summaryrefslogtreecommitdiffstats
path: root/intl/icu/source/i18n/collationdatabuilder.cpp
blob: 7c6f1b881e6b1e99ad7c8cd0fd37be8689372f36 (plain)
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
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
// © 2016 and later: Unicode, Inc. and others.
// License & terms of use: http://www.unicode.org/copyright.html
/*
*******************************************************************************
* Copyright (C) 2012-2015, International Business Machines
* Corporation and others.  All Rights Reserved.
*******************************************************************************
* collationdatabuilder.cpp
*
* (replaced the former ucol_elm.cpp)
*
* created on: 2012apr01
* created by: Markus W. Scherer
*/

#include "unicode/utypes.h"

#if !UCONFIG_NO_COLLATION

#include "unicode/localpointer.h"
#include "unicode/uchar.h"
#include "unicode/ucharstrie.h"
#include "unicode/ucharstriebuilder.h"
#include "unicode/uniset.h"
#include "unicode/unistr.h"
#include "unicode/usetiter.h"
#include "unicode/utf16.h"
#include "cmemory.h"
#include "collation.h"
#include "collationdata.h"
#include "collationdatabuilder.h"
#include "collationfastlatinbuilder.h"
#include "collationiterator.h"
#include "normalizer2impl.h"
#include "utrie2.h"
#include "uvectr32.h"
#include "uvectr64.h"
#include "uvector.h"

U_NAMESPACE_BEGIN

CollationDataBuilder::CEModifier::~CEModifier() {}

/**
 * Build-time context and CE32 for a code point.
 * If a code point has contextual mappings, then the default (no-context) mapping
 * and all conditional mappings are stored in a singly-linked list
 * of ConditionalCE32, sorted by context strings.
 *
 * Context strings sort by prefix length, then by prefix, then by contraction suffix.
 * Context strings must be unique and in ascending order.
 */
struct ConditionalCE32 : public UMemory {
    ConditionalCE32()
            : context(),
              ce32(0), defaultCE32(Collation::NO_CE32), builtCE32(Collation::NO_CE32),
              next(-1) {}
    ConditionalCE32(const UnicodeString &ct, uint32_t ce)
            : context(ct),
              ce32(ce), defaultCE32(Collation::NO_CE32), builtCE32(Collation::NO_CE32),
              next(-1) {}

    inline UBool hasContext() const { return context.length() > 1; }
    inline int32_t prefixLength() const { return context.charAt(0); }

    /**
     * "\0" for the first entry for any code point, with its default CE32.
     *
     * Otherwise one unit with the length of the prefix string,
     * then the prefix string, then the contraction suffix.
     */
    UnicodeString context;
    /**
     * CE32 for the code point and its context.
     * Can be special (e.g., for an expansion) but not contextual (prefix or contraction tag).
     */
    uint32_t ce32;
    /**
     * Default CE32 for all contexts with this same prefix.
     * Initially NO_CE32. Set only while building runtime data structures,
     * and only on one of the nodes of a sub-list with the same prefix.
     */
    uint32_t defaultCE32;
    /**
     * CE32 for the built contexts.
     * When fetching CEs from the builder, the contexts are built into their runtime form
     * so that the normal collation implementation can process them.
     * The result is cached in the list head. It is reset when the contexts are modified.
     * All of these builtCE32 are invalidated by clearContexts(),
     * via incrementing the contextsEra.
     */
    uint32_t builtCE32;
    /**
     * The "era" of building intermediate contexts when the above builtCE32 was set.
     * When the array of cached, temporary contexts overflows, then clearContexts()
     * removes them all and invalidates the builtCE32 that used to point to built tries.
     */
    int32_t era = -1;
    /**
     * Index of the next ConditionalCE32.
     * Negative for the end of the list.
     */
    int32_t next;
    // Note: We could create a separate class for all of the contextual mappings for
    // a code point, with the builtCE32, the era, and a list of the actual mappings.
    // The class that represents one mapping would then not need to
    // store those fields in each element.
};

U_CDECL_BEGIN

void U_CALLCONV
uprv_deleteConditionalCE32(void *obj) {
    delete static_cast<ConditionalCE32 *>(obj);
}

U_CDECL_END

/**
 * Build-time collation element and character iterator.
 * Uses the runtime CollationIterator for fetching CEs for a string
 * but reads from the builder's unfinished data structures.
 * In particular, this class reads from the unfinished trie
 * and has to avoid CollationIterator::nextCE() and redirect other
 * calls to data->getCE32() and data->getCE32FromSupplementary().
 *
 * We do this so that we need not implement the collation algorithm
 * again for the builder and make it behave exactly like the runtime code.
 * That would be more difficult to test and maintain than this indirection.
 *
 * Some CE32 tags (for example, the DIGIT_TAG) do not occur in the builder data,
 * so the data accesses from those code paths need not be modified.
 *
 * This class iterates directly over whole code points
 * so that the CollationIterator does not need the finished trie
 * for handling the LEAD_SURROGATE_TAG.
 */
class DataBuilderCollationIterator : public CollationIterator {
public:
    DataBuilderCollationIterator(CollationDataBuilder &b);

    virtual ~DataBuilderCollationIterator();

    int32_t fetchCEs(const UnicodeString &str, int32_t start, int64_t ces[], int32_t cesLength);

    virtual void resetToOffset(int32_t newOffset) override;
    virtual int32_t getOffset() const override;

    virtual UChar32 nextCodePoint(UErrorCode &errorCode) override;
    virtual UChar32 previousCodePoint(UErrorCode &errorCode) override;

protected:
    virtual void forwardNumCodePoints(int32_t num, UErrorCode &errorCode) override;
    virtual void backwardNumCodePoints(int32_t num, UErrorCode &errorCode) override;

    virtual uint32_t getDataCE32(UChar32 c) const override;
    virtual uint32_t getCE32FromBuilderData(uint32_t ce32, UErrorCode &errorCode) override;

    CollationDataBuilder &builder;
    CollationData builderData;
    uint32_t jamoCE32s[CollationData::JAMO_CE32S_LENGTH];
    const UnicodeString *s;
    int32_t pos;
};

DataBuilderCollationIterator::DataBuilderCollationIterator(CollationDataBuilder &b)
        : CollationIterator(&builderData, /*numeric=*/ false),
          builder(b), builderData(b.nfcImpl),
          s(nullptr), pos(0) {
    builderData.base = builder.base;
    // Set all of the jamoCE32s[] to indirection CE32s.
    for(int32_t j = 0; j < CollationData::JAMO_CE32S_LENGTH; ++j) {  // Count across Jamo types.
        UChar32 jamo = CollationDataBuilder::jamoCpFromIndex(j);
        jamoCE32s[j] = Collation::makeCE32FromTagAndIndex(Collation::BUILDER_DATA_TAG, jamo) |
                CollationDataBuilder::IS_BUILDER_JAMO_CE32;
    }
    builderData.jamoCE32s = jamoCE32s;
}

DataBuilderCollationIterator::~DataBuilderCollationIterator() {}

int32_t
DataBuilderCollationIterator::fetchCEs(const UnicodeString &str, int32_t start,
                                       int64_t ces[], int32_t cesLength) {
    // Set the pointers each time, in case they changed due to reallocation.
    builderData.ce32s = reinterpret_cast<const uint32_t *>(builder.ce32s.getBuffer());
    builderData.ces = builder.ce64s.getBuffer();
    builderData.contexts = builder.contexts.getBuffer();
    // Modified copy of CollationIterator::nextCE() and CollationIterator::nextCEFromCE32().
    reset();
    s = &str;
    pos = start;
    UErrorCode errorCode = U_ZERO_ERROR;
    while(U_SUCCESS(errorCode) && pos < s->length()) {
        // No need to keep all CEs in the iterator buffer.
        clearCEs();
        UChar32 c = s->char32At(pos);
        pos += U16_LENGTH(c);
        uint32_t ce32 = utrie2_get32(builder.trie, c);
        const CollationData *d;
        if(ce32 == Collation::FALLBACK_CE32) {
            d = builder.base;
            ce32 = builder.base->getCE32(c);
        } else {
            d = &builderData;
        }
        appendCEsFromCE32(d, c, ce32, /*forward=*/ true, errorCode);
        U_ASSERT(U_SUCCESS(errorCode));
        for(int32_t i = 0; i < getCEsLength(); ++i) {
            int64_t ce = getCE(i);
            if(ce != 0) {
                if(cesLength < Collation::MAX_EXPANSION_LENGTH) {
                    ces[cesLength] = ce;
                }
                ++cesLength;
            }
        }
    }
    return cesLength;
}

void
DataBuilderCollationIterator::resetToOffset(int32_t newOffset) {
    reset();
    pos = newOffset;
}

int32_t
DataBuilderCollationIterator::getOffset() const {
    return pos;
}

UChar32
DataBuilderCollationIterator::nextCodePoint(UErrorCode & /*errorCode*/) {
    if(pos == s->length()) {
        return U_SENTINEL;
    }
    UChar32 c = s->char32At(pos);
    pos += U16_LENGTH(c);
    return c;
}

