@delali/narsil 0.3.1-next.128 → 0.3.1-next.130

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (108) hide show
  1. package/dist/adapters/filesystem.mjs +1 -1
  2. package/dist/adapters/indexeddb.mjs +1 -1
  3. package/dist/browser/chunk-26A5JFT2.mjs +89 -0
  4. package/dist/browser/{chunk-4Y2EGA3U.mjs → chunk-2Y3CO4EQ.mjs} +878 -310
  5. package/dist/browser/{chunk-2CBQ2PCC.mjs → chunk-AIIJDWMG.mjs} +1 -1
  6. package/dist/browser/{chunk-P7C2S5JX.mjs → chunk-D7WQF6DV.mjs} +2 -2
  7. package/dist/browser/{chunk-V5DCQIOT.mjs → chunk-M5UAAXGU.mjs} +2 -2
  8. package/dist/browser/{chunk-B65QVTST.mjs → chunk-NSCSR6MA.mjs} +24 -9
  9. package/dist/browser/{chunk-LJS6PREA.mjs → chunk-UDGB2MWT.mjs} +7 -28
  10. package/dist/browser/{chunk-6POI2KNI.mjs → chunk-VBBLTD7O.mjs} +4 -3
  11. package/dist/browser/chunk-VEGHFP2P.mjs +30 -0
  12. package/dist/browser/{durability-wiring-PUC4QSCR.mjs → durability-wiring-FNZSRH4N.mjs} +8 -6
  13. package/dist/browser/{module-ACWDC7YI.mjs → module-4HIMOQM5.mjs} +6 -5
  14. package/dist/browser/phrase-module-WESYVU7K.mjs +327 -0
  15. package/dist/browser/{restore-queue-MLUEUEAN.mjs → restore-queue-GEZ54CA5.mjs} +1 -1
  16. package/dist/browser/{runtime-BX6XRFYR.mjs → runtime-3X7ZJ5T3.mjs} +2 -1
  17. package/dist/browser/{snapshot-U5Y346XG.mjs → snapshot-V7VF6BN6.mjs} +25 -10
  18. package/dist/browser/{write-GZHR6AVG.mjs → write-YIC2MKRY.mjs} +7 -5
  19. package/dist/{capabilities-9YwwCcqP.d.ts → capabilities-Bl8G9taj.d.ts} +10 -3
  20. package/dist/{chunk-VK45J47R.mjs → chunk-2C3WXBQS.mjs} +1 -1
  21. package/dist/{chunk-QSCVJJW5.mjs → chunk-2JWNJSNE.mjs} +8 -8
  22. package/dist/{chunk-IBGMFR5Y.mjs → chunk-2MKX2SCO.mjs} +481 -121
  23. package/dist/{chunk-P34VZEXB.mjs → chunk-2SEL6SBS.mjs} +4 -95
  24. package/dist/{chunk-7ZZQNOIY.mjs → chunk-2WBY4Y4M.mjs} +1 -1
  25. package/dist/{chunk-HFJXEFFV.mjs → chunk-4GDLFDPZ.mjs} +3 -3
  26. package/dist/chunk-4UMZ7KQL.mjs +17 -0
  27. package/dist/{chunk-ZR63OGT2.mjs → chunk-5LX6MOIB.mjs} +2 -2
  28. package/dist/{chunk-UKXMSO5H.mjs → chunk-6M6SDH3S.mjs} +189 -45
  29. package/dist/{chunk-IR4KQ4WB.mjs → chunk-6SSVIW7Y.mjs} +49 -31
  30. package/dist/{chunk-4IHMMDRF.mjs → chunk-7OIQ2SIO.mjs} +4 -4
  31. package/dist/{chunk-6ODIY6ZB.mjs → chunk-7XY6VD7B.mjs} +1 -1
  32. package/dist/{chunk-OBWUWZRA.mjs → chunk-CZ7ISLXA.mjs} +3 -0
  33. package/dist/{chunk-OFYKBKKV.mjs → chunk-D5TFRB5G.mjs} +3 -15
  34. package/dist/{chunk-KCQG2PJT.mjs → chunk-EUCX5K74.mjs} +4 -4
  35. package/dist/{chunk-HWQDZB2U.mjs → chunk-F6AAXI7T.mjs} +3 -3
  36. package/dist/{chunk-E74YLAIG.mjs → chunk-FQBFZAWG.mjs} +4 -1
  37. package/dist/{chunk-U6NGTWI3.mjs → chunk-HCGJKDZV.mjs} +1 -1
  38. package/dist/{chunk-SJEIIR56.mjs → chunk-JJ6EUMXB.mjs} +2 -2
  39. package/dist/{chunk-N57XXGPG.mjs → chunk-JT3YZGCE.mjs} +23 -4
  40. package/dist/chunk-JUKYYXSP.mjs +93 -0
  41. package/dist/{chunk-JIXO442E.mjs → chunk-K77FJXKS.mjs} +26 -10
  42. package/dist/{chunk-4PQNDHSA.mjs → chunk-NMXRXFGZ.mjs} +2 -2
  43. package/dist/{chunk-6PN3U6XW.mjs → chunk-NNO7CO3A.mjs} +1 -1
  44. package/dist/{chunk-542TD4BL.mjs → chunk-PATMDIMN.mjs} +1 -1
  45. package/dist/{chunk-DG62H7OD.mjs → chunk-PU6SKH7O.mjs} +2 -2
  46. package/dist/chunk-Q2AP2FSN.mjs +15 -0
  47. package/dist/{chunk-22CK2LQO.mjs → chunk-QQTXT65X.mjs} +6 -5
  48. package/dist/{chunk-LXWDQLTW.mjs → chunk-RNSC4NFI.mjs} +1 -1
  49. package/dist/{chunk-JEXJDCQH.mjs → chunk-RWESFSJI.mjs} +28 -13
  50. package/dist/{chunk-SEG4CL6W.mjs → chunk-SOHMMNCH.mjs} +4 -4
  51. package/dist/{chunk-OGFSUXYY.mjs → chunk-SZGRYEIC.mjs} +4 -4
  52. package/dist/{chunk-LOBYK2FG.mjs → chunk-T4OFG4IQ.mjs} +20 -9
  53. package/dist/{chunk-ENBN5VQ6.mjs → chunk-TGHNXN7W.mjs} +150 -52
  54. package/dist/{chunk-Z4DZ4KK2.mjs → chunk-UJWGQIXU.mjs} +1 -0
  55. package/dist/{chunk-ANFNYCQG.mjs → chunk-WVRDRLIC.mjs} +6 -6
  56. package/dist/{chunk-62IPM62F.mjs → chunk-X5CNUKD5.mjs} +2 -2
  57. package/dist/{chunk-IE43SRL3.mjs → chunk-XABMCBHA.mjs} +12 -6
  58. package/dist/{chunk-OCB7TMUQ.mjs → chunk-Y2U5VVWX.mjs} +1 -1
  59. package/dist/{chunk-MFPJVITN.mjs → chunk-Y4BVTQ53.mjs} +2 -2
  60. package/dist/{chunk-N6J6EJW3.mjs → chunk-YBWSV5DV.mjs} +2 -2
  61. package/dist/{chunk-FUUP45IS.mjs → chunk-ZR4VM4A6.mjs} +1 -1
  62. package/dist/{client-CsuhZ13c.d.ts → client-Cx-TiqPE.d.ts} +19 -3
  63. package/dist/client.d.ts +8 -8
  64. package/dist/client.mjs +10 -2
  65. package/dist/{config-CtTfxbpV.d.ts → config-CzB9YvF_.d.ts} +2 -2
  66. package/dist/{config-DLVnlByZ.d.ts → config-hOWGBeC6.d.ts} +7 -5
  67. package/dist/direct-executor-HQ4QVDX7.mjs +28 -0
  68. package/dist/distribution/coordinator/etcd.d.ts +2 -2
  69. package/dist/distribution/coordinator/etcd.mjs +18 -1
  70. package/dist/distribution/coordinator/in-memory.d.ts +2 -2
  71. package/dist/distribution/coordinator/in-memory.mjs +33 -1
  72. package/dist/distribution/transport/grpc.mjs +1 -1
  73. package/dist/distribution.d.ts +71 -43
  74. package/dist/distribution.mjs +411 -127
  75. package/dist/{durability-wiring-UGGUNATA.mjs → durability-wiring-S7OX2MAI.mjs} +28 -26
  76. package/dist/embeddings/openai.mjs +2 -2
  77. package/dist/{engine-Bq5wz6c1.d.ts → engine-B_W9sdZK.d.ts} +38 -8
  78. package/dist/{events-x2aVu0_o.d.ts → events-DErop1LM.d.ts} +23 -2
  79. package/dist/{index-Cl5qQauj.d.ts → index-CKI1fDEl.d.ts} +37 -3
  80. package/dist/index.browser.mjs +5 -3
  81. package/dist/index.d.ts +7 -7
  82. package/dist/index.mjs +21 -19
  83. package/dist/{long-query-thread-5BIPZVS6.mjs → long-query-thread-WVSMIBYP.mjs} +17 -15
  84. package/dist/{module-7OHZ2OPI.mjs → module-2E4XYSGT.mjs} +11 -11
  85. package/dist/partial-words.mjs +7 -6
  86. package/dist/pattern.mjs +4 -3
  87. package/dist/persistence/durability/checkpoint-worker.d.ts +3 -1
  88. package/dist/persistence/durability/checkpoint-worker.mjs +24 -22
  89. package/dist/phrase-module-JERF2IDG.mjs +329 -0
  90. package/dist/react.d.ts +7 -7
  91. package/dist/react.mjs +3 -0
  92. package/dist/{rerank-NYX5S7OG.mjs → rerank-DMMT4FZA.mjs} +6 -6
  93. package/dist/rerankers/http.mjs +2 -2
  94. package/dist/{restore-queue-NVMWJMHI.mjs → restore-queue-JAMNRC6T.mjs} +1 -1
  95. package/dist/{runtime-XNYW6AWX.mjs → runtime-7MLQ7J64.mjs} +7 -7
  96. package/dist/{schema-4W_7tk1f.d.ts → schema-CS-clLun.d.ts} +54 -28
  97. package/dist/server.d.ts +10 -10
  98. package/dist/server.mjs +27 -15
  99. package/dist/{snapshot-P7IQGNNQ.mjs → snapshot-DOQLGGEX.mjs} +45 -30
  100. package/dist/synonyms.mjs +4 -3
  101. package/dist/{tasks-DLQIfXqy.d.ts → tasks-BfF-VRhm.d.ts} +4 -6
  102. package/dist/{thread-QPETQEPK.mjs → thread-DNK5XWXG.mjs} +16 -14
  103. package/dist/vector/search-worker.d.ts +1 -1
  104. package/dist/vector/search-worker.mjs +4 -4
  105. package/dist/workers/entry.mjs +8 -8
  106. package/dist/{write-WZVM7ZL6.mjs → write-ZKR7SNRT.mjs} +23 -21
  107. package/package.json +7 -7
  108. package/dist/direct-executor-6KCVVHYT.mjs +0 -26
@@ -1,15 +1,17 @@
1
- import { PATTERN_EXPRESSION_OPERATORS, PATTERN_TEST_OPERATORS, PATTERN_ELEMENT_OPERATORS, requireTextRangeField, isFiniteNumber, invalid, requireFieldKind, kindOfValue, requireListLength, isObjectRecord, patternSearchModule, DEFAULT_PATTERN_WORK_CAP, bitsetHas, createBitSet, bitsetSet, bitsetIsEmpty, bitsetOrInto, bitsetAndNotInto, bitsetAndInto, TYPO_CLAIM_SLOTS_PER_WORD, bitsetFromSet, PATTERN_WORK_FLUSH_UNITS, MATCHER_STATE_BUDGET_BYTES } from './chunk-P34VZEXB.mjs';
1
+ import { PATTERN_EXPRESSION_OPERATORS, PATTERN_TEST_OPERATORS, PATTERN_ELEMENT_OPERATORS, requireTextRangeField, isFiniteNumber, invalid, requireFieldKind, kindOfValue, requireListLength, isObjectRecord, patternSearchModule, DEFAULT_PATTERN_WORK_CAP, TYPO_CLAIM_SLOTS_PER_WORD, PATTERN_WORK_FLUSH_UNITS, MATCHER_STATE_BUDGET_BYTES } from './chunk-2SEL6SBS.mjs';
2
2
  import { GEO_UNIT_METRES, MAX_FILTER_DEPTH, MAX_FILTER_TESTS, FILTER_OPERAND_COPY_DEPTH } from './chunk-WQBCBKY3.mjs';
3
