
# This is the code producing joined subqueries like:
# SELECT me.*, other.* FROM ( SELECT me.* FROM ... ) JOIN other ON ...
#
sub _adjust_select_args_for_complex_prefetch {
  my ($self, $attrs) = @_;

  $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute') unless (
    ref $attrs->{from} eq 'ARRAY'
      and
    @{$attrs->{from}} > 1
      and
    ref $attrs->{from}[0] eq 'HASH'
      and
    ref $attrs->{from}[1] eq 'ARRAY'
  );

  my $root_alias = $attrs->{alias};

  # generate inner/outer attribute lists, remove stuff that doesn't apply
  my $outer_attrs = { %$attrs };
  delete @{$outer_attrs}{qw(from bind rows offset group_by _grouped_by_distinct having)};

  my $inner_attrs = { %$attrs, _simple_passthrough_construction => 1 };
  delete @{$inner_attrs}{qw(for collapse select as)};

  # there is no point of ordering the insides if there is no limit
  delete $inner_attrs->{order_by} if (
    delete $inner_attrs->{_order_is_artificial}
      or
    ! $inner_attrs->{rows}
  );

  # generate the inner/outer select lists
  # for inside we consider only stuff *not* brought in by the prefetch
  # on the outside we substitute any function for its alias
  $outer_attrs->{select} = [ @{$attrs->{select}} ];

  my ($root_node, $root_node_offset);

  for my $i (0 .. $#{$inner_attrs->{from}}) {
    my $node = $inner_attrs->{from}[$i];
    my $h = (ref $node eq 'HASH')                                ? $node
          : (ref $node  eq 'ARRAY' and ref $node->[0] eq 'HASH') ? $node->[0]
          : next
    ;

    if ( ($h->{-alias}||'') eq $root_alias and $h->{-rsrc} ) {
      $root_node = $h;
      $root_node_offset = $i;
      last;
    }
  }

  $self->throw_exception ('Complex prefetches are not supported on resultsets with a custom from attribute')
    unless $root_node;

  # use the heavy duty resolver to take care of aliased/nonaliased naming
  my $colinfo = $self->_resolve_column_info($inner_attrs->{from});
  my $selected_root_columns;

  for my $i (0 .. $#{$outer_attrs->{select}}) {
    my $sel = $outer_attrs->{select}->[$i];

    next if (
      $colinfo->{$sel} and $colinfo->{$sel}{-source_alias} ne $root_alias
    );

    if (ref $sel eq 'HASH' ) {
      $sel->{-as} ||= $attrs->{as}[$i];
      $outer_attrs->{select}->[$i] = join ('.', $root_alias, ($sel->{-as} || "inner_column_$i") );
    }
    elsif (! ref $sel and my $ci = $colinfo->{$sel}) {
      $selected_root_columns->{$ci->{-colname}} = 1;
    }

    push @{$inner_attrs->{select}}, $sel;

    push @{$inner_attrs->{as}}, $attrs->{as}[$i];
  }

  # We will need to fetch all native columns in the inner subquery, which may
  # be a part of an *outer* join condition, or an order_by (which needs to be
  # preserved outside), or wheres. In other words everything but the inner
  # selector
  # We can not just fetch everything because a potential has_many restricting
  # join collapse *will not work* on heavy data types.
  my $connecting_aliastypes = $self->_resolve_aliastypes_from_select_args({
    %$inner_attrs,
    select => [],
  });

  for (sort map { keys %{$_->{-seen_columns}||{}} } map { values %$_ } values %$connecting_aliastypes) {
    my $ci = $colinfo->{$_} or next;
    if (
      $ci->{-source_alias} eq $root_alias
        and
      ! $selected_root_columns->{$ci->{-colname}}++
    ) {
      # adding it to both to keep limits not supporting dark selectors happy
      push @{$inner_attrs->{select}}, $ci->{-fq_colname};
      push @{$inner_attrs->{as}}, $ci->{-fq_colname};
    }
  }

  # construct the inner {from} and lock it in a subquery
  # we need to prune first, because this will determine if we need a group_by below
  # throw away all non-selecting, non-restricting multijoins
  # (since we def. do not care about multiplication of the contents of the subquery)
  my $inner_subq = do {

    # must use it here regardless of user requests (vastly gentler on optimizer)
    local $self->{_use_join_optimizer} = 1;

    # throw away multijoins since we def. do not care about those inside the subquery
    ($inner_attrs->{from}, my $inner_aliastypes) = $self->_prune_unused_joins ({
      %$inner_attrs, _force_prune_multiplying_joins => 1
    });

    # uh-oh a multiplier (which is not us) left in, this is a problem for limits
    # we will need to add a group_by to collapse the resultset for proper counts
    if (
      grep { $_ ne $root_alias } keys %{ $inner_aliastypes->{multiplying} || {} }
        and
      # if there are user-supplied groups - assume user knows wtf they are up to
      ( ! $inner_aliastypes->{grouping} or $inner_attrs->{_grouped_by_distinct} )
    ) {

      my $cur_sel = { map { $_ => 1 } @{$inner_attrs->{select}} };

      # *possibly* supplement the main selection with pks if not already
      # there, as they will have to be a part of the group_by to collapse
      # things properly
      my $inner_select_with_extras;
      my @pks = map { "$root_alias.$_" } $root_node->{-rsrc}->primary_columns
        or $self->throw_exception( sprintf
          'Unable to perform complex limited prefetch off %s without declared primary key',
          $root_node->{-rsrc}->source_name,
        );
      for my $col (@pks) {
        push @{ $inner_select_with_extras ||= [ @{$inner_attrs->{select}} ] }, $col
          unless $cur_sel->{$col}++;
      }

      ($inner_attrs->{group_by}, $inner_attrs->{order_by}) = $self->_group_over_selection({
        %$inner_attrs,
        $inner_select_with_extras ? ( select => $inner_select_with_extras ) : (),
        _aliastypes => $inner_aliastypes,
      });
    }