UChar32
DataBuilderCollationIterator::previousCodePoint(UErrorCode & /*errorCode*/) {
    if(pos == 0) {
        return U_SENTINEL;
    }
    UChar32 c = s->char32At(pos - 1);
    pos -= U16_LENGTH(c);
    return c;
}

void
DataBuilderCollationIterator::forwardNumCodePoints(int32_t num, UErrorCode & /*errorCode*/) {
    pos = s->moveIndex32(pos, num);
}

void
DataBuilderCollationIterator::backwardNumCodePoints(int32_t num, UErrorCode & /*errorCode*/) {
    pos = s->moveIndex32(pos, -num);
}

uint32_t
DataBuilderCollationIterator::getDataCE32(UChar32 c) const {
    return utrie2_get32(builder.trie, c);
}

uint32_t
DataBuilderCollationIterator::getCE32FromBuilderData(uint32_t ce32, UErrorCode &errorCode) {
    if (U_FAILURE(errorCode)) { return 0; }
    U_ASSERT(Collation::hasCE32Tag(ce32, Collation::BUILDER_DATA_TAG));
    if((ce32 & CollationDataBuilder::IS_BUILDER_JAMO_CE32) != 0) {
        UChar32 jamo = Collation::indexFromCE32(ce32);
        return utrie2_get32(builder.trie, jamo);
    } else {
        ConditionalCE32 *cond = builder.getConditionalCE32ForCE32(ce32);
        if (cond == nullptr) {
            errorCode = U_INTERNAL_PROGRAM_ERROR;
            // TODO: ICU-21531 figure out why this happens.
            return 0;
        }
        if(cond->builtCE32 == Collation::NO_CE32 || cond->era != builder.contextsEra) {
            // Build the context-sensitive mappings into their runtime form and cache the result.
            cond->builtCE32 = builder.buildContext(cond, errorCode);
            if(errorCode == U_BUFFER_OVERFLOW_ERROR) {
                errorCode = U_ZERO_ERROR;
                builder.clearContexts();
                cond->builtCE32 = builder.buildContext(cond, errorCode);
            }
            cond->era = builder.contextsEra;
            builderData.contexts = builder.contexts.getBuffer();
        }
        return cond->builtCE32;
    }
}

// ------------------------------------------------------------------------- ***

CollationDataBuilder::CollationDataBuilder(UBool icu4xMode, UErrorCode &errorCode)
        : nfcImpl(*Normalizer2Factory::getNFCImpl(errorCode)),
          base(nullptr), baseSettings(nullptr),
          trie(nullptr),
          ce32s(errorCode), ce64s(errorCode), conditionalCE32s(errorCode),
          modified(false),
          icu4xMode(icu4xMode),
          fastLatinEnabled(false), fastLatinBuilder(nullptr),
          collIter(nullptr) {
    // Reserve the first CE32 for U+0000.
    if (!icu4xMode) {
        ce32s.addElement(0, errorCode);
    }
    conditionalCE32s.setDeleter(uprv_deleteConditionalCE32);
}

CollationDataBuilder::~CollationDataBuilder() {
    utrie2_close(trie);
    delete fastLatinBuilder;
    delete collIter;
}

void
CollationDataBuilder::initForTailoring(const CollationData *b, UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return; }
    if(trie != nullptr) {
        errorCode = U_INVALID_STATE_ERROR;
        return;
    }
    if(b == nullptr) {
        errorCode = U_ILLEGAL_ARGUMENT_ERROR;
        return;
    }
    base = b;

    // For a tailoring, the default is to fall back to the base.
    // For ICU4X, use the same value for fallback as for the default
    // to avoid having to have different blocks for the two.
    trie = utrie2_open(Collation::FALLBACK_CE32, icu4xMode ? Collation::FALLBACK_CE32 : Collation::FFFD_CE32, &errorCode);

    if (!icu4xMode) {
        // Set the Latin-1 letters block so that it is allocated first in the data array,
        // to try to improve locality of reference when sorting Latin-1 text.
        // Do not use utrie2_setRange32() since that will not actually allocate blocks
        // that are filled with the default value.
        // ASCII (0..7F) is already preallocated anyway.
        for(UChar32 c = 0xc0; c <= 0xff; ++c) {
            utrie2_set32(trie, c, Collation::FALLBACK_CE32, &errorCode);
        }

        // Hangul syllables are not tailorable (except via tailoring Jamos).
        // Always set the Hangul tag to help performance.
        // Do this here, rather than in buildMappings(),
        // so that we see the HANGUL_TAG in various assertions.
        uint32_t hangulCE32 = Collation::makeCE32FromTagAndIndex(Collation::HANGUL_TAG, 0);
        utrie2_setRange32(trie, Hangul::HANGUL_BASE, Hangul::HANGUL_END, hangulCE32, true, &errorCode);

        // Copy the set contents but don't copy/clone the set as a whole because
        // that would copy the isFrozen state too.
        unsafeBackwardSet.addAll(*b->unsafeBackwardSet);
    }

    if(U_FAILURE(errorCode)) { return; }
}

UBool
CollationDataBuilder::maybeSetPrimaryRange(UChar32 start, UChar32 end,
                                           uint32_t primary, int32_t step,
                                           UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return false; }
    U_ASSERT(start <= end);
    // TODO: Do we need to check what values are currently set for start..end?
    // An offset range is worth it only if we can achieve an overlap between
    // adjacent UTrie2 blocks of 32 code points each.
    // An offset CE is also a little more expensive to look up and compute
    // than a simple CE.
    // If the range spans at least three UTrie2 block boundaries (> 64 code points),
    // then we take it.
    // If the range spans one or two block boundaries and there are
    // at least 4 code points on either side, then we take it.
    // (We could additionally require a minimum range length of, say, 16.)
    int32_t blockDelta = (end >> 5) - (start >> 5);
    if(2 <= step && step <= 0x7f &&
            (blockDelta >= 3 ||
            (blockDelta > 0 && (start & 0x1f) <= 0x1c && (end & 0x1f) >= 3))) {
        int64_t dataCE = ((int64_t)primary << 32) | (start << 8) | step;
        if(isCompressiblePrimary(primary)) { dataCE |= 0x80; }
        int32_t index = addCE(dataCE, errorCode);
        if(U_FAILURE(errorCode)) { return 0; }
        if(index > Collation::MAX_INDEX) {
            errorCode = U_BUFFER_OVERFLOW_ERROR;
            return 0;
        }
        uint32_t offsetCE32 = Collation::makeCE32FromTagAndIndex(Collation::OFFSET_TAG, index);
        utrie2_setRange32(trie, start, end, offsetCE32, true, &errorCode);
        modified = true;
        return true;
    } else {
        return false;
    }
}

uint32_t
CollationDataBuilder::setPrimaryRangeAndReturnNext(UChar32 start, UChar32 end,
                                                   uint32_t primary, int32_t step,
                                                   UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return 0; }
    UBool isCompressible = isCompressiblePrimary(primary);
    if(maybeSetPrimaryRange(start, end, primary, step, errorCode)) {
        return Collation::incThreeBytePrimaryByOffset(primary, isCompressible,
                                                      (end - start + 1) * step);
    } else {
        // Short range: Set individual CE32s.
        for(;;) {
            utrie2_set32(trie, start, Collation::makeLongPrimaryCE32(primary), &errorCode);
            ++start;
            primary = Collation::incThreeBytePrimaryByOffset(primary, isCompressible, step);
            if(start > end) { return primary; }
        }
        modified = true;
    }
}

uint32_t
CollationDataBuilder::getCE32FromOffsetCE32(UBool fromBase, UChar32 c, uint32_t ce32) const {
    int32_t i = Collation::indexFromCE32(ce32);
    int64_t dataCE = fromBase ? base->ces[i] : ce64s.elementAti(i);
    uint32_t p = Collation::getThreeBytePrimaryForOffsetData(c, dataCE);
    return Collation::makeLongPrimaryCE32(p);
}

UBool
CollationDataBuilder::isCompressibleLeadByte(uint32_t b) const {
    return base->isCompressibleLeadByte(b);
}