- import { expandTermPrefix, suggestDisplayTerms, getNestedValue, textFieldsChanged, getFlatSchema, tokenizeOptions, partialWordSearchModule, getOrCreateFieldNameIndex, partitionDocumentFrequency } from './chunk-KCQG2PJT.mjs';
4
- import { isLatitude, MAX_LATITUDE_DEGREES, isLongitude, MAX_LONGITUDE_DEGREES, MIN_POLYGON_POINTS, partialFieldsOf, schemaFieldsOf, flattenSchema, validateDocument, validateDocumentStrict, ANTIMERIDIAN_SPAN_DEGREES, FULL_TURN_DEGREES, LOCATION_CODE_BITS_PER_AXIS, MAX_COVERING_CELLS, VINCENTY_CONVERGENCE_THRESHOLD, EARTH_RADIUS_METERS, RADIUS_BOX_MARGIN_RATIO, RADIUS_BOX_MARGIN_RADIANS, DEG_TO_RAD, VINCENTY_MAX_ITERATIONS } from './chunk-DG62H7OD.mjs';
5
- import { MAX_SORT_FIELDS, MAX_SORT_FIELD_NAME_LENGTH, MAX_CURSOR_LENGTH, MAX_CURSOR_ANCHOR_CODE_POINTS, RESULT_WINDOW, DEFAULT_FUZZY_PREFIX_LENGTH, DEFAULT_PAGE_SIZE, PREFIX_MAX_EXPANSIONS } from './chunk-E74YLAIG.mjs';
3
+ import { bitsetHas, createBitSet, bitsetSet, bitsetIsEmpty, bitsetOrInto, bitsetAndNotInto, bitsetAndInto, bitsetFromSet } from './chunk-JUKYYXSP.mjs';
4
+ import { expandTermPrefix, suggestDisplayTerms, getNestedValue, textFieldsChanged, getFlatSchema, tokenizeOptions, partialWordSearchModule, getOrCreateFieldNameIndex, partitionDocumentFrequency } from './chunk-EUCX5K74.mjs';
5
+ import { isLatitude, MAX_LATITUDE_DEGREES, isLongitude, MAX_LONGITUDE_DEGREES, MIN_POLYGON_POINTS, partialFieldsOf, schemaFieldsOf, flattenSchema, validateDocument, validateDocumentStrict, ANTIMERIDIAN_SPAN_DEGREES, FULL_TURN_DEGREES, LOCATION_CODE_BITS_PER_AXIS, MAX_COVERING_CELLS, VINCENTY_CONVERGENCE_THRESHOLD, EARTH_RADIUS_METERS, RADIUS_BOX_MARGIN_RATIO, RADIUS_BOX_MARGIN_RADIANS, DEG_TO_RAD, VINCENTY_MAX_ITERATIONS } from './chunk-PU6SKH7O.mjs';
6
+ import { MAX_SORT_FIELDS, MAX_SORT_FIELD_NAME_LENGTH, MAX_CURSOR_LENGTH, MAX_CURSOR_ANCHOR_CODE_POINTS, RESULT_WINDOW, DEFAULT_FUZZY_PREFIX_LENGTH, DEFAULT_PAGE_SIZE, PREFIX_MAX_EXPANSIONS } from './chunk-FQBFZAWG.mjs';
6
7
  import { fuzzyPrefixOf, boundedLevenshtein, fuzzyTermMatches } from './chunk-GES7U2PM.mjs';
7
- import { parsedTypeOf, isStringValuedBase, withSortableOption, isWordIndexedBase, isWordIndexedFieldType, isListBase } from './chunk-6ODIY6ZB.mjs';
8
- import { tokenize, SCORE_BUFFER_INITIAL_TOUCHED_CAPACITY, PRUNING_REJECTION_SLACK, visitTokens, SORT_COLUMN_MINIMUM_REBUILD_THRESHOLD, TARGET_BLOCK_ENTRIES, SORT_COLUMN_REBUILD_FRACTION_SHIFT } from './chunk-OFYKBKKV.mjs';
8
+ import { parsedTypeOf, isStringValuedBase, withSortableOption, isWordIndexedBase, isWordIndexedFieldType, isListBase } from './chunk-7XY6VD7B.mjs';
9
+ import { tokenize, visitTokens } from './chunk-D5TFRB5G.mjs';
10
+ import { SCORE_BUFFER_INITIAL_TOUCHED_CAPACITY, PRUNING_REJECTION_SLACK, SORT_COLUMN_MINIMUM_REBUILD_THRESHOLD, TARGET_BLOCK_ENTRIES, SORT_COLUMN_REBUILD_FRACTION_SHIFT } from './chunk-4UMZ7KQL.mjs';
9
11
  import { singleFoldTable, appendFoldedCodePoint, sortModeOf, defaultSortMode, SORT_VALUE_MAX_CODE_POINTS, toReducedSortValue, readSortField, compareComparableKeys, toComparableSortValue, isSortMode, SORT_MODES, compareSortValues, compareComparableValues, truncateSortString } from './chunk-R6A773WK.mjs';
10
12
  import { compareCodePoints, exceedsCodePoints, codePointCount, holdsUnpairedSurrogate } from './chunk-YTJCGKM4.mjs';
11
13
  import { isGeopointOrVectorType, PROTOTYPE_POLLUTION_KEYS } from './chunk-L5ORED7V.mjs';
12
- import { NarsilError, ErrorCodes } from './chunk-OBWUWZRA.mjs';
14
+ import { NarsilError, ErrorCodes } from './chunk-CZ7ISLXA.mjs';
13
15
 
14
16
  // src/core/partition/partial-words/platform.ts
15
17
  function platformPartialWordSearch() {
@@ -165,6 +167,17 @@ function requirePatternSearch(fieldPath) {
165
167
  );
166
168
  }
167
169
 
170
+ // src/core/constants.ts
171
+ var POSTING_LIST_COMPACTION_THRESHOLD = 0.3;
172
+ var MAX_TERM_FREQUENCY = 65535;
173
+ var ARRAY_POSITION_GAP = 100;
174
+ var DEFAULT_BM25_K1 = 1.2;
175
+ var DEFAULT_BM25_B = 0.75;
176
+ var MAX_BM25_K1 = 1e3;
177
+ var MAX_BM25_B = 1;
178
+ var PENDING_REMOVAL_MERGE_FLOOR = 64;
179
+ var PENDING_REMOVAL_MERGE_DIVISOR = 256;
180
+
168
181
  // src/core/projection.ts
169
182
  function requireFieldPathList(paths, option) {
170
183
  if (paths === void 0) return;
@@ -594,11 +607,64 @@ function createDocumentStore() {
594
607
  };
595
608
  }
596
609
 
597
- // src/core/constants.ts
598
- var POSTING_LIST_COMPACTION_THRESHOLD = 0.3;
599
- var MAX_TERM_FREQUENCY = 65535;
600
- var PENDING_REMOVAL_MERGE_FLOOR = 64;
601
- var PENDING_REMOVAL_MERGE_DIVISOR = 256;
610
+ // src/search/quotes.ts
611
+ var QUOTE_CHARACTER = /["“”„]/;
612
+ function holdsQuoteCharacter(term) {
613
+ return QUOTE_CHARACTER.test(term);
614
+ }
615
+ function holdsQuotedText(term) {
616
+ if (!QUOTE_CHARACTER.test(term)) return false;
617
+ const parts = term.split(QUOTE_CHARACTER);
618
+ for (let index = 1; index < parts.length; index += 2) {
619
+ if (parts[index].trim().length > 0) return true;
620
+ }
621
+ return false;
622
+ }
623
+
624
+ // src/search/phrase-code.ts
625
+ var loaded = null;
626
+ var loading = null;
627
+ function requireLoadedPhraseCode() {
628
+ if (loaded === null) {
629
+ throw new NarsilError(ErrorCodes.SEARCH_PHRASE_NOT_READY, "The phrase code has not loaded. Try the query again");
630
+ }
631
+ return loaded;
632
+ }
633
+ function loadPhraseCode() {
634
+ if (loading === null) {
635
+ loading = import('./phrase-module-JERF2IDG.mjs').then(
636
+ (code) => {
637
+ loaded = code;
638
+ return code;
639
+ },
640
+ (err) => {
641
+ loading = null;
642
+ throw new NarsilError(
643
+ ErrorCodes.SEARCH_PHRASE_NOT_READY,
644
+ "The phrase code failed to load. Remove the quotes, or try again once it can load",
645
+ { cause: err instanceof Error ? err.message : String(err) }
646
+ );
647
+ }
648
+ );
649
+ }
650
+ return loading;
651
+ }
652
+ function loadPhraseCodeFor(term) {
653
+ if (loaded !== null || term === void 0 || !holdsQuoteCharacter(term)) return null;
654
+ return loadPhraseCode();
655
+ }
656
+ function analyseQueryPhrases(term, language, analysis) {
657
+ if (!holdsQuoteCharacter(term)) return null;
658
+ return requireLoadedPhraseCode().analyseQueryPhrases(
659
+ term.split(QUOTE_CHARACTER),
660
+ (text) => tokenize(text, language, {
661
+ stem: true,
662
+ removeStopWords: true,
663
+ customTokenizer: analysis.customTokenizer,
664
+ stopWordOverride: analysis.stopWords
665
+ }).tokens
666
+ );
667
+ }
602
668
 
603
669
  // src/core/inverted-index/constants.ts
604
670
  var TERM_TABLE_START_CAPACITY = 1 << 10;
@@ -1030,6 +1096,13 @@ var LivePostingView = class {
1030
1096
  const starts = this.store.pool.positionStarts;
1031
1097
  return starts === null ? 0 : this.store.positions.countAt(starts[this.syncedStart + row]);
1032
1098
  }
1099
+ get positionStore() {
1100
+ return this.holdsPositions ? this.store.positions.values : null;
1101
+ }
1102
+ positionStartAt(row) {
1103
+ const starts = this.store.pool.positionStarts;
1104
+ return starts === null ? 0 : starts[this.syncedStart + row] + 1;
1105
+ }
1033
1106
  copyPositionsAt(row, target, offset) {
1034
1107
  if (!this.holdsPositions) return 0;
1035
1108
  const starts = this.store.pool.positionStarts;
@@ -3022,7 +3095,7 @@ function indexStringArrayField(state, internalId, fieldPath, arr, language, opti
3022
3095
  if (state.trackPositions) {
3023
3096
  const tokenFreqs = /* @__PURE__ */ new Map();
3024
3097
  let positionOffset = 0;
3025
- let itemTokenCount = 0;
3098
+ let lastItemPosition = -1;
3026
3099
  const addToken = (token, position, surface) => {
3027
3100
  const existing = tokenFreqs.get(token);
3028
3101
  if (existing) {
@@ -3033,12 +3106,12 @@ function indexStringArrayField(state, internalId, fieldPath, arr, language, opti
3033
3106
  }
3034
3107
  fieldTokenList.push(token);
3035
3108
  countOccurrence(surfaceCounts, spellings, surface, token);
3036
- itemTokenCount++;
3109
+ lastItemPosition = position;
3037
3110
  };
3038
3111
  for (const item of arr) {
3039
- itemTokenCount = 0;
3112
+ lastItemPosition = -1;
3040
3113
  visitTokens(item, language, opts, addToken);
3041
- positionOffset += itemTokenCount;
