code-gauge 4.4.0 → 4.5.0

package/dist/duplication.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"duplication.cjs","names":[],"sources":["../src/duplication.ts"],"sourcesContent":["import type { DuplicationOptions } from './types.js';\n\n/**\n * Project-level duplication machinery operating on normalized token streams. Tokenization itself\n * (parsing, identifier anonymization, literal normalization) happens in the Rust addon, which\n * serializes each file's Token stream and statement structure; the helpers here match statement\n * windows across files, merge gap-adjacent groups, and count duplicated lines over that data.\n */\n\nexport const defaultDuplicationOptions: Required<DuplicationOptions> = {\n  minTokens: 40,\n  maxGapTokens: 30,\n  minSimilarityPercent: 70,\n};\n\n/**\n * Fills defaults for absent settings, applying the same normalization as the native boundary's\n * clampToU32 — NaN (e.g. `Number(unsetEnvVariable)`) counts as absent, and other values truncate\n * and clamp to [0, u32::MAX] — so the TypeScript half of cross-file matching cannot diverge from\n * the natively collected candidates on such input.\n */\nexport function resolveDuplicationOptions(options?: DuplicationOptions): Required<DuplicationOptions> {\n  return {\n    minTokens: resolveOption(options?.minTokens, defaultDuplicationOptions.minTokens),\n    maxGapTokens: resolveOption(options?.maxGapTokens, defaultDuplicationOptions.maxGapTokens),\n    minSimilarityPercent: resolveOption(options?.minSimilarityPercent, defaultDuplicationOptions.minSimilarityPercent),\n  };\n}\n\nfunction resolveOption(value: number | undefined, fallback: number): number {\n  return value === undefined || Number.isNaN(value)\n    ? fallback\n    : Math.min(Math.max(Math.trunc(value), 0), 0xFF_FF_FF_FF);\n}\n\n/** Minimum consecutive statements for a statement-sequence duplicate candidate. */\nconst minSequenceStatementCount = 2;\n/**\n * Caps the window length so statement-sequence enumeration stays linear in the statement count.\n * Heterogeneous clones longer than the cap are reported as capped windows (a deliberate\n * conservative undercount trading completeness for bounded discovery cost).\n */\nconst maxSequenceStatementCount = 100;\n\n/**\n * A region whose normalized tokens are at least 20% literal values is data-like (a lookup table, a\n * constant list, a value-mapping switch), not logic: literal values re-enter its fingerprint so\n * tables that merely share their shape stop counting as copy-paste. Compared in integer math\n * (5 * literals >= total) so the TypeScript and native sides cannot disagree on the boundary.\n */\nfunction isLiteralDense(literalCount: number, tokenCount: number): boolean {\n  return literalCount * 5 >= tokenCount;\n}\n\nexport interface Token {\n  /** Normalization target: identifiers to anonymize, literal kind tags, or the raw token text. */\n  kind: 'id' | 'text';\n  text: string;\n  /**\n   * Two INDEPENDENT hashes of `text` (djb2 and FNV-1a), precomputed so fingerprinting nested\n   * regions never re-hashes a token. Feeding the same per-token hash to both fingerprint\n   * accumulators would collapse the key to 32 effective bits: one djb2 collision between two\n   * token texts would then equate whole regions.\n   */\n  textHash: number;\n  textHash2: number;\n  /** Hash pair of a value-carrying literal's value, folded into data-like region fingerprints. */\n  literalHash?: number;\n  literalHash2?: number;\n  /**\n   * True for verbatim-kept NAMES (member/callee/type names, named grammar leaves): together with\n   * value-carrying literals these are the content-bearing tokens the near-miss content gate\n   * counts. Keywords, operators, and punctuation come from unnamed nodes and stay false.\n   */\n  isName?: boolean;\n  /** 0-based source rows the token occupies, so line coverage counts only matched-token lines. */\n  startRow: number;\n  endRow: number;\n}\n\n/** A token span with its source position; plain data so cross-file matching can retain it. */\nexport interface TokenRange {\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\n/** A contiguous run of matched tokens; gapped (merged) duplicates carry several per occurrence. */\ninterface TokenSegment {\n  startTokenIndex: number;\n  endTokenIndex: number;\n}\n\ninterface DuplicateCandidate {\n  fingerprint: string;\n  tokenCount: number;\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\nexport interface CountedOccurrence {\n  /** Matched token runs; more than one once gapped groups are merged. */\n  segments: TokenSegment[];\n  /**\n   * Set on occurrences whose span another reported group already counts: a retained group's\n   * occurrences that a partial gapped merge also paired into a merged group, and cross-file copies\n   * nested inside a larger group's region. Block counting must not count them again.\n   */\n  spanCountedElsewhere?: boolean;\n  /**\n   * Set on cross-file copies nested inside a larger group's region (they also set\n   * `spanCountedElsewhere`). They never pair in gapped merging, and they do not keep their group's\n   * standalone copies from merging, which they are not copies of. They do keep the merged group\n   * from taking their group's place, since only the original group reports the nesting.\n   */\n  nestedInLargerGroup?: boolean;\n  /** Sum of segment token counts (the gap tokens are not matched content). */\n  tokenCount: number;\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\n/** A duplicate region found in one file, exported for cross-file matching by fingerprint. */\nexport interface CrossFileDuplicateCandidate {\n  /** Content key: equal fingerprints mean equal normalized token sequences (up to hash collision). */\n  fingerprint: string;\n  tokenCount: number;\n  /** Token positions within the owning file, for cross-file gapped (Type-3) merging. */\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\n/**\n * One file's contribution to cross-file clone detection: catalogued candidates plus the normalized\n * token stream and statement structure, so the project-level pass can match partial statement runs\n * (windows that a single file cannot know repeat elsewhere) and merge gap-adjacent groups.\n */\nexport interface CrossFileDuplicationFileData {\n  candidates: CrossFileDuplicateCandidate[];\n  tokens: Token[];\n  containerStatements: TokenRange[][];\n  /**\n   * 1-based lines that are neither blank nor comment-only, so cross-file line coverage counts only\n   * code lines (blank rows inside multi-row tokens such as template literals carry no content).\n   * Optional for backward compatibility; without it, every matched-token row counts.\n   */\n  codeLineNumbers?: Set<number>;\n}\n\n/** literalCountPrefix[i] = value-carrying literal tokens in tokens[0..i), for O(1) density checks. */\nexport function buildLiteralCountPrefix(tokens: Token[]): Int32Array {\n  const prefix = new Int32Array(tokens.length + 1);\n  for (const [index, token] of tokens.entries()) {\n    prefix[index + 1] = (prefix[index] ?? 0) + (token.literalHash === undefined ? 0 : 1);\n  }\n  return prefix;\n}\n\ninterface WindowOccurrences {\n  count: number;\n  /** -1 once occurrences span more than one container. */\n  containerIndex: number;\n  /** -1 once occurrences span more than one context (file). */\n  contextIndex: number;\n  minStart: number;\n  maxStart: number;\n}\n\ninterface SequenceWindow {\n  containerIndex: number;\n  start: number;\n  length: number;\n}\n\n/** One file's token stream and statement containers, as a window-matching context. */\nexport interface SequenceWindowContext {\n  tokens: Token[];\n  literalCountPrefix: Int32Array;\n  containers: TokenRange[][];\n}\n\nexport interface ContextualSequenceCandidate {\n  candidate: DuplicateCandidate;\n  contextIndex: number;\n}\n\n/**\n * Enumerates runs of consecutive sibling statements over one or more contexts (files). Every\n * container statement participates; only the window length is capped, so enumeration stays linear\n * in the statement count. Windows are grouped by a cheap rolling hash of per-statement\n * fingerprints, and only locally maximal repeated windows — those whose one-statement extensions\n * stop repeating — become candidates with an exact (window-consistent) fingerprint. Without the\n * maximality filter a degenerate file of near-identical statements would fingerprint every\n * sub-window of every repeated region. With `requireMultipleContexts` a window only counts as\n * repeated when its occurrences span at least two contexts (CPD-style cross-file matching): a\n * repeat confined to one file is that file's own concern, and emitting it here would flood the\n * project-level selection with unusable single-file groups.\n */\nexport function collectSequenceWindowCandidates(\n  contexts: SequenceWindowContext[],\n  minTokens: number,\n  requireMultipleContexts: boolean\n): ContextualSequenceCandidate[] {\n  const candidates: ContextualSequenceCandidate[] = [];\n  const contextIndexByContainer: number[] = [];\n  const containers: TokenRange[][] = [];\n  for (const [contextIndex, context] of contexts.entries()) {\n    for (const statements of context.containers) {\n      contextIndexByContainer.push(contextIndex);\n      containers.push(statements);\n    }\n  }\n  const contextAt = (containerIndex: number): SequenceWindowContext | undefined =>\n    contexts[contextIndexByContainer[containerIndex] ?? 0];\n  const occurrencesByWindowKey = new Map<number, WindowOccurrences>();\n  const containerWindows = containers.map((statements, containerIndex) =>\n    enumerateContainerWindows(contextAt(containerIndex)?.tokens ?? [], statements, minTokens)\n  );\n  for (const [containerIndex, windows] of containerWindows.entries()) {\n    const contextIndex = contextIndexByContainer[containerIndex] ?? 0;\n    for (const [start, row] of windows.windowKeysByStart.entries()) {\n      for (const windowKey of row) {\n        if (windowKey === undefined) {\n          continue;\n        }\n        const occurrences = occurrencesByWindowKey.get(windowKey);\n        if (occurrences) {\n          occurrences.count += 1;\n          if (occurrences.containerIndex !== containerIndex) {\n            occurrences.containerIndex = -1;\n          }\n          if (occurrences.contextIndex !== contextIndex) {\n            occurrences.contextIndex = -1;\n          }\n          occurrences.minStart = Math.min(occurrences.minStart, start);\n          occurrences.maxStart = Math.max(occurrences.maxStart, start);\n        } else {\n          occurrencesByWindowKey.set(windowKey, {\n            count: 1,\n            containerIndex,\n            contextIndex,\n            minStart: start,\n            maxStart: start,\n          });\n        }\n      }\n    }\n  }\n\n  // A window only \"repeats\" when two of its occurrences can coexist without overlapping: sliding\n  // matches inside a homogeneous run (start spread smaller than the window length) can never both\n  // be counted and must neither qualify a window nor dominate its sub-windows. Cross-context\n  // matching instead requires occurrences in two contexts, which coexist by construction.\n  const repeats = (windowKey: number | undefined, length: number): boolean => {\n    if (windowKey === undefined) {\n      return false;\n    }\n    const occurrences = occurrencesByWindowKey.get(windowKey);\n    if (occurrences === undefined || occurrences.count < 2) {\n      return false;\n    }\n    if (requireMultipleContexts) {\n      return occurrences.contextIndex === -1;\n    }\n    return occurrences.containerIndex === -1 || occurrences.maxStart - occurrences.minStart >= length;\n  };\n\n  // A window whose statements all share one normalized shape (sixteen `let x = 0;` declarations,\n  // a constant table) is a homogeneous preamble, not a copy-paste: requiring two distinct\n  // per-statement shapes keeps such runs out of duplicate groups and the duplication ratio.\n  const hasDistinctStatements = (window: SequenceWindow): boolean => {\n    const hashes = containerWindows[window.containerIndex]?.statementHashes ?? [];\n    const firstHash = hashes[window.start];\n    for (let index = window.start + 1; index < window.start + window.length; index += 1) {\n      if (hashes[index] !== firstHash) {\n        return true;\n      }\n    }\n    return false;\n  };\n\n  const maximalWindows: SequenceWindow[] = [];\n  for (const [containerIndex, windows] of containerWindows.entries()) {\n    for (const [start, row] of windows.windowKeysByStart.entries()) {\n      for (const [length, windowKey] of row.entries()) {\n        if (!repeats(windowKey, length) || !hasDistinctStatements({ containerIndex, start, length })) {\n          continue;\n        }\n        // Dominated windows are skipped: the one-statement extension also repeats, so a larger\n        // candidate covering this window exists.\n        const extendedRight = windows.windowKeysByStart[start]?.[length + 1];\n        const extendedLeft = windows.windowKeysByStart[start - 1]?.[length + 1];\n        if (repeats(extendedRight, length + 1) || repeats(extendedLeft, length + 1)) {\n          continue;\n        }\n        maximalWindows.push({ containerIndex, start, length });\n      }\n    }\n  }\n\n  // The rolling hash anonymizes identifiers per statement, so a window can look repeated coarsely\n  // while its exact (window-consistent) fingerprints differ, and a longer window's match can\n  // dominate sub-windows that other copies still need (three copies where only two extend one\n  // statement further). Every emitted window therefore exposes its repeating, unvisited\n  // sub-windows; `visited` bounds the worklist and lengths strictly decrease, so it terminates.\n  const visited = new Set(maximalWindows.map(windowId));\n  let frontier = maximalWindows;\n  while (frontier.length > 0) {\n    const emitted: SequenceWindow[] = [];\n    for (const window of frontier) {\n      const statements = containers[window.containerIndex];\n      const first = statements?.[window.start];\n      const last = statements?.[window.start + window.length - 1];\n      const context = contextAt(window.containerIndex);\n      if (!first || !last || !context) {\n        continue;\n      }\n      const fingerprint = `s:${fingerprintKey(context.tokens, context.literalCountPrefix, first.startTokenIndex, last.endTokenIndex)}`;\n      candidates.push({\n        candidate: toCandidate(fingerprint, first.startTokenIndex, last.endTokenIndex, first, last),\n        contextIndex: contextIndexByContainer[window.containerIndex] ?? 0,\n      });\n      emitted.push(window);\n    }\n    frontier = [];\n    for (const window of emitted) {\n      for (const start of [window.start, window.start + 1]) {\n        const subWindow = { containerIndex: window.containerIndex, start, length: window.length - 1 };\n        const subWindowKey = containerWindows[window.containerIndex]?.windowKeysByStart[start]?.