    # we already optimized $inner_attrs->{from} above
    # and already local()ized
    $self->{_use_join_optimizer} = 0;

    # generate the subquery
    $self->_select_args_to_query (
      @{$inner_attrs}{qw(from select where)},
      $inner_attrs,
    );
  };

  # Generate the outer from - this is relatively easy (really just replace
  # the join slot with the subquery), with a major caveat - we can not
  # join anything that is non-selecting (not part of the prefetch), but at
  # the same time is a multi-type relationship, as it will explode the result.
  #
  # There are two possibilities here
  # - either the join is non-restricting, in which case we simply throw it away
  # - it is part of the restrictions, in which case we need to collapse the outer
  #   result by tackling yet another group_by to the outside of the query

  # work on a shallow copy
  my @orig_from = @{$attrs->{from}};


  $outer_attrs->{from} = \ my @outer_from;

  # we may not be the head
  if ($root_node_offset) {
    # first generate the outer_from, up to the substitution point
    @outer_from = splice @orig_from, 0, $root_node_offset;

    # substitute the subq at the right spot
    push @outer_from, [
      {
        -alias => $root_alias,
        -rsrc => $root_node->{-rsrc},
        $root_alias => $inner_subq,
      },
      # preserve attrs from what is now the head of the from after the splice
      @{$orig_from[0]}[1 .. $#{$orig_from[0]}],
    ];
  }
  else {
    @outer_from = {
      -alias => $root_alias,
      -rsrc => $root_node->{-rsrc},
      $root_alias => $inner_subq,
    };
  }

  shift @orig_from; # what we just replaced above

  # scan the *remaining* from spec against different attributes, and see which joins are needed
  # in what role
  my $outer_aliastypes = $outer_attrs->{_aliastypes} =
    $self->_resolve_aliastypes_from_select_args({ %$outer_attrs, from => \@orig_from });

  # unroll parents
  my ($outer_select_chain, @outer_nonselecting_chains) = map { +{
    map { $_ => 1 } map { values %$_} map { @{$_->{-parents}} } values %{ $outer_aliastypes->{$_} || {} }
  } } qw/selecting restricting grouping ordering/;

  # see what's left - throw away if not selecting/restricting
  my $may_need_outer_group_by;
  while (my $j = shift @orig_from) {
    my $alias = $j->[0]{-alias};

    if (
      $outer_select_chain->{$alias}
    ) {
      push @outer_from, $j
    }
    elsif (grep { $_->{$alias} } @outer_nonselecting_chains ) {
      push @outer_from, $j;
      $may_need_outer_group_by ||= $outer_aliastypes->{multiplying}{$alias} ? 1 : 0;
    }
  }

  # also throw in a synthetic group_by if a non-selecting multiplier,
  # to guard against cross-join explosions
  # the logic is somewhat fragile, but relies on the idea that if a user supplied
  # a group by on their own - they know what they were doing
  if ( $may_need_outer_group_by and $attrs->{_grouped_by_distinct} ) {
    ($outer_attrs->{group_by}, $outer_attrs->{order_by}) = $self->_group_over_selection ({
      %$outer_attrs,
      from => \@outer_from,
    });
  }

  # FIXME: The {where} ends up in both the inner and outer query, i.e. *twice*
  #
  # This is rather horrific, and while we currently *do* have enough
  # introspection tooling available to attempt a stab at properly deciding
  # whether or not to include the where condition on the outside, the
  # machinery is still too slow to apply it here.
  # Thus for the time being we do not attempt any sanitation of the where
  # clause and just pass it through on both sides of the subquery. This *will*
  # be addressed at a later stage, most likely after folding the SQL generator
  # into SQLMaker proper
  #
  # OTOH it can be seen as a plus: <ash> (notes that this query would make a DBA cry ;)
  #
  return $outer_attrs;
}

#
# This is probably the ickiest, yet most relied upon part of the codebase:
# this is the place where we take arbitrary SQL input and break it into its
# constituent parts, making sure we know which *sources* are used in what
# *capacity* ( selecting / restricting / grouping / ordering / joining, etc. )
# Although the method is pretty horrific, the worst thing that can happen is
# for a classification failure, which in turn will result in a vocal exception,
# and will lead to a relatively prompt fix.
# The code has been slowly improving and is covered with a formiddable battery
# of tests, so can be considered "reliably stable" at this point (Oct 2015).
#
# A note to implementors attempting to "replace" this - keep in mind that while
# there are multiple optimization avenues, the actual "scan literal elements"
# part *MAY NEVER BE REMOVED*, even if in the future it is limited to only AST
# nodes that are deemed opaque (i.e. contain literal expressions). The use and
# comprehension of blackbox literals is at this point firmly a user-facing API,
# and is one of *the* reasons DBIC remains as flexible as it is.
#
# In other words, when working on this keep in mind that the following is both
# a widespread and *encouraged* way of using DBIC in the wild when push comes
# to shove:
#
# $rs->search( {}, {
#   select => \[ $random, @stuff],
#   from => \[ $random, @stuff ],
#   where => \[ $random, @stuff ],
#   group_by => \[ $random, @stuff ],
#   order_by => \[ $random, @stuff ],
# } )
#
# Various incarnations of the above are reflected in many of the tests. If one
# gets to fail, or if a user complains: you get to fix it. A stance amounting
# to "this is crazy, nobody does that" is not acceptable going forward.
#
sub _resolve_aliastypes_from_select_args {
  my ( $self, $attrs ) = @_;

  $self->throw_exception ('Unable to analyze custom {from}')
    if ref $attrs->{from} ne 'ARRAY';

  # what we will return
  my $aliases_by_type;

  # see what aliases are there to work with
  # and record who is a multiplier and who is premultiplied
  my $alias_list;
  for my $node (@{$attrs->{from}}) {

    my $j = $node;
    $j = $j->[0] if ref $j eq 'ARRAY';
    my $al = $j->{-alias}
      or next;

    $alias_list->{$al} = $j;

    $aliases_by_type->{multiplying}{$al} ||= { -parents => $j->{-join_path}||[] }
      # not array == {from} head == can't be multiplying
      if ref($node) eq 'ARRAY' and ! $j->{-is_single};