UBool
CollationDataBuilder::isAssigned(UChar32 c) const {
    return Collation::isAssignedCE32(utrie2_get32(trie, c));
}

uint32_t
CollationDataBuilder::getLongPrimaryIfSingleCE(UChar32 c) const {
    uint32_t ce32 = utrie2_get32(trie, c);
    if(Collation::isLongPrimaryCE32(ce32)) {
        return Collation::primaryFromLongPrimaryCE32(ce32);
    } else {
        return 0;
    }
}

int64_t
CollationDataBuilder::getSingleCE(UChar32 c, UErrorCode &errorCode) const {
    if(U_FAILURE(errorCode)) { return 0; }
    // Keep parallel with CollationData::getSingleCE().
    UBool fromBase = false;
    uint32_t ce32 = utrie2_get32(trie, c);
    if(ce32 == Collation::FALLBACK_CE32) {
        fromBase = true;
        ce32 = base->getCE32(c);
    }
    while(Collation::isSpecialCE32(ce32)) {
        switch(Collation::tagFromCE32(ce32)) {
        case Collation::LATIN_EXPANSION_TAG:
        case Collation::BUILDER_DATA_TAG:
        case Collation::PREFIX_TAG:
        case Collation::CONTRACTION_TAG:
        case Collation::HANGUL_TAG:
        case Collation::LEAD_SURROGATE_TAG:
            errorCode = U_UNSUPPORTED_ERROR;
            return 0;
        case Collation::FALLBACK_TAG:
        case Collation::RESERVED_TAG_3:
            errorCode = U_INTERNAL_PROGRAM_ERROR;
            return 0;
        case Collation::LONG_PRIMARY_TAG:
            return Collation::ceFromLongPrimaryCE32(ce32);
        case Collation::LONG_SECONDARY_TAG:
            return Collation::ceFromLongSecondaryCE32(ce32);
        case Collation::EXPANSION32_TAG:
            if(Collation::lengthFromCE32(ce32) == 1) {
                int32_t i = Collation::indexFromCE32(ce32);
                ce32 = fromBase ? base->ce32s[i] : ce32s.elementAti(i);
                break;
            } else {
                errorCode = U_UNSUPPORTED_ERROR;
                return 0;
            }
        case Collation::EXPANSION_TAG: {
            if(Collation::lengthFromCE32(ce32) == 1) {
                int32_t i = Collation::indexFromCE32(ce32);
                return fromBase ? base->ces[i] : ce64s.elementAti(i);
            } else {
                errorCode = U_UNSUPPORTED_ERROR;
                return 0;
            }
        }
        case Collation::DIGIT_TAG:
            // Fetch the non-numeric-collation CE32 and continue.
            ce32 = ce32s.elementAti(Collation::indexFromCE32(ce32));
            break;
        case Collation::U0000_TAG:
            U_ASSERT(c == 0);
            // Fetch the normal ce32 for U+0000 and continue.
            ce32 = fromBase ? base->ce32s[0] : ce32s.elementAti(0);
            break;
        case Collation::OFFSET_TAG:
            ce32 = getCE32FromOffsetCE32(fromBase, c, ce32);
            break;
        case Collation::IMPLICIT_TAG:
            return Collation::unassignedCEFromCodePoint(c);
        }
    }
    return Collation::ceFromSimpleCE32(ce32);
}

int32_t
CollationDataBuilder::addCE(int64_t ce, UErrorCode &errorCode) {
    int32_t length = ce64s.size();
    for(int32_t i = 0; i < length; ++i) {
        if(ce == ce64s.elementAti(i)) { return i; }
    }
    ce64s.addElement(ce, errorCode);
    return length;
}

int32_t
CollationDataBuilder::addCE32(uint32_t ce32, UErrorCode &errorCode) {
    int32_t length = ce32s.size();
    for(int32_t i = 0; i < length; ++i) {
        if(ce32 == (uint32_t)ce32s.elementAti(i)) { return i; }
    }
    ce32s.addElement((int32_t)ce32, errorCode);  
    return length;
}

int32_t
CollationDataBuilder::addConditionalCE32(const UnicodeString &context, uint32_t ce32,
                                         UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return -1; }
    U_ASSERT(!context.isEmpty());
    int32_t index = conditionalCE32s.size();
    if(index > Collation::MAX_INDEX) {
        errorCode = U_BUFFER_OVERFLOW_ERROR;
        return -1;
    }
    LocalPointer<ConditionalCE32> cond(new ConditionalCE32(context, ce32), errorCode);
    conditionalCE32s.adoptElement(cond.orphan(), errorCode);
    if(U_FAILURE(errorCode)) {
        return -1;
    }
    return index;
}

void
CollationDataBuilder::add(const UnicodeString &prefix, const UnicodeString &s,
                          const int64_t ces[], int32_t cesLength,
                          UErrorCode &errorCode) {
    uint32_t ce32 = encodeCEs(ces, cesLength, errorCode);
    addCE32(prefix, s, ce32, errorCode);
}

void
CollationDataBuilder::addCE32(const UnicodeString &prefix, const UnicodeString &s,
                              uint32_t ce32, UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return; }
    if(s.isEmpty()) {
        errorCode = U_ILLEGAL_ARGUMENT_ERROR;
        return;
    }
    if(trie == nullptr || utrie2_isFrozen(trie)) {
        errorCode = U_INVALID_STATE_ERROR;
        return;
    }
    UChar32 c = s.char32At(0);
    int32_t cLength = U16_LENGTH(c);
    uint32_t oldCE32 = utrie2_get32(trie, c);
    UBool hasContext = !prefix.isEmpty() || s.length() > cLength;

    if (icu4xMode) {
        if (base && c >= 0x1100 && c < 0x1200) {
            // Omit jamo tailorings.
            // TODO(https://github.com/unicode-org/icu4x/issues/1941).
        }
        const Normalizer2* nfdNormalizer = Normalizer2::getNFDInstance(errorCode);
        UnicodeString sInNfd;
        nfdNormalizer->normalize(s, sInNfd, errorCode);
        if (s != sInNfd) {
            // s is not in NFD, so it cannot match in ICU4X, since ICU4X only
            // does NFD lookups.
            // Now check that we're only rejecting known cases.
            if (s.length() == 2) {
                char16_t second = s.charAt(1);
                if (second == 0x0F73 || second == 0x0F75 || second == 0x0F81) {
                    // Second is a special decomposing Tibetan vowel sign.
                    // These also get added in the decomposed form, so ignoring
                    // this instance is OK.
                    return;
                }
                if (c == 0xFDD1 && second == 0xAC00) {
                    // This strange contraction exists in the root and
                    // doesn't have a decomposed counterpart there.
                    // This won't match in ICU4X anyway and is very strange:
                    // Unassigned Arabic presentation form contracting with
                    // the very first Hangul syllable. Let's ignore this
                    // explicitly.
                    return;
                }
            }
            // Unknown case worth investigating if ever found.
            errorCode = U_UNSUPPORTED_ERROR;
            return;
        }

        if (!prefix.isEmpty()) {
            UnicodeString prefixInNfd;
            nfdNormalizer->normalize(prefix, prefixInNfd, errorCode);
            if (prefix != prefixInNfd) {
                errorCode = U_UNSUPPORTED_ERROR;
                return;
            }

            int32_t count = prefix.countChar32();
            if (count > 2) {
                // Prefix too long for ICU4X.
                errorCode = U_UNSUPPORTED_ERROR;
                return;
            }
            UChar32 utf32[4];
            int32_t len = prefix.toUTF32(utf32, 4, errorCode);
            if (len != count) {
                errorCode = U_INVALID_STATE_ERROR;
                return;
            }
            UChar32 c = utf32[0];
            if (u_getCombiningClass(c)) {
                // Prefix must start with as starter for ICU4X.
                errorCode = U_UNSUPPORTED_ERROR;
                return;
            }
            // XXX: Korean searchjl has jamo in prefix, so commenting out this
            // check for now. ICU4X currently ignores non-root jamo tables anyway.
            // searchjl was added in
            // https://unicode-org.atlassian.net/browse/CLDR-3560
            // Contractions were changed to prefixes in
            // https://unicode-org.atlassian.net/browse/CLDR-6546
            //
            // if ((c >= 0x1100 && c < 0x1200) || (c >= 0xAC00 && c < 0xD7A4)) {
            //     errorCode = U_UNSUPPORTED_ERROR;
            //     return;
            // }
            if ((len > 1) && !(utf32[1] == 0x3099 || utf32[1] == 0x309A)) {
                // Second character in prefix, if present, must be a kana voicing mark for ICU4X.
                errorCode = U_UNSUPPORTED_ERROR;
                return;
            }
        }

        if (s.length() > cLength) {
            // Check that there's no modern Hangul in contractions.
            for (int32_t i = 0; i < s.length(); ++i) {
                char16_t c = s.charAt(i);
                if ((c >= 0x1100 && c < 0x1100 + 19) || (c >= 0x1161 && c < 0x1161 + 21) || (c >= 0x11A7 && c < 0x11A7 + 28) || (c >= 0xAC00 && c < 0xD7A4)) {
                    errorCode = U_UNSUPPORTED_ERROR;
                    return;
                }
            }
        }
    }