[subWindow.length];\n        if (\n          visited.has(windowId(subWindow)) ||\n          !repeats(subWindowKey, subWindow.length) ||\n          !hasDistinctStatements(subWindow)\n        ) {\n          continue;\n        }\n        visited.add(windowId(subWindow));\n        frontier.push(subWindow);\n      }\n    }\n  }\n  return candidates;\n}\n\nfunction windowId(window: SequenceWindow): string {\n  return `${window.containerIndex}:${window.start}:${window.length}`;\n}\n\ninterface ContainerWindows {\n  /** windowKeysByStart[start][length] is the rolling-hash key of the window, or undefined if it is below the size thresholds. */\n  windowKeysByStart: (number | undefined)[][];\n  /** Per-statement fingerprint hashes, for the distinct-shape requirement on windows. */\n  statementHashes: number[];\n}\n\nfunction enumerateContainerWindows(tokens: Token[], statements: TokenRange[], minTokens: number): ContainerWindows {\n  const statementHashes = statements.map((statement) =>\n    fingerprintHash(tokens, statement.startTokenIndex, statement.endTokenIndex)\n  );\n  const windowKeysByStart: (number | undefined)[][] = [];\n  for (let start = 0; start < statements.length; start += 1) {\n    const row: (number | undefined)[] = [];\n    let hash = 5381;\n    let tokenCount = 0;\n    const maxEnd = Math.min(statements.length, start + maxSequenceStatementCount);\n    for (let end = start; end < maxEnd; end += 1) {\n      const statement = statements[end];\n      const statementHash = statementHashes[end];\n      if (!statement || statementHash === undefined) {\n        break;\n      }\n      hash = combineHashes(hash, statementHash);\n      tokenCount += statement.endTokenIndex - statement.startTokenIndex;\n      const statementCount = end - start + 1;\n      row[statementCount] =\n        statementCount >= minSequenceStatementCount && tokenCount >= minTokens\n          ? combineHashes(hash, statementCount)\n          : undefined;\n    }\n    windowKeysByStart.push(row);\n  }\n  return { windowKeysByStart, statementHashes };\n}\n\n/**\n * Longest-common-subsequence LENGTH of two symbol sequences via the Allison–Dix bit-parallel\n * recurrence (O(|a|/32 · |b|) words): per symbol of `b`, `x = match | v` and\n * `v = x & ~(x - ((v << 1) | 1))` over multi-word bit vectors; the set bits of `v` count the LCS.\n */\nexport function lcsLength(a: Int32Array, b: Int32Array): number {\n  const wordCount = (a.length + 31) >>> 5;\n  const positionMasks = new Map<number, Uint32Array>();\n  for (const [index, symbol] of a.entries()) {\n    let mask = positionMasks.get(symbol);\n    if (!mask) {\n      mask = new Uint32Array(wordCount);\n      positionMasks.set(symbol, mask);\n    }\n    const word = index >>> 5;\n    // The Uint32Array store wraps the signed int32 bit pattern to unsigned. The `?? 0` guards in\n    // this function are required by noUncheckedIndexedAccess (typed-array reads type as\n    // `number | undefined`), not redundancy: every index is in bounds.\n    mask[word] = (mask[word] ?? 0) | (1 << (index & 31));\n  }\n\n  const v = new Uint32Array(wordCount);\n  for (const symbol of b) {\n    const matchMask = positionMasks.get(symbol);\n    // `(v << 1) | 1` shifts a carry bit across words; subtraction borrows across words.\n    let shiftCarry = 1;\n    let borrow = 0;\n    for (let word = 0; word < wordCount; word += 1) {\n      const previous = v[word] ?? 0;\n      // oxlint-disable-next-line unicorn/prefer-math-trunc -- `>>> 0` reinterprets the signed int32 bit pattern as unsigned so the borrow subtraction below compares magnitudes; Math.trunc would keep it negative.\n      const x = ((matchMask?.[word] ?? 0) | previous) >>> 0;\n      // oxlint-disable-next-line unicorn/prefer-math-trunc -- same unsigned reinterpretation as `x`.\n      const shifted = ((previous << 1) | shiftCarry) >>> 0;\n      shiftCarry = previous >>> 31;\n      const difference = x - shifted - borrow;\n      borrow = difference < 0 ? 1 : 0;\n      // The Uint32Array store wraps the signed int32 bit pattern to unsigned.\n      v[word] = x & ~difference;\n    }\n  }\n\n  let length = 0;\n  for (const word of v) {\n    length += popCount(word);\n  }\n  return length;\n}\n\nfunction popCount(value: number): number {\n  let count = value - ((value >>> 1) & 0x55_55_55_55);\n  count = (count & 0x33_33_33_33) + ((count >>> 2) & 0x33_33_33_33);\n  return (Math.imul((count + (count >>> 4)) & 0x0F_0F_0F_0F, 0x01_01_01_01) >>> 24) & 0xFF;\n}\n\nfunction toCandidate(\n  fingerprint: string,\n  startTokenIndex: number,\n  endTokenIndex: number,\n  first: TokenRange,\n  last: TokenRange\n): DuplicateCandidate {\n  return {\n    fingerprint,\n    tokenCount: endTokenIndex - startTokenIndex,\n    startTokenIndex,\n    endTokenIndex,\n    startIndex: first.startIndex,\n    endIndex: last.endIndex,\n    startLine: first.startLine,\n    endLine: last.endLine,\n  };\n}\n\n/** Caches of hashText/hashText2 over '$0', '$1', ... so anonymized identifiers hash without allocating. */\nconst anonymizedIndexHashes: number[] = [];\nconst anonymizedIndexHashes2: number[] = [];\n\nfunction anonymizedIndexHash(index: number): number {\n  let hash = anonymizedIndexHashes[index];\n  if (hash === undefined) {\n    hash = hashText(`$${index}`);\n    anonymizedIndexHashes[index] = hash;\n  }\n  return hash;\n}\n\nfunction anonymizedIndexHash2(index: number): number {\n  let hash = anonymizedIndexHashes2[index];\n  if (hash === undefined) {\n    hash = hashText2(`$${index}`);\n    anonymizedIndexHashes2[index] = hash;\n  }\n  return hash;\n}\n\n/**\n * Content key of a token range: two independent 32-bit hashes over the normalized token sequence\n * (identifiers anonymized consistently by first-occurrence order) plus the token count. Regions\n * with equal keys are treated as equal content; a collision would need both 32-bit hashes and the\n * length to coincide, which is negligible for a metrics report. The format and arithmetic match\n * fingerprint_key in native/src/duplication.rs exactly, so window candidates fingerprinted here\n * group together with the per-file candidates the addon catalogues.\n */\nfunction fingerprintKey(\n  tokens: Token[],\n  literalCountPrefix: Int32Array,\n  startTokenIndex: number,\n  endTokenIndex: number\n): string {\n  const literalCount = (literalCountPrefix[endTokenIndex] ?? 0) - (literalCountPrefix[startTokenIndex] ?? 0);\n  const literalDense = isLiteralDense(literalCount, endTokenIndex - startTokenIndex);\n  const [primary, secondary] = fingerprintHashPair(tokens, startTokenIndex, endTokenIndex, literalDense);\n  return `${primary}:${secondary}:${endTokenIndex - startTokenIndex}`;\n}\n\n/**\n * A single 32-bit summary of a range for the coarse rolling-hash phase. Deliberately\n * density-agnostic: density is a property of the final candidate REGION, and folding literal\n * values into per-statement hashes would make a dense statement inside a logic-heavy window\n * (`const weights = [1, 2, 3];`) block the window from ever being enumerated. The coarse phase\n * over-approximates on shape alone; the exact region fingerprint still applies the density rule.\n */\nfunction fingerprintHash(tokens: Token[], startTokenIndex: number, endTokenIndex: number): number {\n  const [primary, secondary] = fingerprintHashPair(tokens, startTokenIndex, endTokenIndex, false);\n  // XOR already coerces to int32, matching the native side's i32 arithmetic.\n  return primary ^ Math.imul(secondary, 31);\n}\n\nfunction fingerprintHashPair(\n  tokens: Token[],\n  startTokenIndex: number,\n  endTokenIndex: number,\n  foldLiteralValues: boolean\n): [number, number] {\n  const indexByIdentifier = new Map<string, number>();\n  let primary = 5381;\n  let secondary = 52_711;\n  for (let index = startTokenIndex; index < endTokenIndex; index += 1) {\n    const token = tokens[index];\n    if (!token) {\n      continue;\n    }\n    // Each accumulator consumes its own independent per-token hash: sharing one would collapse\n    // the key to 32 effective bits (a single djb2 collision would equate whole regions).\n    let part: number;\n    let part2: number;\n    if (token.kind === 'id') {\n      let identifierIndex = indexByIdentifier.get(token.text);\n      if (identifierIndex === undefined) {\n        identifierIndex = indexByIdentifier.size;\n        indexByIdentifier.set(token.text, identifierIndex);\n      }\n      part = anonymizedIndexHash(identifierIndex);\n      part2 = anonymizedIndexHash2(identifierIndex);\n    } else {\n      part = token.textHash;\n      part2 = token.textHash2;\n    }\n    // oxlint-disable-next-line unicorn/prefer-math-trunc -- `| 0` wraps the sum to int32 (Math.trunc does not), which must match the native side's wrapping i32 arithmetic.\n    primary = (Math.imul(primary, 31) + part) | 0;\n    secondary = Math.imul(secondary, 37) ^ part2;\n    if (foldLiteralValues && token.literalHash !== undefined && token.literalHash2 !== undefined) {\n      // oxlint-disable-next-line unicorn/prefer-math-trunc -- `| 0` wraps the sum to int32 (Math.trunc does not), which must match the native side's wrapping i32 arithmetic.\n      primary = (Math.imul(primary, 31) + token.literalHash) | 0;\n      secondary = Math.imul(secondary, 37) ^ token.literalHash2;\n    }\n  }\n  return [primary, secondary];\n}\n\n/** djb2-style hash; XOR keeps the value in signed 32-bit range, which is fine for a grouping key. */\nfunction hashText(text: string): number {\n  let hash = 5381;\n  for (let index = 0; index < text.length; index += 1) {\n    // oxlint-disable-next-line unicorn/prefer-code-point -- djb2 hashes UTF-16 code units; codePointAt would hash surrogate pairs twice (full code point, then the lone low surrogate).\n    hash = Math.imul(hash, 33) ^ text.charCodeAt(index);\n  }\n  return hash;\n}\n\n/** FNV-1a over UTF-16 code units: independent of hashText so the two accumulators never share input. */\nfunction hashText2(text: string): number {\n  let hash = -2_128_831_035; // 2166136261 as int32 (the FNV-1a offset basis)\n  for (let index = 0; index < text.length; index += 1) {\n    // oxlint-disable-next-line unicorn/prefer-code-point -- hashes UTF-16 code units like hashText.\n    hash = Math.imul(hash ^ text.charCodeAt(index), 16_777_619);\n  }\n  return hash;\n}\n\nfunction combineHashes(hash: number, value: number): number {\n  return Math.imul(hash, 31) + value;\n}\n\n/**\n * Merges duplicate groups separated by a small token gap into one gapped (Type-3) clone group: a\n * copy edited in one spot splits into two exact groups whose occurrences sit side by side in the\n * same order. Occurrences are paired greedily in source order; a merge happens when the pairing\n * fully pairs at least one group with at least two pairs. Equal-cardinality groups whose\n * occurrences all pair merge into one group as before. When cardinalities differ (a fragment also\n * occurs standalone: prefix ×3, suffix ×2), the fully-paired group is subsumed into the merged\n * gapped group while the other group is RETAINED with ALL its occurrences: dropping the leftover\n * would lose duplicated-line coverage, and reporting it alone would make a single-occurrence group\n * (contradicting duplicateBlockGroupCount's \"appears more than once\" meaning). Line coverage\n * unions ranges, so the overlap between the retained exact group and the merged group is harmless.\n * Merging repeats to a fixpoint so a clone edited in several spots still reassembles; it\n * terminates because a merge either removes the input groups it replaces or marks their paired\n * occurrences as counted elsewhere, and only unmarked occurrences of remaining groups pair, so the\n * number of pairable occurrences strictly decreases with every merge. Gap tokens are not matched\n * content: line coverage and sizes count only the matched segments. Generic so cross-file merging\n * can thread file identity through occurrences.\n */\nexport function mergeAdjacentGroups<T extends CountedOccurrence>(\n  groups: T[][],\n  maxGapTokens: number,\n  isReportableGroup: (group: T[]) => boolean = () => true\n): T[][] {\n  if (maxGapTokens <= 0 || groups.length < 2) {\n    return groups;\n  }\n  // Deterministic processing order (mirrored by the native side): by first occurrence position.\n  groups.sort(compareGroups);\n  for (let restart = true; restart;) {\n    restart = false;\n    for (let leftIndex = 0; leftIndex < groups.length && !restart; leftIndex += 1) {\n      for (let rightIndex = leftIndex + 1; rightIndex < groups.length; rightIndex += 1) {\n        const left = groups[leftIndex];\n        const right = groups[rightIndex];\n        if (!left || !right) {\n          continue;\n        }\n        const forward = mergeGroups(left, right, maxGapTokens, isReportableGroup);\n        const result = forward ?? mergeGroups(right, left, maxGapTokens, isReportableGroup);\n        if (!result) {\n          continue;\n        }\n        const leftReplaced = forward ? result.firstReplaced : result.secondReplaced;\n        const rightReplaced = forward ? result.secondReplaced : result.firstReplaced;\n        if (leftReplaced && rightReplaced) {\n          groups[leftIndex] = result.merged;\n          groups.splice(rightIndex, 1);\n        } else if (rightReplaced) {\n          groups[rightIndex] = result.merged;\n        } else if (leftReplaced) {\n          groups[leftIndex] = result.merged;\n        } else {\n          // Both groups stay (each is only partly paired, or reports nested copies of its own), so\n          // the merged group joins them instead of taking a place.\n          groups.push(result.merged);\n        }\n        // A group that stays keeps ALL its occurrences (line coverage must not shrink, and a\n        // reported group must keep >= 2 occurrences), so its paired occurrences now also live\n        // inside the merged group's occurrences: mark them so duplicateBlockCount counts each\n        // token span once, and so the same pair cannot merge again.\n        for (const occurrence of result.pairedRetained) {\n          occurrence.spanCountedElsewhere = true;\n        }\n        groups.sort(compareGroups);\n        restart = true;\n        break;\n      }\n    }\n  }\n  return groups;\n}\n\nfunction compareGroups(left: CountedOccurrence[], right: CountedOccurrence[]): number {\n  const leftFirst = left[0];\n  const rightFirst = right[0];\n  return (\n    (leftFirst?.startTokenIndex ?? 0) - (rightFirst?.startTokenIndex ?? 0) ||\n    (leftFirst?.endTokenIndex ?? 0) - (rightFirst?.endTokenIndex ?? 0)\n  );\n}\n\ninterface MergeResult<T> {\n  merged: T[];\n  /**\n   * Whether the merged group takes the respective input group's place: its standalone occurrences\n   * were all paired and it holds no nested copies that only it can report.\n   */\n  firstReplaced: boolean;\n  secondReplaced: boolean;\n  /** The occurrences of groups that stay, which the merged group's spans now also cover. */\n  pairedRetained: T[];\n}\n\n/**\n * Pairs `second` occurrences with gap-preceding `first` occurrences, greedily in source order:\n * each trailing occurrence takes the earliest unused leading occurrence within the gap, and a\n * pair's leading must start at or after the previous pair's trailing end so merged spans never\n * overlap. A merge needs at least two pairs (a merged group must still mean \"appears more than\n * once\") and must fully consume at least one group; for equal cardinalities this reduces to the\n * strict all-pairs merge, so pre-partial-merge behavior is unchanged there.\n */\nfunction mergeGroups<T extends CountedOccurrence>(\n  first: T[],\n  second: T[],\n  maxGapTokens: number,\n  isReportableGroup: (group: T[]) => boolean\n): MergeResult<T> | undefined {\n  // Occurrences a previous partial merge already paired into a merged group must not pair again:\n  // their spans already live inside that merged group, so re-pairing them would assemble a second,\n  // competing merged group instead of letting the existing merged group extend (and would count\n  // the same span twice). Consumption is still judged against the FULL group, so a group holding\n  // shared occurrences is never subsumed away; nested copies, which the merged span would not\n  // cover anyway, are left out of that judgment so they cannot veto a merge of the standalone\n  // copies.\n  const [leadings, firstLength] = pairableOccurrences(first);\n  const [trailings, secondLength] = pairableOccurrences(second);\n  const pairs: [T, T][] = [];\n  let leadingIndex = 0;\n  let previousTrailingEnd = -1;\n  for (const trailing of trailings) {\n    // Leadings ending too far before this trailing can never pair a later (even farther) one.