    $aliases_by_type->{premultiplied}{$al} ||= { -parents => $j->{-join_path}||[] }
      # parts of the path that are not us but are multiplying
      if grep { $aliases_by_type->{multiplying}{$_} }
          grep { $_ ne $al }
           map { values %$_ }
            @{ $j->{-join_path}||[] }
  }

  # get a column to source/alias map (including unambiguous unqualified ones)
  my $colinfo = $self->_resolve_column_info ($attrs->{from});

  # set up a botched SQLMaker
  my $sql_maker = $self->sql_maker;

  # these are throw away results, do not pollute the bind stack
  local $sql_maker->{where_bind};
  local $sql_maker->{group_bind};
  local $sql_maker->{having_bind};
  local $sql_maker->{from_bind};

  # we can't scan properly without any quoting (\b doesn't cut it
  # everywhere), so unless there is proper quoting set - use our
  # own weird impossible character.
  # Also in the case of no quoting, we need to explicitly disable
  # name_sep, otherwise sorry nasty legacy syntax like
  # { 'count(foo.id)' => { '>' => 3 } } will stop working >:(
  local $sql_maker->{quote_char} = $sql_maker->{quote_char};
  local $sql_maker->{name_sep} = $sql_maker->{name_sep};

  unless (defined $sql_maker->{quote_char} and length $sql_maker->{quote_char}) {
    $sql_maker->{quote_char} = ["\x00", "\xFF"];
    # if we don't unset it we screw up retarded but unfortunately working
    # 'MAX(foo.bar)' => { '>', 3 }
    $sql_maker->{name_sep} = '';
  }

  my ($lquote, $rquote, $sep) = map { quotemeta $_ } ($sql_maker->_quote_chars, $sql_maker->name_sep);

  # generate sql chunks
  my $to_scan = {
    restricting => [
      ($sql_maker->_recurse_where ($attrs->{where}))[0],
      $sql_maker->_parse_rs_attrs ({ having => $attrs->{having} }),
    ],
    grouping => [
      $sql_maker->_parse_rs_attrs ({ group_by => $attrs->{group_by} }),
    ],
    joining => [
      $sql_maker->_recurse_from (
        ref $attrs->{from}[0] eq 'ARRAY' ? $attrs->{from}[0][0] : $attrs->{from}[0],
        @{$attrs->{from}}[1 .. $#{$attrs->{from}}],
      ),
    ],
    selecting => [
      map { ($sql_maker->_recurse_fields($_))[0] } @{$attrs->{select}},
    ],
    ordering => [
      map { $_->[0] } $self->_extract_order_criteria ($attrs->{order_by}, $sql_maker),
    ],
  };

  # throw away empty chunks and all 2-value arrayrefs: the thinking is that these are
  # bind value specs left in by the sloppy renderer above. It is ok to do this
  # at this point, since we are going to end up rewriting this crap anyway
  for my $v (values %$to_scan) {
    my @nv;
    for (@$v) {
      next if (
        ! defined $_
          or
        (
          ref $_ eq 'ARRAY'
            and
          ( @$_ == 0 or @$_ == 2 )
        )
      );

      if (ref $_) {
        require Data::Dumper::Concise;
        $self->throw_exception("Unexpected ref in scan-plan: " . Data::Dumper::Concise::Dumper($v) );
      }

      push @nv, $_;
    }

    $v = \@nv;
  }

  # kill all selectors which look like a proper subquery
  # this is a sucky heuristic *BUT* - if we get it wrong the query will simply
  # fail to run, so we are relatively safe
  $to_scan->{selecting} = [ grep {
    $_ !~ / \A \s* \( \s* SELECT \s+ .+? \s+ FROM \s+ .+? \) \s* \z /xsi
  } @{ $to_scan->{selecting} || [] } ];

  # first see if we have any exact matches (qualified or unqualified)
  for my $type (keys %$to_scan) {
    for my $piece (@{$to_scan->{$type}}) {
      if ($colinfo->{$piece} and my $alias = $colinfo->{$piece}{-source_alias}) {
        $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
        $aliases_by_type->{$type}{$alias}{-seen_columns}{$colinfo->{$piece}{-fq_colname}} = $piece;
      }
    }
  }

  # now loop through all fully qualified columns and get the corresponding
  # alias (should work even if they are in scalarrefs)
  for my $alias (keys %$alias_list) {
    my $al_re = qr/
      $lquote $alias $rquote $sep (?: $lquote ([^$rquote]+) $rquote )?
        |
      \b $alias \. ([^\s\)\($rquote]+)?
    /x;

    for my $type (keys %$to_scan) {
      for my $piece (@{$to_scan->{$type}}) {
        if (my @matches = $piece =~ /$al_re/g) {
          $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
          $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = "$alias.$_"
            for grep { defined $_ } @matches;
        }
      }
    }
  }

  # now loop through unqualified column names, and try to locate them within
  # the chunks
  for my $col (keys %$colinfo) {
    next if $col =~ / \. /x;   # if column is qualified it was caught by the above

    my $col_re = qr/ $lquote ($col) $rquote /x;

    for my $type (keys %$to_scan) {
      for my $piece (@{$to_scan->{$type}}) {
        if ( my @matches = $piece =~ /$col_re/g) {
          my $alias = $colinfo->{$col}{-source_alias};
          $aliases_by_type->{$type}{$alias} ||= { -parents => $alias_list->{$alias}{-join_path}||[] };
          $aliases_by_type->{$type}{$alias}{-seen_columns}{"$alias.$_"} = $_
            for grep { defined $_ } @matches;
        }
      }
    }
  }