3114
+ if (lastItemPosition >= 0) positionOffset += lastItemPosition + 1 + ARRAY_POSITION_GAP;
3042
3115
  }
3043
3116
  fieldLengths[fieldPath] = fieldTokenList.length;
3044
3117
  for (const [token, freq] of tokenFreqs) {
@@ -4732,19 +4805,63 @@ function partitionFilterMatches(state, filters, schema, meter) {
4732
4805
  }
4733
4806
 
4734
4807
  // src/core/scorer.ts
4735
- var DEFAULT_K1 = 1.2;
4736
- var DEFAULT_B = 0.75;
4737
4808
  function computeIDF(docFrequency, totalDocs) {
4738
4809
  if (totalDocs === 0) return 0;
4739
4810
  return Math.log((totalDocs - docFrequency + 0.5) / (docFrequency + 0.5) + 1);
4740
4811
  }
4812
+ function pairKeepsBoundSound(k1, b) {
4813
+ return k1 >= 0 && k1 <= MAX_BM25_K1 && b >= 0 && b <= MAX_BM25_B;
4814
+ }
4741
4815
  function bm25PruningSound(params) {
4742
- const k1 = params?.k1 ?? DEFAULT_K1;
4743
- const b = params?.b ?? DEFAULT_B;
4744
- return Number.isFinite(k1) && k1 >= 0 && b >= 0 && b <= 1;
4816
+ const k1 = params?.k1 ?? DEFAULT_BM25_K1;
4817
+ const b = params?.b ?? DEFAULT_BM25_B;
4818
+ if (!pairKeepsBoundSound(k1, b)) return false;
4819
+ const fields = params?.fields;
4820
+ if (fields === void 0) return true;
4821
+ for (const name of Object.keys(fields)) {
4822
+ if (!pairKeepsBoundSound(fields[name].k1 ?? k1, fields[name].b ?? b)) return false;
4823
+ }
4824
+ return true;
4825
+ }
4826
+ function bm25SetsValue(params) {
4827
+ if (typeof params !== "object" || params === null) return params !== void 0;
4828
+ if (params.k1 !== void 0 || params.b !== void 0) return true;
4829
+ const fields = params.fields;
4830
+ if (fields === void 0) return false;
4831
+ if (typeof fields !== "object" || fields === null) return true;
4832
+ for (const name of Object.keys(fields)) {
4833
+ const pair = fields[name];
4834
+ if (typeof pair !== "object" || pair === null) return true;
4835
+ if (pair.k1 !== void 0 || pair.b !== void 0) return true;
4836
+ }
4837
+ return false;
4745
4838
  }
4746
- function resolveBM25Params(params) {
4747
- return { k1: params?.k1 ?? DEFAULT_K1, b: params?.b ?? DEFAULT_B };
4839
+ function ownField(fields, name) {
4840
+ return fields !== void 0 && Object.hasOwn(fields, name) ? fields[name] : void 0;
4841
+ }
4842
+ function mergeBM25Levels(search, index) {
4843
+ if (search === void 0 || !bm25SetsValue(search)) return index;
4844
+ const k1 = search.k1 ?? index?.k1;
4845
+ const b = search.b ?? index?.b;
4846
+ const merged = {
4847
+ ...k1 !== void 0 ? { k1 } : {},
4848
+ ...b !== void 0 ? { b } : {}
4849
+ };
4850
+ const names = [...Object.keys(search.fields ?? {}), ...Object.keys(index?.fields ?? {})];
4851
+ if (names.length === 0) return merged;
4852
+ const fields = /* @__PURE__ */ Object.create(null);
4853
+ for (const name of names) {
4854
+ const fromSearch = ownField(search.fields, name);
4855
+ const fromIndex = ownField(index?.fields, name);
4856
+ const fieldK1 = fromSearch?.k1 ?? search.k1 ?? fromIndex?.k1;
4857
+ const fieldB = fromSearch?.b ?? search.b ?? fromIndex?.b;
4858
+ fields[name] = {
4859
+ ...fieldK1 !== void 0 ? { k1: fieldK1 } : {},
4860
+ ...fieldB !== void 0 ? { b: fieldB } : {}
4861
+ };
4862
+ }
4863
+ merged.fields = fields;
4864
+ return merged;
4748
4865
  }
4749
4866
  function computeBM25WithIDF(termFrequency, idf, fieldLength, avgFieldLength, k1, b) {
4750
4867
  if (avgFieldLength === 0) return 0;
@@ -4753,29 +4870,49 @@ function computeBM25WithIDF(termFrequency, idf, fieldLength, avgFieldLength, k1,
4753
4870
  if (denominator === 0) return 0;
4754
4871
  return idf * (numerator / denominator);
4755
4872
  }
4756
- function computeBM25(termFrequency, docFrequency, totalDocs, fieldLength, avgFieldLength, params) {
4757
- if (totalDocs === 0) return 0;
4758
- const k1 = params?.k1 ?? DEFAULT_K1;
4759
- const b = params?.b ?? DEFAULT_B;
4760
- return computeBM25WithIDF(termFrequency, computeIDF(docFrequency, totalDocs), fieldLength, avgFieldLength, k1, b);
4761
- }
4762
4873
 
4763
4874
  // src/core/partition/field-scoring.ts
4764
- function loadFieldScoring(state, fields, boost, averageFieldLengths) {
4875
+ var NO_FIELD_VALUES = new Float64Array(0);
4876
+ function differentPairsOf(searchable, k1s, bs) {
4877
+ const pairs = [];
4878
+ for (let fieldIndex = 0; fieldIndex < searchable.length; fieldIndex++) {
4879
+ if (searchable[fieldIndex] !== 1) continue;
4880
+ let seen = false;
4881
+ for (let at = 0; at < pairs.length && !seen; at += 2) {
4882
+ seen = pairs[at] === k1s[fieldIndex] && pairs[at + 1] === bs[fieldIndex];
4883
+ }
4884
+ if (!seen) pairs.push(k1s[fieldIndex], bs[fieldIndex]);
4885
+ }
4886
+ return pairs;
4887
+ }
4888
+ function loadFieldScoring(state, fields, boost, averageFieldLengths, bm25) {
4765
4889
  const fieldNames = state.fieldNameTable.names;
4890
+ const fieldCount = fieldNames.length;
4891
+ const k1 = bm25?.k1 ?? DEFAULT_BM25_K1;
4892
+ const b = bm25?.b ?? DEFAULT_BM25_B;
4893
+ const perField = bm25?.fields;
4766
4894
  const scoring = {
4767
- searchable: new Uint8Array(fieldNames.length),
4768
- boosts: new Float64Array(fieldNames.length),
4769
- averageLengths: new Float64Array(fieldNames.length),
4770
- lengthColumns: new Array(fieldNames.length).fill(null)
4895
+ searchable: new Uint8Array(fieldCount),
4896
+ boosts: new Float64Array(fieldCount),
4897
+ averageLengths: new Float64Array(fieldCount),
4898
+ lengthColumns: new Array(fieldCount).fill(null),
4899
+ k1s: perField === void 0 ? NO_FIELD_VALUES : new Float64Array(fieldCount),
4900
+ bs: perField === void 0 ? NO_FIELD_VALUES : new Float64Array(fieldCount),
4901
+ boundPairs: [k1, b]
4771
4902
  };
4772
- for (let fieldIndex = 0; fieldIndex < fieldNames.length; fieldIndex++) {
4903
+ for (let fieldIndex = 0; fieldIndex < fieldCount; fieldIndex++) {
4773
4904
  const fieldName = fieldNames[fieldIndex];
4774
4905
  scoring.searchable[fieldIndex] = fields === void 0 || fields.includes(fieldName) ? 1 : 0;
4775
4906
  scoring.boosts[fieldIndex] = boost?.[fieldName] ?? 1;
4776
4907
  scoring.averageLengths[fieldIndex] = averageFieldLengths[fieldName] ?? 1;
4777
4908
  scoring.lengthColumns[fieldIndex] = state.docStore.fieldLengthColumn(fieldName);
4909
+ if (perField !== void 0) {
4910
+ const fieldPair = Object.hasOwn(perField, fieldName) ? perField[fieldName] : void 0;
4911
+ scoring.k1s[fieldIndex] = fieldPair?.k1 ?? k1;
4912
+ scoring.bs[fieldIndex] = fieldPair?.b ?? b;
4913
+ }
4778
4914
  }
4915
+ if (perField !== void 0) scoring.boundPairs = differentPairsOf(scoring.searchable, scoring.k1s, scoring.bs);
4779
4916
  return scoring;
4780
4917
  }
4781
4918
  function fieldLengthOf(columns, fieldIndex, internalId, averageLength) {
@@ -4791,6 +4928,15 @@ function bestSearchable(searchable, values) {
4791
4928
  }
4792
4929
  return best;
4793
4930
  }
4931
+ function blockEntryBound(fields, maxTermFrequency, idf, minFieldLength, maxAverageLength) {
4932
+ const pairs = fields.boundPairs;
4933
+ let best = 0;
4934
+ for (let at = 0; at < pairs.length; at += 2) {
4935
+ const bound = computeBM25WithIDF(maxTermFrequency, idf, minFieldLength, maxAverageLength, pairs[at], pairs[at + 1]);
4936
+ if (bound > best) best = bound;
4937
+ }
4938
+ return best;
4939
+ }
4794
4940
 
4795
4941
  // src/core/partition/posting-columns.ts
4796
4942
  function postingColumns(list) {
@@ -4884,8 +5030,9 @@ function groupPostingCount(matches) {
4884
5030
  return total;
4885
5031
  }
4886
5032
  function computeGroupContributions(context, matches, collect) {
4887
- const { fields, fieldNames, filterBitset, partialMarks, k1, b } = context;
5033
+ const { fields, fieldNames, filterBitset, partialMarks } = context;
4888
5034
  const scoresAreZero = context.totalDocs === 0;
5035
+ const onePair = fields.boundPairs.length === 2;
4889
5036
  const best = /* @__PURE__ */ new Map();
4890
5037
  for (const match of matches) {
4891
5038
  const perTerm = /* @__PURE__ */ new Map();
@@ -4905,7 +5052,14 @@ function computeGroupContributions(context, matches, collect) {
4905
5052
  const fieldBoost = fields.boosts[fieldIndex];