\n    while (leadingIndex < leadings.length) {\n      const leading = leadings[leadingIndex];\n      if (leading && leading.endTokenIndex + maxGapTokens < trailing.startTokenIndex) {\n        leadingIndex += 1;\n      } else {\n        break;\n      }\n    }\n    const leading = leadings[leadingIndex];\n    if (\n      leading &&\n      leading.endTokenIndex <= trailing.startTokenIndex &&\n      leading.startTokenIndex >= previousTrailingEnd\n    ) {\n      pairs.push([leading, trailing]);\n      previousTrailingEnd = trailing.endTokenIndex;\n      leadingIndex += 1;\n    }\n  }\n  const firstFullyPaired = pairs.length === firstLength;\n  const secondFullyPaired = pairs.length === secondLength;\n  if (pairs.length < 2 || (!firstFullyPaired && !secondFullyPaired)) {\n    return undefined;\n  }\n  // A group holding nested copies stays even when all its standalone copies pair: the merged\n  // group's content is larger than what those copies matched, so only the original group can\n  // report which files share the matched fragment.\n  const firstReplaced = firstFullyPaired && !first.some((occurrence) => occurrence.nestedInLargerGroup);\n  const secondReplaced = secondFullyPaired && !second.some((occurrence) => occurrence.nestedInLargerGroup);\n  const merged = pairs.map(([leading, trailing]) => ({\n    ...leading,\n    // A merged occurrence is a fresh span combination; it never inherits shared-span marks.\n    spanCountedElsewhere: undefined,\n    nestedInLargerGroup: undefined,\n    segments: [...leading.segments, ...trailing.segments],\n    tokenCount: leading.tokenCount + trailing.tokenCount,\n    endTokenIndex: trailing.endTokenIndex,\n    endIndex: trailing.endIndex,\n    endLine: trailing.endLine,\n  }));\n  // Only occurrences of one file pair (file offsets exceed the gap), so a merged group spans the\n  // files its pairs sit in: with nested copies left out of pairing, that can be fewer files than\n  // the input groups covered, and a merged group that is no longer reportable must not form.\n  if (!isReportableGroup(merged)) {\n    return undefined;\n  }\n  const pairedRetained = [\n    ...(firstReplaced ? [] : pairs.map(([leading]) => leading)),\n    ...(secondReplaced ? [] : pairs.map(([, trailing]) => trailing)),\n  ];\n  return { merged, firstReplaced, secondReplaced, pairedRetained };\n}\n\n/** One pass over a group: its pairable occurrences and its non-nested occurrence count. */\nfunction pairableOccurrences<T extends CountedOccurrence>(group: T[]): [T[], number] {\n  const pairable: T[] = [];\n  let length = 0;\n  for (const occurrence of group) {\n    if (!occurrence.nestedInLargerGroup) {\n      length += 1;\n    }\n    // Either flag keeps an occurrence out of pairing: its span is already counted elsewhere, or it\n    // is a nested copy of content the merged span would not cover.\n    if (!occurrence.spanCountedElsewhere && !occurrence.nestedInLargerGroup) {\n      pairable.push(occurrence);\n    }\n  }\n  return [pairable, length];\n}\n\n/**\n * Redundant copies one group adds to duplicateBlockCount. Each redundant occurrence contributes\n * one count per matched fragment, so merging a gapped clone's fragments into one group does not\n * halve duplicateBlockCount: an edited two-fragment pair still counts 2, exactly as its unmerged\n * fragments did. Occurrence shapes can differ within one group (a\n * gap-merged exact pair plus an appended whole-block near-miss copy), so every occurrence's\n * fragments are summed and one representative — the largest — is deducted, keeping the count\n * independent of source order. Occurrences a partial gapped merge also paired into a merged group\n * are skipped: their spans are already counted there, and such a retained group deducts no\n * representative of its own — the merged group's representative already stands for the shared\n * content — so no token span contributes to the count twice.\n */\nexport function countRedundantFragments(group: CountedOccurrence[]): number {\n  let fragmentCount = 0;\n  let maxFragmentCount = 0;\n  let hasSharedOccurrence = false;\n  for (const occurrence of group) {\n    if (occurrence.spanCountedElsewhere) {\n      hasSharedOccurrence = true;\n      continue;\n    }\n    fragmentCount += occurrence.segments.length;\n    maxFragmentCount = Math.max(maxFragmentCount, occurrence.segments.length);\n  }\n  return hasSharedOccurrence ? fragmentCount : fragmentCount - maxFragmentCount;\n}\n\n/** Adds the 1-based code lines the segment's matched tokens cover; shared with cross-file coverage. */\nexport function collectSegmentLines(\n  segment: { startTokenIndex: number; endTokenIndex: number },\n  tokens: Token[],\n  codeLineNumbers: Set<number> | undefined,\n  duplicatedLines: Set<number>\n): void {\n  for (let index = segment.startTokenIndex; index < segment.endTokenIndex; index += 1) {\n    const token = tokens[index];\n    for (let row = token?.startRow ?? 0; row <= (token?.endRow ?? -1); row += 1) {\n      if (!codeLineNumbers || codeLineNumbers.has(row + 1)) {\n        duplicatedLines.add(row + 1);\n      }\n    }\n  }\n}\n"],"mappings":"aASA,MAAa,EAA0D,CACrE,UAAW,GACX,aAAc,GACd,qBAAsB,EACxB,EAQA,SAAgB,EAA0B,EAA4D,CACpG,MAAO,CACL,UAAW,EAAc,GAAS,UAAW,EAA0B,SAAS,EAChF,aAAc,EAAc,GAAS,aAAc,EAA0B,YAAY,EACzF,qBAAsB,EAAc,GAAS,qBAAsB,EAA0B,oBAAoB,CACnH,CACF,CAEA,SAAS,EAAc,EAA2B,EAA0B,CAC1E,OAAO,IAAU,IAAA,IAAa,OAAO,MAAM,CAAK,EAC5C,EACA,KAAK,IAAI,KAAK,IAAI,KAAK,MAAM,CAAK,EAAG,CAAC,EAAG,UAAa,CAC5D,CAiBA,SAAS,EAAe,EAAsB,EAA6B,CACzE,OAAO,EAAe,GAAK,CAC7B,CAiHA,SAAgB,EAAwB,EAA6B,CACnE,IAAM,EAAS,IAAI,WAAW,EAAO,OAAS,CAAC,EAC/C,IAAK,GAAM,CAAC,EAAO,KAAU,EAAO,QAAQ,EAC1C,EAAO,EAAQ,IAAM,EAAO,IAAU,IAAM,EAAM,cAAgB,IAAA,GAAY,EAAI,GAEpF,OAAO,CACT,CA0CA,SAAgB,EACd,EACA,EACA,EAC+B,CAC/B,IAAM,EAA4C,CAAC,EAC7C,EAAoC,CAAC,EACrC,EAA6B,CAAC,EACpC,IAAK,GAAM,CAAC,EAAc,KAAY,EAAS,QAAQ,EACrD,IAAK,IAAM,KAAc,EAAQ,WAC/B,EAAwB,KAAK,CAAY,EACzC,EAAW,KAAK,CAAU,EAG9B,IAAM,EAAa,GACjB,EAAS,EAAwB,IAAmB,GAChD,EAAyB,IAAI,IAC7B,EAAmB,EAAW,KAAK,EAAY,IACnD,EAA0B,EAAU,CAAc,CAAC,EAAE,QAAU,CAAC,EAAG,EAAY,CAAS,CAC1F,EACA,IAAK,GAAM,CAAC,EAAgB,KAAY,EAAiB,QAAQ,EAAG,CAClE,IAAM,EAAe,EAAwB,IAAmB,EAChE,IAAK,GAAM,CAAC,EAAO,KAAQ,EAAQ,kBAAkB,QAAQ,EAC3D,IAAK,IAAM,KAAa,EAAK,CAC3B,GAAI,IAAc,IAAA,GAChB,SAEF,IAAM,EAAc,EAAuB,IAAI,CAAS,EACpD,GACF,EAAY,OAAS,EACjB,EAAY,iBAAmB,IACjC,EAAY,eAAiB,IAE3B,EAAY,eAAiB,IAC/B,EAAY,aAAe,IAE7B,EAAY,SAAW,KAAK,IAAI,EAAY,SAAU,CAAK,EAC3D,EAAY,SAAW,KAAK,IAAI,EAAY,SAAU,CAAK,GAE3D,EAAuB,IAAI,EAAW,CACpC,MAAO,EACP,iBACA,eACA,SAAU,EACV,SAAU,CACZ,CAAC,CAEL,CAEJ,CAMA,IAAM,GAAW,EAA+B,IAA4B,CAC1E,GAAI,IAAc,IAAA,GAChB,MAAO,GAET,IAAM,EAAc,EAAuB,IAAI,CAAS,EAOxD,OANI,IAAgB,IAAA,IAAa,EAAY,MAAQ,EAC5C,GAEL,EACK,EAAY,eAAiB,GAE/B,EAAY,iBAAmB,IAAM,EAAY,SAAW,EAAY,UAAY,CAC7F,EAKM,EAAyB,GAAoC,CACjE,IAAM,EAAS,EAAiB,EAAO,eAAe,EAAE,iBAAmB,CAAC,EACtE,EAAY,EAAO,EAAO,OAChC,IAAK,IAAI,EAAQ,EAAO,MAAQ,EAAG,EAAQ,EAAO,MAAQ,EAAO,OAAQ,GAAS,EAChF,GAAI,EAAO,KAAW,EACpB,MAAO,GAGX,MAAO,EACT,EAEM,EAAmC,CAAC,EAC1C,IAAK,GAAM,CAAC,EAAgB,KAAY,EAAiB,QAAQ,EAC/D,IAAK,GAAM,CAAC,EAAO,KAAQ,EAAQ,kBAAkB,QAAQ,EAC3D,IAAK,GAAM,CAAC,EAAQ,KAAc,EAAI,QAAQ,EAAG,CAC/C,GAAI,CAAC,EAAQ,EAAW,CAAM,GAAK,CAAC,EAAsB,CAAE,iBAAgB,QAAO,QAAO,CAAC,EACzF,SAIF,IAAM,EAAgB,EAAQ,kBAAkB,EAAM,GAAG,EAAS,GAC5D,EAAe,EAAQ,kBAAkB,EAAQ,EAAE,GAAG,EAAS,GACjE,EAAQ,EAAe,EAAS,CAAC,GAAK,EAAQ,EAAc,EAAS,CAAC,GAG1E,EAAe,KAAK,CAAE,iBAAgB,QAAO,QAAO,CAAC,CACvD,CASJ,IAAM,EAAU,IAAI,IAAI,EAAe,IAAI,CAAQ,CAAC,EAChD,EAAW,EACf,KAAO,EAAS,OAAS,GAAG,CAC1B,IAAM,EAA4B,CAAC,EACnC,IAAK,IAAM,KAAU,EAAU,CAC7B,IAAM,EAAa,EAAW,EAAO,gBAC/B,EAAQ,IAAa,EAAO,OAC5B,EAAO,IAAa,EAAO,MAAQ,EAAO,OAAS,GACnD,EAAU,EAAU,EAAO,cAAc,EAC/C,GAAI,CAAC,GAAS,CAAC,GAAQ,CAAC,EACtB,SAEF,IAAM,EAAc,KAAK,EAAe,EAAQ,OAAQ,EAAQ,mBAAoB,EAAM,gBAAiB,EAAK,aAAa,IAC7H,EAAW,KAAK,CACd,UAAW,EAAY,EAAa,EAAM,gBAAiB,EAAK,cAAe,EAAO,CAAI,EAC1F,aAAc,EAAwB,EAAO,iBAAmB,CAClE,CAAC,EACD,EAAQ,KAAK,CAAM,CACrB,CACA,EAAW,CAAC,EACZ,IAAK,IAAM,KAAU,EACnB,IAAK,IAAM,IAAS,CAAC,EAAO,MAAO,EAAO,MAAQ,CAAC,EAAG,CACpD,IAAM,EAAY,CAAE,eAAgB,EAAO,eAAgB,QAAO,OAAQ,EAAO,OAAS,CAAE,EACtF,EAAe,EAAiB,EAAO,eAAe,EAAE,kBAAkB,EAAM,GAAG,EAAU,QAEjG,GAAQ,IAAI,EAAS,CAAS,CAAC,GAC9B,EAAQ,EAAc,EAAU,MAAM,GACtC,EAAsB,CAAS,IAIlC,EAAQ,IAAI,EAAS,CAAS,CAAC,EAC/B,EAAS,KAAK,CAAS,EACzB,CAEJ,CACA,OAAO,CACT,CAEA,SAAS,EAAS,EAAgC,CAChD,MAAO,GAAG,EAAO,eAAe,GAAG,EAAO,MAAM,GAAG,EAAO,QAC5D,CASA,SAAS,EAA0B,EAAiB,EAA0B,EAAqC,CACjH,IAAM,EAAkB,EAAW,IAAK,GACtC,EAAgB,EAAQ,EAAU,gBAAiB,EAAU,aAAa,CAC5E,EACM,EAA8C,CAAC,EACrD,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAW,OAAQ,GAAS,EAAG,CACzD,IAAM,EAA8B,CAAC,EACjC,EAAO,KACP,EAAa,EACX,EAAS,KAAK,IAAI,EAAW,OAAQ,EAAQ,GAAyB,EAC5E,IAAK,IAAI,EAAM,EAAO,EAAM,EAAQ,GAAO,EAAG,CAC5C,IAAM,EAAY,EAAW,GACvB,EAAgB,EAAgB,GACtC,GAAI,CAAC,GAAa,IAAkB,IAAA,GAClC,MAEF,EAAO,EAAc,EAAM,CAAa,EACxC,GAAc,EAAU,cAAgB,EAAU,gBAClD,IAAM,EAAiB,EAAM,EAAQ,EACrC,EAAI,GACF,GAAkB,GAA6B,GAAc,EACzD,EAAc,EAAM,CAAc,EAClC,IAAA,EACR,CACA,EAAkB,KAAK,CAAG,CAC5B,CACA,MAAO,CAAE,oBAAmB,iBAAgB,CAC9C,CAOA,SAAgB,EAAU,EAAe,EAAuB,CAC9D,IAAM,EAAa,EAAE,OAAS,KAAQ,EAChC,EAAgB,IAAI,IAC1B,IAAK,GAAM,CAAC,EAAO,KAAW,EAAE,QAAQ,EAAG,CACzC,IAAI,EAAO,EAAc,IAAI,CAAM,EAC9B,IACH,EAAO,IAAI,YAAY,CAAS,EAChC,EAAc,IAAI,EAAQ,CAAI,GAEhC,IAAM,EAAO,IAAU,EAIvB,EAAK,IAAS,EAAK,IAAS,GAAM,IAAM,EAAQ,GAClD,CAEA,IAAM,EAAI,IAAI,YAAY,CAAS,EACnC,IAAK,IAAM,KAAU,EAAG,CACtB,IAAM,EAAY,EAAc,IAAI,CAAM,EAEtC,EAAa,EACb,EAAS,EACb,IAAK,IAAI,EAAO,EAAG,EAAO,EAAW,GAAQ,EAAG,CAC9C,IAAM,EAAW,EAAE,IAAS,EAEtB,IAAM,IAAY,IAAS,GAAK,KAAc,EAE9C,GAAY,GAAY,EAAK,KAAgB,EACnD,EAAa,IAAa,GAC1B,IAAM,EAAa,EAAI,EAAU,EACjC,EAAS,IAAa,GAEtB,EAAE,GAAQ,EAAI,CAAC,CACjB,CACF,CAEA,IAAI,EAAS,EACb,IAAK,IAAM,KAAQ,EACjB,GAAU,EAAS,CAAI,EAEzB,OAAO,CACT,CAEA,SAAS,EAAS,EAAuB,CACvC,IAAI,EAAQ,GAAU,IAAU,EAAK,YAErC,MADA,IAAS,EAAQ,YAAmB,IAAU,EAAK,WAC3C,KAAK,KAAM,GAAS,IAAU,GAAM,UAAe,QAAa,IAAM,GAAM,GACtF,CAEA,SAAS,EACP,EACA,EACA,EACA,EACA,EACoB,CACpB,MAAO,CACL,cACA,WAAY,EAAgB,EAC5B,kBACA,gBACA,WAAY,EAAM,WAClB,SAAU,EAAK,SACf,UAAW,EAAM,UACjB,QAAS,EAAK,OAChB,CACF,CAGA,MAAM,EAAkC,CAAC,EACnC,EAAmC,CAAC,EAE1C,SAAS,EAAoB,EAAuB,CAClD,IAAI,EAAO,EAAsB,GAKjC,OAJI,IAAS,IAAA,KACX,EAAO,EAAS,IAAI,GAAO,EAC3B,EAAsB,GAAS,GAE1B,CACT,CAEA,SAAS,EAAqB,EAAuB,CACnD,IAAI,EAAO,EAAuB,GAKlC,OAJI,IAAS,IAAA,KACX,EAAO,EAAU,IAAI,GAAO,EAC5B,EAAuB,GAAS,GAE3B,CACT,CAUA,SAAS,EACP,EACA,EACA,EACA,EACQ,CAGR,GAAM,CAAC,EAAS,GAAa,EAAoB,EAAQ,EAAiB,EADrD,GADC,EAAmB,IAAkB,IAAM,EAAmB,IAAoB,GACtD,EAAgB,CACkC,CAAC,EACrG,MAAO,GAAG,EAAQ,GAAG,EAAU,GAAG,EAAgB,GACpD,CASA,SAAS,EAAgB,EAAiB,EAAyB,EAA+B,CAChG,GAAM,CAAC,EAAS,GAAa,EAAoB,EAAQ,EAAiB,EAAe,EAAK,EAE9F,OAAO,EAAU,KAAK,KAAK,EAAW,EAAE,CAC1C,CAEA,SAAS,EACP,EACA,EACA,EACA,EACkB,CAClB,IAAM,EAAoB,IAAI,IAC1B,EAAU,KACV,EAAY,MAChB,IAAK,IAAI,EAAQ,EAAiB,EAAQ,EAAe,GAAS,EAAG,CACnE,IAAM,EAAQ,EAAO,GACrB,GAAI,CAAC,EACH,SAIF,IAAI,EACA,EACJ,GAAI,EAAM,OAAS,KAAM,CACvB,IAAI,EAAkB,EAAkB,IAAI,EAAM,IAAI,EAClD,IAAoB,IAAA,KACtB,EAAkB,EAAkB,KACpC,EAAkB,IAAI,EAAM,KAAM,CAAe,GAEnD,EAAO,EAAoB,CAAe,EAC1C,EAAQ,EAAqB,CAAe,CAC9C,KACE,GAAO,EAAM,SACb,EAAQ,EAAM,UAGhB,EAAW,KAAK,KAAK,EAAS,EAAE,EAAI,EAAQ,EAC5C,EAAY,KAAK,KAAK,EAAW,EAAE,EAAI,EACnC,GAAqB,EAAM,cAAgB,IAAA,IAAa,EAAM,eAAiB,IAAA,KAEjF,EAAW,KAAK,KAAK,EAAS,EAAE,EAAI,EAAM,YAAe,EACzD,EAAY,KAAK,KAAK,EAAW,EAAE,EAAI,EAAM,aAEjD,CACA,MAAO,CAAC,EAAS,CAAS,CAC5B,CAGA,SAAS,EAAS,EAAsB,CACtC,IAAI,EAAO,KACX,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAK,OAAQ,GAAS,EAEhD,EAAO,KAAK,KAAK,EAAM,EAAE,EAAI,EAAK,WAAW,CAAK,EAEpD,OAAO,CACT,CAGA,SAAS,EAAU,EAAsB,CACvC,IAAI,EAAO,YACX,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAK,OAAQ,GAAS,EAEhD,EAAO,KAAK,KAAK,EAAO,EAAK,WAAW,CAAK,EAAG,QAAU,EAE5D,OAAO,CACT,CAEA,SAAS,EAAc,EAAc,EAAuB,CAC1D,OAAO,KAAK,KAAK,EAAM,EAAE,EAAI,CAC/B,CAoBA,SAAgB,EACd,EACA,EACA,MAAmD,GAC5C,CACP,GAAI,GAAgB,GAAK,EAAO,OAAS,EACvC,OAAO,EAGT,EAAO,KAAK,CAAa,EACzB,IAAK,IAAI,EAAU,GAAM,GAAU,CACjC,EAAU,GACV,IAAK,IAAI,EAAY,EAAG,EAAY,EAAO,QAAU,CAAC,EAAS,GAAa,EAC1E,IAAK,IAAI,EAAa,EAAY,EAAG,EAAa,EAAO,OAAQ,GAAc,EAAG,CAChF,IAAM,EAAO,EAAO,GACd,EAAQ,EAAO,GACrB,GAAI,CAAC,GAAQ,CAAC,EACZ,SAEF,IAAM,EAAU,EAAY,EAAM,EAAO,EAAc,CAAiB,EAClE,EAAS,GAAW,EAAY,EAAO,EAAM,EAAc,CAAiB,EAClF,GAAI,CAAC,EACH,SAEF,IAAM,EAAe,EAAU,EAAO,cAAgB,EAAO,eACvD,EAAgB,EAAU,EAAO,eAAiB,EAAO,cAC3D,GAAgB,GAClB,EAAO,GAAa,EAAO,OAC3B,EAAO,OAAO,EAAY,CAAC,GAClB,EACT,EAAO,GAAc,EAAO,OACnB,EACT,EAAO,GAAa,EAAO,OAI3B,EAAO,KAAK,EAAO,MAAM,EAM3B,IAAK,IAAM,KAAc,EAAO,eAC9B,EAAW,qBAAuB,GAEpC,EAAO,KAAK,CAAa,EACzB,EAAU,GACV,KACF,CAEJ,CACA,OAAO,CACT,CAEA,SAAS,EAAc,EAA2B,EAAoC,CACpF,IAAM,EAAY,EAAK,GACjB,EAAa,EAAM,GACzB,OACG,GAAW,iBAAmB,IAAM,GAAY,iBAAmB,KACnE,GAAW,eAAiB,IAAM,GAAY,eAAiB,EAEpE,CAsBA,SAAS,EACP,EACA,EACA,EACA,EAC4B,CAQ5B,GAAM,CAAC,EAAU,GAAe,EAAoB,CAAK,EACnD,CAAC,EAAW,GAAgB,EAAoB,CAAM,EACtD,EAAkB,CAAC,EACrB,EAAe,EACf,EAAsB,GAC1B,IAAK,IAAM,KAAY,EAAW,CAEhC,KAAO,EAAe,EAAS,QAAQ,CACrC,IAAM,EAAU,EAAS,GACzB,GAAI,GAAW,EAAQ,cAAgB,EAAe,EAAS,gBAC7D,GAAgB,OAEhB,KAEJ,CACA,IAAM,EAAU,EAAS,GAEvB,GACA,EAAQ,eAAiB,EAAS,iBAClC,EAAQ,iBAAmB,IAE3B,EAAM,KAAK,CAAC,EAAS,CAAQ,CAAC,EAC9B,EAAsB,EAAS,cAC/B,GAAgB,EAEpB,CACA,IAAM,EAAmB,EAAM,SAAW,EACpC,EAAoB,EAAM,SAAW,EAC3C,GAAI,EAAM,OAAS,GAAM,CAAC,GAAoB,CAAC,EAC7C,OAKF,IAAM,EAAgB,GAAoB,CAAC,EAAM,KAAM,GAAe,EAAW,mBAAmB,EAC9F,EAAiB,GAAqB,CAAC,EAAO,KAAM,GAAe,EAAW,mBAAmB,EACjG,EAAS,EAAM,KAAK,CAAC,EAAS,MAAe,CACjD,GAAG,EAEH,qBAAsB,IAAA,GACtB,oBAAqB,IAAA,GACrB,SAAU,CAAC,GAAG,EAAQ,SAAU,GAAG,EAAS,QAAQ,EACpD,WAAY,EAAQ,WAAa,EAAS,WAC1C,cAAe,EAAS,cACxB,SAAU,EAAS,SACnB,QAAS,EAAS,OACpB,EAAE,EAIG,KAAkB,CAAM,EAO7B,MAAO,CAAE,SAAQ,gBAAe,iBAAgB,eAAA,CAH9C,GAAI,EAAgB,CAAC,EAAI,EAAM,KAAK,CAAC,KAAa,CAAO,EACzD,GAAI,EAAiB,CAAC,EAAI,EAAM,KAAK,EAAG,KAAc,CAAQ,CAEH,CAAE,CACjE,CAGA,SAAS,EAAiD,EAA2B,CACnF,IAAM,EAAgB,CAAC,EACnB,EAAS,EACb,IAAK,IAAM,KAAc,EAClB,EAAW,sBACd,GAAU,GAIR,CAAC,EAAW,sBAAwB,CAAC,EAAW,qBAClD,EAAS,KAAK,CAAU,EAG5B,MAAO,CAAC,EAAU,CAAM,CAC1B,CAcA,SAAgB,EAAwB,EAAoC,CAC1E,IAAI,EAAgB,EAChB,EAAmB,EACnB,EAAsB,GAC1B,IAAK,IAAM,KAAc,EAAO,CAC9B,GAAI,EAAW,qBAAsB,CACnC,EAAsB,GACtB,QACF,CACA,GAAiB,EAAW,SAAS,OACrC,EAAmB,KAAK,IAAI,EAAkB,EAAW,SAAS,MAAM,CAC1E,CACA,OAAO,EAAsB,EAAgB,EAAgB,CAC/D,CAGA,SAAgB,EACd,EACA,EACA,EACA,EACM,CACN,IAAK,IAAI,EAAQ,EAAQ,gBAAiB,EAAQ,EAAQ,cAAe,GAAS,EAAG,CACnF,IAAM,EAAQ,EAAO,GACrB,IAAK,IAAI,EAAM,GAAO,UAAY,EAAG,IAAQ,GAAO,QAAU,IAAK,GAAO,GACpE,CAAC,GAAmB,EAAgB,IAAI,EAAM,CAAC,IACjD,EAAgB,IAAI,EAAM,CAAC,CAGjC,CACF"}