    if(oldCE32 == Collation::FALLBACK_CE32) {
        // First tailoring for c.
        // If c has contextual base mappings or if we add a contextual mapping,
        // then copy the base mappings.
        // Otherwise we just override the base mapping.
        uint32_t baseCE32 = base->getFinalCE32(base->getCE32(c));
        if(hasContext || Collation::ce32HasContext(baseCE32)) {
            oldCE32 = copyFromBaseCE32(c, baseCE32, true, errorCode);
            utrie2_set32(trie, c, oldCE32, &errorCode);
            if(U_FAILURE(errorCode)) { return; }
        }
    }
    if(!hasContext) {
        // No prefix, no contraction.
        if(!isBuilderContextCE32(oldCE32)) {
            utrie2_set32(trie, c, ce32, &errorCode);
        } else {
            ConditionalCE32 *cond = getConditionalCE32ForCE32(oldCE32);
            cond->builtCE32 = Collation::NO_CE32;
            cond->ce32 = ce32;
        }
    } else {
        ConditionalCE32 *cond;
        if(!isBuilderContextCE32(oldCE32)) {
            // Replace the simple oldCE32 with a builder context CE32
            // pointing to a new ConditionalCE32 list head.
            int32_t index = addConditionalCE32(UnicodeString((char16_t)0), oldCE32, errorCode);
            if(U_FAILURE(errorCode)) { return; }
            uint32_t contextCE32 = makeBuilderContextCE32(index);
            utrie2_set32(trie, c, contextCE32, &errorCode);
            contextChars.add(c);
            cond = getConditionalCE32(index);
        } else {
            cond = getConditionalCE32ForCE32(oldCE32);
            cond->builtCE32 = Collation::NO_CE32;
        }
        UnicodeString suffix(s, cLength);
        UnicodeString context((char16_t)prefix.length());
        context.append(prefix).append(suffix);
        unsafeBackwardSet.addAll(suffix);
        for(;;) {
            // invariant: context > cond->context
            int32_t next = cond->next;
            if(next < 0) {
                // Append a new ConditionalCE32 after cond.
                int32_t index = addConditionalCE32(context, ce32, errorCode);
                if(U_FAILURE(errorCode)) { return; }
                cond->next = index;
                break;
            }
            ConditionalCE32 *nextCond = getConditionalCE32(next);
            int8_t cmp = context.compare(nextCond->context);
            if(cmp < 0) {
                // Insert a new ConditionalCE32 between cond and nextCond.
                int32_t index = addConditionalCE32(context, ce32, errorCode);
                if(U_FAILURE(errorCode)) { return; }
                cond->next = index;
                getConditionalCE32(index)->next = next;
                break;
            } else if(cmp == 0) {
                // Same context as before, overwrite its ce32.
                nextCond->ce32 = ce32;
                break;
            }
            cond = nextCond;
        }
    }
    modified = true;
}

uint32_t
CollationDataBuilder::encodeOneCEAsCE32(int64_t ce) {
    uint32_t p = (uint32_t)(ce >> 32);
    uint32_t lower32 = (uint32_t)ce;
    uint32_t t = (uint32_t)(ce & 0xffff);
    U_ASSERT((t & 0xc000) != 0xc000);  // Impossible case bits 11 mark special CE32s.
    if((ce & INT64_C(0xffff00ff00ff)) == 0) {
        // normal form ppppsstt
        return p | (lower32 >> 16) | (t >> 8);
    } else if((ce & INT64_C(0xffffffffff)) == Collation::COMMON_SEC_AND_TER_CE) {
        // long-primary form ppppppC1
        return Collation::makeLongPrimaryCE32(p);
    } else if(p == 0 && (t & 0xff) == 0) {
        // long-secondary form ssssttC2
        return Collation::makeLongSecondaryCE32(lower32);
    }
    return Collation::NO_CE32;
}

uint32_t
CollationDataBuilder::encodeOneCE(int64_t ce, UErrorCode &errorCode) {
    // Try to encode one CE as one CE32.
    uint32_t ce32 = encodeOneCEAsCE32(ce);
    if(ce32 != Collation::NO_CE32) { return ce32; }
    int32_t index = addCE(ce, errorCode);
    if(U_FAILURE(errorCode)) { return 0; }
    if(index > Collation::MAX_INDEX) {
        errorCode = U_BUFFER_OVERFLOW_ERROR;
        return 0;
    }
    return Collation::makeCE32FromTagIndexAndLength(Collation::EXPANSION_TAG, index, 1);
}

uint32_t
CollationDataBuilder::encodeCEs(const int64_t ces[], int32_t cesLength,
                                UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return 0; }
    if(cesLength < 0 || cesLength > Collation::MAX_EXPANSION_LENGTH) {
        errorCode = U_ILLEGAL_ARGUMENT_ERROR;
        return 0;
    }
    if(trie == nullptr || utrie2_isFrozen(trie)) {
        errorCode = U_INVALID_STATE_ERROR;
        return 0;
    }
    if(cesLength == 0) {
        // Convenience: We cannot map to nothing, but we can map to a completely ignorable CE.
        // Do this here so that callers need not do it.
        return encodeOneCEAsCE32(0);
    } else if(cesLength == 1) {
        return encodeOneCE(ces[0], errorCode);
    } else if(cesLength == 2 && !icu4xMode) {
        // Try to encode two CEs as one CE32.
        // Turn this off for ICU4X, because without the canonical closure
        // these are so rare that it doesn't make sense to spend a branch
        // on checking this tag when using the data.
        int64_t ce0 = ces[0];
        int64_t ce1 = ces[1];
        uint32_t p0 = (uint32_t)(ce0 >> 32);
        if((ce0 & INT64_C(0xffffffffff00ff)) == Collation::COMMON_SECONDARY_CE &&
                (ce1 & INT64_C(0xffffffff00ffffff)) == Collation::COMMON_TERTIARY_CE &&
                p0 != 0) {
            // Latin mini expansion
            return
                p0 |
                (((uint32_t)ce0 & 0xff00u) << 8) |
                (uint32_t)(ce1 >> 16) |
                Collation::SPECIAL_CE32_LOW_BYTE |
                Collation::LATIN_EXPANSION_TAG;
        }
    }
    // Try to encode two or more CEs as CE32s.
    int32_t newCE32s[Collation::MAX_EXPANSION_LENGTH];
    for(int32_t i = 0;; ++i) {
        if(i == cesLength) {
            return encodeExpansion32(newCE32s, cesLength, errorCode);
        }
        uint32_t ce32 = encodeOneCEAsCE32(ces[i]);
        if(ce32 == Collation::NO_CE32) { break; }
        newCE32s[i] = (int32_t)ce32;
    }
    return encodeExpansion(ces, cesLength, errorCode);
}

uint32_t
CollationDataBuilder::encodeExpansion(const int64_t ces[], int32_t length, UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return 0; }
    // See if this sequence of CEs has already been stored.
    int64_t first = ces[0];
    int32_t ce64sMax = ce64s.size() - length;
    for(int32_t i = 0; i <= ce64sMax; ++i) {
        if(first == ce64s.elementAti(i)) {
            if(i > Collation::MAX_INDEX) {
                errorCode = U_BUFFER_OVERFLOW_ERROR;
                return 0;
            }
            for(int32_t j = 1;; ++j) {
                if(j == length) {
                    return Collation::makeCE32FromTagIndexAndLength(
                            Collation::EXPANSION_TAG, i, length);
                }
                if(ce64s.elementAti(i + j) != ces[j]) { break; }
            }
        }
    }
    // Store the new sequence.
    int32_t i = ce64s.size();
    if(i > Collation::MAX_INDEX) {
        errorCode = U_BUFFER_OVERFLOW_ERROR;
        return 0;
    }
    for(int32_t j = 0; j < length; ++j) {
        ce64s.addElement(ces[j], errorCode);
    }
    return Collation::makeCE32FromTagIndexAndLength(Collation::EXPANSION_TAG, i, length);
}

uint32_t
CollationDataBuilder::encodeExpansion32(const int32_t newCE32s[], int32_t length,
                                        UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return 0; }
    // See if this sequence of CE32s has already been stored.
    int32_t first = newCE32s[0];
    int32_t ce32sMax = ce32s.size() - length;
    for(int32_t i = 0; i <= ce32sMax; ++i) {
        if(first == ce32s.elementAti(i)) {
            if(i > Collation::MAX_INDEX) {
                errorCode = U_BUFFER_OVERFLOW_ERROR;
                return 0;
            }
            for(int32_t j = 1;; ++j) {
                if(j == length) {
                    return Collation::makeCE32FromTagIndexAndLength(
                            Collation::EXPANSION32_TAG, i, length);
                }
                if(ce32s.elementAti(i + j) != newCE32s[j]) { break; }
            }
        }
    }
    // Store the new sequence.
    int32_t i = ce32s.size();
    if(i > Collation::MAX_INDEX) {
        errorCode = U_BUFFER_OVERFLOW_ERROR;
        return 0;
    }
    for(int32_t j = 0; j < length; ++j) {
        ce32s.addElement(newCE32s[j], errorCode);
    }
    return Collation::makeCE32FromTagIndexAndLength(Collation::EXPANSION32_TAG, i, length);
}