4906
5053
  const avgLen = fields.averageLengths[fieldIndex];
4907
5054
  const actualFieldLength = fieldLengthOf(fields.lengthColumns, fieldIndex, internalId, avgLen);
4908
- const termScore = (scoresAreZero ? 0 : computeBM25WithIDF(termFrequency, match.idf, actualFieldLength, avgLen, k1, b)) * fieldBoost * factor;
5055
+ const termScore = (scoresAreZero ? 0 : computeBM25WithIDF(
5056
+ termFrequency,
5057
+ match.idf,
5058
+ actualFieldLength,
5059
+ avgLen,
5060
+ onePair ? fields.boundPairs[0] : fields.k1s[fieldIndex],
5061
+ onePair ? fields.boundPairs[1] : fields.bs[fieldIndex]
5062
+ )) * fieldBoost * factor;
4909
5063
  const existing = perTerm.get(internalId);
4910
5064
  if (existing) {
4911
5065
  existing.score += termScore;
@@ -4962,11 +5116,20 @@ function searchedFieldMarks(fieldNames, fields) {
4962
5116
  function searchFulltextMatches(state, params) {
4963
5117
  const { queryTokens, termGroups, fields, tolerance = 0, prefixLength = 2, exact = false, filterBitset } = params;
4964
5118
  const fieldNames = state.fieldNameTable.names;
4965
- const matched = createBitSet(state.docStore.internalIdCapacity());
4966
5119
  const searched = searchedFieldMarks(fieldNames, fields);
4967
5120
  const partial = partialFieldMarks(fieldNames, params.partialFields);
4968
5121
  const searchedPartial = searched.map((mark, fieldIndex) => mark & partial[fieldIndex]);
4969
- for (const qt of queryTokens) {
5122
+ const capacity = state.docStore.internalIdCapacity();
5123
+ const matched = params.phrases === void 0 ? createBitSet(capacity) : requireLoadedPhraseCode().matchPhrases(
5124
+ state.invertedIdx,
5125
+ params.phrases,
5126
+ searched,
5127
+ filterBitset,
5128
+ capacity,
5129
+ params.phraseWork,
5130
+ false
5131
+ ).matched;
5132
+ for (const qt of params.phrases === void 0 ? queryTokens : []) {
4970
5133
  const group = termGroupOf(termGroups, qt.token);
4971
5134
  for (const match of typoMatchesFor(state.invertedIdx, qt.token, group, exact, tolerance, prefixLength)) {
4972
5135
  markPostingList(matched, match.postingList, searched, filterBitset);
@@ -5469,7 +5632,7 @@ function boostsKeepScoresNonNegative(fields) {
5469
5632
  }
5470
5633
  return true;
5471
5634
  }
5472
- function termUpperBound(scan2, list, maxBoost, maxAverageLength) {
5635
+ function samePairUpperBound(scan2, list, maxBoost, maxAverageLength) {
5473
5636
  if (scan2.scoresAreZero) return 0;
5474
5637
  const bounds = blockBoundsFor(list, scan2.fields.lengthColumns);
5475
5638
  let best = 0;
@@ -5487,7 +5650,24 @@ function termUpperBound(scan2, list, maxBoost, maxAverageLength) {
5487
5650
  }
5488
5651
  return best;
5489
5652
  }
5490
- function entryScore(scan2, entry, internalId) {
5653
+ function fieldPairUpperBound(scan2, list, maxBoost, maxAverageLength) {
5654
+ if (scan2.scoresAreZero) return 0;
5655
+ const bounds = blockBoundsFor(list, scan2.fields.lengthColumns);
5656
+ let best = 0;
5657
+ for (let block = 0; block < bounds.blockCount; block++) {
5658
+ const bestEntry = blockEntryBound(
5659
+ scan2.fields,
5660
+ bounds.maxTermFrequency[block],
5661
+ scan2.idf,
5662
+ bounds.minFieldLength[block],
5663
+ maxAverageLength
5664
+ );
5665
+ const blockBound = bestEntry * maxBoost * bounds.maxEntriesPerDocument[block];
5666
+ if (blockBound > best) best = blockBound;
5667
+ }
5668
+ return best;
5669
+ }
5670
+ function samePairEntryScore(scan2, entry, internalId) {
5491
5671
  const { fields, columns } = scan2;
5492
5672
  const fieldIndex = columns.fieldNameIndices[entry];
5493
5673
  const averageLength = fields.averageLengths[fieldIndex];
@@ -5495,17 +5675,42 @@ function entryScore(scan2, entry, internalId) {
5495
5675
  const bm25 = scan2.scoresAreZero ? 0 : computeBM25WithIDF(columns.termFrequencies[entry], scan2.idf, fieldLength, averageLength, scan2.k1, scan2.b);
5496
5676
  return bm25 * fields.boosts[fieldIndex];
5497
5677
  }
5498
- function scoreEveryDocument(scan2, buffer) {
5678
+ function fieldPairEntryScore(scan2, entry, internalId) {
5679
+ const { fields, columns } = scan2;
5680
+ const fieldIndex = columns.fieldNameIndices[entry];
5681
+ const averageLength = fields.averageLengths[fieldIndex];
5682
+ const fieldLength = fieldLengthOf(fields.lengthColumns, fieldIndex, internalId, averageLength);
5683
+ const bm25 = scan2.scoresAreZero ? 0 : computeBM25WithIDF(
5684
+ columns.termFrequencies[entry],
5685
+ scan2.idf,
5686
+ fieldLength,
5687
+ averageLength,
5688
+ fields.k1s[fieldIndex],
5689
+ fields.bs[fieldIndex]
5690
+ );
5691
+ return bm25 * fields.boosts[fieldIndex];
5692
+ }
5693
+ function samePairScoreEveryDocument(scan2, buffer) {
5694
+ const { docIds, fieldNameIndices, deletedDocs, hasDeleted, count } = scan2.columns;
5695
+ const searchable = scan2.fields.searchable;
5696
+ for (let entry = 0; entry < count; entry++) {
5697
+ const internalId = docIds[entry];
5698
+ if (hasDeleted && deletedDocs.has(internalId)) continue;
5699
+ if (searchable[fieldNameIndices[entry]] !== 1) continue;
5700
+ addScore(buffer, internalId, samePairEntryScore(scan2, entry, internalId));
5701
+ }
5702
+ }
5703
+ function fieldPairScoreEveryDocument(scan2, buffer) {
5499
5704
  const { docIds, fieldNameIndices, deletedDocs, hasDeleted, count } = scan2.columns;
5500
5705
  const searchable = scan2.fields.searchable;
5501
5706
  for (let entry = 0; entry < count; entry++) {
5502
5707
  const internalId = docIds[entry];
5503
5708
  if (hasDeleted && deletedDocs.has(internalId)) continue;
5504
5709
  if (searchable[fieldNameIndices[entry]] !== 1) continue;
5505
- addScore(buffer, internalId, entryScore(scan2, entry, internalId));
5710
+ addScore(buffer, internalId, fieldPairEntryScore(scan2, entry, internalId));
5506
5711
  }
5507
5712
  }
5508
- function scoreKnownDocumentsAndCountTheRest(scan2, buffer) {
5713
+ function samePairScoreKnownDocuments(scan2, buffer) {
5509
5714
  const { docIds, fieldNameIndices, deletedDocs, hasDeleted, count } = scan2.columns;
5510
5715
  const searchable = scan2.fields.searchable;
5511
5716
  let counted = 0;
@@ -5513,31 +5718,71 @@ function scoreKnownDocumentsAndCountTheRest(scan2, buffer) {
5513
5718
  const internalId = docIds[entry];
5514
5719
  if (hasDeleted && deletedDocs.has(internalId)) continue;
5515
5720
  if (searchable[fieldNameIndices[entry]] !== 1) continue;
5516
- if (hasScore(buffer, internalId)) addScore(buffer, internalId, entryScore(scan2, entry, internalId));
5721
+ if (hasScore(buffer, internalId)) addScore(buffer, internalId, samePairEntryScore(scan2, entry, internalId));
5517
5722
  else if (markCounted(buffer, internalId)) counted++;
5518
5723
  }
5519
5724
  return counted;
5520
5725
  }
5521
- function queryOrderScore(scans, internalId) {
5726
+ function fieldPairScoreKnownDocuments(scan2, buffer) {
5727
+ const { docIds, fieldNameIndices, deletedDocs, hasDeleted, count } = scan2.columns;
5728
+ const searchable = scan2.fields.searchable;
5729
+ let counted = 0;
5730
+ for (let entry = 0; entry < count; entry++) {
5731
+ const internalId = docIds[entry];
5732
+ if (hasDeleted && deletedDocs.has(internalId)) continue;
5733
+ if (searchable[fieldNameIndices[entry]] !== 1) continue;
5734
+ if (hasScore(buffer, internalId)) addScore(buffer, internalId, fieldPairEntryScore(scan2, entry, internalId));
5735
+ else if (markCounted(buffer, internalId)) counted++;
5736
+ }
5737
+ return counted;
5738
+ }
5739
+ function firstEntryOf(docIds, count, internalId) {
5740
+ let low = 0;
5741
+ let high = count;
5742
+ while (low < high) {
5743
+ const middle = low + high >>> 1;
5744
+ if (docIds[middle] < internalId) low = middle + 1;
5745
+ else high = middle;
5746
+ }
5747
+ return low;
5748
+ }
5749
+ function samePairQueryOrderScore(scans, internalId) {
5522
5750
  let score = 0;
5523
5751
  for (const scan2 of scans) {
5524
5752
  const { docIds, deletedDocs, hasDeleted, count } = scan2.columns;
5525
5753
  if (hasDeleted && deletedDocs.has(internalId)) continue;
5526
- let low = 0;
5527
- let high = count;
5528
- while (low < high) {