+{"version":3,"file":"duplication.cjs","names":[],"sources":["../src/duplication.ts"],"sourcesContent":["import type { DuplicationOptions } from './types.js';\n\n/**\n * Project-level duplication machinery operating on normalized token streams. Tokenization itself\n * (parsing, identifier anonymization, literal normalization) happens in the Rust addon, which\n * serializes each file's Token stream and statement structure; the helpers here match statement\n * windows across files, merge gap-adjacent groups, and count duplicated lines over that data.\n */\n\nexport const defaultDuplicationOptions: Required<DuplicationOptions> = {\n  minTokens: 40,\n  maxGapTokens: 30,\n  minSimilarityPercent: 70,\n};\n\n/**\n * Fills defaults for absent settings, applying the same normalization as the native boundary's\n * clampToU32 — NaN (e.g. `Number(unsetEnvVariable)`) counts as absent, and other values truncate\n * and clamp to [0, u32::MAX] — so the TypeScript half of cross-file matching cannot diverge from\n * the natively collected candidates on such input.\n */\nexport function resolveDuplicationOptions(options?: DuplicationOptions): Required<DuplicationOptions> {\n  return {\n    minTokens: resolveOption(options?.minTokens, defaultDuplicationOptions.minTokens),\n    maxGapTokens: resolveOption(options?.maxGapTokens, defaultDuplicationOptions.maxGapTokens),\n    minSimilarityPercent: resolveOption(options?.minSimilarityPercent, defaultDuplicationOptions.minSimilarityPercent),\n  };\n}\n\nfunction resolveOption(value: number | undefined, fallback: number): number {\n  return value === undefined || Number.isNaN(value)\n    ? fallback\n    : Math.min(Math.max(Math.trunc(value), 0), 0xFF_FF_FF_FF);\n}\n\n/** Minimum consecutive statements for a statement-sequence duplicate candidate. */\nconst minSequenceStatementCount = 2;\n/**\n * Caps the window length so statement-sequence enumeration stays linear in the statement count.\n * Heterogeneous clones longer than the cap are reported as capped windows (a deliberate\n * conservative undercount trading completeness for bounded discovery cost).\n */\nconst maxSequenceStatementCount = 100;\n\n/**\n * A region whose normalized tokens are at least 20% literal values is data-like (a lookup table, a\n * constant list, a value-mapping switch), not logic: literal values re-enter its fingerprint so\n * tables that merely share their shape stop counting as copy-paste. Compared in integer math\n * (5 * literals >= total) so the TypeScript and native sides cannot disagree on the boundary.\n */\nfunction isLiteralDense(literalCount: number, tokenCount: number): boolean {\n  return literalCount * 5 >= tokenCount;\n}\n\nexport interface Token {\n  /** Normalization target: identifiers to anonymize, literal kind tags, or the raw token text. */\n  kind: 'id' | 'text';\n  text: string;\n  /**\n   * Two INDEPENDENT hashes of `text` (djb2 and FNV-1a), precomputed so fingerprinting nested\n   * regions never re-hashes a token. Feeding the same per-token hash to both fingerprint\n   * accumulators would collapse the key to 32 effective bits: one djb2 collision between two\n   * token texts would then equate whole regions.\n   */\n  textHash: number;\n  textHash2: number;\n  /** Hash pair of a value-carrying literal's value, folded into data-like region fingerprints. */\n  literalHash?: number;\n  literalHash2?: number;\n  /**\n   * True for verbatim-kept NAMES (member/callee/type names, named grammar leaves): together with\n   * value-carrying literals these are the content-bearing tokens the near-miss content gate\n   * counts. Keywords, operators, and punctuation come from unnamed nodes and stay false.\n   */\n  isName?: boolean;\n  /** 0-based source rows the token occupies, so line coverage counts only matched-token lines. */\n  startRow: number;\n  endRow: number;\n}\n\n/** A token span with its source position; plain data so cross-file matching can retain it. */\nexport interface TokenRange {\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\n/** A contiguous run of matched tokens; gapped (merged) duplicates carry several per occurrence. */\ninterface TokenSegment {\n  startTokenIndex: number;\n  endTokenIndex: number;\n}\n\ninterface DuplicateCandidate {\n  fingerprint: string;\n  tokenCount: number;\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\nexport interface CountedOccurrence {\n  /** Matched token runs; more than one once gapped groups are merged. */\n  segments: TokenSegment[];\n  /**\n   * Set on occurrences whose span another reported group already counts: a retained group's\n   * occurrences that a partial gapped merge also paired into a merged group, and cross-file copies\n   * nested inside a larger group's region. Block counting must not count them again.\n   */\n  spanCountedElsewhere?: boolean;\n  /**\n   * Set on cross-file copies nested inside a larger group's region (they also set\n   * `spanCountedElsewhere`). They never pair in gapped merging, and they do not keep their group's\n   * standalone copies from merging, which they are not copies of. They do keep the merged group\n   * from taking their group's place, since only the original group reports the nesting.\n   */\n  nestedInLargerGroup?: boolean;\n  /** Sum of segment token counts (the gap tokens are not matched content). */\n  tokenCount: number;\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\n/** A duplicate region found in one file, exported for cross-file matching by fingerprint. */\nexport interface CrossFileDuplicateCandidate {\n  /** Content key: equal fingerprints mean equal normalized token sequences (up to hash collision). */\n  fingerprint: string;\n  tokenCount: number;\n  /** Token positions within the owning file, for cross-file gapped (Type-3) merging. */\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\n/**\n * One file's contribution to cross-file clone detection: catalogued candidates plus the normalized\n * token stream and statement structure, so the project-level pass can match partial statement runs\n * (windows that a single file cannot know repeat elsewhere) and merge gap-adjacent groups.\n */\nexport interface CrossFileDuplicationFileData {\n  candidates: CrossFileDuplicateCandidate[];\n  tokens: Token[];\n  containerStatements: TokenRange[][];\n  /**\n   * The mutually disjoint blocks (at least `minTokens` long, not literal-dense) compared for\n   * cross-file near-miss (Type-3) clones. Optional for backward compatibility; without it, the\n   * file takes part in exact and gapped cross-file matching only.\n   */\n  nearMissBlocks?: TokenRange[];\n  /**\n   * 1-based lines that are neither blank nor comment-only, so cross-file line coverage counts only\n   * code lines (blank rows inside multi-row tokens such as template literals carry no content).\n   * Optional for backward compatibility; without it, every matched-token row counts.\n   */\n  codeLineNumbers?: Set<number>;\n}\n\n/** literalCountPrefix[i] = value-carrying literal tokens in tokens[0..i), for O(1) density checks. */\nexport function buildLiteralCountPrefix(tokens: Token[]): Int32Array {\n  const prefix = new Int32Array(tokens.length + 1);\n  for (const [index, token] of tokens.entries()) {\n    prefix[index + 1] = (prefix[index] ?? 0) + (token.literalHash === undefined ? 0 : 1);\n  }\n  return prefix;\n}\n\ninterface WindowOccurrences {\n  count: number;\n  /** -1 once occurrences span more than one container. */\n  containerIndex: number;\n  /** -1 once occurrences span more than one context (file). */\n  contextIndex: number;\n  minStart: number;\n  maxStart: number;\n}\n\ninterface SequenceWindow {\n  containerIndex: number;\n  start: number;\n  length: number;\n}\n\n/** One file's token stream and statement containers, as a window-matching context. */\nexport interface SequenceWindowContext {\n  tokens: Token[];\n  literalCountPrefix: Int32Array;\n  containers: TokenRange[][];\n}\n\nexport interface ContextualSequenceCandidate {\n  candidate: DuplicateCandidate;\n  contextIndex: number;\n}\n\n/**\n * Enumerates runs of consecutive sibling statements over one or more contexts (files). Every\n * container statement participates; only the window length is capped, so enumeration stays linear\n * in the statement count. Windows are grouped by a cheap rolling hash of per-statement\n * fingerprints, and only locally maximal repeated windows — those whose one-statement extensions\n * stop repeating — become candidates with an exact (window-consistent) fingerprint. Without the\n * maximality filter a degenerate file of near-identical statements would fingerprint every\n * sub-window of every repeated region. With `requireMultipleContexts` a window only counts as\n * repeated when its occurrences span at least two contexts (CPD-style cross-file matching): a\n * repeat confined to one file is that file's own concern, and emitting it here would flood the\n * project-level selection with unusable single-file groups.\n */\nexport function collectSequenceWindowCandidates(\n  contexts: SequenceWindowContext[],\n  minTokens: number,\n  requireMultipleContexts: boolean\n): ContextualSequenceCandidate[] {\n  const candidates: ContextualSequenceCandidate[] = [];\n  const contextIndexByContainer: number[] = [];\n  const containers: TokenRange[][] = [];\n  for (const [contextIndex, context] of contexts.entries()) {\n    for (const statements of context.containers) {\n      contextIndexByContainer.push(contextIndex);\n      containers.push(statements);\n    }\n  }\n  const contextAt = (containerIndex: number): SequenceWindowContext | undefined =>\n    contexts[contextIndexByContainer[containerIndex] ?? 0];\n  const occurrencesByWindowKey = new Map<number, WindowOccurrences>();\n  const containerWindows = containers.map((statements, containerIndex) =>\n    enumerateContainerWindows(contextAt(containerIndex)?.tokens ?? [], statements, minTokens)\n  );\n  for (const [containerIndex, windows] of containerWindows.entries()) {\n    const contextIndex = contextIndexByContainer[containerIndex] ?? 0;\n    for (const [start, row] of windows.windowKeysByStart.entries()) {\n      for (const windowKey of row) {\n        if (windowKey === undefined) {\n          continue;\n        }\n        const occurrences = occurrencesByWindowKey.get(windowKey);\n        if (occurrences) {\n          occurrences.count += 1;\n          if (occurrences.containerIndex !== containerIndex) {\n            occurrences.containerIndex = -1;\n          }\n          if (occurrences.contextIndex !== contextIndex) {\n            occurrences.contextIndex = -1;\n          }\n          occurrences.minStart = Math.min(occurrences.minStart, start);\n          occurrences.maxStart = Math.max(occurrences.maxStart, start);\n        } else {\n          occurrencesByWindowKey.set(windowKey, {\n            count: 1,\n            containerIndex,\n            contextIndex,\n            minStart: start,\n            maxStart: start,\n          });\n        }\n      }\n    }\n  }\n\n  // A window only \"repeats\" when two of its occurrences can coexist without overlapping: sliding\n  // matches inside a homogeneous run (start spread smaller than the window length) can never both\n  // be counted and must neither qualify a window nor dominate its sub-windows. Cross-context\n  // matching instead requires occurrences in two contexts, which coexist by construction.\n  const repeats = (windowKey: number | undefined, length: number): boolean => {\n    if (windowKey === undefined) {\n      return false;\n    }\n    const occurrences = occurrencesByWindowKey.get(windowKey);\n    if (occurrences === undefined || occurrences.count < 2) {\n      return false;\n    }\n    if (requireMultipleContexts) {\n      return occurrences.contextIndex === -1;\n    }\n    return occurrences.containerIndex === -1 || occurrences.maxStart - occurrences.minStart >= length;\n  };\n\n  // A window whose statements all share one normalized shape (sixteen `let x = 0;` declarations,\n  // a constant table) is a homogeneous preamble, not a copy-paste: requiring two distinct\n  // per-statement shapes keeps such runs out of duplicate groups and the duplication ratio.\n  const hasDistinctStatements = (window: SequenceWindow): boolean => {\n    const hashes = containerWindows[window.containerIndex]?.statementHashes ?? [];\n    const firstHash = hashes[window.start];\n    for (let index = window.start + 1; index < window.start + window.length; index += 1) {\n      if (hashes[index] !== firstHash) {\n        return true;\n      }\n    }\n    return false;\n  };\n\n  const maximalWindows: SequenceWindow[] = [];\n  for (const [containerIndex, windows] of containerWindows.entries()) {\n    for (const [start, row] of windows.windowKeysByStart.entries()) {\n      for (const [length, windowKey] of row.entries()) {\n        if (!repeats(windowKey, length) || !hasDistinctStatements({ containerIndex, start, length })) {\n          continue;\n        }\n        // Dominated windows are skipped: the one-statement extension also repeats, so a larger\n        // candidate covering this window exists.\n        const extendedRight = windows.windowKeysByStart[start]?.[length + 1];\n        const extendedLeft = windows.windowKeysByStart[start - 1]?.[length + 1];\n        if (repeats(extendedRight, length + 1) || repeats(extendedLeft, length + 1)) {\n          continue;\n        }\n        maximalWindows.push({ containerIndex, start, length });\n      }\n    }\n  }\n\n  // The rolling hash anonymizes identifiers per statement, so a window can look repeated coarsely\n  // while its exact (window-consistent) fingerprints differ, and a longer window's match can\n  // dominate sub-windows that other copies still need (three copies where only two extend one\n  // statement further). Every emitted window therefore exposes its repeating, unvisited\n  // sub-windows; `visited` bounds the worklist and lengths strictly decrease, so it terminates.\n  const visited = new Set(maximalWindows.map(windowId));\n  let frontier = maximalWindows;\n  while (frontier.length > 0) {\n    const emitted: SequenceWindow[] = [];\n    for (const window of frontier) {\n      const statements = containers[window.containerIndex];\n      const first = statements?.[window.start];\n      const last = statements?.[window.start + window.length - 1];\n      const context = contextAt(window.containerIndex);\n      if (!first || !last || !context) {\n        continue;\n      }\n      const fingerprint = `s:${fingerprintKey(context.tokens, context.literalCountPrefix, first.startTokenIndex, last.endTokenIndex)}`;\n      candidates.push({\n        candidate: toCandidate(fingerprint, first.startTokenIndex, last.endTokenIndex, first, last),\n        contextIndex: contextIndexByContainer[window.containerIndex] ?? 0,\n      });\n      emitted.push(window);\n    }\n    frontier = [];\n    for (const window of emitted) {\n      for (const start of [window.start, window.start + 1]) {\n        const subWindow = { containerIndex: window.containerIndex, start, length: window.length - 1 };\n        const subWindowKey = containerWindows[window.containerIndex]?.windowKeysByStart[start]?.[subWindow.length];\n        if (\n          visited.has(windowId(subWindow)) ||\n          !repeats(subWindowKey, subWindow.length) ||\n          !hasDistinctStatements(subWindow)\n        ) {\n          continue;\n        }\n        visited.add(windowId(subWindow));\n        frontier.push(subWindow);\n      }\n    }\n  }\n  return candidates;\n}\n\nfunction windowId(window: SequenceWindow): string {\n  return `${window.containerIndex}:${window.start}:${window.length}`;\n}\n\ninterface ContainerWindows {\n  /** windowKeysByStart[start][length] is the rolling-hash key of the window, or undefined if it is below the size thresholds. */\n  windowKeysByStart: (number | undefined)[][];\n  /** Per-statement fingerprint hashes, for the distinct-shape requirement on windows. */\n  statementHashes: number[];\n}\n\nfunction enumerateContainerWindows(tokens: Token[], statements: TokenRange[], minTokens: number): ContainerWindows {\n  const statementHashes = statements.map((statement) =>\n    fingerprintHash(tokens, statement.startTokenIndex, statement.endTokenIndex)\n  );\n  const windowKeysByStart: (number | undefined)[][] = [];\n  for (let start = 0; start < statements.length; start += 1) {\n    const row: (number | undefined)[] = [];\n    let hash = 5381;\n    let tokenCount = 0;\n    const maxEnd = Math.min(statements.length, start + maxSequenceStatementCount);\n    for (let end = start; end < maxEnd; end += 1) {\n      const statement = statements[end];\n      const statementHash = statementHashes[end];\n      if (!statement || statementHash === undefined) {\n        break;\n      }\n      hash = combineHashes(hash, statementHash);\n      tokenCount += statement.endTokenIndex - statement.startTokenIndex;\n      const statementCount = end - start + 1;\n      row[statementCount] =\n        statementCount >= minSequenceStatementCount && tokenCount >= minTokens\n          ? combineHashes(hash, statementCount)\n          : undefined;\n    }\n    windowKeysByStart.push(row);\n  }\n  return { windowKeysByStart, statementHashes };\n}\n\n/**\n * Longest-common-subsequence LENGTH of two symbol sequences via the Allison–Dix bit-parallel\n * recurrence (O(|a|/32 · |b|) words): per symbol of `b`, `x = match | v` and\n * `v = x & ~(x - ((v << 1) | 1))` over multi-word bit vectors; the set bits of `v` count the LCS.\n */\nexport function lcsLength(a: Int32Array, b: Int32Array): number {\n  return createLcsLengthCounter(a)(b);\n}\n\n/**\n * lcsLength against a fixed `a`, whose per-symbol position masks are built once and reused for\n * every `b` compared with it.\n */\nexport function createLcsLengthCounter(a: Int32Array): (b: Int32Array) => number {\n  const wordCount = (a.length + 31) >>> 5;\n  const positionMasks = new Map<number, Uint32Array>();\n  for (const [index, symbol] of a.entries()) {\n    let mask = positionMasks.get(symbol);\n    if (!mask) {\n      mask = new Uint32Array(wordCount);\n      positionMasks.set(symbol, mask);\n    }\n    const word = index >>> 5;\n    // The Uint32Array store wraps the signed int32 bit pattern to unsigned. The `?? 