uint32_t
CollationDataBuilder::copyFromBaseCE32(UChar32 c, uint32_t ce32, UBool withContext,
                                       UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return 0; }
    if(!Collation::isSpecialCE32(ce32)) { return ce32; }
    switch(Collation::tagFromCE32(ce32)) {
    case Collation::LONG_PRIMARY_TAG:
    case Collation::LONG_SECONDARY_TAG:
    case Collation::LATIN_EXPANSION_TAG:
        // copy as is
        break;
    case Collation::EXPANSION32_TAG: {
        const uint32_t *baseCE32s = base->ce32s + Collation::indexFromCE32(ce32);
        int32_t length = Collation::lengthFromCE32(ce32);
        ce32 = encodeExpansion32(
            reinterpret_cast<const int32_t *>(baseCE32s), length, errorCode);
        break;
    }
    case Collation::EXPANSION_TAG: {
        const int64_t *baseCEs = base->ces + Collation::indexFromCE32(ce32);
        int32_t length = Collation::lengthFromCE32(ce32);
        ce32 = encodeExpansion(baseCEs, length, errorCode);
        break;
    }
    case Collation::PREFIX_TAG: {
        // Flatten prefixes and nested suffixes (contractions)
        // into a linear list of ConditionalCE32.
        const char16_t *p = base->contexts + Collation::indexFromCE32(ce32);
        ce32 = CollationData::readCE32(p);  // Default if no prefix match.
        if(!withContext) {
            return copyFromBaseCE32(c, ce32, false, errorCode);
        }
        ConditionalCE32 head;
        UnicodeString context((char16_t)0);
        int32_t index;
        if(Collation::isContractionCE32(ce32)) {
            index = copyContractionsFromBaseCE32(context, c, ce32, &head, errorCode);
        } else {
            ce32 = copyFromBaseCE32(c, ce32, true, errorCode);
            head.next = index = addConditionalCE32(context, ce32, errorCode);
        }
        if(U_FAILURE(errorCode)) { return 0; }
        ConditionalCE32 *cond = getConditionalCE32(index);  // the last ConditionalCE32 so far
        UCharsTrie::Iterator prefixes(p + 2, 0, errorCode);
        while(prefixes.next(errorCode)) {
            context = prefixes.getString();
            context.reverse();
            context.insert(0, (char16_t)context.length());
            ce32 = (uint32_t)prefixes.getValue();
            if(Collation::isContractionCE32(ce32)) {
                index = copyContractionsFromBaseCE32(context, c, ce32, cond, errorCode);
            } else {
                ce32 = copyFromBaseCE32(c, ce32, true, errorCode);
                cond->next = index = addConditionalCE32(context, ce32, errorCode);
            }
            if(U_FAILURE(errorCode)) { return 0; }
            cond = getConditionalCE32(index);
        }
        ce32 = makeBuilderContextCE32(head.next);
        contextChars.add(c);
        break;
    }
    case Collation::CONTRACTION_TAG: {
        if(!withContext) {
            const char16_t *p = base->contexts + Collation::indexFromCE32(ce32);
            ce32 = CollationData::readCE32(p);  // Default if no suffix match.
            return copyFromBaseCE32(c, ce32, false, errorCode);
        }
        ConditionalCE32 head;
        UnicodeString context((char16_t)0);
        copyContractionsFromBaseCE32(context, c, ce32, &head, errorCode);
        ce32 = makeBuilderContextCE32(head.next);
        contextChars.add(c);
        break;
    }
    case Collation::HANGUL_TAG:
        errorCode = U_UNSUPPORTED_ERROR;  // We forbid tailoring of Hangul syllables.
        break;
    case Collation::OFFSET_TAG:
        ce32 = getCE32FromOffsetCE32(true, c, ce32);
        break;
    case Collation::IMPLICIT_TAG:
        ce32 = encodeOneCE(Collation::unassignedCEFromCodePoint(c), errorCode);
        break;
    default:
        UPRV_UNREACHABLE_EXIT;  // require ce32 == base->getFinalCE32(ce32)
    }
    return ce32;
}

int32_t
CollationDataBuilder::copyContractionsFromBaseCE32(UnicodeString &context, UChar32 c, uint32_t ce32,
                                                   ConditionalCE32 *cond, UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return 0; }
    const char16_t *p = base->contexts + Collation::indexFromCE32(ce32);
    int32_t index;
    if((ce32 & Collation::CONTRACT_SINGLE_CP_NO_MATCH) != 0) {
        // No match on the single code point.
        // We are underneath a prefix, and the default mapping is just
        // a fallback to the mappings for a shorter prefix.
        U_ASSERT(context.length() > 1);
        index = -1;
    } else {
        ce32 = CollationData::readCE32(p);  // Default if no suffix match.
        U_ASSERT(!Collation::isContractionCE32(ce32));
        ce32 = copyFromBaseCE32(c, ce32, true, errorCode);
        cond->next = index = addConditionalCE32(context, ce32, errorCode);
        if(U_FAILURE(errorCode)) { return 0; }
        cond = getConditionalCE32(index);
    }

    int32_t suffixStart = context.length();
    UCharsTrie::Iterator suffixes(p + 2, 0, errorCode);
    while(suffixes.next(errorCode)) {
        context.append(suffixes.getString());
        ce32 = copyFromBaseCE32(c, (uint32_t)suffixes.getValue(), true, errorCode);
        cond->next = index = addConditionalCE32(context, ce32, errorCode);
        if(U_FAILURE(errorCode)) { return 0; }
        // No need to update the unsafeBackwardSet because the tailoring set
        // is already a copy of the base set.
        cond = getConditionalCE32(index);
        context.truncate(suffixStart);
    }
    U_ASSERT(index >= 0);
    return index;
}

class CopyHelper {
public:
    CopyHelper(const CollationDataBuilder &s, CollationDataBuilder &d,
               const CollationDataBuilder::CEModifier &m, UErrorCode &initialErrorCode)
            : src(s), dest(d), modifier(m),
              errorCode(initialErrorCode) {}

    UBool copyRangeCE32(UChar32 start, UChar32 end, uint32_t ce32) {
        ce32 = copyCE32(ce32);
        utrie2_setRange32(dest.trie, start, end, ce32, true, &errorCode);
        if(CollationDataBuilder::isBuilderContextCE32(ce32)) {
            dest.contextChars.add(start, end);
        }
        return U_SUCCESS(errorCode);
    }