5529
- const middle = low + high >>> 1;
5530
- if (docIds[middle] < internalId) low = middle + 1;
5531
- else high = middle;
5754
+ for (let entry = firstEntryOf(docIds, count, internalId); entry < count && docIds[entry] === internalId; entry++) {
5755
+ if (scan2.fields.searchable[scan2.columns.fieldNameIndices[entry]] !== 1) continue;
5756
+ score += samePairEntryScore(scan2, entry, internalId);
5532
5757
  }
5533
- for (let entry = low; entry < count && docIds[entry] === internalId; entry++) {
5758
+ }
5759
+ return score;
5760
+ }
5761
+ function fieldPairQueryOrderScore(scans, internalId) {
5762
+ let score = 0;
5763
+ for (const scan2 of scans) {
5764
+ const { docIds, deletedDocs, hasDeleted, count } = scan2.columns;
5765
+ if (hasDeleted && deletedDocs.has(internalId)) continue;
5766
+ for (let entry = firstEntryOf(docIds, count, internalId); entry < count && docIds[entry] === internalId; entry++) {
5534
5767
  if (scan2.fields.searchable[scan2.columns.fieldNameIndices[entry]] !== 1) continue;
5535
- score += entryScore(scan2, entry, internalId);
5768
+ score += fieldPairEntryScore(scan2, entry, internalId);
5536
5769
  }
5537
5770
  }
5538
5771
  return score;
5539
5772
  }
5540
- function pageFromBuffer(scans, buffer, selection, resolver) {
5773
+ var SAME_PAIR_SCORER = {
5774
+ upperBound: samePairUpperBound,
5775
+ scoreEveryDocument: samePairScoreEveryDocument,
5776
+ scoreKnownDocumentsAndCountTheRest: samePairScoreKnownDocuments,
5777
+ queryOrderScore: samePairQueryOrderScore
5778
+ };
5779
+ var FIELD_PAIR_SCORER = {
5780
+ upperBound: fieldPairUpperBound,
5781
+ scoreEveryDocument: fieldPairScoreEveryDocument,
5782
+ scoreKnownDocumentsAndCountTheRest: fieldPairScoreKnownDocuments,
5783
+ queryOrderScore: fieldPairQueryOrderScore
5784
+ };
5785
+ function pageFromBuffer(scorer, scans, buffer, selection, resolver) {
5541
5786
  const approximateKth = kthBestScore(buffer, selection, resolver);
5542
5787
  const gatherFrom = approximateKth - Math.abs(approximateKth) * PRUNING_REJECTION_SLACK;
5543
5788
  const heap = [];
@@ -5549,7 +5794,7 @@ function pageFromBuffer(scans, buffer, selection, resolver) {
5549
5794
  if (scores[internalId] < gatherFrom) continue;
5550
5795
  const externalId = resolver.toExternal(internalId);
5551
5796
  if (externalId === void 0) continue;
5552
- const score = queryOrderScore(scans, internalId);
5797
+ const score = scorer.queryOrderScore(scans, internalId);
5553
5798
  if (full && score < threshold) continue;
5554
5799
  const candidate = { internalId, externalId, score };
5555
5800
  if (!full) {
@@ -5579,18 +5824,28 @@ function pageFromBuffer(scans, buffer, selection, resolver) {
5579
5824
  return scored;
5580
5825
  }
5581
5826
  function multiTermTopK(request) {
5582
- const { terms, docFrequencies, totalDocs, bm25Params, limit, fields, resolver, buffer } = request;
5827
+ const { terms, docFrequencies, totalDocs, limit, fields, resolver, buffer } = request;
5583
5828
  if (!boostsKeepScoresNonNegative(fields)) return null;
5584
- const { k1, b } = resolveBM25Params(bm25Params);
5829
+ const onePair = fields.boundPairs.length === 2;
5830
+ const scorer = onePair ? SAME_PAIR_SCORER : FIELD_PAIR_SCORER;
5831
+ const k1 = onePair ? fields.boundPairs[0] : 0;
5832
+ const b = onePair ? fields.boundPairs[1] : 0;
5585
5833
  const scans = new Array(terms.length);
5586
5834
  for (let term = 0; term < terms.length; term++) {
5587
5835
  const idf = computeIDF(docFrequencies[term], totalDocs);
5588
5836
  if (!(idf >= 0)) return null;
5589
- scans[term] = { columns: postingColumns(terms[term].list), idf, k1, b, scoresAreZero: totalDocs === 0, fields };
5837
+ scans[term] = {
5838
+ columns: postingColumns(terms[term].list),
5839
+ idf,
5840
+ scoresAreZero: totalDocs === 0,
5841
+ fields,
5842
+ k1,
5843
+ b
5844
+ };
5590
5845
  }
5591
5846
  const maxBoost = bestSearchable(fields.searchable, fields.boosts);
5592
5847
  const maxAverageLength = bestSearchable(fields.searchable, fields.averageLengths);
5593
- const upperBounds = scans.map((scan2, term) => termUpperBound(scan2, terms[term].list, maxBoost, maxAverageLength));
5848
+ const upperBounds = scans.map((scan2, term) => scorer.upperBound(scan2, terms[term].list, maxBoost, maxAverageLength));
5594
5849
  const order = scans.map((_, term) => term).sort((left, right) => upperBounds[right] - upperBounds[left]);
5595
5850
  const remainingBound = new Float64Array(order.length + 1);
5596
5851
  for (let position2 = order.length - 1; position2 >= 0; position2--) {
@@ -5605,15 +5860,15 @@ function multiTermTopK(request) {
5605
5860
  const threshold = kthBestScore(buffer, selection, resolver);
5606
5861
  if (remaining < threshold - threshold * PRUNING_REJECTION_SLACK) break;
5607
5862
  }
5608
- scoreEveryDocument(scans[order[position]], buffer);
5863
+ scorer.scoreEveryDocument(scans[order[position]], buffer);
5609
5864
  }
5610
5865
  let countedOnly = 0;
5611
5866
  for (let rest = position; rest < order.length; rest++) {
5612
- countedOnly += scoreKnownDocumentsAndCountTheRest(scans[order[rest]], buffer);
5867
+ countedOnly += scorer.scoreKnownDocumentsAndCountTheRest(scans[order[rest]], buffer);
5613
5868
  }
5614
5869
  const totalMatched = buffer.touchedCount + countedOnly;
5615
5870
  if (wanted === 0) return { scored: [], totalMatched };
5616
- return { scored: pageFromBuffer(scans, buffer, selection, resolver), totalMatched };
5871
+ return { scored: pageFromBuffer(scorer, scans, buffer, selection, resolver), totalMatched };
5617
5872
  }
5618
5873
 
5619
5874
  // src/core/partition/single-term-topk.ts
@@ -5641,20 +5896,25 @@ function prunableSingleTermList(params, index, fieldNames) {
5641
5896
  return list;
5642
5897
  }
5643
5898
  function singleTermTopK(request) {
5644
- const { list, docFrequency, totalDocs, bm25Params, limit, resolver } = request;
5899
+ const { list, docFrequency, totalDocs, limit, resolver } = request;
5645
5900
  const {
5646
5901
  searchable: fieldSearchable,
5647
5902
  boosts: fieldBoosts,
5648
5903
  averageLengths: fieldAvgLengths,
5649
- lengthColumns: fieldLengthColumns
5904
+ lengthColumns: fieldLengthColumns,
5905
+ k1s,
5906
+ bs,
5907
+ boundPairs
5650
5908
  } = request.fields;
5909
+ const onePair = boundPairs.length === 2;
5910
+ const onlyK1 = boundPairs[0];
5911
+ const onlyB = boundPairs[1];
5651
5912
  const wanted = Number.isFinite(limit) ? Math.max(0, Math.floor(limit)) : 0;
5652
5913
  const { docIds, termFrequencies, fieldNameIndices, deletedDocs: deleted, hasDeleted } = postingColumns(list);
5653
5914
  const bounds = blockBoundsFor(list, fieldLengthColumns);
5654
5915
  const maxBoost = bestSearchable(fieldSearchable, fieldBoosts);
5655
5916
  const maxAverageLength = bestSearchable(fieldSearchable, fieldAvgLengths);
5656
5917
  const idf = computeIDF(docFrequency, totalDocs);
5657
- const { k1, b } = resolveBM25Params(bm25Params);
5658
5918
  const scoresAreZero = totalDocs === 0;
5659
5919
  const heap = [];
5660
5920
  let full = false;
@@ -5664,14 +5924,12 @@ function singleTermTopK(request) {
5664
5924
  const start = block === 0 ? 0 : bounds.entryEnd[block - 1];
5665
5925
  const end = bounds.entryEnd[block];
5666
5926
  if (full) {
5667
- const bestEntry = computeBM25(
5668
- bounds.maxTermFrequency[block],
5669
- docFrequency,
5670
- totalDocs,
5671
- bounds.minFieldLength[block],
5672
- maxAverageLength,
5673
- bm25Params
5674
- );
5927
+ let bestEntry = 0;
5928
+ if (!scoresAreZero) {
5929
+ const blockMaxTf = bounds.maxTermFrequency[block];
5930
+ const blockMinLength = bounds.minFieldLength[block];
5931
+ bestEntry = onePair ? computeBM25WithIDF(blockMaxTf, idf, blockMinLength, maxAverageLength, onlyK1, onlyB) : blockEntryBound(request.fields, blockMaxTf, idf, blockMinLength, maxAverageLength);
5932
+ }
5675
5933
  if (bestEntry * maxBoost * bounds.maxEntriesPerDocument[block] < threshold) {
5676
5934
  totalMatched += bounds.documentCount[block];
5677
5935
  continue;
@@ -5690,7 +5948,14 @@ function singleTermTopK(request) {
5690
5948