0` guards in\n    // this function are required by noUncheckedIndexedAccess (typed-array reads type as\n    // `number | undefined`), not redundancy: every index is in bounds.\n    mask[word] = (mask[word] ?? 0) | (1 << (index & 31));\n  }\n\n  const v = new Uint32Array(wordCount);\n  return (b) => {\n    v.fill(0);\n    for (const symbol of b) {\n      const matchMask = positionMasks.get(symbol);\n      // `(v << 1) | 1` shifts a carry bit across words; subtraction borrows across words.\n      let shiftCarry = 1;\n      let borrow = 0;\n      for (let word = 0; word < wordCount; word += 1) {\n        const previous = v[word] ?? 0;\n        // oxlint-disable-next-line unicorn/prefer-math-trunc -- `>>> 0` reinterprets the signed int32 bit pattern as unsigned so the borrow subtraction below compares magnitudes; Math.trunc would keep it negative.\n        const x = ((matchMask?.[word] ?? 0) | previous) >>> 0;\n        // oxlint-disable-next-line unicorn/prefer-math-trunc -- same unsigned reinterpretation as `x`.\n        const shifted = ((previous << 1) | shiftCarry) >>> 0;\n        shiftCarry = previous >>> 31;\n        const difference = x - shifted - borrow;\n        borrow = difference < 0 ? 1 : 0;\n        // The Uint32Array store wraps the signed int32 bit pattern to unsigned.\n        v[word] = x & ~difference;\n      }\n    }\n\n    let length = 0;\n    for (const word of v) {\n      length += popCount(word);\n    }\n    return length;\n  };\n}\n\nfunction popCount(value: number): number {\n  let count = value - ((value >>> 1) & 0x55_55_55_55);\n  count = (count & 0x33_33_33_33) + ((count >>> 2) & 0x33_33_33_33);\n  return (Math.imul((count + (count >>> 4)) & 0x0F_0F_0F_0F, 0x01_01_01_01) >>> 24) & 0xFF;\n}\n\nfunction toCandidate(\n  fingerprint: string,\n  startTokenIndex: number,\n  endTokenIndex: number,\n  first: TokenRange,\n  last: TokenRange\n): DuplicateCandidate {\n  return {\n    fingerprint,\n    tokenCount: endTokenIndex - startTokenIndex,\n    startTokenIndex,\n    endTokenIndex,\n    startIndex: first.startIndex,\n    endIndex: last.endIndex,\n    startLine: first.startLine,\n    endLine: last.endLine,\n  };\n}\n\n/** Caches of hashText/hashText2 over '$0', '$1', ... so anonymized identifiers hash without allocating. */\nconst anonymizedIndexHashes: number[] = [];\nconst anonymizedIndexHashes2: number[] = [];\n\nfunction anonymizedIndexHash(index: number): number {\n  let hash = anonymizedIndexHashes[index];\n  if (hash === undefined) {\n    hash = hashText(`$${index}`);\n    anonymizedIndexHashes[index] = hash;\n  }\n  return hash;\n}\n\nfunction anonymizedIndexHash2(index: number): number {\n  let hash = anonymizedIndexHashes2[index];\n  if (hash === undefined) {\n    hash = hashText2(`$${index}`);\n    anonymizedIndexHashes2[index] = hash;\n  }\n  return hash;\n}\n\n/**\n * Content key of a token range: two independent 32-bit hashes over the normalized token sequence\n * (identifiers anonymized consistently by first-occurrence order) plus the token count. Regions\n * with equal keys are treated as equal content; a collision would need both 32-bit hashes and the\n * length to coincide, which is negligible for a metrics report. The format and arithmetic match\n * fingerprint_key in native/src/duplication.rs exactly, so window candidates fingerprinted here\n * group together with the per-file candidates the addon catalogues.\n */\nfunction fingerprintKey(\n  tokens: Token[],\n  literalCountPrefix: Int32Array,\n  startTokenIndex: number,\n  endTokenIndex: number\n): string {\n  const literalCount = (literalCountPrefix[endTokenIndex] ?? 0) - (literalCountPrefix[startTokenIndex] ?? 0);\n  const literalDense = isLiteralDense(literalCount, endTokenIndex - startTokenIndex);\n  const [primary, secondary] = fingerprintHashPair(tokens, startTokenIndex, endTokenIndex, literalDense);\n  return `${primary}:${secondary}:${endTokenIndex - startTokenIndex}`;\n}\n\n/**\n * A single 32-bit summary of a range for the coarse rolling-hash phase. Deliberately\n * density-agnostic: density is a property of the final candidate REGION, and folding literal\n * values into per-statement hashes would make a dense statement inside a logic-heavy window\n * (`const weights = [1, 2, 3];`) block the window from ever being enumerated. The coarse phase\n * over-approximates on shape alone; the exact region fingerprint still applies the density rule.\n */\nfunction fingerprintHash(tokens: Token[], startTokenIndex: number, endTokenIndex: number): number {\n  const [primary, secondary] = fingerprintHashPair(tokens, startTokenIndex, endTokenIndex, false);\n  // XOR already coerces to int32, matching the native side's i32 arithmetic.\n  return primary ^ Math.imul(secondary, 31);\n}\n\nfunction fingerprintHashPair(\n  tokens: Token[],\n  startTokenIndex: number,\n  endTokenIndex: number,\n  foldLiteralValues: boolean\n): [number, number] {\n  const indexByIdentifier = new Map<string, number>();\n  let primary = 5381;\n  let secondary = 52_711;\n  for (let index = startTokenIndex; index < endTokenIndex; index += 1) {\n    const token = tokens[index];\n    if (!token) {\n      continue;\n    }\n    // Each accumulator consumes its own independent per-token hash: sharing one would collapse\n    // the key to 32 effective bits (a single djb2 collision would equate whole regions).\n    let part: number;\n    let part2: number;\n    if (token.kind === 'id') {\n      let identifierIndex = indexByIdentifier.get(token.text);\n      if (identifierIndex === undefined) {\n        identifierIndex = indexByIdentifier.size;\n        indexByIdentifier.set(token.text, identifierIndex);\n      }\n      part = anonymizedIndexHash(identifierIndex);\n      part2 = anonymizedIndexHash2(identifierIndex);\n    } else {\n      part = token.textHash;\n      part2 = token.textHash2;\n    }\n    // oxlint-disable-next-line unicorn/prefer-math-trunc -- `| 0` wraps the sum to int32 (Math.trunc does not), which must match the native side's wrapping i32 arithmetic.\n    primary = (Math.imul(primary, 31) + part) | 0;\n    secondary = Math.imul(secondary, 37) ^ part2;\n    if (foldLiteralValues && token.literalHash !== undefined && token.literalHash2 !== undefined) {\n      // oxlint-disable-next-line unicorn/prefer-math-trunc -- `| 0` wraps the sum to int32 (Math.trunc does not), which must match the native side's wrapping i32 arithmetic.\n      primary = (Math.imul(primary, 31) + token.literalHash) | 0;\n      secondary = Math.imul(secondary, 37) ^ token.literalHash2;\n    }\n  }\n  return [primary, secondary];\n}\n\n/** djb2-style hash; XOR keeps the value in signed 32-bit range, which is fine for a grouping key. */\nfunction hashText(text: string): number {\n  let hash = 5381;\n  for (let index = 0; index < text.length; index += 1) {\n    // oxlint-disable-next-line unicorn/prefer-code-point -- djb2 hashes UTF-16 code units; codePointAt would hash surrogate pairs twice (full code point, then the lone low surrogate).\n    hash = Math.imul(hash, 33) ^ text.charCodeAt(index);\n  }\n  return hash;\n}\n\n/** FNV-1a over UTF-16 code units: independent of hashText so the two accumulators never share input. */\nfunction hashText2(text: string): number {\n  let hash = -2_128_831_035; // 2166136261 as int32 (the FNV-1a offset basis)\n  for (let index = 0; index < text.length; index += 1) {\n    // oxlint-disable-next-line unicorn/prefer-code-point -- hashes UTF-16 code units like hashText.\n    hash = Math.imul(hash ^ text.charCodeAt(index), 16_777_619);\n  }\n  return hash;\n}\n\nfunction combineHashes(hash: number, value: number): number {\n  return Math.imul(hash, 31) + value;\n}\n\n/**\n * Merges duplicate groups separated by a small token gap into one gapped (Type-3) clone group: a\n * copy edited in one spot splits into two exact groups whose occurrences sit side by side in the\n * same order. Occurrences are paired greedily in source order; a merge happens when the pairing\n * fully pairs at least one group with at least two pairs. Equal-cardinality groups whose\n * occurrences all pair merge into one group as before. When cardinalities differ (a fragment also\n * occurs standalone: prefix ×3, suffix ×2), the fully-paired group is subsumed into the merged\n * gapped group while the other group is RETAINED with ALL its occurrences: dropping the leftover\n * would lose duplicated-line coverage, and reporting it alone would make a single-occurrence group\n * (contradicting duplicateBlockGroupCount's \"appears more than once\" meaning). Line coverage\n * unions ranges, so the overlap between the retained exact group and the merged group is harmless.\n * Merging repeats to a fixpoint so a clone edited in several spots still reassembles; it\n * terminates because a merge either removes the input groups it replaces or marks their paired\n * occurrences as counted elsewhere, and only unmarked occurrences of remaining groups pair, so the\n * number of pairable occurrences strictly decreases with every merge. Gap tokens are not matched\n * content: line coverage and sizes count only the matched segments. Generic so cross-file merging\n * can thread file identity through occurrences.\n */\nexport function mergeAdjacentGroups<T extends CountedOccurrence>(\n  groups: T[][],\n  maxGapTokens: number,\n  isReportableGroup: (group: T[]) => boolean = () => true\n): T[][] {\n  if (maxGapTokens <= 0 || groups.length < 2) {\n    return groups;\n  }\n  // Deterministic processing order (mirrored by the native side): by first occurrence position.\n  groups.sort(compareGroups);\n  for (let restart = true; restart;) {\n    restart = false;\n    const partnersByGroup = collectGapAdjacentPartners(groups, maxGapTokens);\n    for (let leftIndex = 0; leftIndex < groups.length && !restart; leftIndex += 1) {\n      for (const rightIndex of partnersByGroup[leftIndex] ?? []) {\n        const left = groups[leftIndex];\n        const right = groups[rightIndex];\n        if (!left || !right) {\n          continue;\n        }\n        const forward = mergeGroups(left, right, maxGapTokens, isReportableGroup);\n        const result = forward ?? mergeGroups(right, left, maxGapTokens, isReportableGroup);\n        if (!result) {\n          continue;\n        }\n        const leftReplaced = forward ? result.firstReplaced : result.secondReplaced;\n        const rightReplaced = forward ? result.secondReplaced : result.firstReplaced;\n        if (leftReplaced && rightReplaced) {\n          groups[leftIndex] = result.merged;\n          groups.splice(rightIndex, 1);\n        } else if (rightReplaced) {\n          groups[rightIndex] = result.merged;\n        } else if (leftReplaced) {\n          groups[leftIndex] = result.merged;\n        } else {\n          // Both groups stay (each is only partly paired, or reports nested copies of its own), so\n          // the merged group joins them instead of taking a place.\n          groups.push(result.merged);\n        }\n        // A group that stays keeps ALL its occurrences (line coverage must not shrink, and a\n        // reported group must keep >= 2 occurrences), so its paired occurrences now also live\n        // inside the merged group's occurrences: mark them so duplicateBlockCount counts each\n        // token span once, and so the same pair cannot merge again.\n        for (const occurrence of result.pairedRetained) {\n          occurrence.spanCountedElsewhere = true;\n        }\n        groups.sort(compareGroups);\n        restart = true;\n        break;\n      }\n    }\n  }\n  return groups;\n}\n\n/**\n * Per group index, the ascending indexes of later groups that mergeGroups can pair with it in\n * either direction: an occurrence of one starts within `maxGapTokens` after an occurrence of the\n * other ends. Every other pair fails mergeGroups, so skipping it leaves the fixpoint unchanged\n * while each restart stops costing a comparison per pair of groups.\n */\nfunction collectGapAdjacentPartners(groups: CountedOccurrence[][], maxGapTokens: number): number[][] {\n  const starts: { startTokenIndex: number; groupIndex: number }[] = [];\n  for (const [groupIndex, group] of groups.entries()) {\n    for (const occurrence of group) {\n      starts.push({ startTokenIndex: occurrence.startTokenIndex, groupIndex });\n    }\n  }\n  starts.sort((left, right) => left.startTokenIndex - right.startTokenIndex);\n  const partners = groups.map(() => new Set<number>());\n  for (const [groupIndex, group] of groups.entries()) {\n    for (const occurrence of group) {\n      for (\n        let index = lowerBoundByStart(starts, occurrence.endTokenIndex);\n        index < starts.length && (starts[index]?.startTokenIndex ?? 0) <= occurrence.endTokenIndex + maxGapTokens;\n        index += 1\n      ) {\n        const other = starts[index]?.groupIndex ?? groupIndex;\n        if (other !== groupIndex) {\n          partners[Math.min(other, groupIndex)]?.add(Math.max(other, groupIndex));\n        }\n      }\n    }\n  }\n  return partners.map((set) => [...set].toSorted((left, right) => left - right));\n}\n\nfunction lowerBoundByStart(sorted: { startTokenIndex: number }[], target: number): number {\n  let low = 0;\n  let high = sorted.length;\n  while (low < high) {\n    const middle = (low + high) >>> 1;\n    if ((sorted[middle]?.startTokenIndex ?? 0) < target) {\n      low = middle + 1;\n    } else {\n      high = middle;\n    }\n  }\n  return low;\n}\n\nfunction compareGroups(left: CountedOccurrence[], right: CountedOccurrence[]): number {\n  const leftFirst = left[0];\n  const rightFirst = right[0];\n  return (\n    (leftFirst?.startTokenIndex ?? 0) - (rightFirst?.startTokenIndex ?? 0) ||\n    (leftFirst?.endTokenIndex ?? 0) - (rightFirst?.endTokenIndex ?? 0)\n  );\n}\n\ninterface MergeResult<T> {\n  merged: T[];\n  /**\n   * Whether the merged group takes the respective input group's place: its standalone occurrences\n   * were all paired and it holds no nested copies that only it can report.\n   */\n  firstReplaced: boolean;\n  secondReplaced: boolean;\n  /** The occurrences of groups that stay, which the merged group's spans now also cover. */\n  pairedRetained: T[];\n}\n\n/**\n * Pairs `second` occurrences with gap-preceding `first` occurrences, greedily in source order:\n * each trailing occurrence takes the earliest unused leading occurrence within the gap, and a\n * pair's leading must start at or after the previous pair's trailing end so merged spans never\n * overlap. A merge needs at least two pairs (a merged group must still mean \"appears more than\n * once\") and must fully consume at least one group; for equal cardinalities this reduces to the\n * strict all-pairs merge, so pre-partial-merge behavior is unchanged there.\n */\nfunction mergeGroups<T extends CountedOccurrence>(\n  first: T[],\n  second: T[],\n  maxGapTokens: number,\n  isReportableGroup: (group: T[]) => boolean\n): MergeResult<T> | undefined {\n  // Occurrences a previous partial merge already paired into a merged group must not pair again:\n  // their spans already live inside that merged group, so re-pairing them would assemble a second,\n  // competing merged group instead of letting the existing merged group extend (and would count\n  // the same span twice). Consumption is still judged against the FULL group, so a group holding\n  // shared occurrences is never subsumed away; nested copies, which the merged span would not\n  // cover anyway, are left out of that judgment so they cannot veto a merge of the standalone\n  // copies.\n  const [leadings, firstLength] = pairableOccurrences(first);\n  const [trailings, secondLength] = pairableOccurrences(second);\n  const pairs: [T, T][] = [];\n  let leadingIndex = 0;\n  let previousTrailingEnd = -1;\n  for (const trailing of trailings) {\n    // Leadings ending too far before this trailing can never pair a later (even farther) one.