    uint32_t copyCE32(uint32_t ce32) {
        if(!Collation::isSpecialCE32(ce32)) {
            int64_t ce = modifier.modifyCE32(ce32);
            if(ce != Collation::NO_CE) {
                ce32 = dest.encodeOneCE(ce, errorCode);
            }
        } else {
            int32_t tag = Collation::tagFromCE32(ce32);
            if(tag == Collation::EXPANSION32_TAG) {
                const uint32_t *srcCE32s = reinterpret_cast<uint32_t *>(src.ce32s.getBuffer());
                srcCE32s += Collation::indexFromCE32(ce32);
                int32_t length = Collation::lengthFromCE32(ce32);
                // Inspect the source CE32s. Just copy them if none are modified.
                // Otherwise copy to modifiedCEs, with modifications.
                UBool isModified = false;
                for(int32_t i = 0; i < length; ++i) {
                    ce32 = srcCE32s[i];
                    int64_t ce;
                    if(Collation::isSpecialCE32(ce32) ||
                            (ce = modifier.modifyCE32(ce32)) == Collation::NO_CE) {
                        if(isModified) {
                            modifiedCEs[i] = Collation::ceFromCE32(ce32);
                        }
                    } else {
                        if(!isModified) {
                            for(int32_t j = 0; j < i; ++j) {
                                modifiedCEs[j] = Collation::ceFromCE32(srcCE32s[j]);
                            }
                            isModified = true;
                        }
                        modifiedCEs[i] = ce;
                    }
                }
                if(isModified) {
                    ce32 = dest.encodeCEs(modifiedCEs, length, errorCode);
                } else {
                    ce32 = dest.encodeExpansion32(
                        reinterpret_cast<const int32_t *>(srcCE32s), length, errorCode);
                }
            } else if(tag == Collation::EXPANSION_TAG) {
                const int64_t *srcCEs = src.ce64s.getBuffer();
                srcCEs += Collation::indexFromCE32(ce32);
                int32_t length = Collation::lengthFromCE32(ce32);
                // Inspect the source CEs. Just copy them if none are modified.
                // Otherwise copy to modifiedCEs, with modifications.
                UBool isModified = false;
                for(int32_t i = 0; i < length; ++i) {
                    int64_t srcCE = srcCEs[i];
                    int64_t ce = modifier.modifyCE(srcCE);
                    if(ce == Collation::NO_CE) {
                        if(isModified) {
                            modifiedCEs[i] = srcCE;
                        }
                    } else {
                        if(!isModified) {
                            for(int32_t j = 0; j < i; ++j) {
                                modifiedCEs[j] = srcCEs[j];
                            }
                            isModified = true;
                        }
                        modifiedCEs[i] = ce;
                    }
                }
                if(isModified) {
                    ce32 = dest.encodeCEs(modifiedCEs, length, errorCode);
                } else {
                    ce32 = dest.encodeExpansion(srcCEs, length, errorCode);
                }
            } else if(tag == Collation::BUILDER_DATA_TAG) {
                // Copy the list of ConditionalCE32.
                ConditionalCE32 *cond = src.getConditionalCE32ForCE32(ce32);
                U_ASSERT(!cond->hasContext());
                int32_t destIndex = dest.addConditionalCE32(
                        cond->context, copyCE32(cond->ce32), errorCode);
                ce32 = CollationDataBuilder::makeBuilderContextCE32(destIndex);
                while(cond->next >= 0) {
                    cond = src.getConditionalCE32(cond->next);
                    ConditionalCE32 *prevDestCond = dest.getConditionalCE32(destIndex);
                    destIndex = dest.addConditionalCE32(
                            cond->context, copyCE32(cond->ce32), errorCode);
                    int32_t suffixStart = cond->prefixLength() + 1;
                    dest.unsafeBackwardSet.addAll(cond->context.tempSubString(suffixStart));
                    prevDestCond->next = destIndex;
                }
            } else {
                // Just copy long CEs and Latin mini expansions (and other expected values) as is,
                // assuming that the modifier would not modify them.
                U_ASSERT(tag == Collation::LONG_PRIMARY_TAG ||
                        tag == Collation::LONG_SECONDARY_TAG ||
                        tag == Collation::LATIN_EXPANSION_TAG ||
                        tag == Collation::HANGUL_TAG);
            }
        }
        return ce32;
    }

    const CollationDataBuilder &src;
    CollationDataBuilder &dest;
    const CollationDataBuilder::CEModifier &modifier;
    int64_t modifiedCEs[Collation::MAX_EXPANSION_LENGTH];
    UErrorCode errorCode;
};

U_CDECL_BEGIN

static UBool U_CALLCONV
enumRangeForCopy(const void *context, UChar32 start, UChar32 end, uint32_t value) {
    return
        value == Collation::UNASSIGNED_CE32 || value == Collation::FALLBACK_CE32 ||
        ((CopyHelper *)context)->copyRangeCE32(start, end, value);
}

U_CDECL_END

void
CollationDataBuilder::copyFrom(const CollationDataBuilder &src, const CEModifier &modifier,
                               UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return; }
    if(trie == nullptr || utrie2_isFrozen(trie)) {
        errorCode = U_INVALID_STATE_ERROR;
        return;
    }
    CopyHelper helper(src, *this, modifier, errorCode);
    utrie2_enum(src.trie, nullptr, enumRangeForCopy, &helper);
    errorCode = helper.errorCode;
    // Update the contextChars and the unsafeBackwardSet while copying,
    // in case a character had conditional mappings in the source builder
    // and they were removed later.
    modified |= src.modified;
}

void
CollationDataBuilder::optimize(const UnicodeSet &set, UErrorCode &errorCode) {
    if(U_FAILURE(errorCode) || set.isEmpty()) { return; }
    UnicodeSetIterator iter(set);
    while(iter.next() && !iter.isString()) {
        UChar32 c = iter.getCodepoint();
        uint32_t ce32 = utrie2_get32(trie, c);
        if(ce32 == Collation::FALLBACK_CE32) {
            ce32 = base->getFinalCE32(base->getCE32(c));
            ce32 = copyFromBaseCE32(c, ce32, true, errorCode);
            utrie2_set32(trie, c, ce32, &errorCode);
        }
    }
    modified = true;
}

void
CollationDataBuilder::suppressContractions(const UnicodeSet &set, UErrorCode &errorCode) {
    if(U_FAILURE(errorCode) || set.isEmpty()) { return; }
    UnicodeSetIterator iter(set);
    while(iter.next() && !iter.isString()) {
        UChar32 c = iter.getCodepoint();
        uint32_t ce32 = utrie2_get32(trie, c);
        if(ce32 == Collation::FALLBACK_CE32) {
            ce32 = base->getFinalCE32(base->getCE32(c));
            if(Collation::ce32HasContext(ce32)) {
                ce32 = copyFromBaseCE32(c, ce32, false /* without context */, errorCode);
                utrie2_set32(trie, c, ce32, &errorCode);
            }
        } else if(isBuilderContextCE32(ce32)) {
            ce32 = getConditionalCE32ForCE32(ce32)->ce32;
            // Simply abandon the list of ConditionalCE32.
            // The caller will copy this builder in the end,
            // eliminating unreachable data.
            utrie2_set32(trie, c, ce32, &errorCode);
            contextChars.remove(c);
        }
    }
    modified = true;
}

UBool
CollationDataBuilder::getJamoCE32s(uint32_t jamoCE32s[], UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return false; }
    UBool anyJamoAssigned = base == nullptr;  // always set jamoCE32s in the base data
    UBool needToCopyFromBase = false;
    for(int32_t j = 0; j < CollationData::JAMO_CE32S_LENGTH; ++j) {  // Count across Jamo types.
        UChar32 jamo = jamoCpFromIndex(j);
        UBool fromBase = false;
        uint32_t ce32 = utrie2_get32(trie, jamo);
        anyJamoAssigned |= Collation::isAssignedCE32(ce32);
        // TODO: Try to prevent [optimize [Jamo]] from counting as anyJamoAssigned.
        // (As of CLDR 24 [2013] the Korean tailoring does not optimize conjoining Jamo.)
        if(ce32 == Collation::FALLBACK_CE32) {
            fromBase = true;
            ce32 = base->getCE32(jamo);
        }
        if(Collation::isSpecialCE32(ce32)) {
            switch(Collation::tagFromCE32(ce32)) {
            case Collation::LONG_PRIMARY_TAG:
            case Collation::LONG_SECONDARY_TAG:
            case Collation::LATIN_EXPANSION_TAG:
                // Copy the ce32 as-is.
                break;
            case Collation::EXPANSION32_TAG:
            case Collation::EXPANSION_TAG:
            case Collation::PREFIX_TAG:
            case Collation::CONTRACTION_TAG:
                if(fromBase) {
                    // Defer copying until we know if anyJamoAssigned.
                    ce32 = Collation::FALLBACK_CE32;
                    needToCopyFromBase = true;
                }
                break;
            case Collation::IMPLICIT_TAG:
                // An unassigned Jamo should only occur in tests with incomplete bases.
                U_ASSERT(fromBase);
                ce32 = Collation::FALLBACK_CE32;
                needToCopyFromBase = true;
                break;
            case Collation::OFFSET_TAG:
                ce32 = getCE32FromOffsetCE32(fromBase, jamo, ce32);
                break;
            case Collation::FALLBACK_TAG:
            case Collation::RESERVED_TAG_3:
            case Collation::BUILDER_DATA_TAG:
            case Collation::DIGIT_TAG:
            case Collation::U0000_TAG:
            case Collation::HANGUL_TAG:
            case Collation::LEAD_SURROGATE_TAG:
                errorCode = U_INTERNAL_PROGRAM_ERROR;
                return false;
            }
        }
        jamoCE32s[j] = ce32;
    }
    if(anyJamoAssigned && needToCopyFromBase) {
        for(int32_t j = 0; j < CollationData::JAMO_CE32S_LENGTH; ++j) {
            if(jamoCE32s[j] == Collation::FALLBACK_CE32) {
                UChar32 jamo = jamoCpFromIndex(j);
                jamoCE32s[j] = copyFromBaseCE32(jamo, base->getCE32(jamo),
                                                /*withContext=*/ true, errorCode);
            }
        }
    }
    return anyJamoAssigned && U_SUCCESS(errorCode);
}

void
CollationDataBuilder::setDigitTags(UErrorCode &errorCode) {
    UnicodeSet digits(UNICODE_STRING_SIMPLE("[:Nd:]"), errorCode);
    if(U_FAILURE(errorCode)) { return; }
    UnicodeSetIterator iter(digits);
    while(iter.next()) {
        U_ASSERT(!iter.isString());
        UChar32 c = iter.getCodepoint();
        uint32_t ce32 = utrie2_get32(trie, c);
        if(ce32 != Collation::FALLBACK_CE32 && ce32 != Collation::UNASSIGNED_CE32) {
            int32_t index = addCE32(ce32, errorCode);
            if(U_FAILURE(errorCode)) { return; }
            if(index > Collation::MAX_INDEX) {
                errorCode = U_BUFFER_OVERFLOW_ERROR;
                return;
            }
            ce32 = Collation::makeCE32FromTagIndexAndLength(
                    Collation::DIGIT_TAG, index, u_charDigitValue(c));
            utrie2_set32(trie, c, ce32, &errorCode);
        }
    }
}