  # Add any non-left joins to the restriction list (such joins are indeed restrictions)
  for my $j (values %$alias_list) {
    my $alias = $j->{-alias} or next;
    $aliases_by_type->{restricting}{$alias} ||= { -parents => $j->{-join_path}||[] } if (
      (not $j->{-join_type})
        or
      ($j->{-join_type} !~ /^left (?: \s+ outer)? $/xi)
    );
  }

  for (keys %$aliases_by_type) {
    delete $aliases_by_type->{$_} unless keys %{$aliases_by_type->{$_}};
  }

  return $aliases_by_type;
}

# This is the engine behind { distinct => 1 } and the general
# complex prefetch grouper
sub _group_over_selection {
  my ($self, $attrs) = @_;

  my $colinfos = $self->_resolve_column_info ($attrs->{from});

  my (@group_by, %group_index);

  # the logic is: if it is a { func => val } we assume an aggregate,
  # otherwise if \'...' or \[...] we assume the user knows what is
  # going on thus group over it
  for (@{$attrs->{select}}) {
    if (! ref($_) or ref ($_) ne 'HASH' ) {
      push @group_by, $_;
      $group_index{$_}++;
      if ($colinfos->{$_} and $_ !~ /\./ ) {
        # add a fully qualified version as well
        $group_index{"$colinfos->{$_}{-source_alias}.$_"}++;
      }
    }
  }

  my @order_by = $self->_extract_order_criteria($attrs->{order_by})
    or return (\@group_by, $attrs->{order_by});

  # add any order_by parts that are not already present in the group_by
  # to maintain SQL cross-compatibility and general sanity
  #
  # also in case the original selection is *not* unique, or in case part
  # of the ORDER BY refers to a multiplier - we will need to replace the
  # skipped order_by elements with their MIN/MAX equivalents as to maintain
  # the proper overall order without polluting the group criteria (and
  # possibly changing the outcome entirely)

  my ($leftovers, $sql_maker, @new_order_by, $order_chunks, $aliastypes);

  my $group_already_unique = $self->_columns_comprise_identifying_set($colinfos, \@group_by);

  for my $o_idx (0 .. $#order_by) {

    # if the chunk is already a min/max function - there is nothing left to touch
    next if $order_by[$o_idx][0] =~ /^ (?: min | max ) \s* \( .+ \) $/ix;

    # only consider real columns (for functions the user got to do an explicit group_by)
    my $chunk_ci;
    if (
      @{$order_by[$o_idx]} != 1
        or
      # only declare an unknown *plain* identifier as "leftover" if we are called with
      # aliastypes to examine. If there are none - we are still in _resolve_attrs, and
      # can just assume the user knows what they want
      ( ! ( $chunk_ci = $colinfos->{$order_by[$o_idx][0]} ) and $attrs->{_aliastypes} )
    ) {
      push @$leftovers, $order_by[$o_idx][0];
    }

    next unless $chunk_ci;

    # no duplication of group criteria
    next if $group_index{$chunk_ci->{-fq_colname}};

    $aliastypes ||= (
      $attrs->{_aliastypes}
        or
      $self->_resolve_aliastypes_from_select_args({
        from => $attrs->{from},
        order_by => $attrs->{order_by},
      })
    ) if $group_already_unique;

    # check that we are not ordering by a multiplier (if a check is requested at all)
    if (
      $group_already_unique
        and
      ! $aliastypes->{multiplying}{$chunk_ci->{-source_alias}}
        and
      ! $aliastypes->{premultiplied}{$chunk_ci->{-source_alias}}
    ) {
      push @group_by, $chunk_ci->{-fq_colname};
      $group_index{$chunk_ci->{-fq_colname}}++
    }
    else {
      # We need to order by external columns without adding them to the group
      # (eiehter a non-unique selection, or a multi-external)
      #
      # This doesn't really make sense in SQL, however from DBICs point
      # of view is rather valid (e.g. order the leftmost objects by whatever
      # criteria and get the offset/rows many). There is a way around
      # this however in SQL - we simply tae the direction of each piece
      # of the external order and convert them to MIN(X) for ASC or MAX(X)
      # for DESC, and group_by the root columns. The end result should be
      # exactly what we expect
      #

      # both populated on the first loop over $o_idx
      $sql_maker ||= $self->sql_maker;
      $order_chunks ||= [
        map { ref $_ eq 'ARRAY' ? $_ : [ $_ ] } $sql_maker->_order_by_chunks($attrs->{order_by})
      ];

      my ($chunk, $is_desc) = $sql_maker->_split_order_chunk($order_chunks->[$o_idx][0]);

      $new_order_by[$o_idx] = \[
        sprintf( '%s( %s )%s',
          $self->_minmax_operator_for_datatype($chunk_ci->{data_type}, $is_desc),
          $chunk,
          ($is_desc ? ' DESC' : ''),
        ),
        @ {$order_chunks->[$o_idx]} [ 1 .. $#{$order_chunks->[$o_idx]} ]
      ];
    }
  }

  $self->throw_exception ( sprintf
    'Unable to programatically derive a required group_by from the supplied '
  . 'order_by criteria. To proceed either add an explicit group_by, or '
  . 'simplify your order_by to only include plain columns '
  . '(supplied order_by: %s)',
    join ', ', map { "'$_'" } @$leftovers,
  ) if $leftovers;

  # recreate the untouched order parts
  if (@new_order_by) {
    $new_order_by[$_] ||= \ $order_chunks->[$_] for ( 0 .. $#$order_chunks );
  }

  return (
    \@group_by,
    (@new_order_by ? \@new_order_by : $attrs->{order_by} ),  # same ref as original == unchanged
  );
}

sub _minmax_operator_for_datatype {
  #my ($self, $datatype, $want_max) = @_;

  $_[2] ? 'MAX' : 'MIN';
}

sub _resolve_ident_sources {
  my ($self, $ident) = @_;

  my $alias2source = {};

  # the reason this is so contrived is that $ident may be a {from}
  # structure, specifying multiple tables to join
  if ( blessed $ident && $ident->isa("DBIx::Class::ResultSource") ) {
    # this is compat mode for insert/update/delete which do not deal with aliases
    $alias2source->{me} = $ident;
  }
  elsif (ref $ident eq 'ARRAY') {

    for (@$ident) {
      my $tabinfo;
      if (ref $_ eq 'HASH') {
        $tabinfo = $_;
      }
      if (ref $_ eq 'ARRAY' and ref $_->[0] eq 'HASH') {
        $tabinfo = $_->[0];
      }