\n    while (leadingIndex < leadings.length) {\n      const leading = leadings[leadingIndex];\n      if (leading && leading.endTokenIndex + maxGapTokens < trailing.startTokenIndex) {\n        leadingIndex += 1;\n      } else {\n        break;\n      }\n    }\n    const leading = leadings[leadingIndex];\n    if (\n      leading &&\n      leading.endTokenIndex <= trailing.startTokenIndex &&\n      leading.startTokenIndex >= previousTrailingEnd\n    ) {\n      pairs.push([leading, trailing]);\n      previousTrailingEnd = trailing.endTokenIndex;\n      leadingIndex += 1;\n    }\n  }\n  const firstFullyPaired = pairs.length === firstLength;\n  const secondFullyPaired = pairs.length === secondLength;\n  if (pairs.length < 2 || (!firstFullyPaired && !secondFullyPaired)) {\n    return undefined;\n  }\n  // A group holding nested copies stays even when all its standalone copies pair: the merged\n  // group's content is larger than what those copies matched, so only the original group can\n  // report which files share the matched fragment.\n  const firstReplaced = firstFullyPaired && !first.some((occurrence) => occurrence.nestedInLargerGroup);\n  const secondReplaced = secondFullyPaired && !second.some((occurrence) => occurrence.nestedInLargerGroup);\n  const merged = pairs.map(([leading, trailing]) => ({\n    ...leading,\n    // A merged occurrence is a fresh span combination; it never inherits shared-span marks.\n    spanCountedElsewhere: undefined,\n    nestedInLargerGroup: undefined,\n    segments: [...leading.segments, ...trailing.segments],\n    tokenCount: leading.tokenCount + trailing.tokenCount,\n    endTokenIndex: trailing.endTokenIndex,\n    endIndex: trailing.endIndex,\n    endLine: trailing.endLine,\n  }));\n  // Only occurrences of one file pair (file offsets exceed the gap), so a merged group spans the\n  // files its pairs sit in: with nested copies left out of pairing, that can be fewer files than\n  // the input groups covered, and a merged group that is no longer reportable must not form.\n  if (!isReportableGroup(merged)) {\n    return undefined;\n  }\n  const pairedRetained = [\n    ...(firstReplaced ? [] : pairs.map(([leading]) => leading)),\n    ...(secondReplaced ? [] : pairs.map(([, trailing]) => trailing)),\n  ];\n  return { merged, firstReplaced, secondReplaced, pairedRetained };\n}\n\n/** One pass over a group: its pairable occurrences and its non-nested occurrence count. */\nfunction pairableOccurrences<T extends CountedOccurrence>(group: T[]): [T[], number] {\n  const pairable: T[] = [];\n  let length = 0;\n  for (const occurrence of group) {\n    if (!occurrence.nestedInLargerGroup) {\n      length += 1;\n    }\n    // Either flag keeps an occurrence out of pairing: its span is already counted elsewhere, or it\n    // is a nested copy of content the merged span would not cover.\n    if (!occurrence.spanCountedElsewhere && !occurrence.nestedInLargerGroup) {\n      pairable.push(occurrence);\n    }\n  }\n  return [pairable, length];\n}\n\n/**\n * Redundant copies one group adds to duplicateBlockCount. Each redundant occurrence contributes\n * one count per matched fragment, so merging a gapped clone's fragments into one group does not\n * halve duplicateBlockCount: an edited two-fragment pair still counts 2, exactly as its unmerged\n * fragments did. Occurrence shapes can differ within one group (a\n * gap-merged exact pair plus an appended whole-block near-miss copy), so every occurrence's\n * fragments are summed and one representative — the largest — is deducted, keeping the count\n * independent of source order. Occurrences a partial gapped merge also paired into a merged group\n * are skipped: their spans are already counted there, and such a retained group deducts no\n * representative of its own — the merged group's representative already stands for the shared\n * content — so no token span contributes to the count twice.\n */\nexport function countRedundantFragments(group: CountedOccurrence[]): number {\n  let fragmentCount = 0;\n  let maxFragmentCount = 0;\n  let hasSharedOccurrence = false;\n  for (const occurrence of group) {\n    if (occurrence.spanCountedElsewhere) {\n      hasSharedOccurrence = true;\n      continue;\n    }\n    fragmentCount += occurrence.segments.length;\n    maxFragmentCount = Math.max(maxFragmentCount, occurrence.segments.length);\n  }\n  return hasSharedOccurrence ? fragmentCount : fragmentCount - maxFragmentCount;\n}\n\n/** Adds the 1-based code lines the segment's matched tokens cover; shared with cross-file coverage. */\nexport function collectSegmentLines(\n  segment: { startTokenIndex: number; endTokenIndex: number },\n  tokens: Token[],\n  codeLineNumbers: Set<number> | undefined,\n  duplicatedLines: Set<number>\n): void {\n  for (let index = segment.startTokenIndex; index < segment.endTokenIndex; index += 1) {\n    const token = tokens[index];\n    for (let row = token?.startRow ?? 0; row <= (token?.endRow ?? -1); row += 1) {\n      if (!codeLineNumbers || codeLineNumbers.has(row + 1)) {\n        duplicatedLines.add(row + 1);\n      }\n    }\n  }\n}\n"],"mappings":"aASA,MAAa,EAA0D,CACrE,UAAW,GACX,aAAc,GACd,qBAAsB,EACxB,EAQA,SAAgB,EAA0B,EAA4D,CACpG,MAAO,CACL,UAAW,EAAc,GAAS,UAAW,EAA0B,SAAS,EAChF,aAAc,EAAc,GAAS,aAAc,EAA0B,YAAY,EACzF,qBAAsB,EAAc,GAAS,qBAAsB,EAA0B,oBAAoB,CACnH,CACF,CAEA,SAAS,EAAc,EAA2B,EAA0B,CAC1E,OAAO,IAAU,IAAA,IAAa,OAAO,MAAM,CAAK,EAC5C,EACA,KAAK,IAAI,KAAK,IAAI,KAAK,MAAM,CAAK,EAAG,CAAC,EAAG,UAAa,CAC5D,CAiBA,SAAS,EAAe,EAAsB,EAA6B,CACzE,OAAO,EAAe,GAAK,CAC7B,CAuHA,SAAgB,EAAwB,EAA6B,CACnE,IAAM,EAAS,IAAI,WAAW,EAAO,OAAS,CAAC,EAC/C,IAAK,GAAM,CAAC,EAAO,KAAU,EAAO,QAAQ,EAC1C,EAAO,EAAQ,IAAM,EAAO,IAAU,IAAM,EAAM,cAAgB,IAAA,GAAY,EAAI,GAEpF,OAAO,CACT,CA0CA,SAAgB,EACd,EACA,EACA,EAC+B,CAC/B,IAAM,EAA4C,CAAC,EAC7C,EAAoC,CAAC,EACrC,EAA6B,CAAC,EACpC,IAAK,GAAM,CAAC,EAAc,KAAY,EAAS,QAAQ,EACrD,IAAK,IAAM,KAAc,EAAQ,WAC/B,EAAwB,KAAK,CAAY,EACzC,EAAW,KAAK,CAAU,EAG9B,IAAM,EAAa,GACjB,EAAS,EAAwB,IAAmB,GAChD,EAAyB,IAAI,IAC7B,EAAmB,EAAW,KAAK,EAAY,IACnD,EAA0B,EAAU,CAAc,CAAC,EAAE,QAAU,CAAC,EAAG,EAAY,CAAS,CAC1F,EACA,IAAK,GAAM,CAAC,EAAgB,KAAY,EAAiB,QAAQ,EAAG,CAClE,IAAM,EAAe,EAAwB,IAAmB,EAChE,IAAK,GAAM,CAAC,EAAO,KAAQ,EAAQ,kBAAkB,QAAQ,EAC3D,IAAK,IAAM,KAAa,EAAK,CAC3B,GAAI,IAAc,IAAA,GAChB,SAEF,IAAM,EAAc,EAAuB,IAAI,CAAS,EACpD,GACF,EAAY,OAAS,EACjB,EAAY,iBAAmB,IACjC,EAAY,eAAiB,IAE3B,EAAY,eAAiB,IAC/B,EAAY,aAAe,IAE7B,EAAY,SAAW,KAAK,IAAI,EAAY,SAAU,CAAK,EAC3D,EAAY,SAAW,KAAK,IAAI,EAAY,SAAU,CAAK,GAE3D,EAAuB,IAAI,EAAW,CACpC,MAAO,EACP,iBACA,eACA,SAAU,EACV,SAAU,CACZ,CAAC,CAEL,CAEJ,CAMA,IAAM,GAAW,EAA+B,IAA4B,CAC1E,GAAI,IAAc,IAAA,GAChB,MAAO,GAET,IAAM,EAAc,EAAuB,IAAI,CAAS,EAOxD,OANI,IAAgB,IAAA,IAAa,EAAY,MAAQ,EAC5C,GAEL,EACK,EAAY,eAAiB,GAE/B,EAAY,iBAAmB,IAAM,EAAY,SAAW,EAAY,UAAY,CAC7F,EAKM,EAAyB,GAAoC,CACjE,IAAM,EAAS,EAAiB,EAAO,eAAe,EAAE,iBAAmB,CAAC,EACtE,EAAY,EAAO,EAAO,OAChC,IAAK,IAAI,EAAQ,EAAO,MAAQ,EAAG,EAAQ,EAAO,MAAQ,EAAO,OAAQ,GAAS,EAChF,GAAI,EAAO,KAAW,EACpB,MAAO,GAGX,MAAO,EACT,EAEM,EAAmC,CAAC,EAC1C,IAAK,GAAM,CAAC,EAAgB,KAAY,EAAiB,QAAQ,EAC/D,IAAK,GAAM,CAAC,EAAO,KAAQ,EAAQ,kBAAkB,QAAQ,EAC3D,IAAK,GAAM,CAAC,EAAQ,KAAc,EAAI,QAAQ,EAAG,CAC/C,GAAI,CAAC,EAAQ,EAAW,CAAM,GAAK,CAAC,EAAsB,CAAE,iBAAgB,QAAO,QAAO,CAAC,EACzF,SAIF,IAAM,EAAgB,EAAQ,kBAAkB,EAAM,GAAG,EAAS,GAC5D,EAAe,EAAQ,kBAAkB,EAAQ,EAAE,GAAG,EAAS,GACjE,EAAQ,EAAe,EAAS,CAAC,GAAK,EAAQ,EAAc,EAAS,CAAC,GAG1E,EAAe,KAAK,CAAE,iBAAgB,QAAO,QAAO,CAAC,CACvD,CASJ,IAAM,EAAU,IAAI,IAAI,EAAe,IAAI,CAAQ,CAAC,EAChD,EAAW,EACf,KAAO,EAAS,OAAS,GAAG,CAC1B,IAAM,EAA4B,CAAC,EACnC,IAAK,IAAM,KAAU,EAAU,CAC7B,IAAM,EAAa,EAAW,EAAO,gBAC/B,EAAQ,IAAa,EAAO,OAC5B,EAAO,IAAa,EAAO,MAAQ,EAAO,OAAS,GACnD,EAAU,EAAU,EAAO,cAAc,EAC/C,GAAI,CAAC,GAAS,CAAC,GAAQ,CAAC,EACtB,SAEF,IAAM,EAAc,KAAK,EAAe,EAAQ,OAAQ,EAAQ,mBAAoB,EAAM,gBAAiB,EAAK,aAAa,IAC7H,EAAW,KAAK,CACd,UAAW,EAAY,EAAa,EAAM,gBAAiB,EAAK,cAAe,EAAO,CAAI,EAC1F,aAAc,EAAwB,EAAO,iBAAmB,CAClE,CAAC,EACD,EAAQ,KAAK,CAAM,CACrB,CACA,EAAW,CAAC,EACZ,IAAK,IAAM,KAAU,EACnB,IAAK,IAAM,IAAS,CAAC,EAAO,MAAO,EAAO,MAAQ,CAAC,EAAG,CACpD,IAAM,EAAY,CAAE,eAAgB,EAAO,eAAgB,QAAO,OAAQ,EAAO,OAAS,CAAE,EACtF,EAAe,EAAiB,EAAO,eAAe,EAAE,kBAAkB,EAAM,GAAG,EAAU,QAEjG,GAAQ,IAAI,EAAS,CAAS,CAAC,GAC9B,EAAQ,EAAc,EAAU,MAAM,GACtC,EAAsB,CAAS,IAIlC,EAAQ,IAAI,EAAS,CAAS,CAAC,EAC/B,EAAS,KAAK,CAAS,EACzB,CAEJ,CACA,OAAO,CACT,CAEA,SAAS,EAAS,EAAgC,CAChD,MAAO,GAAG,EAAO,eAAe,GAAG,EAAO,MAAM,GAAG,EAAO,QAC5D,CASA,SAAS,EAA0B,EAAiB,EAA0B,EAAqC,CACjH,IAAM,EAAkB,EAAW,IAAK,GACtC,EAAgB,EAAQ,EAAU,gBAAiB,EAAU,aAAa,CAC5E,EACM,EAA8C,CAAC,EACrD,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAW,OAAQ,GAAS,EAAG,CACzD,IAAM,EAA8B,CAAC,EACjC,EAAO,KACP,EAAa,EACX,EAAS,KAAK,IAAI,EAAW,OAAQ,EAAQ,GAAyB,EAC5E,IAAK,IAAI,EAAM,EAAO,EAAM,EAAQ,GAAO,EAAG,CAC5C,IAAM,EAAY,EAAW,GACvB,EAAgB,EAAgB,GACtC,GAAI,CAAC,GAAa,IAAkB,IAAA,GAClC,MAEF,EAAO,EAAc,EAAM,CAAa,EACxC,GAAc,EAAU,cAAgB,EAAU,gBAClD,IAAM,EAAiB,EAAM,EAAQ,EACrC,EAAI,GACF,GAAkB,GAA6B,GAAc,EACzD,EAAc,EAAM,CAAc,EAClC,IAAA,EACR,CACA,EAAkB,KAAK,CAAG,CAC5B,CACA,MAAO,CAAE,oBAAmB,iBAAgB,CAC9C,CAOA,SAAgB,EAAU,EAAe,EAAuB,CAC9D,OAAO,EAAuB,CAAC,CAAC,CAAC,CAAC,CACpC,CAMA,SAAgB,EAAuB,EAA0C,CAC/E,IAAM,EAAa,EAAE,OAAS,KAAQ,EAChC,EAAgB,IAAI,IAC1B,IAAK,GAAM,CAAC,EAAO,KAAW,EAAE,QAAQ,EAAG,CACzC,IAAI,EAAO,EAAc,IAAI,CAAM,EAC9B,IACH,EAAO,IAAI,YAAY,CAAS,EAChC,EAAc,IAAI,EAAQ,CAAI,GAEhC,IAAM,EAAO,IAAU,EAIvB,EAAK,IAAS,EAAK,IAAS,GAAM,IAAM,EAAQ,GAClD,CAEA,IAAM,EAAI,IAAI,YAAY,CAAS,EACnC,MAAQ,IAAM,CACZ,EAAE,KAAK,CAAC,EACR,IAAK,IAAM,KAAU,EAAG,CACtB,IAAM,EAAY,EAAc,IAAI,CAAM,EAEtC,EAAa,EACb,EAAS,EACb,IAAK,IAAI,EAAO,EAAG,EAAO,EAAW,GAAQ,EAAG,CAC9C,IAAM,EAAW,EAAE,IAAS,EAEtB,IAAM,IAAY,IAAS,GAAK,KAAc,EAE9C,GAAY,GAAY,EAAK,KAAgB,EACnD,EAAa,IAAa,GAC1B,IAAM,EAAa,EAAI,EAAU,EACjC,EAAS,IAAa,GAEtB,EAAE,GAAQ,EAAI,CAAC,CACjB,CACF,CAEA,IAAI,EAAS,EACb,IAAK,IAAM,KAAQ,EACjB,GAAU,EAAS,CAAI,EAEzB,OAAO,CACT,CACF,CAEA,SAAS,EAAS,EAAuB,CACvC,IAAI,EAAQ,GAAU,IAAU,EAAK,YAErC,MADA,IAAS,EAAQ,YAAmB,IAAU,EAAK,WAC3C,KAAK,KAAM,GAAS,IAAU,GAAM,UAAe,QAAa,IAAM,GAAM,GACtF,CAEA,SAAS,EACP,EACA,EACA,EACA,EACA,EACoB,CACpB,MAAO,CACL,cACA,WAAY,EAAgB,EAC5B,kBACA,gBACA,WAAY,EAAM,WAClB,SAAU,EAAK,SACf,UAAW,EAAM,UACjB,QAAS,EAAK,OAChB,CACF,CAGA,MAAM,EAAkC,CAAC,EACnC,EAAmC,CAAC,EAE1C,SAAS,EAAoB,EAAuB,CAClD,IAAI,EAAO,EAAsB,GAKjC,OAJI,IAAS,IAAA,KACX,EAAO,EAAS,IAAI,GAAO,EAC3B,EAAsB,GAAS,GAE1B,CACT,CAEA,SAAS,EAAqB,EAAuB,CACnD,IAAI,EAAO,EAAuB,GAKlC,OAJI,IAAS,IAAA,KACX,EAAO,EAAU,IAAI,GAAO,EAC5B,EAAuB,GAAS,GAE3B,CACT,CAUA,SAAS,EACP,EACA,EACA,EACA,EACQ,CAGR,GAAM,CAAC,EAAS,GAAa,EAAoB,EAAQ,EAAiB,EADrD,GADC,EAAmB,IAAkB,IAAM,EAAmB,IAAoB,GACtD,EAAgB,CACkC,CAAC,EACrG,MAAO,GAAG,EAAQ,GAAG,EAAU,GAAG,EAAgB,GACpD,CASA,SAAS,EAAgB,EAAiB,EAAyB,EAA+B,CAChG,GAAM,CAAC,EAAS,GAAa,EAAoB,EAAQ,EAAiB,EAAe,EAAK,EAE9F,OAAO,EAAU,KAAK,KAAK,EAAW,EAAE,CAC1C,CAEA,SAAS,EACP,EACA,EACA,EACA,EACkB,CAClB,IAAM,EAAoB,IAAI,IAC1B,EAAU,KACV,EAAY,MAChB,IAAK,IAAI,EAAQ,EAAiB,EAAQ,EAAe,GAAS,EAAG,CACnE,IAAM,EAAQ,EAAO,GACrB,GAAI,CAAC,EACH,SAIF,IAAI,EACA,EACJ,GAAI,EAAM,OAAS,KAAM,CACvB,IAAI,EAAkB,EAAkB,IAAI,EAAM,IAAI,EAClD,IAAoB,IAAA,KACtB,EAAkB,EAAkB,KACpC,EAAkB,IAAI,EAAM,KAAM,CAAe,GAEnD,EAAO,EAAoB,CAAe,EAC1C,EAAQ,EAAqB,CAAe,CAC9C,KACE,GAAO,EAAM,SACb,EAAQ,EAAM,UAGhB,EAAW,KAAK,KAAK,EAAS,EAAE,EAAI,EAAQ,EAC5C,EAAY,KAAK,KAAK,EAAW,EAAE,EAAI,EACnC,GAAqB,EAAM,cAAgB,IAAA,IAAa,EAAM,eAAiB,IAAA,KAEjF,EAAW,KAAK,KAAK,EAAS,EAAE,EAAI,EAAM,YAAe,EACzD,EAAY,KAAK,KAAK,EAAW,EAAE,EAAI,EAAM,aAEjD,CACA,MAAO,CAAC,EAAS,CAAS,CAC5B,CAGA,SAAS,EAAS,EAAsB,CACtC,IAAI,EAAO,KACX,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAK,OAAQ,GAAS,EAEhD,EAAO,KAAK,KAAK,EAAM,EAAE,EAAI,EAAK,WAAW,CAAK,EAEpD,OAAO,CACT,CAGA,SAAS,EAAU,EAAsB,CACvC,IAAI,EAAO,YACX,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAK,OAAQ,GAAS,EAEhD,EAAO,KAAK,KAAK,EAAO,EAAK,WAAW,CAAK,EAAG,QAAU,EAE5D,OAAO,CACT,CAEA,SAAS,EAAc,EAAc,EAAuB,CAC1D,OAAO,KAAK,KAAK,EAAM,EAAE,EAAI,CAC/B,CAoBA,SAAgB,EACd,EACA,EACA,MAAmD,GAC5C,CACP,GAAI,GAAgB,GAAK,EAAO,OAAS,EACvC,OAAO,EAGT,EAAO,KAAK,CAAa,EACzB,IAAK,IAAI,EAAU,GAAM,GAAU,CACjC,EAAU,GACV,IAAM,EAAkB,EAA2B,EAAQ,CAAY,EACvE,IAAK,IAAI,EAAY,EAAG,EAAY,EAAO,QAAU,CAAC,EAAS,GAAa,EAC1E,IAAK,IAAM,KAAc,EAAgB,IAAc,CAAC,EAAG,CACzD,IAAM,EAAO,EAAO,GACd,EAAQ,EAAO,GACrB,GAAI,CAAC,GAAQ,CAAC,EACZ,SAEF,IAAM,EAAU,EAAY,EAAM,EAAO,EAAc,CAAiB,EAClE,EAAS,GAAW,EAAY,EAAO,EAAM,EAAc,CAAiB,EAClF,GAAI,CAAC,EACH,SAEF,IAAM,EAAe,EAAU,EAAO,cAAgB,EAAO,eACvD,EAAgB,EAAU,EAAO,eAAiB,EAAO,cAC3D,GAAgB,GAClB,EAAO,GAAa,EAAO,OAC3B,EAAO,OAAO,EAAY,CAAC,GAClB,EACT,EAAO,GAAc,EAAO,OACnB,EACT,EAAO,GAAa,EAAO,OAI3B,EAAO,KAAK,EAAO,MAAM,EAM3B,IAAK,IAAM,KAAc,EAAO,eAC9B,EAAW,qBAAuB,GAEpC,EAAO,KAAK,CAAa,EACzB,EAAU,GACV,KACF,CAEJ,CACA,OAAO,CACT,CAQA,SAAS,EAA2B,EAA+B,EAAkC,CACnG,IAAM,EAA4D,CAAC,EACnE,IAAK,GAAM,CAAC,EAAY,KAAU,EAAO,QAAQ,EAC/C,IAAK,IAAM,KAAc,EACvB,EAAO,KAAK,CAAE,gBAAiB,EAAW,gBAAiB,YAAW,CAAC,EAG3E,EAAO,MAAM,EAAM,IAAU,EAAK,gBAAkB,EAAM,eAAe,EACzE,IAAM,EAAW,EAAO,QAAU,IAAI,GAAa,EACnD,IAAK,GAAM,CAAC,EAAY,KAAU,EAAO,QAAQ,EAC/C,IAAK,IAAM,KAAc,EACvB,IACE,IAAI,EAAQ,EAAkB,EAAQ,EAAW,aAAa,EAC9D,EAAQ,EAAO,SAAW,EAAO,EAAM,EAAE,iBAAmB,IAAM,EAAW,cAAgB,EAC7F,GAAS,EACT,CACA,IAAM,EAAQ,EAAO,EAAM,EAAE,YAAc,EACvC,IAAU,GACZ,EAAS,KAAK,IAAI,EAAO,CAAU,EAAE,EAAE,IAAI,KAAK,IAAI,EAAO,CAAU,CAAC,CAE1E,CAGJ,OAAO,EAAS,IAAK,GAAQ,CAAC,GAAG,CAAG,CAAC,CAAC,UAAU,EAAM,IAAU,EAAO,CAAK,CAAC,CAC/E,CAEA,SAAS,EAAkB,EAAuC,EAAwB,CACxF,IAAI,EAAM,EACN,EAAO,EAAO,OAClB,KAAO,EAAM,GAAM,CACjB,IAAM,EAAU,EAAM,IAAU,GAC3B,EAAO,EAAO,EAAE,iBAAmB,GAAK,EAC3C,EAAM,EAAS,EAEf,EAAO,CAEX,CACA,OAAO,CACT,CAEA,SAAS,EAAc,EAA2B,EAAoC,CACpF,IAAM,EAAY,EAAK,GACjB,EAAa,EAAM,GACzB,OACG,GAAW,iBAAmB,IAAM,GAAY,iBAAmB,KACnE,GAAW,eAAiB,IAAM,GAAY,eAAiB,EAEpE,CAsBA,SAAS,EACP,EACA,EACA,EACA,EAC4B,CAQ5B,GAAM,CAAC,EAAU,GAAe,EAAoB,CAAK,EACnD,CAAC,EAAW,GAAgB,EAAoB,CAAM,EACtD,EAAkB,CAAC,EACrB,EAAe,EACf,EAAsB,GAC1B,IAAK,IAAM,KAAY,EAAW,CAEhC,KAAO,EAAe,EAAS,QAAQ,CACrC,IAAM,EAAU,EAAS,GACzB,GAAI,GAAW,EAAQ,cAAgB,EAAe,EAAS,gBAC7D,GAAgB,OAEhB,KAEJ,CACA,IAAM,EAAU,EAAS,GAEvB,GACA,EAAQ,eAAiB,EAAS,iBAClC,EAAQ,iBAAmB,IAE3B,EAAM,KAAK,CAAC,EAAS,CAAQ,CAAC,EAC9B,EAAsB,EAAS,cAC/B,GAAgB,EAEpB,CACA,IAAM,EAAmB,EAAM,SAAW,EACpC,EAAoB,EAAM,SAAW,EAC3C,GAAI,EAAM,OAAS,GAAM,CAAC,GAAoB,CAAC,EAC7C,OAKF,IAAM,EAAgB,GAAoB,CAAC,EAAM,KAAM,GAAe,EAAW,mBAAmB,EAC9F,EAAiB,GAAqB,CAAC,EAAO,KAAM,GAAe,EAAW,mBAAmB,EACjG,EAAS,EAAM,KAAK,CAAC,EAAS,MAAe,CACjD,GAAG,EAEH,qBAAsB,IAAA,GACtB,oBAAqB,IAAA,GACrB,SAAU,CAAC,GAAG,EAAQ,SAAU,GAAG,EAAS,QAAQ,EACpD,WAAY,EAAQ,WAAa,EAAS,WAC1C,cAAe,EAAS,cACxB,SAAU,EAAS,SACnB,QAAS,EAAS,OACpB,EAAE,EAIG,KAAkB,CAAM,EAO7B,MAAO,CAAE,SAAQ,gBAAe,iBAAgB,eAAA,CAH9C,GAAI,EAAgB,CAAC,EAAI,EAAM,KAAK,CAAC,KAAa,CAAO,EACzD,GAAI,EAAiB,CAAC,EAAI,EAAM,KAAK,EAAG,KAAc,CAAQ,CAEH,CAAE,CACjE,CAGA,SAAS,EAAiD,EAA2B,CACnF,IAAM,EAAgB,CAAC,EACnB,EAAS,EACb,IAAK,IAAM,KAAc,EAClB,EAAW,sBACd,GAAU,GAIR,CAAC,EAAW,sBAAwB,CAAC,EAAW,qBAClD,EAAS,KAAK,CAAU,EAG5B,MAAO,CAAC,EAAU,CAAM,CAC1B,CAcA,SAAgB,EAAwB,EAAoC,CAC1E,IAAI,EAAgB,EAChB,EAAmB,EACnB,EAAsB,GAC1B,IAAK,IAAM,KAAc,EAAO,CAC9B,GAAI,EAAW,qBAAsB,CACnC,EAAsB,GACtB,QACF,CACA,GAAiB,EAAW,SAAS,OACrC,EAAmB,KAAK,IAAI,EAAkB,EAAW,SAAS,MAAM,CAC1E,CACA,OAAO,EAAsB,EAAgB,EAAgB,CAC/D,CAGA,SAAgB,EACd,EACA,EACA,EACA,EACM,CACN,IAAK,IAAI,EAAQ,EAAQ,gBAAiB,EAAQ,EAAQ,cAAe,GAAS,EAAG,CACnF,IAAM,EAAQ,EAAO,GACrB,IAAK,IAAI,EAAM,GAAO,UAAY,EAAG,IAAQ,GAAO,QAAU,IAAK,GAAO,GACpE,CAAC,GAAmB,EAAgB,IAAI,EAAM,CAAC,IACjD,EAAgB,IAAI,EAAM,CAAC,CAGjC,CACF"}

package/dist/duplication.d.ts

@@ -109,6 +109,12 @@ export interface CrossFileDuplicationFileData {
     candidates: CrossFileDuplicateCandidate[];
     tokens: Token[];
     containerStatements: TokenRange[][];
+    /**
+     * The mutually disjoint blocks (at least `minTokens` long, not literal-dense) compared for
+     * cross-file near-miss (Type-3) clones. Optional for backward compatibility; without it, the
+     * file takes part in exact and gapped cross-file matching only.
+     */
+    nearMissBlocks?: TokenRange[];
     /**
      * 1-based lines that are neither blank nor comment-only, so cross-file line coverage counts only
      * code lines (blank rows inside multi-row tokens such as template literals carry no content).
@@ -147,6 +153,11 @@ export declare function collectSequenceWindowCandidates(contexts: SequenceWindow
  * `v = x & ~(x - ((v << 1) | 1))` over multi-word bit vectors; the set bits of `v` count the LCS.
  */
 export declare function lcsLength(a: Int32Array, b: Int32Array): number;
+/**
+ * lcsLength against a fixed `a`, whose per-symbol position masks are built once and reused for
+ * every `b` compared with it.