U_CDECL_BEGIN

static UBool U_CALLCONV
enumRangeLeadValue(const void *context, UChar32 /*start*/, UChar32 /*end*/, uint32_t value) {
    int32_t *pValue = (int32_t *)context;
    if(value == Collation::UNASSIGNED_CE32) {
        value = Collation::LEAD_ALL_UNASSIGNED;
    } else if(value == Collation::FALLBACK_CE32) {
        value = Collation::LEAD_ALL_FALLBACK;
    } else {
        *pValue = Collation::LEAD_MIXED;
        return false;
    }
    if(*pValue < 0) {
        *pValue = (int32_t)value;
    } else if(*pValue != (int32_t)value) {
        *pValue = Collation::LEAD_MIXED;
        return false;
    }
    return true;
}

U_CDECL_END

void
CollationDataBuilder::setLeadSurrogates(UErrorCode &errorCode) {
    for(char16_t lead = 0xd800; lead < 0xdc00; ++lead) {
        int32_t value = -1;
        utrie2_enumForLeadSurrogate(trie, lead, nullptr, enumRangeLeadValue, &value);
        utrie2_set32ForLeadSurrogateCodeUnit(
            trie, lead,
            Collation::makeCE32FromTagAndIndex(Collation::LEAD_SURROGATE_TAG, 0) | (uint32_t)value,
            &errorCode);
    }
}

void
CollationDataBuilder::build(CollationData &data, UErrorCode &errorCode) {
    buildMappings(data, errorCode);
    if(base != nullptr) {
        data.numericPrimary = base->numericPrimary;
        data.compressibleBytes = base->compressibleBytes;
        data.numScripts = base->numScripts;
        data.scriptsIndex = base->scriptsIndex;
        data.scriptStarts = base->scriptStarts;
        data.scriptStartsLength = base->scriptStartsLength;
    }
    buildFastLatinTable(data, errorCode);
}

void
CollationDataBuilder::buildMappings(CollationData &data, UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return; }
    if(trie == nullptr || utrie2_isFrozen(trie)) {
        errorCode = U_INVALID_STATE_ERROR;
        return;
    }

    buildContexts(errorCode);

    uint32_t jamoCE32s[CollationData::JAMO_CE32S_LENGTH];
    int32_t jamoIndex = -1;
    if(getJamoCE32s(jamoCE32s, errorCode)) {
        jamoIndex = ce32s.size();
        for(int32_t i = 0; i < CollationData::JAMO_CE32S_LENGTH; ++i) {
            ce32s.addElement((int32_t)jamoCE32s[i], errorCode);
        }
        // Small optimization: Use a bit in the Hangul ce32
        // to indicate that none of the Jamo CE32s are isSpecialCE32()
        // (as it should be in the root collator).
        // It allows CollationIterator to avoid recursive function calls and per-Jamo tests.
        // In order to still have good trie compression and keep this code simple,
        // we only set this flag if a whole block of 588 Hangul syllables starting with
        // a common leading consonant (Jamo L) has this property.
        UBool isAnyJamoVTSpecial = false;
        for(int32_t i = Hangul::JAMO_L_COUNT; i < CollationData::JAMO_CE32S_LENGTH; ++i) {
            if(Collation::isSpecialCE32(jamoCE32s[i])) {
                isAnyJamoVTSpecial = true;
                break;
            }
        }
        uint32_t hangulCE32 = Collation::makeCE32FromTagAndIndex(Collation::HANGUL_TAG, 0);
        UChar32 c = Hangul::HANGUL_BASE;
        for(int32_t i = 0; i < Hangul::JAMO_L_COUNT; ++i) {  // iterate over the Jamo L
            uint32_t ce32 = hangulCE32;
            if(!isAnyJamoVTSpecial && !Collation::isSpecialCE32(jamoCE32s[i])) {
                ce32 |= Collation::HANGUL_NO_SPECIAL_JAMO;
            }
            UChar32 limit = c + Hangul::JAMO_VT_COUNT;
            utrie2_setRange32(trie, c, limit - 1, ce32, true, &errorCode);
            c = limit;
        }
    } else {
        // Copy the Hangul CE32s from the base in blocks per Jamo L,
        // assuming that HANGUL_NO_SPECIAL_JAMO is set or not set for whole blocks.
        for(UChar32 c = Hangul::HANGUL_BASE; c < Hangul::HANGUL_LIMIT;) {
            uint32_t ce32 = base->getCE32(c);
            U_ASSERT(Collation::hasCE32Tag(ce32, Collation::HANGUL_TAG));
            UChar32 limit = c + Hangul::JAMO_VT_COUNT;
            utrie2_setRange32(trie, c, limit - 1, ce32, true, &errorCode);
            c = limit;
        }
    }

    setDigitTags(errorCode);
    setLeadSurrogates(errorCode);

    if (!icu4xMode) {
        // For U+0000, move its normal ce32 into CE32s[0] and set U0000_TAG.
        ce32s.setElementAt((int32_t)utrie2_get32(trie, 0), 0);
        utrie2_set32(trie, 0, Collation::makeCE32FromTagAndIndex(Collation::U0000_TAG, 0), &errorCode);
    }

    utrie2_freeze(trie, UTRIE2_32_VALUE_BITS, &errorCode);
    if(U_FAILURE(errorCode)) { return; }

    // Mark each lead surrogate as "unsafe"
    // if any of its 1024 associated supplementary code points is "unsafe".
    UChar32 c = 0x10000;
    for(char16_t lead = 0xd800; lead < 0xdc00; ++lead, c += 0x400) {
        if(unsafeBackwardSet.containsSome(c, c + 0x3ff)) {
            unsafeBackwardSet.add(lead);
        }
    }
    unsafeBackwardSet.freeze();

    data.trie = trie;
    data.ce32s = reinterpret_cast<const uint32_t *>(ce32s.getBuffer());
    data.ces = ce64s.getBuffer();
    data.contexts = contexts.getBuffer();

    data.ce32sLength = ce32s.size();
    data.cesLength = ce64s.size();
    data.contextsLength = contexts.length();

    data.base = base;
    if(jamoIndex >= 0) {
        data.jamoCE32s = data.ce32s + jamoIndex;
    } else {
        data.jamoCE32s = base->jamoCE32s;
    }
    data.unsafeBackwardSet = &unsafeBackwardSet;
}

void
CollationDataBuilder::clearContexts() {
    contexts.remove();
    // Incrementing the contexts build "era" invalidates all of the builtCE32
    // from before this clearContexts() call.
    // Simpler than finding and resetting all of those fields.
    ++contextsEra;
}

void
CollationDataBuilder::buildContexts(UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return; }
    // Ignore abandoned lists and the cached builtCE32,
    // and build all contexts from scratch.
    clearContexts();
    UnicodeSetIterator iter(contextChars);
    while(U_SUCCESS(errorCode) && iter.next()) {
        U_ASSERT(!iter.isString());
        UChar32 c = iter.getCodepoint();
        uint32_t ce32 = utrie2_get32(trie, c);
        if(!isBuilderContextCE32(ce32)) {
            // Impossible: No context data for c in contextChars.
            errorCode = U_INTERNAL_PROGRAM_ERROR;
            return;
        }
        ConditionalCE32 *cond = getConditionalCE32ForCE32(ce32);
        ce32 = buildContext(cond, errorCode);
        utrie2_set32(trie, c, ce32, &errorCode);
    }
}