      $alias2source->{$tabinfo->{-alias}} = $tabinfo->{-rsrc}
        if ($tabinfo->{-rsrc});
    }
  }

  return $alias2source;
}

# Takes $ident, \@column_names
#
# returns { $column_name => \%column_info, ... }
# also note: this adds -result_source => $rsrc to the column info
#
# If no columns_names are supplied returns info about *all* columns
# for all sources
sub _resolve_column_info {
  my ($self, $ident, $colnames) = @_;

  return {} if $colnames and ! @$colnames;

  my $alias2src = $self->_resolve_ident_sources($ident);

  my (%seen_cols, @auto_colnames);

  # compile a global list of column names, to be able to properly
  # disambiguate unqualified column names (if at all possible)
  for my $alias (keys %$alias2src) {
    my $rsrc = $alias2src->{$alias};
    for my $colname ($rsrc->columns) {
      push @{$seen_cols{$colname}}, $alias;
      push @auto_colnames, "$alias.$colname" unless $colnames;
    }
  }

  $colnames ||= [
    @auto_colnames,
    grep { @{$seen_cols{$_}} == 1 } (keys %seen_cols),
  ];

  my (%return, $colinfos);
  foreach my $col (@$colnames) {
    my ($source_alias, $colname) = $col =~ m/^ (?: ([^\.]+) \. )? (.+) $/x;

    # if the column was seen exactly once - we know which rsrc it came from
    $source_alias ||= $seen_cols{$colname}[0]
      if ($seen_cols{$colname} and @{$seen_cols{$colname}} == 1);

    next unless $source_alias;

    my $rsrc = $alias2src->{$source_alias}
      or next;

    $return{$col} = {
      %{
          ( $colinfos->{$source_alias} ||= $rsrc->columns_info )->{$colname}
            ||
          $self->throw_exception(
            "No such column '$colname' on source " . $rsrc->source_name
          );
      },
      -result_source => $rsrc,
      -source_alias => $source_alias,
      -fq_colname => $col eq $colname ? "$source_alias.$col" : $col,
      -colname => $colname,
    };

    $return{"$source_alias.$colname"} = $return{$col} if $col eq $colname;
  }

  return \%return;
}

# The DBIC relationship chaining implementation is pretty simple - every
# new related_relationship is pushed onto the {from} stack, and the {select}
# window simply slides further in. This means that when we count somewhere
# in the middle, we got to make sure that everything in the join chain is an
# actual inner join, otherwise the count will come back with unpredictable
# results (a resultset may be generated with _some_ rows regardless of if
# the relation which the $rs currently selects has rows or not). E.g.
# $artist_rs->cds->count - normally generates:
# SELECT COUNT( * ) FROM artist me LEFT JOIN cd cds ON cds.artist = me.artistid
# which actually returns the number of artists * (number of cds || 1)
#
# So what we do here is crawl {from}, determine if the current alias is at
# the top of the stack, and if not - make sure the chain is inner-joined down
# to the root.
#
sub _inner_join_to_node {
  my ($self, $from, $alias) = @_;

  my $switch_branch = $self->_find_join_path_to_node($from, $alias);

  return $from unless @{$switch_branch||[]};

  # So it looks like we will have to switch some stuff around.
  # local() is useless here as we will be leaving the scope
  # anyway, and deep cloning is just too fucking expensive
  # So replace the first hashref in the node arrayref manually
  my @new_from = ($from->[0]);
  my $sw_idx = { map { (values %$_), 1 } @$switch_branch }; #there's one k/v per join-path

  for my $j (@{$from}[1 .. $#$from]) {
    my $jalias = $j->[0]{-alias};

    if ($sw_idx->{$jalias}) {
      my %attrs = %{$j->[0]};
      delete $attrs{-join_type};
      push @new_from, [
        \%attrs,
        @{$j}[ 1 .. $#$j ],
      ];
    }
    else {
      push @new_from, $j;
    }
  }

  return \@new_from;
}

sub _find_join_path_to_node {
  my ($self, $from, $target_alias) = @_;

  # subqueries and other oddness are naturally not supported
  return undef if (
    ref $from ne 'ARRAY'
      ||
    ref $from->[0] ne 'HASH'
      ||
    ! defined $from->[0]{-alias}
  );

  # no path - the head is the alias
  return [] if $from->[0]{-alias} eq $target_alias;

  for my $i (1 .. $#$from) {
    return $from->[$i][0]{-join_path} if ( ($from->[$i][0]{-alias}||'') eq $target_alias );
  }

  # something else went quite wrong
  return undef;
}

sub _extract_order_criteria {
  my ($self, $order_by, $sql_maker) = @_;

  my $parser = sub {
    my ($sql_maker, $order_by, $orig_quote_chars) = @_;

    return scalar $sql_maker->_order_by_chunks ($order_by)
      unless wantarray;

    my ($lq, $rq, $sep) = map { quotemeta($_) } (
      ($orig_quote_chars ? @$orig_quote_chars : $sql_maker->_quote_chars),
      $sql_maker->name_sep
    );

    my @chunks;
    for ($sql_maker->_order_by_chunks ($order_by) ) {
      my $chunk = ref $_ ? [ @$_ ] : [ $_ ];
      ($chunk->[0]) = $sql_maker->_split_order_chunk($chunk->[0]);

      # order criteria may have come back pre-quoted (literals and whatnot)
      # this is fragile, but the best we can currently do
      $chunk->[0] =~ s/^ $lq (.+?) $rq $sep $lq (.+?) $rq $/"$1.$2"/xe
        or $chunk->[0] =~ s/^ $lq (.+) $rq $/$1/x;

      push @chunks, $chunk;
    }

    return @chunks;
  };

  if ($sql_maker) {
    return $parser->($sql_maker, $order_by);
  }
  else {
    $sql_maker = $self->sql_maker;