  const fieldIndex = fieldNameIndices[entry];
5691
5949
  if (fieldSearchable[fieldIndex] === 1) {
5692
5950
  const fieldLength = fieldLengthOf(fieldLengthColumns, fieldIndex, internalId, fieldAvgLengths[fieldIndex]);
5693
- score += (scoresAreZero ? 0 : computeBM25WithIDF(termFrequencies[entry], idf, fieldLength, fieldAvgLengths[fieldIndex], k1, b)) * fieldBoosts[fieldIndex];
5951
+ score += (scoresAreZero ? 0 : computeBM25WithIDF(
5952
+ termFrequencies[entry],
5953
+ idf,
5954
+ fieldLength,
5955
+ fieldAvgLengths[fieldIndex],
5956
+ onePair ? onlyK1 : k1s[fieldIndex],
5957
+ onePair ? onlyB : bs[fieldIndex]
5958
+ )) * fieldBoosts[fieldIndex];
5694
5959
  scored2 = true;
5695
5960
  }
5696
5961
  entry++;
@@ -5750,15 +6015,14 @@ function searchFulltext(state, params) {
5750
6015
  globalStats,
5751
6016
  maxResults,
5752
6017
  termMatch,
5753
- filterBitset
6018
+ phrases
5754
6019
  } = params;
5755
6020
  const collectComponents = params.collectComponents !== false;
5756
- if (queryTokens.length === 0) {
6021
+ if (queryTokens.length === 0 && phrases === void 0) {
5757
6022
  return { scored: [], totalMatched: 0 };
5758
6023
  }
5759
6024
  const totalDocs = globalStats?.totalDocuments ?? state.stats.totalDocuments;
5760
6025
  const avgFieldLengths = globalStats?.averageFieldLengths ?? state.stats.averageFieldLengths;
5761
- const { k1, b } = resolveBM25Params(bm25Params);
5762
6026
  const scoresAreZero = totalDocs === 0;
5763
6027
  if (state.scoreBuffer === null) state.scoreBuffer = createScoreBuffer(state.docStore.internalIdCapacity());
5764
6028
  const scoreBuffer = state.scoreBuffer;
@@ -5767,19 +6031,59 @@ function searchFulltext(state, params) {
5767
6031
  const useIntersection = termMatch === "all" && queryTokens.length > 1;
5768
6032
  const fieldNames = state.fieldNameTable.names;
5769
6033
  const resolver = state.docStore.resolver();
5770
- const fieldScoring = loadFieldScoring(state, fields, boost, avgFieldLengths);
5771
- const { searchable, boosts, averageLengths, lengthColumns } = fieldScoring;
6034
+ const fieldScoring = loadFieldScoring(state, fields, boost, avgFieldLengths, bm25Params);
6035
+ const { searchable, boosts, averageLengths, lengthColumns, k1s, bs, boundPairs } = fieldScoring;
6036
+ const onePair = boundPairs.length === 2;
6037
+ const onlyK1 = boundPairs[0];
6038
+ const onlyB = boundPairs[1];
6039
+ const phraseMatches = phrases === void 0 ? null : requireLoadedPhraseCode().matchPhrases(
6040
+ state.invertedIdx,
6041
+ phrases,
6042
+ searchable,
6043
+ params.filterBitset,
6044
+ state.docStore.internalIdCapacity(),
6045
+ params.phraseWork,
6046
+ true
6047
+ );
6048
+ if (phraseMatches !== null && bitsetIsEmpty(phraseMatches.matched)) {
6049
+ return { scored: [], totalMatched: 0 };
6050
+ }
6051
+ const filterBitset = phraseMatches?.matched ?? params.filterBitset;
5772
6052
  function documentFrequency(token, list) {
5773
6053
  return globalStats ? globalDocFreqFor(globalStats.docFrequencies, token, list.docIdSet.size) : list.docIdSet.size;
5774
6054
  }
6055
+ function scorePhraseRows(rows, matched) {
6056
+ let idf = 0;
6057
+ for (let entry = 0; entry < rows.phrase.tokens.length; entry++) {
6058
+ idf += computeIDF(documentFrequency(rows.phrase.tokens[entry], rows.lists[entry]), totalDocs);
6059
+ }
6060
+ for (let row = 0; row < rows.docIds.length; row++) {
6061
+ const internalId = rows.docIds.values[row];
6062
+ if (!bitsetHas(matched, internalId)) continue;
6063
+ const fieldIndex = rows.fieldIndices.values[row];
6064
+ const frequency = rows.frequencies.values[row];
6065
+ const avgLen = averageLengths[fieldIndex];
6066
+ const fieldLength = fieldLengthOf(lengthColumns, fieldIndex, internalId, avgLen);
6067
+ const score = scoresAreZero ? 0 : computeBM25WithIDF(
6068
+ frequency,
6069
+ idf,
6070
+ fieldLength,
6071
+ avgLen,
6072
+ onePair ? onlyK1 : k1s[fieldIndex],
6073
+ onePair ? onlyB : bs[fieldIndex]
6074
+ );
6075
+ addScore(scoreBuffer, internalId, score * boosts[fieldIndex]);
6076
+ if (components !== null) {
6077
+ recordComponents(components, internalId, fieldNames[fieldIndex], rows.phrase.key, frequency, fieldLength, idf);
6078
+ }
6079
+ }
6080
+ }
5775
6081
  const groupContext = {
5776
6082
  index: state.invertedIdx,
5777
6083
  fields: fieldScoring,
5778
6084
  partialMarks: partialFieldMarks(fieldNames, partialFields),
5779
6085
  fieldNames,
5780
6086
  totalDocs,
5781
- k1,
5782
- b,
5783
6087
  filterBitset,
5784
6088
  globalFrequency(token) {
5785
6089
  if (globalStats === void 0 || !Object.hasOwn(globalStats.docFrequencies, token)) return void 0;
@@ -5805,7 +6109,14 @@ function searchFulltext(state, params) {
5805
6109
  if (searchable[fieldIndex] === 0) continue;
5806
6110
  const avgLen = averageLengths[fieldIndex];
5807
6111
  const actualFieldLength = fieldLengthOf(lengthColumns, fieldIndex, internalId, avgLen);
5808
- const termScore = scoresAreZero ? 0 : computeBM25WithIDF(termFrequencies[pi], match.idf, actualFieldLength, avgLen, k1, b);
6112
+ const termScore = scoresAreZero ? 0 : computeBM25WithIDF(
6113
+ termFrequencies[pi],
6114
+ match.idf,
6115
+ actualFieldLength,
6116
+ avgLen,
6117
+ onePair ? onlyK1 : k1s[fieldIndex],
6118
+ onePair ? onlyB : bs[fieldIndex]
6119
+ );
5809
6120
  addScore(scoreBuffer, internalId, termScore * boosts[fieldIndex]);
5810
6121
  }
5811
6122
  return;
@@ -5820,7 +6131,14 @@ function searchFulltext(state, params) {
5820
6131
  const termFrequency = termFrequencies[pi];
5821
6132
  const avgLen = averageLengths[fieldIndex];
5822
6133
  const actualFieldLength = fieldLengthOf(lengthColumns, fieldIndex, internalId, avgLen);
5823
- const termScore = scoresAreZero ? 0 : computeBM25WithIDF(termFrequency, match.idf, actualFieldLength, avgLen, k1, b);
6134
+ const termScore = scoresAreZero ? 0 : computeBM25WithIDF(
6135
+ termFrequency,
6136
+ match.idf,
6137
+ actualFieldLength,
6138
+ avgLen,
6139
+ onePair ? onlyK1 : k1s[fieldIndex],
6140
+ onePair ? onlyB : bs[fieldIndex]
6141
+ );
5824
6142
  addScore(scoreBuffer, internalId, termScore * boosts[fieldIndex]);
5825
6143
  if (components !== null) {
5826
6144
  recordComponents(
@@ -5842,25 +6160,23 @@ function searchFulltext(state, params) {
5842
6160
  mergeGroupContribution(internalId, contribution);
5843
6161
  }
5844
6162
  }
5845
- const prunableList = prunableSingleTermList(params, state.invertedIdx, fieldNames);
6163
+ const prunableList = phrases === void 0 ? prunableSingleTermList(params, state.invertedIdx, fieldNames) : null;
5846
6164
  if (prunableList !== null && maxResults !== void 0) {
5847
6165
  return singleTermTopK({
5848
6166
  list: prunableList,
5849
6167
  docFrequency: documentFrequency(queryTokens[0].token, prunableList),
5850
6168
  totalDocs,
5851
- bm25Params,
5852
6169
  limit: maxResults,
5853
6170
  fields: fieldScoring,
5854
6171
  resolver
5855
6172
  });
5856
6173
  }
5857
- const prunableLists = prunableMultiTermLists(params, state.invertedIdx);
6174
+ const prunableLists = phrases === void 0 ? prunableMultiTermLists(params, state.invertedIdx) : null;
5858
6175
  if (prunableLists !== null && maxResults !== void 0) {
5859
6176
  const pruned = multiTermTopK({
5860
6177
  terms: prunableLists,
5861
6178
  docFrequencies: prunableLists.map((term) => documentFrequency(term.token, term.list)),
5862
6179
  totalDocs,
5863
- bm25Params,
5864
6180
  limit: maxResults,
5865
6181
  fields: fieldScoring,
5866
6182
  resolver,
@@ -5881,7 +6197,7 @@ function searchFulltext(state, params) {
5881
6197
  }
5882
6198
  resolved.push({ token: qt.token, matches, totalPostings, group: groupMatches });
5883
6199
  }
5884
- resolved.sort((a, b2) => a.totalPostings - b2.totalPostings);
6200
+ resolved.sort((a, b) => a.totalPostings - b.totalPostings);
5885
6201
  for (let tokenIndex = 0; tokenIndex < resolved.length; tokenIndex++) {
5886
6202
  const onlyScored = tokenIndex > 0;
5887
6203
  for (const match of resolved[tokenIndex].matches) scorePostings(match, onlyScored);
@@ -5896,6 +6212,9 @@ function searchFulltext(state, params) {