+ */
+export declare function createLcsLengthCounter(a: Int32Array): (b: Int32Array) => number;
 /**
  * Merges duplicate groups separated by a small token gap into one gapped (Type-3) clone group: a
  * copy edited in one spot splits into two exact groups whose occurrences sit side by side in the

package/dist/duplication.js

@@ -1,2 +1,2 @@
-const e={minTokens:40,maxGapTokens:30,minSimilarityPercent:70};function t(t){return{minTokens:n(t?.minTokens,e.minTokens),maxGapTokens:n(t?.maxGapTokens,e.maxGapTokens),minSimilarityPercent:n(t?.minSimilarityPercent,e.minSimilarityPercent)}}function n(e,t){return e===void 0||Number.isNaN(e)?t:Math.min(Math.max(Math.trunc(e),0),4294967295)}function r(e,t){return e*5>=t}function i(e){let t=new Int32Array(e.length+1);for(let[n,r]of e.entries())t[n+1]=(t[n]??0)+(r.literalHash===void 0?0:1);return t}function a(e,t,n){let r=[],i=[],a=[];for(let[t,n]of e.entries())for(let e of n.containers)i.push(t),a.push(e);let c=t=>e[i[t]??0],l=new Map,d=a.map((e,n)=>s(c(n)?.tokens??[],e,t));for(let[e,t]of d.entries()){let n=i[e]??0;for(let[r,i]of t.windowKeysByStart.entries())for(let t of i){if(t===void 0)continue;let i=l.get(t);i?(i.count+=1,i.containerIndex!==e&&(i.containerIndex=-1),i.contextIndex!==n&&(i.contextIndex=-1),i.minStart=Math.min(i.minStart,r),i.maxStart=Math.max(i.maxStart,r)):l.set(t,{count:1,containerIndex:e,contextIndex:n,minStart:r,maxStart:r})}}let f=(e,t)=>{if(e===void 0)return!1;let r=l.get(e);return r===void 0||r.count<2?!1:n?r.contextIndex===-1:r.containerIndex===-1||r.maxStart-r.minStart>=t},p=e=>{let t=d[e.containerIndex]?.statementHashes??[],n=t[e.start];for(let r=e.start+1;r<e.start+e.length;r+=1)if(t[r]!==n)return!0;return!1},m=[];for(let[e,t]of d.entries())for(let[n,r]of t.windowKeysByStart.entries())for(let[i,a]of r.entries()){if(!f(a,i)||!p({containerIndex:e,start:n,length:i}))continue;let r=t.windowKeysByStart[n]?.[i+1],o=t.windowKeysByStart[n-1]?.[i+1];f(r,i+1)||f(o,i+1)||m.push({containerIndex:e,start:n,length:i})}let g=new Set(m.map(o)),_=m;for(;_.length>0;){let e=[];for(let t of _){let n=a[t.containerIndex],o=n?.[t.start],s=n?.[t.start+t.length-1],l=c(t.containerIndex);if(!o||!s||!l)continue;let d=`s:${h(l.tokens,l.literalCountPrefix,o.startTokenIndex,s.endTokenIndex)}`;r.push({candidate:u(d,o.startTokenIndex,s.endTokenIndex,o,s),contextIndex:i[t.containerIndex]??0}),e.push(t)}_=[];for(let t of e)for(let e of[t.start,t.start+1]){let n={containerIndex:t.containerIndex,start:e,length:t.length-1},r=d[t.containerIndex]?.windowKeysByStart[e]?.[n.length];!g.has(o(n))&&f(r,n.length)&&p(n)&&(g.add(o(n)),_.push(n))}}return r}function o(e){return`${e.containerIndex}:${e.start}:${e.length}`}function s(e,t,n){let r=t.map(t=>g(e,t.startTokenIndex,t.endTokenIndex)),i=[];for(let e=0;e<t.length;e+=1){let a=[],o=5381,s=0,c=Math.min(t.length,e+100);for(let i=e;i<c;i+=1){let c=t[i],l=r[i];if(!c||l===void 0)break;o=b(o,l),s+=c.endTokenIndex-c.startTokenIndex;let u=i-e+1;a[u]=u>=2&&s>=n?b(o,u):void 0}i.push(a)}return{windowKeysByStart:i,statementHashes:r}}function c(e,t){let n=e.length+31>>>5,r=new Map;for(let[t,i]of e.entries()){let e=r.get(i);e||(e=new Uint32Array(n),r.set(i,e));let a=t>>>5;e[a]=(e[a]??0)|1<<(t&31)}let i=new Uint32Array(n);for(let e of t){let t=r.get(e),a=1,o=0;for(let e=0;e<n;e+=1){let n=i[e]??0,r=((t?.[e]??0)|n)>>>0,s=(n<<1|a)>>>0;a=n>>>31;let c=r-s-o;o=+(c<0),i[e]=r&~c}}let a=0;for(let e of i)a+=l(e);return a}function l(e){let t=e-(e>>>1&1431655765);return t=(t&858993459)+(t>>>2&858993459),Math.imul(t+(t>>>4)&252645135,16843009)>>>24&255}function u(e,t,n,r,i){return{fingerprint:e,tokenCount:n-t,startTokenIndex:t,endTokenIndex:n,startIndex:r.startIndex,endIndex:i.endIndex,startLine:r.startLine,endLine:i.endLine}}const d=[],f=[];function p(e){let t=d[e];return t===void 0&&(t=v(`$${e}`),d[e]=t),t}function m(e){let t=f[e];return t===void 0&&(t=y(`$${e}`),f[e]=t),t}function h(e,t,n,i){let[a,o]=_(e,n,i,r((t[i]??0)-(t[n]??0),i-n));return`${a}:${o}:${i-n}`}function g(e,t,n){let[r,i]=_(e,t,n,!1);return r^Math.imul(i,31)}function _(e,t,n,r){let i=new Map,a=5381,o=52711;for(let s=t;s<n;s+=1){let t=e[s];if(!t)continue;let n,c;if(t.kind===`id`){let e=i.get(t.text);e===void 0&&(e=i.size,i.set(t.text,e)),n=p(e),c=m(e)}else n=t.textHash,c=t.textHash2;a=Math.imul(a,31)+n|0,o=Math.imul(o,37)^c,r&&t.literalHash!==void 0&&t.literalHash2!==void 0&&(a=Math.imul(a,31)+t.literalHash|0,o=Math.imul(o,37)^t.literalHash2)}return[a,o]}function v(e){let t=5381;for(let n=0;n<e.length;n+=1)t=Math.imul(t,33)^e.charCodeAt(n);return t}function y(e){let t=-2128831035;for(let n=0;n<e.length;n+=1)t=Math.imul(t^e.charCodeAt(n),16777619);return t}function b(e,t){return Math.imul(e,31)+t}function x(e,t,n=()=>!0){if(t<=0||e.length<2)return e;e.sort(S);for(let r=!0;r;){r=!1;for(let i=0;i<e.length&&!r;i+=1)for(let a=i+1;a<e.length;a+=1){let o=e[i],s=e[a];if(!o||!s)continue;let c=C(o,s,t,n),l=c??C(s,o,t,n);if(!l)continue;let u=c?l.firstReplaced:l.secondReplaced,d=c?l.secondReplaced:l.firstReplaced;u&&d?(e[i]=l.merged,e.splice(a,1)):d?e[a]=l.merged:u?e[i]=l.merged:e.push(l.merged);for(let e of l.pairedRetained)e.spanCountedElsewhere=!0;e.sort(S),r=!0;break}}return e}function S(e,t){let n=e[0],r=t[0];return(n?.startTokenIndex??0)-(r?.startTokenIndex??0)||(n?.endTokenIndex??0)-(r?.endTokenIndex??0)}function C(e,t,n,r){let[i,a]=w(e),[o,s]=w(t),c=[],l=0,u=-1;for(let e of o){for(;l<i.length;){let t=i[l];if(t&&t.endTokenIndex+n<e.startTokenIndex)l+=1;else break}let t=i[l];t&&t.endTokenIndex<=e.startTokenIndex&&t.startTokenIndex>=u&&(c.push([t,e]),u=e.endTokenIndex,l+=1)}let d=c.length===a,f=c.length===s;if(c.length<2||!d&&!f)return;let p=d&&!e.some(e=>e.nestedInLargerGroup),m=f&&!t.some(e=>e.nestedInLargerGroup),h=c.map(([e,t])=>({...e,spanCountedElsewhere:void 0,nestedInLargerGroup:void 0,segments:[...e.segments,...t.segments],tokenCount:e.tokenCount+t.tokenCount,endTokenIndex:t.endTokenIndex,endIndex:t.endIndex,endLine:t.endLine}));if(r(h))return{merged:h,firstReplaced:p,secondReplaced:m,pairedRetained:[...p?[]:c.map(([e])=>e),...m?[]:c.map(([,e])=>e)]}}function w(e){let t=[],n=0;for(let r of e)r.nestedInLargerGroup||(n+=1),!r.spanCountedElsewhere&&!r.nestedInLargerGroup&&t.push(r);return[t,n]}function T(e){let t=0,n=0,r=!1;for(let i of e){if(i.spanCountedElsewhere){r=!0;continue}t+=i.segments.length,n=Math.max(n,i.segments.length)}return r?t:t-n}function E(e,t,n,r){for(let i=e.startTokenIndex;i<e.endTokenIndex;i+=1){let e=t[i];for(let t=e?.startRow??0;t<=(e?.endRow??-1);t+=1)(!n||n.has(t+1))&&r.add(t+1)}}export{i as buildLiteralCountPrefix,E as collectSegmentLines,a as collectSequenceWindowCandidates,T as countRedundantFragments,e as defaultDuplicationOptions,c as lcsLength,x as mergeAdjacentGroups,t as resolveDuplicationOptions};
+const e={minTokens:40,maxGapTokens:30,minSimilarityPercent:70};function t(t){return{minTokens:n(t?.minTokens,e.minTokens),maxGapTokens:n(t?.maxGapTokens,e.maxGapTokens),minSimilarityPercent:n(t?.minSimilarityPercent,e.minSimilarityPercent)}}function n(e,t){return e===void 0||Number.isNaN(e)?t:Math.min(Math.max(Math.trunc(e),0),4294967295)}function r(e,t){return e*5>=t}function i(e){let t=new Int32Array(e.length+1);for(let[n,r]of e.entries())t[n+1]=(t[n]??0)+(r.literalHash===void 0?0:1);return t}function a(e,t,n){let r=[],i=[],a=[];for(let[t,n]of e.entries())for(let e of n.containers)i.push(t),a.push(e);let c=t=>e[i[t]??0],l=new Map,u=a.map((e,n)=>s(c(n)?.tokens??[],e,t));for(let[e,t]of u.entries()){let n=i[e]??0;for(let[r,i]of t.windowKeysByStart.entries())for(let t of i){if(t===void 0)continue;let i=l.get(t);i?(i.count+=1,i.containerIndex!==e&&(i.containerIndex=-1),i.contextIndex!==n&&(i.contextIndex=-1),i.minStart=Math.min(i.minStart,r),i.maxStart=Math.max(i.maxStart,r)):l.set(t,{count:1,containerIndex:e,contextIndex:n,minStart:r,maxStart:r})}}let f=(e,t)=>{if(e===void 0)return!1;let r=l.get(e);return r===void 0||r.count<2?!1:n?r.contextIndex===-1:r.containerIndex===-1||r.maxStart-r.minStart>=t},p=e=>{let t=u[e.containerIndex]?.statementHashes??[],n=t[e.start];for(let r=e.start+1;r<e.start+e.length;r+=1)if(t[r]!==n)return!0;return!1},m=[];for(let[e,t]of u.entries())for(let[n,r]of t.windowKeysByStart.entries())for(let[i,a]of r.entries()){if(!f(a,i)||!p({containerIndex:e,start:n,length:i}))continue;let r=t.windowKeysByStart[n]?.[i+1],o=t.windowKeysByStart[n-1]?.[i+1];f(r,i+1)||f(o,i+1)||m.push({containerIndex:e,start:n,length:i})}let h=new Set(m.map(o)),_=m;for(;_.length>0;){let e=[];for(let t of _){let n=a[t.containerIndex],o=n?.[t.start],s=n?.[t.start+t.length-1],l=c(t.containerIndex);if(!o||!s||!l)continue;let u=`s:${g(l.tokens,l.literalCountPrefix,o.startTokenIndex,s.endTokenIndex)}`;r.push({candidate:d(u,o.startTokenIndex,s.endTokenIndex,o,s),contextIndex:i[t.containerIndex]??0}),e.push(t)}_=[];for(let t of e)for(let e of[t.start,t.start+1]){let n={containerIndex:t.containerIndex,start:e,length:t.length-1},r=u[t.containerIndex]?.windowKeysByStart[e]?.[n.length];!h.has(o(n))&&f(r,n.length)&&p(n)&&(h.add(o(n)),_.push(n))}}return r}function o(e){return`${e.containerIndex}:${e.start}:${e.length}`}function s(e,t,n){let r=t.map(t=>_(e,t.startTokenIndex,t.endTokenIndex)),i=[];for(let e=0;e<t.length;e+=1){let a=[],o=5381,s=0,c=Math.min(t.length,e+100);for(let i=e;i<c;i+=1){let c=t[i],l=r[i];if(!c||l===void 0)break;o=x(o,l),s+=c.endTokenIndex-c.startTokenIndex;let u=i-e+1;a[u]=u>=2&&s>=n?x(o,u):void 0}i.push(a)}return{windowKeysByStart:i,statementHashes:r}}function c(e,t){return l(e)(t)}function l(e){let t=e.length+31>>>5,n=new Map;for(let[r,i]of e.entries()){let e=n.get(i);e||(e=new Uint32Array(t),n.set(i,e));let a=r>>>5;e[a]=(e[a]??0)|1<<(r&31)}let r=new Uint32Array(t);return e=>{r.fill(0);for(let i of e){let e=n.get(i),a=1,o=0;for(let n=0;n<t;n+=1){let t=r[n]??0,i=((e?.[n]??0)|t)>>>0,s=(t<<1|a)>>>0;a=t>>>31;let c=i-s-o;o=+(c<0),r[n]=i&~c}}let i=0;for(let e of r)i+=u(e);return i}}function u(e){let t=e-(e>>>1&1431655765);return t=(t&858993459)+(t>>>2&858993459),Math.imul(t+(t>>>4)&252645135,16843009)>>>24&255}function d(e,t,n,r,i){return{fingerprint:e,tokenCount:n-t,startTokenIndex:t,endTokenIndex:n,startIndex:r.startIndex,endIndex:i.endIndex,startLine:r.startLine,endLine:i.endLine}}const f=[],p=[];function m(e){let t=f[e];return t===void 0&&(t=y(`$${e}`),f[e]=t),t}function h(e){let t=p[e];return t===void 0&&(t=b(`$${e}`),p[e]=t),t}function g(e,t,n,i){let[a,o]=v(e,n,i,r((t[i]??0)-(t[n]??0),i-n));return`${a}:${o}:${i-n}`}function _(e,t,n){let[r,i]=v(e,t,n,!1);return r^Math.imul(i,31)}function v(e,t,n,r){let i=new Map,a=5381,o=52711;for(let s=t;s<n;s+=1){let t=e[s];if(!t)continue;let n,c;if(t.kind===`id`){let e=i.get(t.text);e===void 0&&(e=i.size,i.set(t.text,e)),n=m(e),c=h(e)}else n=t.textHash,c=t.textHash2;a=Math.imul(a,31)+n|0,o=Math.imul(o,37)^c,r&&t.literalHash!==void 0&&t.literalHash2!==void 0&&(a=Math.imul(a,31)+t.literalHash|0,o=Math.imul(o,37)^t.literalHash2)}return[a,o]}function y(e){let t=5381;for(let n=0;n<e.length;n+=1)t=Math.imul(t,33)^e.charCodeAt(n);return t}function b(e){let t=-2128831035;for(let n=0;n<e.length;n+=1)t=Math.imul(t^e.charCodeAt(n),16777619);return t}function x(e,t){return Math.imul(e,31)+t}function S(e,t,n=()=>!0){if(t<=0||e.length<2)return e;e.sort(T);for(let r=!0;r;){r=!1;let i=C(e,t);for(let a=0;a<e.length&&!r;a+=1)for(let o of i[a]??[]){let i=e[a],s=e[o];if(!i||!s)continue;let c=E(i,s,t,n),l=c??E(s,i,t,n);if(!l)continue;let u=c?l.firstReplaced:l.secondReplaced,d=c?l.secondReplaced:l.firstReplaced;u&&d?(e[a]=l.merged,e.splice(o,1)):d?e[o]=l.merged:u?e[a]=l.merged:e.push(l.merged);for(let e of l.pairedRetained)e.spanCountedElsewhere=!0;e.sort(T),r=!0;break}}return e}function C(e,t){let n=[];for(let[t,r]of e.entries())for(let e of r)n.push({startTokenIndex:e.startTokenIndex,groupIndex:t});n.sort((e,t)=>e.startTokenIndex-t.startTokenIndex);let r=e.map(()=>new Set);for(let[i,a]of e.entries())for(let e of a)for(let a=w(n,e.endTokenIndex);a<n.length&&(n[a]?.startTokenIndex??0)<=e.endTokenIndex+t;a+=1){let e=n[a]?.groupIndex??i;e!==i&&r[Math.min(e,i)]?.add(Math.max(e,i))}return r.map(e=>[...e].toSorted((e,t)=>e-t))}function w(e,t){let n=0,r=e.length;for(;n<r;){let i=n+r>>>1;(e[i]?.startTokenIndex??0)<t?n=i+1:r=i}return n}function T(e,t){let n=e[0],r=t[0];return(n?.startTokenIndex??0)-(r?.startTokenIndex??0)||(n?.endTokenIndex??0)-(r?.endTokenIndex??0)}function E(e,t,n,r){let[i,a]=D(e),[o,s]=D(t),c=[],l=0,u=-1;for(let e of o){for(;l<i.length;){let t=i[l];if(t&&t.endTokenIndex+n<e.startTokenIndex)l+=1;else break}let t=i[l];t&&t.endTokenIndex<=e.startTokenIndex&&t.startTokenIndex>=u&&(c.push([t,e]),u=e.endTokenIndex,l+=1)}let d=c.length===a,f=c.length===s;if(c.length<2||!d&&!f)return;let p=d&&!e.some(e=>e.nestedInLargerGroup),m=f&&!t.some(e=>e.nestedInLargerGroup),h=c.map(([e,t])=>({...e,spanCountedElsewhere:void 0,nestedInLargerGroup:void 0,segments:[...e.segments,...t.segments],tokenCount:e.tokenCount+t.tokenCount,endTokenIndex:t.endTokenIndex,endIndex:t.endIndex,endLine:t.endLine}));if(r(h))return{merged:h,firstReplaced:p,secondReplaced:m,pairedRetained:[...p?[]:c.map(([e])=>e),...m?[]:c.map(([,e])=>e)]}}function D(e){let t=[],n=0;for(let r of e)r.nestedInLargerGroup||(n+=1),!r.spanCountedElsewhere&&!r.nestedInLargerGroup&&t.push(r);return[t,n]}function O(e){let t=0,n=0,r=!1;for(let i of e){if(i.spanCountedElsewhere){r=!0;continue}t+=i.segments.length,n=Math.max(n,i.segments.length)}return r?t:t-n}function k(e,t,n,r){for(let i=e.startTokenIndex;i<e.endTokenIndex;i+=1){let e=t[i];for(let t=e?.startRow??0;t<=(e?.endRow??-1);t+=1)(!n||n.has(t+1))&&r.add(t+1)}}export{i as buildLiteralCountPrefix,k as collectSegmentLines,a as collectSequenceWindowCandidates,O as countRedundantFragments,l as createLcsLengthCounter,e as defaultDuplicationOptions,c as lcsLength,S as mergeAdjacentGroups,t as resolveDuplicationOptions};
 //# sourceMappingURL=duplication.js.map

package/dist/duplication.js.map

@@ -1 +1 @@
-{"version":3,"file":"duplication.js","names":[],"sources":["../src/duplication.ts"],"sourcesContent":["import type { DuplicationOptions } from './types.js';\n\n/**\n * Project-level duplication machinery operating on normalized token streams. Tokenization itself\n * (parsing, identifier anonymization, literal normalization) happens in the Rust addon, which\n * serializes each file's Token stream and statement structure; the helpers here match statement\n * windows across files, merge gap-adjacent groups, and count duplicated lines over that data.\n */\n\nexport const defaultDuplicationOptions: Required<DuplicationOptions> = {\n  minTokens: 40,\n  maxGapTokens: 30,\n  minSimilarityPercent: 70,\n};\n\n/**\n * Fills defaults for absent settings, applying the same normalization as the native boundary's\n * clampToU32 — NaN (e.g. `Number(unsetEnvVariable)`) counts as absent, and other values truncate\n * and clamp to [0, u32::MAX] — so the TypeScript half of cross-file matching cannot diverge from\n * the natively collected candidates on such input.\n */\nexport function resolveDuplicationOptions(options?: DuplicationOptions): Required<DuplicationOptions> {\n  return {\n    minTokens: resolveOption(options?.minTokens, defaultDuplicationOptions.minTokens),\n    maxGapTokens: resolveOption(options?.maxGapTokens, defaultDuplicationOptions.maxGapTokens),\n    minSimilarityPercent: resolveOption(options?.minSimilarityPercent, defaultDuplicationOptions.minSimilarityPercent),\n  };\n}\n\nfunction resolveOption(value: number | undefined, fallback: number): number {\n  return value === undefined || Number.isNaN(value)\n    ? fallback\n    : Math.min(Math.max(Math.trunc(value), 0), 0xFF_FF_FF_FF);\n}\n\n/** Minimum consecutive statements for a statement-sequence duplicate candidate. */\nconst minSequenceStatementCount = 2;\n/**\n * Caps the window length so statement-sequence enumeration stays linear in the statement count.\n * Heterogeneous clones longer than the cap are reported as capped windows (a deliberate\n * conservative undercount trading completeness for bounded discovery cost).\n */\nconst maxSequenceStatementCount = 100;\n\n/**\n * A region whose normalized tokens are at least 20% literal values is data-like (a lookup table, a\n * constant list, a value-mapping switch), not logic: literal values re-enter its fingerprint so\n * tables that merely share their shape stop counting as copy-paste. Compared in integer math\n * (5 * literals >= total) so the TypeScript and native sides cannot disagree on the boundary.\n */\nfunction isLiteralDense(literalCount: number, tokenCount: number): boolean {\n  return literalCount * 5 >= tokenCount;\n}\n\nexport interface Token {\n  /** Normalization target: identifiers to anonymize, literal kind tags, or the raw token text. */\n  kind: 'id' | 'text';\n  text: string;\n  /**\n   * Two INDEPENDENT hashes of `text` (djb2 and FNV-1a), precomputed so fingerprinting nested\n   * regions never re-hashes a token. Feeding the same per-token hash to both fingerprint\n   * accumulators would collapse the key to 32 effective bits: one djb2 collision between two\n   * token texts would then equate whole regions.\n   */\n  textHash: number;\n  textHash2: number;\n  /** Hash pair of a value-carrying literal's value, folded into data-like region fingerprints. */\n  literalHash?: number;\n  literalHash2?: number;\n  /**\n   * True for verbatim-kept NAMES (member/callee/type names, named grammar leaves): together with\n   * value-carrying literals these are the content-bearing tokens the near-miss content gate\n   * counts. Keywords, operators, and punctuation come from unnamed nodes and stay false.