    # pass these in to deal with literals coming from
    # the user or the deep guts of prefetch
    my $orig_quote_chars = [$sql_maker->_quote_chars];

    local $sql_maker->{quote_char};
    return $parser->($sql_maker, $order_by, $orig_quote_chars);
  }
}

sub _order_by_is_stable {
  my ($self, $ident, $order_by, $where) = @_;

  my @cols = (
    ( map { $_->[0] } $self->_extract_order_criteria($order_by) ),
    ( $where ? keys %{ $self->_extract_fixed_condition_columns($where) } : () ),
  ) or return 0;

  my $colinfo = $self->_resolve_column_info($ident, \@cols);

  return keys %$colinfo
    ? $self->_columns_comprise_identifying_set( $colinfo,  \@cols )
    : 0
  ;
}

sub _columns_comprise_identifying_set {
  my ($self, $colinfo, $columns) = @_;

  my $cols_per_src;
  $cols_per_src -> {$_->{-source_alias}} -> {$_->{-colname}} = $_
    for grep { defined $_ } @{$colinfo}{@$columns};

  for (values %$cols_per_src) {
    my $src = (values %$_)[0]->{-result_source};
    return 1 if $src->_identifying_column_set($_);
  }

  return 0;
}

# this is almost similar to _order_by_is_stable, except it takes
# a single rsrc, and will succeed only if the first portion of the order
# by is stable.
# returns that portion as a colinfo hashref on success
sub _extract_colinfo_of_stable_main_source_order_by_portion {
  my ($self, $attrs) = @_;

  my $nodes = $self->_find_join_path_to_node($attrs->{from}, $attrs->{alias});

  return unless defined $nodes;

  my @ord_cols = map
    { $_->[0] }
    ( $self->_extract_order_criteria($attrs->{order_by}) )
  ;
  return unless @ord_cols;

  my $valid_aliases = { map { $_ => 1 } (
    $attrs->{from}[0]{-alias},
    map { values %$_ } @$nodes,
  ) };

  my $colinfos = $self->_resolve_column_info($attrs->{from});

  my ($colinfos_to_return, $seen_main_src_cols);

  for my $col (@ord_cols) {
    # if order criteria is unresolvable - there is nothing we can do
    my $colinfo = $colinfos->{$col} or last;

    # if we reached the end of the allowed aliases - also nothing we can do
    last unless $valid_aliases->{$colinfo->{-source_alias}};

    $colinfos_to_return->{$col} = $colinfo;

    $seen_main_src_cols->{$colinfo->{-colname}} = 1
      if $colinfo->{-source_alias} eq $attrs->{alias};
  }

  # FIXME the condition may be singling out things on its own, so we
  # conceivable could come back wi "stable-ordered by nothing"
  # not confient enough in the parser yet, so punt for the time being
  return unless $seen_main_src_cols;

  my $main_src_fixed_cols_from_cond = [ $attrs->{where}
    ? (
      map
      {
        ( $colinfos->{$_} and $colinfos->{$_}{-source_alias} eq $attrs->{alias} )
          ? $colinfos->{$_}{-colname}
          : ()
      }
      keys %{ $self->_extract_fixed_condition_columns($attrs->{where}) }
    )
    : ()
  ];

  return $attrs->{result_source}->_identifying_column_set([
    keys %$seen_main_src_cols,
    @$main_src_fixed_cols_from_cond,
  ]) ? $colinfos_to_return : ();
}

# Attempts to flatten a passed in SQLAC condition as much as possible towards
# a plain hashref, *without* altering its semantics. Required by
# create/populate being able to extract definitive conditions from preexisting
# resultset {where} stacks
#
# FIXME - while relatively robust, this is still imperfect, one of the first
# things to tackle when we get access to a formalized AST. Note that this code
# is covered by a *ridiculous* amount of tests, so starting with porting this
# code would be a rather good exercise
sub _collapse_cond {
  my ($self, $where, $where_is_anded_array) = @_;

  my $fin;

  if (! $where) {
    return;
  }
  elsif ($where_is_anded_array or ref $where eq 'HASH') {

    my @pairs;

    my @pieces = $where_is_anded_array ? @$where : $where;
    while (@pieces) {
      my $chunk = shift @pieces;

      if (ref $chunk eq 'HASH') {
        for (sort keys %$chunk) {

          # Match SQLAC 1.79 behavior
          if ($_ eq '') {
            is_literal_value($chunk->{$_})
              ? carp 'Hash-pairs consisting of an empty string with a literal are deprecated, use -and => [ $literal ] instead'
              : $self->throw_exception("Supplying an empty left hand side argument is not supported in hash-pairs")
            ;
          }

          push @pairs, $_ => $chunk->{$_};
        }
      }
      elsif (ref $chunk eq 'ARRAY') {
        push @pairs, -or => $chunk
          if @$chunk;
      }
      elsif ( ! length ref $chunk) {

        # Match SQLAC 1.79 behavior
        $self->throw_exception("Supplying an empty left hand side argument is not supported in array-pairs")
          if $where_is_anded_array and (! defined $chunk or $chunk eq '');

        push @pairs, $chunk, shift @pieces;
      }
      else {
        push @pairs, '', $chunk;
      }
    }

    return unless @pairs;

    my @conds = $self->_collapse_cond_unroll_pairs(\@pairs)
      or return;

    # Consolidate various @conds back into something more compact
    for my $c (@conds) {
      if (ref $c ne 'HASH') {
        push @{$fin->{-and}}, $c;
      }
      else {
        for my $col (sort keys %$c) {