5896
6212
  if (group !== void 0) scoreGroup(resolveGroupMatches(groupContext, group), false);
5897
6213
  }
5898
6214
  }
6215
+ if (phraseMatches !== null) {
6216
+ for (const rows of phraseMatches.rows) scorePhraseRows(rows, phraseMatches.matched);
6217
+ }
5899
6218
  const totalMatched = scoreBuffer.touchedCount;
5900
6219
  const k = maxResults === void 0 ? totalMatched : Math.max(0, Math.min(maxResults, totalMatched));
5901
6220
  const scored = topKFromBuffer(scoreBuffer, k, resolver, components);
@@ -7131,6 +7450,9 @@ function heapPop(heap) {
7131
7450
  function collectCandidateTerms(subs, params) {
7132
7451
  const { queryTokens, termGroups, tolerance = 0, prefixLength = 2, exact = false } = params;
7133
7452
  const terms = /* @__PURE__ */ new Set();
7453
+ for (const phrase of params.phrases ?? []) {
7454
+ for (const token of phrase.tokens) terms.add(token);
7455
+ }
7134
7456
  for (const qt of queryTokens) {
7135
7457
  const group = termGroupOf(termGroups, qt.token);
7136
7458
  if (group !== void 0) {
@@ -7232,7 +7554,7 @@ function searchSingleSub(subs, layout, params, index) {
7232
7554
  return { ...result, matchedOrdinalBitset };
7233
7555
  }
7234
7556
  function compositeSearchFulltext(subs, layout, params) {
7235
- if (params.queryTokens.length === 0) {
7557
+ if (params.queryTokens.length === 0 && params.phrases === void 0) {
7236
7558
  return { scored: [], totalMatched: 0 };
7237
7559
  }
7238
7560
  const single = onlyPopulatedSub(subs);
@@ -7661,7 +7983,7 @@ var PREPARED_TYPO_TEST_BYTES = Math.ceil(MAX_FILTER_TESTS / TYPO_CLAIM_SLOTS_PER
7661
7983
  function capExceeded(cap) {
7662
7984
  return new NarsilError(
7663
7985
  ErrorCodes.SEARCH_WORK_CAP_EXCEEDED,
7664
- `The engine stopped this search before it checked every candidate, because the work on its pattern tests passed the cap of ${cap} units. Narrow the tests, or raise the patternWorkCap engine setting`,
7986
+ `The engine stopped this search before it checked every candidate, because the work on its pattern tests and phrases passed the cap of ${cap} units. Narrow the tests or the phrases, or raise the patternWorkCap engine setting`,
7665
7987
  { cap }
7666
7988
  );
7667
7989
  }
@@ -7892,15 +8214,19 @@ function addPartitionTokens(tokens, addedMembers, partition, term, analysed, exp
7892
8214
  }
7893
8215
  }
7894
8216
  }
7895
- function addReachedTokens(tokens, partitions, term, expansion, language, analysis) {
8217
+ function addReachedTokens(tokens, partitions, term, requested, language, analysis) {
8218
+ if (requested.options.exact === true) return [];
8219
+ const quoted = analyseQueryPhrases(term, language, analysis);
8220
+ const words = quoted?.words ?? term;
8221
+ if (quoted !== null && words.trim().length === 0) return [];
8222
+ const expansion = quoted?.endsWithPhrase === true ? { schema: requested.schema, options: { ...requested.options, prefix: false } } : requested;
7896
8223
  const { options } = expansion;
7897
- if (options.exact === true) return [];
7898
8224
  if (!((options.tolerance ?? 0) > 0 || options.prefix === true || options.partialWords === true)) return [];
7899
- const analysed = queryTokensOf(term, language, analysis);
8225
+ const analysed = queryTokensOf(words, language, analysis);
7900
8226
  if (analysed === null) return [];
7901
8227
  const addedMembers = /* @__PURE__ */ new Set();
7902
8228
  for (const partition of partitions) {
7903
- addPartitionTokens(tokens, addedMembers, partition, term, analysed, expansion, language, analysis);
8229
+ addPartitionTokens(tokens, addedMembers, partition, words, analysed, expansion, language, analysis);
7904
8230
  }
7905
8231
  for (const token of addedMembers) tokens.add(token);
7906
8232
  return [...addedMembers];
@@ -7923,17 +8249,24 @@ function partitionsIn(manager, partitionIds) {
7923
8249
 
7924
8250
  // src/partitioning/distributed-scoring.ts
7925
8251
  function analysedQueryTokens(term, language, analysis) {
7926
- const { tokens } = tokenize(term, language, {
7927
- stem: true,
7928
- removeStopWords: true,
7929
- customTokenizer: analysis.customTokenizer,
7930
- stopWordOverride: analysis.stopWords
7931
- });
8252
+ const quoted = analyseQueryPhrases(term, language, analysis);
8253
+ const words = quoted?.words ?? term;
7932
8254
  const unique = /* @__PURE__ */ new Set();
7933
- for (const { token } of tokens) {
7934
- unique.add(token);
8255
+ if (quoted === null || words.trim().length > 0) {
8256
+ const { tokens } = tokenize(words, language, {
8257
+ stem: true,
8258
+ removeStopWords: true,
8259
+ customTokenizer: analysis.customTokenizer,
8260
+ stopWordOverride: analysis.stopWords
8261
+ });
8262
+ for (const { token } of tokens) {
8263
+ unique.add(token);
8264
+ }
7935
8265
  }
7936
8266
  for (const synonym of analysis.synonyms?.expand(unique)?.tokens ?? []) unique.add(synonym);
8267
+ for (const phrase of quoted?.phrases ?? []) {
8268
+ for (const token of phrase.tokens) unique.add(token);
8269
+ }
7937
8270
  return unique;
7938
8271
  }
7939
8272
  function copyNumericRecord(source) {
@@ -8336,6 +8669,14 @@ function applyPagination(results, limit, offset, binding, cursor, sort, pinnedId
8336
8669
  }
8337
8670
 
8338
8671
  // src/search/fulltext.ts
8672
+ function analysedWords(term, language, options) {
8673
+ return tokenize(term, language, {
8674
+ stem: true,
8675
+ removeStopWords: true,
8676
+ customTokenizer: options?.customTokenizer,
8677
+ stopWordOverride: options?.stopWords
8678
+ }).tokens;
8679
+ }
8339
8680
  function prepareFulltextQuery(partition, params, language, schema, options) {
8340
8681
  if (!params.term || params.term.trim().length === 0) return null;
8341
8682
  if (params.fields && params.fields.length === 0) return null;
@@ -8343,14 +8684,16 @@ function prepareFulltextQuery(partition, params, language, schema, options) {
8343
8684
  if (params.fields) {
8344
8685
  validateSearchFields(params.fields, flatSchema);
8345
8686
  }
8346
- const queryTokenResult = tokenize(params.term, language, {
8347
- stem: true,
8348
- removeStopWords: true,
8349
- customTokenizer: options?.customTokenizer,
8350
- stopWordOverride: options?.stopWords
8687
+ const quoted = analyseQueryPhrases(params.term, language, {
8688
+ stopWords: options?.stopWords,
8689
+ customTokenizer: options?.customTokenizer
8351
8690
  });
8352
- if (queryTokenResult.tokens.length === 0) return null;
8353
- const queryTokens = deduplicateTokens(queryTokenResult.tokens);
8691
+ const wordParams = quoted === null ? params : requireLoadedPhraseCode().withWordsOnly(params, quoted);
8692
+ const phrases = quoted !== null && quoted.phrases.length > 0 ? quoted.phrases : void 0;
8693
+ const words = wordParams.term ?? "";
8694
+ const analysedTokens = quoted !== null && words.trim().length === 0 ? [] : analysedWords(words, language, options);
8695
+ if (analysedTokens.length === 0 && phrases === void 0) return null;
8696
+ const queryTokens = deduplicateTokens(analysedTokens);
8354
8697
  let filterBitset;
8355
8698
  if (params.filters) {
8356
8699
  filterBitset = partition.applyFiltersBitset(
@@ -8360,16 +8703,19 @@ function prepareFulltextQuery(partition, params, language, schema, options) {
8360
8703
  );
8361
8704
  if (bitsetIsEmpty(filterBitset)) return null;
8362
8705
  }
8363
- const lastToken = queryTokenResult.tokens[queryTokenResult.tokens.length - 1];
8706
+ if (analysedTokens.length === 0) return { queryTokens, phrases, filterBitset };
8707
+ const lastToken = analysedTokens[analysedTokens.length - 1];
8364
8708
  const synonyms = params.exact === true ? void 0 : options?.synonyms?.expand(queryTokens.map((qt) => qt.token));
8365
- const groups = resolveTermGroups(partition, params, language, schema, queryTokens, lastToken, options, synonyms);
8366
- return { queryTokens, ...groups, filterBitset };
8709
+ const groups = resolveTermGroups(partition, wordParams, language, schema, queryTokens, lastToken, options, synonyms);
8710
+ return { queryTokens, phrases, ...groups, filterBitset };
8367
8711
  }