uint32_t
CollationDataBuilder::buildContext(ConditionalCE32 *head, UErrorCode &errorCode) {
    if(U_FAILURE(errorCode)) { return 0; }
    // The list head must have no context.
    U_ASSERT(!head->hasContext());
    // The list head must be followed by one or more nodes that all do have context.
    U_ASSERT(head->next >= 0);
    UCharsTrieBuilder prefixBuilder(errorCode);
    UCharsTrieBuilder contractionBuilder(errorCode);
    // This outer loop goes from each prefix to the next.
    // For each prefix it finds the one or more same-prefix entries (firstCond..lastCond).
    // If there are multiple suffixes for the same prefix,
    // then an inner loop builds a contraction trie for them.
    for(ConditionalCE32 *cond = head;; cond = getConditionalCE32(cond->next)) {
        if(U_FAILURE(errorCode)) { return 0; }  // early out for memory allocation errors
        // After the list head, the prefix or suffix can be empty, but not both.
        U_ASSERT(cond == head || cond->hasContext());
        int32_t prefixLength = cond->prefixLength();
        UnicodeString prefix(cond->context, 0, prefixLength + 1);
        // Collect all contraction suffixes for one prefix.
        ConditionalCE32 *firstCond = cond;
        ConditionalCE32 *lastCond;
        do {
            lastCond = cond;
            // Clear the defaultCE32 fields as we go.
            // They are left over from building a previous version of this list of contexts.
            //
            // One of the code paths below may copy a preceding defaultCE32
            // into its emptySuffixCE32.
            // If a new suffix has been inserted before what used to be
            // the firstCond for its prefix, then that previous firstCond could still
            // contain an outdated defaultCE32 from an earlier buildContext() and
            // result in an incorrect emptySuffixCE32.
            // So we reset all defaultCE32 before reading and setting new values.
            cond->defaultCE32 = Collation::NO_CE32;
        } while(cond->next >= 0 &&
                (cond = getConditionalCE32(cond->next))->context.startsWith(prefix));
        uint32_t ce32;
        int32_t suffixStart = prefixLength + 1;  // == prefix.length()
        if(lastCond->context.length() == suffixStart) {
            // One prefix without contraction suffix.
            U_ASSERT(firstCond == lastCond);
            ce32 = lastCond->ce32;
            cond = lastCond;
        } else {
            // Build the contractions trie.
            contractionBuilder.clear();
            // Entry for an empty suffix, to be stored before the trie.
            uint32_t emptySuffixCE32 = 0;
            uint32_t flags = 0;
            if(firstCond->context.length() == suffixStart) {
                // There is a mapping for the prefix and the single character c. (p|c)
                // If no other suffix matches, then we return this value.
                emptySuffixCE32 = firstCond->ce32;
                cond = getConditionalCE32(firstCond->next);
            } else {
                // There is no mapping for the prefix and just the single character.
                // (There is no p|c, only p|cd, p|ce etc.)
                flags |= Collation::CONTRACT_SINGLE_CP_NO_MATCH;
                // When the prefix matches but none of the prefix-specific suffixes,
                // then we fall back to the mappings with the next-longest prefix,
                // and ultimately to mappings with no prefix.
                // Each fallback might be another set of contractions.
                // For example, if there are mappings for ch, p|cd, p|ce, but not for p|c,
                // then in text "pch" we find the ch contraction.
                for(cond = head;; cond = getConditionalCE32(cond->next)) {
                    int32_t length = cond->prefixLength();
                    if(length == prefixLength) { break; }
                    if(cond->defaultCE32 != Collation::NO_CE32 &&
                            (length==0 || prefix.endsWith(cond->context, 1, length))) {
                        emptySuffixCE32 = cond->defaultCE32;
                    }
                }
                cond = firstCond;
            }
            // Optimization: Set a flag when
            // the first character of every contraction suffix has lccc!=0.
            // Short-circuits contraction matching when a normal letter follows.
            flags |= Collation::CONTRACT_NEXT_CCC;
            // Add all of the non-empty suffixes into the contraction trie.
            for(;;) {
                UnicodeString suffix(cond->context, suffixStart);
                uint16_t fcd16 = nfcImpl.getFCD16(suffix.char32At(0));
                if(fcd16 <= 0xff) {
                    flags &= ~Collation::CONTRACT_NEXT_CCC;
                }
                fcd16 = nfcImpl.getFCD16(suffix.char32At(suffix.length() - 1));
                if(fcd16 > 0xff) {
                    // The last suffix character has lccc!=0, allowing for discontiguous contractions.
                    flags |= Collation::CONTRACT_TRAILING_CCC;
                }
                if (icu4xMode && (flags & Collation::CONTRACT_HAS_STARTER) == 0) {
                    for (int32_t i = 0; i < suffix.length();) {
                        UChar32 c = suffix.char32At(i);
                            if (!u_getCombiningClass(c)) {
                                flags |= Collation::CONTRACT_HAS_STARTER;
                                break;
                            }
                        if (c > 0xFFFF) {
                            i += 2;
                        } else {
                            ++i;
                        }
                    }
                }
                contractionBuilder.add(suffix, (int32_t)cond->ce32, errorCode);
                if(cond == lastCond) { break; }
                cond = getConditionalCE32(cond->next);
            }
            int32_t index = addContextTrie(emptySuffixCE32, contractionBuilder, errorCode);
            if(U_FAILURE(errorCode)) { return 0; }
            if(index > Collation::MAX_INDEX) {
                errorCode = U_BUFFER_OVERFLOW_ERROR;
                return 0;
            }
            ce32 = Collation::makeCE32FromTagAndIndex(Collation::CONTRACTION_TAG, index) | flags;
        }
        U_ASSERT(cond == lastCond);
        firstCond->defaultCE32 = ce32;
        if(prefixLength == 0) {
            if(cond->next < 0) {
                // No non-empty prefixes, only contractions.
                return ce32;
            }
        } else {
            prefix.remove(0, 1);  // Remove the length unit.
            prefix.reverse();
            prefixBuilder.add(prefix, (int32_t)ce32, errorCode);
            if(cond->next < 0) { break; }
        }
    }
    U_ASSERT(head->defaultCE32 != Collation::NO_CE32);
    int32_t index = addContextTrie(head->defaultCE32, prefixBuilder, errorCode);
    if(U_FAILURE(errorCode)) { return 0; }
    if(index > Collation::MAX_INDEX) {
        errorCode = U_BUFFER_OVERFLOW_ERROR;
        return 0;
    }
    return Collation::makeCE32FromTagAndIndex(Collation::PREFIX_TAG, index);
}

int32_t
CollationDataBuilder::addContextTrie(uint32_t defaultCE32, UCharsTrieBuilder &trieBuilder,
                                     UErrorCode &errorCode) {
    UnicodeString context;
    context.append((char16_t)(defaultCE32 >> 16)).append((char16_t)defaultCE32);
    UnicodeString trieString;
    context.append(trieBuilder.buildUnicodeString(USTRINGTRIE_BUILD_SMALL, trieString, errorCode));
    if(U_FAILURE(errorCode)) { return -1; }
    int32_t index = contexts.indexOf(context);
    if(index < 0) {
        index = contexts.length();
        contexts.append(context);
    }
    return index;
}

void
CollationDataBuilder::buildFastLatinTable(CollationData &data, UErrorCode &errorCode) {
    if(U_FAILURE(errorCode) || !fastLatinEnabled) { return; }

    delete fastLatinBuilder;
    fastLatinBuilder = new CollationFastLatinBuilder(errorCode);
    if(fastLatinBuilder == nullptr) {
        errorCode = U_MEMORY_ALLOCATION_ERROR;
        return;
    }
    if(fastLatinBuilder->forData(data, errorCode)) {
        const uint16_t *table = fastLatinBuilder->getTable();
        int32_t length = fastLatinBuilder->lengthOfTable();
        if(base != nullptr && length == base->fastLatinTableLength &&
                uprv_memcmp(table, base->fastLatinTable, length * 2) == 0) {
            // Same fast Latin table as in the base, use that one instead.
            delete fastLatinBuilder;
            fastLatinBuilder = nullptr;
            table = base->fastLatinTable;
        }
        data.fastLatinTable = table;
        data.fastLatinTableLength = length;
    } else {
        delete fastLatinBuilder;
        fastLatinBuilder = nullptr;
    }
}

int32_t
CollationDataBuilder::getCEs(const UnicodeString &s, int64_t ces[], int32_t cesLength) {
    return getCEs(s, 0, ces, cesLength);
}

int32_t
CollationDataBuilder::getCEs(const UnicodeString &prefix, const UnicodeString &s,
                             int64_t ces[], int32_t cesLength) {
    int32_t prefixLength = prefix.length();
    if(prefixLength == 0) {
        return getCEs(s, 0, ces, cesLength);
    } else {
        return getCEs(prefix + s, prefixLength, ces, cesLength);
    }
}

int32_t
CollationDataBuilder::getCEs(const UnicodeString &s, int32_t start,
                             int64_t ces[], int32_t cesLength) {
    if(collIter == nullptr) {
        collIter = new DataBuilderCollationIterator(*this);
        if(collIter == nullptr) { return 0; }
    }
    return collIter->fetchCEs(s, start, ces, cesLength);
}

U_NAMESPACE_END

#endif  // !UCONFIG_NO_COLLATION