          # consolidate all -and nodes
          if ($col =~ /^\-and$/i) {
            push @{$fin->{-and}},
              ref $c->{$col} eq 'ARRAY' ? @{$c->{$col}}
            : ref $c->{$col} eq 'HASH' ? %{$c->{$col}}
            : { $col => $c->{$col} }
            ;
          }
          elsif ($col =~ /^\-/) {
            push @{$fin->{-and}}, { $col => $c->{$col} };
          }
          elsif (exists $fin->{$col}) {
            $fin->{$col} = [ -and => map {
              (ref $_ eq 'ARRAY' and ($_->[0]||'') =~ /^\-and$/i )
                ? @{$_}[1..$#$_]
                : $_
              ;
            } ($fin->{$col}, $c->{$col}) ];
          }
          else {
            $fin->{$col} = $c->{$col};
          }
        }
      }
    }
  }
  elsif (ref $where eq 'ARRAY') {
    # we are always at top-level here, it is safe to dump empty *standalone* pieces
    my $fin_idx;

    for (my $i = 0; $i <= $#$where; $i++ ) {

      # Match SQLAC 1.79 behavior
      $self->throw_exception(
        "Supplying an empty left hand side argument is not supported in array-pairs"
      ) if (! defined $where->[$i] or ! length $where->[$i]);

      my $logic_mod = lc ( ($where->[$i] =~ /^(\-(?:and|or))$/i)[0] || '' );

      if ($logic_mod) {
        $i++;
        $self->throw_exception("Unsupported top-level op/arg pair: [ $logic_mod => $where->[$i] ]")
          unless ref $where->[$i] eq 'HASH' or ref $where->[$i] eq 'ARRAY';

        my $sub_elt = $self->_collapse_cond({ $logic_mod => $where->[$i] })
          or next;

        my @keys = keys %$sub_elt;
        if ( @keys == 1 and $keys[0] !~ /^\-/ ) {
          $fin_idx->{ "COL_$keys[0]_" . serialize $sub_elt } = $sub_elt;
        }
        else {
          $fin_idx->{ "SER_" . serialize $sub_elt } = $sub_elt;
        }
      }
      elsif (! length ref $where->[$i] ) {
        my $sub_elt = $self->_collapse_cond({ @{$where}[$i, $i+1] })
          or next;

        $fin_idx->{ "COL_$where->[$i]_" . serialize $sub_elt } = $sub_elt;
        $i++;
      }
      else {
        $fin_idx->{ "SER_" . serialize $where->[$i] } = $self->_collapse_cond( $where->[$i] ) || next;
      }
    }

    if (! $fin_idx) {
      return;
    }
    elsif ( keys %$fin_idx == 1 ) {
      $fin = (values %$fin_idx)[0];
    }
    else {
      my @or;

      # at this point everything is at most one level deep - unroll if needed
      for (sort keys %$fin_idx) {
        if ( ref $fin_idx->{$_} eq 'HASH' and keys %{$fin_idx->{$_}} == 1 ) {
          my ($l, $r) = %{$fin_idx->{$_}};

          if (
            ref $r eq 'ARRAY'
              and
            (
              ( @$r == 1 and $l =~ /^\-and$/i )
                or
              $l =~ /^\-or$/i
            )
          ) {
            push @or, @$r
          }

          elsif (
            ref $r eq 'HASH'
              and
            keys %$r == 1
              and
            $l =~ /^\-(?:and|or)$/i
          ) {
            push @or, %$r;
          }

          else {
            push @or, $l, $r;
          }
        }
        else {
          push @or, $fin_idx->{$_};
        }
      }

      $fin->{-or} = \@or;
    }
  }
  else {
    # not a hash not an array
    $fin = { -and => [ $where ] };
  }

  # unroll single-element -and's
  while (
    $fin->{-and}
      and
    @{$fin->{-and}} < 2
  ) {
    my $and = delete $fin->{-and};
    last if @$and == 0;

    # at this point we have @$and == 1
    if (
      ref $and->[0] eq 'HASH'
        and
      ! grep { exists $fin->{$_} } keys %{$and->[0]}
    ) {
      $fin = {
        %$fin, %{$and->[0]}
      };
    }
    else {
      $fin->{-and} = $and;
      last;
    }
  }

  # compress same-column conds found in $fin
  for my $col ( grep { $_ !~ /^\-/ } keys %$fin ) {
    next unless ref $fin->{$col} eq 'ARRAY' and ($fin->{$col}[0]||'') =~ /^\-and$/i;
    my $val_bag = { map {
      (! defined $_ )                          ? ( UNDEF => undef )
    : ( ! length ref $_ or is_plain_value $_ ) ? ( "VAL_$_" => $_ )
    : ( ( 'SER_' . serialize $_ ) => $_ )
    } @{$fin->{$col}}[1 .. $#{$fin->{$col}}] };

    if (keys %$val_bag == 1 ) {
      ($fin->{$col}) = values %$val_bag;
    }
    else {
      $fin->{$col} = [ -and => map { $val_bag->{$_} } sort keys %$val_bag ];
    }
  }

  return keys %$fin ? $fin : ();
}

sub _collapse_cond_unroll_pairs {
  my ($self, $pairs) = @_;

  my @conds;

  while (@$pairs) {
    my ($lhs, $rhs) = splice @$pairs, 0, 2;

    if ($lhs eq '') {
      push @conds, $self->_collapse_cond($rhs);
    }
    elsif ( $lhs =~ /^\-and$/i ) {
      push @conds, $self->_collapse_cond($rhs, (ref $rhs eq 'ARRAY'));
    }
    elsif ( $lhs =~ /^\-or$/i ) {
      push @conds, $self->_collapse_cond(
        (ref $rhs eq 'HASH') ? [ map { $_ => $rhs->{$_} } sort keys %$rhs ] : $rhs
      );
    }
    else {
      if (ref $rhs eq 'HASH' and ! keys %$rhs) {
        # FIXME - SQLAC seems to be doing... nothing...?
      }
      # normalize top level -ident, for saner extract_fixed_condition_columns code
      elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{-ident}) {
        push @conds, { $lhs => { '=', $rhs } };
      }
      # can't simply use is_plain_value result, as we need to
      # preserve the -value marker where necessary (non-blessed ref)
      elsif (
        ref $rhs eq 'HASH'
          and
        keys %$rhs == 1
          and
        exists $rhs->{-value}
          and
        (
          ! length ref( $rhs->{-value} )
            or
          (
            defined( blessed $rhs->{-value} )
              and
            is_plain_value $rhs->{-value}
          )
        )
      ) {
        push @conds, { $lhs => $rhs->{-value} };
      }
      elsif (ref $rhs eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}) {
        if ( length ref $rhs->{'='} and is_literal_value $rhs->{'='} ) {
          push @conds, { $lhs => $rhs };
        }
        else {
          for my $p ($self->_collapse_cond_unroll_pairs([ $lhs => $rhs->{'='} ])) {