8368
8712
  function fulltextMatches(partition, params, language, schema, options) {
8369
8713
  const prepared = prepareFulltextQuery(partition, params, language, schema, options);
8370
8714
  if (prepared === null) return null;
8371
8715
  return partition.searchFulltextMatches({
8372
8716
  queryTokens: prepared.queryTokens,
8717
+ phrases: prepared.phrases,
8718
+ phraseWork: options?.patternWork,
8373
8719
  termGroups: prepared.termGroups,
8374
8720
  partialFields: prepared.partialFields,
8375
8721
  fields: params.fields,
@@ -8384,13 +8730,15 @@ function fulltextSearch(partition, params, language, schema, options) {
8384
8730
  if (prepared === null) {
8385
8731
  return { scored: [], totalMatched: 0 };
8386
8732
  }
8387
- const { queryTokens, termGroups, partialFields, filterBitset } = prepared;
8733
+ const { queryTokens, phrases, termGroups, partialFields, filterBitset } = prepared;
8388
8734
  const needsAllResults = params.minScore !== void 0 || params.termMatch !== void 0 && params.termMatch !== "any" || params.sort !== void 0 || params.group !== void 0 || params.pinned !== void 0 || params.searchAfter !== void 0;
8389
8735
  const maxResults = needsAllResults ? void 0 : clampRowCount(params.limit, DEFAULT_PAGE_SIZE) + clampRowCount(params.offset, 0) + 1;
8390
8736
  const collectMatchedSet = params.facets !== void 0 && !needsAllResults ? "ordinals" : void 0;
8391
8737
  const collectComponents = params.includeScoreComponents === true || params.termMatch !== void 0 && params.termMatch !== "any";
8392
8738
  const rawResult = partition.searchFulltext({
8393
8739
  queryTokens,
8740
+ phrases,
8741
+ phraseWork: options?.patternWork,
8394
8742
  termGroups,
8395
8743
  partialFields,
8396
8744
  fields: params.fields,
@@ -8408,7 +8756,7 @@ function fulltextSearch(partition, params, language, schema, options) {
8408
8756
  });
8409
8757
  let scored = rawResult.scored;
8410
8758
  const termMatch = params.termMatch ?? "any";
8411
- if (termMatch !== "any") {
8759
+ if (termMatch !== "any" && queryTokens.length > 0) {
8412
8760
  scored = filterByTermCoverage(
8413
8761
  scored,
8414
8762
  queryTokens,
@@ -8417,7 +8765,8 @@ function fulltextSearch(partition, params, language, schema, options) {
8417
8765
  params.prefixLength ?? DEFAULT_FUZZY_PREFIX_LENGTH,
8418
8766
  params.exact ?? false,
8419
8767
  termGroups,
8420
- partialFields ?? []
8768
+ partialFields ?? [],
8769
+ phrases
8421
8770
  );
8422
8771
  }
8423
8772
  if (params.minScore !== void 0 && params.minScore > 0) {
@@ -8459,8 +8808,9 @@ function deduplicateTokens(tokens) {
8459
8808
  }
8460
8809
  return result;
8461
8810
  }
8462
- function filterByTermCoverage(scored, queryTokens, policy, tolerance, prefixLength, exact, termGroups, partialFields) {
8811
+ function filterByTermCoverage(scored, queryTokens, policy, tolerance, prefixLength, exact, termGroups, partialFields, phrases) {
8463
8812
  const requiredCount = policy === "all" ? queryTokens.length : policy;
8813
+ const phraseKeys = phrases === void 0 ? null : new Set(phrases.map((phrase) => phrase.key));
8464
8814
  if (requiredCount <= 0) return scored;
8465
8815
  if (requiredCount > queryTokens.length) return [];
8466
8816
  const groups = /* @__PURE__ */ new Map();
@@ -8476,12 +8826,22 @@ function filterByTermCoverage(scored, queryTokens, policy, tolerance, prefixLeng
8476
8826
  });
8477
8827
  }
8478
8828
  return scored.filter((doc) => {
8479
- const matched = countDocTermMatches(doc, queryTokens, tolerance, prefixLength, exact, groups, partialFields);
8829
+ const matched = countDocTermMatches(
8830
+ doc,
8831
+ queryTokens,
8832
+ tolerance,
8833
+ prefixLength,
8834
+ exact,
8835
+ groups,
8836
+ partialFields,
8837
+ phraseKeys
8838
+ );
8480
8839
  return matched >= requiredCount;
8481
8840
  });
8482
8841
  }
8483
- function countDocTermMatches(doc, queryTokens, tolerance, prefixLength, exact, groups, partialFields) {
8484
- const indexTokens = Object.keys(doc.idf);
8842
+ function countDocTermMatches(doc, queryTokens, tolerance, prefixLength, exact, groups, partialFields, phraseKeys) {
8843
+ const componentKeys = Object.keys(doc.idf);
8844
+ const indexTokens = phraseKeys === null ? componentKeys : componentKeys.filter((key) => !phraseKeys.has(key));
8485
8845
  if (indexTokens.length === 0) return 0;
8486
8846
  let count = 0;
8487
8847
  for (const qt of queryTokens) {
@@ -8634,4 +8994,4 @@ function fanOutMatchCount(manager, params, language, schema, options) {
8634
8994
  return count;
8635
8995
  }
8636
8996
 
8637
- export { FIELD_FILTER_OPERATORS, FILTER_EXPRESSION_KEYS, FNV_OFFSET_BASIS, MAX_TERM_FREQUENCY, PATTERN_HOST, ValueAffixes, applyPagination, applyPartitionFiltersBitset, applyProjection, applySorting, availablePatternSearch, buildPartialWordTableFrom, clampRowCount, cloneProjected, collectQueryTermStats, copyVectorsInto, countPartialSpelling, countsWithoutScores, createBooleanIndex, createCompositePartition, createEnumIndex, createFrozenPartialWords, createGeoIndex, createNumericIndex, createPartitionIndex, createPatternWorkMeter, createSurfaceRegistry, decodePageCursor, documentValueBytes, documentWithOwnBuffers, encodeFieldIndexes, encodePageCursor, encodePartialSpellingsOf, encodeSegmentState, everyValueFacetConfig, fanOutMatchCount, fanOutQuery, filterTestCount, fnv1a, fnv1aContinue, fnv1aContinueTaggedNumbers, fnv1aUint32s, fulltextMatches, isCompositePartition, joinPatternWorkMeter, kWayMerge, mergeFacets, mergePartitionStats, normalizeSort, oversampledFacetConfig, oversampledShardSize, partialSpellingsOfDocuments, partialWordHighlighter, partialWordsToHighlight, partitionStateOf, partitionsIn, patternIndexOf, projectIncludedPaths, projectionKeepsField, pruneStatsToQueryTerms, readProjectedPath, readSortValues, recordValueSize, registerPartialWordSearch, registerPatternSearch, requireMatchingCursor, requirePartialWordSearchFor, requirePatternSearch, requirePatternSearchFor, requireSortableFields, requireWithinResultWindow, resolveProjection, sanitizeGlobalStats, setNestedValue, sortModesOf, sortSignatureEntry, sortSignatureOf, survivorDocumentsOf, undeclaredFieldOnStrictIndex, validatedFilterOf, valueWithOwnBuffers };
8997
+ export { ARRAY_POSITION_GAP, DEFAULT_BM25_B, DEFAULT_BM25_K1, FIELD_FILTER_OPERATORS, FILTER_EXPRESSION_KEYS, FNV_OFFSET_BASIS, MAX_BM25_B, MAX_BM25_K1, MAX_TERM_FREQUENCY, PATTERN_HOST, ValueAffixes, analyseQueryPhrases, applyPagination, applyPartitionFiltersBitset, applyProjection, applySorting, availablePatternSearch, bm25SetsValue, buildPartialWordTableFrom, clampRowCount, cloneProjected, collectQueryTermStats, copyVectorsInto, countPartialSpelling, countsWithoutScores, createBooleanIndex, createCompositePartition, createEnumIndex, createFrozenPartialWords, createGeoIndex, createNumericIndex, createPartitionIndex, createPatternWorkMeter, createSurfaceRegistry, decodePageCursor, documentValueBytes, documentWithOwnBuffers, encodeFieldIndexes, encodePageCursor, encodePartialSpellingsOf, encodeSegmentState, everyValueFacetConfig, fanOutMatchCount, fanOutQuery, filterTestCount, fnv1a, fnv1aContinue, fnv1aContinueTaggedNumbers, fnv1aUint32s, fulltextMatches, holdsQuotedText, isCompositePartition, joinPatternWorkMeter, kWayMerge, loadPhraseCodeFor, mergeBM25Levels, mergeFacets, mergePartitionStats, normalizeSort, oversampledFacetConfig, oversampledShardSize, partialSpellingsOfDocuments, partialWordHighlighter, partialWordsToHighlight, partitionStateOf, partitionsIn, patternIndexOf, projectIncludedPaths, projectionKeepsField, pruneStatsToQueryTerms, readProjectedPath, readSortValues, recordValueSize, registerPartialWordSearch, registerPatternSearch, requireLoadedPhraseCode, requireMatchingCursor, requirePartialWordSearchFor, requirePatternSearch, requirePatternSearchFor, requireSortableFields, requireWithinResultWindow, resolveProjection, sanitizeGlobalStats, setNestedValue, sortModesOf, sortSignatureEntry, sortSignatureOf, survivorDocumentsOf, undeclaredFieldOnStrictIndex, validatedFilterOf, valueWithOwnBuffers };