            # extra sanity check
            if (keys %$p > 1) {
              require Data::Dumper::Concise;
              local $Data::Dumper::Deepcopy = 1;
              $self->throw_exception(
                "Internal error: unexpected collapse unroll:"
              . Data::Dumper::Concise::Dumper { in => { $lhs => $rhs }, out => $p }
              );
            }

            my ($l, $r) = %$p;

            push @conds, (
              ! length ref $r
                or
              # the unroller recursion may return a '=' prepended value already
              ref $r eq 'HASH' and keys %$rhs == 1 and exists $rhs->{'='}
                or
              is_plain_value($r)
            )
              ? { $l => $r }
              : { $l => { '=' => $r } }
            ;
          }
        }
      }
      elsif (ref $rhs eq 'ARRAY') {
        # some of these conditionals encounter multi-values - roll them out using
        # an unshift, which will cause extra looping in the while{} above
        if (! @$rhs ) {
          push @conds, { $lhs => [] };
        }
        elsif ( ($rhs->[0]||'') =~ /^\-(?:and|or)$/i ) {
          $self->throw_exception("Value modifier not followed by any values: $lhs => [ $rhs->[0] ] ")
            if  @$rhs == 1;

          if( $rhs->[0] =~ /^\-and$/i ) {
            unshift @$pairs, map { $lhs => $_ } @{$rhs}[1..$#$rhs];
          }
          # if not an AND then it's an OR
          elsif(@$rhs == 2) {
            unshift @$pairs, $lhs => $rhs->[1];
          }
          else {
            push @conds, { $lhs => [ @{$rhs}[1..$#$rhs] ] };
          }
        }
        elsif (@$rhs == 1) {
          unshift @$pairs, $lhs => $rhs->[0];
        }
        else {
          push @conds, { $lhs => $rhs };
        }
      }
      # unroll func + { -value => ... }
      # can't simply use is_plain_value result, as we need to
      # preserve the -value marker where necessary (non-blessed ref)
      elsif (
        ref $rhs eq 'HASH'
          and
        ( my ($subop) = keys %$rhs ) == 1
          and
        ref( (values %$rhs)[0] ) eq 'HASH'
          and
        keys %{ (values %$rhs)[0] } == 1
          and
        exists( (values %$rhs)[0]->{-value} )
          and
        (
          ! length ref( (values %$rhs)[0]->{-value} )
            or
          (
            defined( blessed( (values %$rhs)[0]->{-value} ) )
              and
            is_plain_value( (values %$rhs)[0]->{-value} )
          )
        )
      ) {
        push @conds, { $lhs => { $subop => (values %$rhs)[0]->{-value} } };
      }
      else {
        push @conds, { $lhs => $rhs };
      }
    }
  }

  return @conds;
}

# Analyzes a given condition and attempts to extract all columns
# with a definitive fixed-condition criteria. Returns a hashref
# of k/v pairs suitable to be passed to set_columns(), with a
# MAJOR CAVEAT - multi-value (contradictory) equalities are still
# represented as a reference to the UNRESOVABLE_CONDITION constant
# The reason we do this is that some codepaths only care about the
# codition being stable, as opposed to actually making sense
#
# The normal mode is used to figure out if a resultset is constrained
# to a column which is part of a unique constraint, which in turn
# allows us to better predict how ordering will behave etc.
#
# With the optional "consider_nulls" boolean argument, the function
# is instead used to infer inambiguous values from conditions
# (e.g. the inheritance of resultset conditions on new_result)
#
sub _extract_fixed_condition_columns {
  my ($self, $where, $consider_nulls) = @_;
  my $where_hash = $self->_collapse_cond($_[1]);

  my $res = {};
  my ($c, $v);
  for $c (keys %$where_hash) {
    my $vals;

    if (!defined ($v = $where_hash->{$c}) ) {
      $vals->{UNDEF} = $v if $consider_nulls
    }
    elsif (
      ref $v eq 'HASH'
        and
      keys %$v == 1
    ) {
      if (exists $v->{-value}) {
        if (defined $v->{-value}) {
          $vals->{"VAL_$v->{-value}"} = $v->{-value}
        }
        elsif( $consider_nulls ) {
          $vals->{UNDEF} = $v->{-value};
        }
      }
      # do not need to check for plain values - _collapse_cond did it for us
      elsif(
        length ref $v->{'='}
          and
        (
          ( ref $v->{'='} eq 'HASH' and keys %{$v->{'='}} == 1 and exists $v->{'='}{-ident} )
            or
          is_literal_value($v->{'='})
        )
       ) {
        $vals->{ 'SER_' . serialize $v->{'='} } = $v->{'='};
      }
    }
    elsif (
      ! length ref $v
        or
      is_plain_value ($v)
    ) {
      $vals->{"VAL_$v"} = $v;
    }
    elsif (ref $v eq 'ARRAY' and ($v->[0]||'') eq '-and') {
      for ( @{$v}[1..$#$v] ) {
        my $subval = $self->_extract_fixed_condition_columns({ $c => $_ }, 'consider nulls');  # always fish nulls out on recursion
        next unless exists $subval->{$c};  # didn't find anything
        $vals->{
          ! defined $subval->{$c}                                        ? 'UNDEF'
        : ( ! length ref $subval->{$c} or is_plain_value $subval->{$c} ) ? "VAL_$subval->{$c}"
        : ( 'SER_' . serialize $subval->{$c} )
        } = $subval->{$c};
      }
    }

    if (keys %$vals == 1) {
      ($res->{$c}) = (values %$vals)
        unless !$consider_nulls and exists $vals->{UNDEF};
    }
    elsif (keys %$vals > 1) {
      $res->{$c} = UNRESOLVABLE_CONDITION;
    }
  }

  $res;
}

1;