\n   */\n  isName?: boolean;\n  /** 0-based source rows the token occupies, so line coverage counts only matched-token lines. */\n  startRow: number;\n  endRow: number;\n}\n\n/** A token span with its source position; plain data so cross-file matching can retain it. */\nexport interface TokenRange {\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\n/** A contiguous run of matched tokens; gapped (merged) duplicates carry several per occurrence. */\ninterface TokenSegment {\n  startTokenIndex: number;\n  endTokenIndex: number;\n}\n\ninterface DuplicateCandidate {\n  fingerprint: string;\n  tokenCount: number;\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\nexport interface CountedOccurrence {\n  /** Matched token runs; more than one once gapped groups are merged. */\n  segments: TokenSegment[];\n  /**\n   * Set on occurrences whose span another reported group already counts: a retained group's\n   * occurrences that a partial gapped merge also paired into a merged group, and cross-file copies\n   * nested inside a larger group's region. Block counting must not count them again.\n   */\n  spanCountedElsewhere?: boolean;\n  /**\n   * Set on cross-file copies nested inside a larger group's region (they also set\n   * `spanCountedElsewhere`). They never pair in gapped merging, and they do not keep their group's\n   * standalone copies from merging, which they are not copies of. They do keep the merged group\n   * from taking their group's place, since only the original group reports the nesting.\n   */\n  nestedInLargerGroup?: boolean;\n  /** Sum of segment token counts (the gap tokens are not matched content). */\n  tokenCount: number;\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\n/** A duplicate region found in one file, exported for cross-file matching by fingerprint. */\nexport interface CrossFileDuplicateCandidate {\n  /** Content key: equal fingerprints mean equal normalized token sequences (up to hash collision). */\n  fingerprint: string;\n  tokenCount: number;\n  /** Token positions within the owning file, for cross-file gapped (Type-3) merging. */\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\n/**\n * One file's contribution to cross-file clone detection: catalogued candidates plus the normalized\n * token stream and statement structure, so the project-level pass can match partial statement runs\n * (windows that a single file cannot know repeat elsewhere) and merge gap-adjacent groups.\n */\nexport interface CrossFileDuplicationFileData {\n  candidates: CrossFileDuplicateCandidate[];\n  tokens: Token[];\n  containerStatements: TokenRange[][];\n  /**\n   * 1-based lines that are neither blank nor comment-only, so cross-file line coverage counts only\n   * code lines (blank rows inside multi-row tokens such as template literals carry no content).\n   * Optional for backward compatibility; without it, every matched-token row counts.\n   */\n  codeLineNumbers?: Set<number>;\n}\n\n/** literalCountPrefix[i] = value-carrying literal tokens in tokens[0..i), for O(1) density checks. */\nexport function buildLiteralCountPrefix(tokens: Token[]): Int32Array {\n  const prefix = new Int32Array(tokens.length + 1);\n  for (const [index, token] of tokens.entries()) {\n    prefix[index + 1] = (prefix[index] ?? 0) + (token.literalHash === undefined ? 0 : 1);\n  }\n  return prefix;\n}\n\ninterface WindowOccurrences {\n  count: number;\n  /** -1 once occurrences span more than one container. */\n  containerIndex: number;\n  /** -1 once occurrences span more than one context (file). */\n  contextIndex: number;\n  minStart: number;\n  maxStart: number;\n}\n\ninterface SequenceWindow {\n  containerIndex: number;\n  start: number;\n  length: number;\n}\n\n/** One file's token stream and statement containers, as a window-matching context. */\nexport interface SequenceWindowContext {\n  tokens: Token[];\n  literalCountPrefix: Int32Array;\n  containers: TokenRange[][];\n}\n\nexport interface ContextualSequenceCandidate {\n  candidate: DuplicateCandidate;\n  contextIndex: number;\n}\n\n/**\n * Enumerates runs of consecutive sibling statements over one or more contexts (files). Every\n * container statement participates; only the window length is capped, so enumeration stays linear\n * in the statement count. Windows are grouped by a cheap rolling hash of per-statement\n * fingerprints, and only locally maximal repeated windows — those whose one-statement extensions\n * stop repeating — become candidates with an exact (window-consistent) fingerprint. Without the\n * maximality filter a degenerate file of near-identical statements would fingerprint every\n * sub-window of every repeated region. With `requireMultipleContexts` a window only counts as\n * repeated when its occurrences span at least two contexts (CPD-style cross-file matching): a\n * repeat confined to one file is that file's own concern, and emitting it here would flood the\n * project-level selection with unusable single-file groups.\n */\nexport function collectSequenceWindowCandidates(\n  contexts: SequenceWindowContext[],\n  minTokens: number,\n  requireMultipleContexts: boolean\n): ContextualSequenceCandidate[] {\n  const candidates: ContextualSequenceCandidate[] = [];\n  const contextIndexByContainer: number[] = [];\n  const containers: TokenRange[][] = [];\n  for (const [contextIndex, context] of contexts.entries()) {\n    for (const statements of context.containers) {\n      contextIndexByContainer.push(contextIndex);\n      containers.push(statements);\n    }\n  }\n  const contextAt = (containerIndex: number): SequenceWindowContext | undefined =>\n    contexts[contextIndexByContainer[containerIndex] ?? 0];\n  const occurrencesByWindowKey = new Map<number, WindowOccurrences>();\n  const containerWindows = containers.map((statements, containerIndex) =>\n    enumerateContainerWindows(contextAt(containerIndex)?.tokens ?? [], statements, minTokens)\n  );\n  for (const [containerIndex, windows] of containerWindows.entries()) {\n    const contextIndex = contextIndexByContainer[containerIndex] ?? 0;\n    for (const [start, row] of windows.windowKeysByStart.entries()) {\n      for (const windowKey of row) {\n        if (windowKey === undefined) {\n          continue;\n        }\n        const occurrences = occurrencesByWindowKey.get(windowKey);\n        if (occurrences) {\n          occurrences.count += 1;\n          if (occurrences.containerIndex !== containerIndex) {\n            occurrences.containerIndex = -1;\n          }\n          if (occurrences.contextIndex !== contextIndex) {\n            occurrences.contextIndex = -1;\n          }\n          occurrences.minStart = Math.min(occurrences.minStart, start);\n          occurrences.maxStart = Math.max(occurrences.maxStart, start);\n        } else {\n          occurrencesByWindowKey.set(windowKey, {\n            count: 1,\n            containerIndex,\n            contextIndex,\n            minStart: start,\n            maxStart: start,\n          });\n        }\n      }\n    }\n  }\n\n  // A window only \"repeats\" when two of its occurrences can coexist without overlapping: sliding\n  // matches inside a homogeneous run (start spread smaller than the window length) can never both\n  // be counted and must neither qualify a window nor dominate its sub-windows. Cross-context\n  // matching instead requires occurrences in two contexts, which coexist by construction.\n  const repeats = (windowKey: number | undefined, length: number): boolean => {\n    if (windowKey === undefined) {\n      return false;\n    }\n    const occurrences = occurrencesByWindowKey.get(windowKey);\n    if (occurrences === undefined || occurrences.count < 2) {\n      return false;\n    }\n    if (requireMultipleContexts) {\n      return occurrences.contextIndex === -1;\n    }\n    return occurrences.containerIndex === -1 || occurrences.maxStart - occurrences.minStart >= length;\n  };\n\n  // A window whose statements all share one normalized shape (sixteen `let x = 0;` declarations,\n  // a constant table) is a homogeneous preamble, not a copy-paste: requiring two distinct\n  // per-statement shapes keeps such runs out of duplicate groups and the duplication ratio.\n  const hasDistinctStatements = (window: SequenceWindow): boolean => {\n    const hashes = containerWindows[window.containerIndex]?.statementHashes ?? [];\n    const firstHash = hashes[window.start];\n    for (let index = window.start + 1; index < window.start + window.length; index += 1) {\n      if (hashes[index] !== firstHash) {\n        return true;\n      }\n    }\n    return false;\n  };\n\n  const maximalWindows: SequenceWindow[] = [];\n  for (const [containerIndex, windows] of containerWindows.entries()) {\n    for (const [start, row] of windows.windowKeysByStart.entries()) {\n      for (const [length, windowKey] of row.entries()) {\n        if (!repeats(windowKey, length) || !hasDistinctStatements({ containerIndex, start, length })) {\n          continue;\n        }\n        // Dominated windows are skipped: the one-statement extension also repeats, so a larger\n        // candidate covering this window exists.\n        const extendedRight = windows.windowKeysByStart[start]?.[length + 1];\n        const extendedLeft = windows.windowKeysByStart[start - 1]?.[length + 1];\n        if (repeats(extendedRight, length + 1) || repeats(extendedLeft, length + 1)) {\n          continue;\n        }\n        maximalWindows.push({ containerIndex, start, length });\n      }\n    }\n  }\n\n  // The rolling hash anonymizes identifiers per statement, so a window can look repeated coarsely\n  // while its exact (window-consistent) fingerprints differ, and a longer window's match can\n  // dominate sub-windows that other copies still need (three copies where only two extend one\n  // statement further). Every emitted window therefore exposes its repeating, unvisited\n  // sub-windows; `visited` bounds the worklist and lengths strictly decrease, so it terminates.\n  const visited = new Set(maximalWindows.map(windowId));\n  let frontier = maximalWindows;\n  while (frontier.length > 0) {\n    const emitted: SequenceWindow[] = [];\n    for (const window of frontier) {\n      const statements = containers[window.containerIndex];\n      const first = statements?.[window.start];\n      const last = statements?.[window.start + window.length - 1];\n      const context = contextAt(window.containerIndex);\n      if (!first || !last || !context) {\n        continue;\n      }\n      const fingerprint = `s:${fingerprintKey(context.tokens, context.literalCountPrefix, first.startTokenIndex, last.endTokenIndex)}`;\n      candidates.push({\n        candidate: toCandidate(fingerprint, first.startTokenIndex, last.endTokenIndex, first, last),\n        contextIndex: contextIndexByContainer[window.containerIndex] ?? 0,\n      });\n      emitted.push(window);\n    }\n    frontier = [];\n    for (const window of emitted) {\n      for (const start of [window.start, window.start + 1]) {\n        const subWindow = { containerIndex: window.containerIndex, start, length: window.length - 1 };\n        const subWindowKey = containerWindows[window.containerIndex]?.windowKeysByStart[start]?.[subWindow.length];\n        if (\n          visited.has(windowId(subWindow)) ||\n          !repeats(subWindowKey, subWindow.length) ||\n          !hasDistinctStatements(subWindow)\n        ) {\n          continue;\n        }\n        visited.add(windowId(subWindow));\n        frontier.push(subWindow);\n      }\n    }\n  }\n  return candidates;\n}\n\nfunction windowId(window: SequenceWindow): string {\n  return `${window.containerIndex}:${window.start}:${window.length}`;\n}\n\ninterface ContainerWindows {\n  /** windowKeysByStart[start][length] is the rolling-hash key of the window, or undefined if it is below the size thresholds. */\n  windowKeysByStart: (number | undefined)[][];\n  /** Per-statement fingerprint hashes, for the distinct-shape requirement on windows. */\n  statementHashes: number[];\n}\n\nfunction enumerateContainerWindows(tokens: Token[], statements: TokenRange[], minTokens: number): ContainerWindows {\n  const statementHashes = statements.map((statement) =>\n    fingerprintHash(tokens, statement.startTokenIndex, statement.endTokenIndex)\n  );\n  const windowKeysByStart: (number | undefined)[][] = [];\n  for (let start = 0; start < statements.length; start += 1) {\n    const row: (number | undefined)[] = [];\n    let hash = 5381;\n    let tokenCount = 0;\n    const maxEnd = Math.min(statements.length, start + maxSequenceStatementCount);\n    for (let end = start; end < maxEnd; end += 1) {\n      const statement = statements[end];\n      const statementHash = statementHashes[end];\n      if (!statement || statementHash === undefined) {\n        break;\n      }\n      hash = combineHashes(hash, statementHash);\n      tokenCount += statement.endTokenIndex - statement.startTokenIndex;\n      const statementCount = end - start + 1;\n      row[statementCount] =\n        statementCount >= minSequenceStatementCount && tokenCount >= minTokens\n          ? combineHashes(hash, statementCount)\n          : undefined;\n    }\n    windowKeysByStart.push(row);\n  }\n  return { windowKeysByStart, statementHashes };\n}\n\n/**\n * Longest-common-subsequence LENGTH of two symbol sequences via the Allison–Dix bit-parallel\n * recurrence (O(|a|/32 · |b|) words): per symbol of `b`, `x = match | v` and\n * `v = x & ~(x - ((v << 1) | 1))` over multi-word bit vectors; the set bits of `v` count the LCS.\n */\nexport function lcsLength(a: Int32Array, b: Int32Array): number {\n  const wordCount = (a.length + 31) >>> 5;\n  const positionMasks = new Map<number, Uint32Array>();\n  for (const [index, symbol] of a.entries()) {\n    let mask = positionMasks.get(symbol);\n    if (!mask) {\n      mask = new Uint32Array(wordCount);\n      positionMasks.set(symbol, mask);\n    }\n    const word = index >>> 5;\n    // The Uint32Array store wraps the signed int32 bit pattern to unsigned. The `?? 0` guards in\n    // this function are required by noUncheckedIndexedAccess (typed-array reads type as\n    // `number | undefined`), not redundancy: every index is in bounds.\n    mask[word] = (mask[word] ?? 0) | (1 << (index & 31));\n  }\n\n  const v = new Uint32Array(wordCount);\n  for (const symbol of b) {\n    const matchMask = positionMasks.get(symbol);\n    // `(v << 1) | 1` shifts a carry bit across words; subtraction borrows across words.\n    let shiftCarry = 1;\n    let borrow = 0;\n    for (let word = 0; word < wordCount; word += 1) {\n      const previous = v[word] ?? 0;\n      // oxlint-disable-next-line unicorn/prefer-math-trunc -- `>>> 0` reinterprets the signed int32 bit pattern as unsigned so the borrow subtraction below compares magnitudes; Math.trunc would keep it negative.\n      const x = ((matchMask?.[word] ?? 0) | previous) >>> 0;\n      // oxlint-disable-next-line unicorn/prefer-math-trunc -- same unsigned reinterpretation as `x`.\n      const shifted = ((previous << 1) | shiftCarry) >>> 0;\n      shiftCarry = previous >>> 31;\n      const difference = x - shifted - borrow;\n      borrow = difference < 0 ? 1 : 0;\n      // The Uint32Array store wraps the signed int32 bit pattern to unsigned.\n      v[word] = x & ~difference;\n    }\n  }\n\n  let length = 0;\n  for (const word of v) {\n    length += popCount(word);\n  }\n  return length;\n}\n\nfunction popCount(value: number): number {\n  let count = value - ((value >>> 1) & 0x55_55_55_55);\n  count = (count & 0x33_33_33_33) + ((count >>> 2) & 0x33_33_33_33);\n  return (Math.imul((count + (count >>> 4)) & 0x0F_0F_0F_0F, 0x01_01_01_01) >>> 24) & 0xFF;\n}\n\nfunction toCandidate(\n  fingerprint: string,\n  startTokenIndex: number,\n  endTokenIndex: number,\n  first: TokenRange,\n  last: TokenRange\n): DuplicateCandidate {\n  return {\n    fingerprint,\n    tokenCount: endTokenIndex - startTokenIndex,\n    startTokenIndex,\n    endTokenIndex,\n    startIndex: first.startIndex,\n    endIndex: last.endIndex,\n    startLine: first.startLine,\n    endLine: last.endLine,\n  };\n}\n\n/** Caches of hashText/hashText2 over '$0', '$1', ... so anonymized identifiers hash without allocating. */\nconst anonymizedIndexHashes: number[] = [];\nconst anonymizedIndexHashes2: number[] = [];\n\nfunction anonymizedIndexHash(index: number): number {\n  let hash = anonymizedIndexHashes[index];\n  if (hash === undefined) {\n    hash = hashText(`$${index}`);\n    anonymizedIndexHashes[index] = hash;\n  }\n  return hash;\n}\n\nfunction anonymizedIndexHash2(index: number): number {\n  let hash = anonymizedIndexHashes2[index];\n  if (hash === undefined) {\n    hash = hashText2(`$${index}`);\n    anonymizedIndexHashes2[index] = hash;\n  }\n  return hash;\n}\n\n/**\n * Content key of a token range: two independent 32-bit hashes over the normalized token sequence\n * (identifiers anonymized consistently by first-occurrence order) plus the token count. Regions\n * with equal keys are treated as equal content; a collision would need both 32-bit hashes and the\n * length to coincide, which is negligible for a metrics report. The format and arithmetic match\n * fingerprint_key in native/src/duplication.rs exactly, so window candidates fingerprinted here\n * group together with the per-file candidates the addon catalogues.\n */\nfunction fingerprintKey(\n  tokens: Token[],\n  literalCountPrefix: Int32Array,\n  startTokenIndex: number,\n  endTokenIndex: number\n): string {\n  const literalCount = (literalCountPrefix[endTokenIndex] ?? 0) - (literalCountPrefix[startTokenIndex] ?? 0);\n  const literalDense = isLiteralDense(literalCount, endTokenIndex - startTokenIndex);\n  const [primary, secondary] = fingerprintHashPair(tokens, startTokenIndex, endTokenIndex, literalDense);\n  return `${primary}:${secondary}:${endTokenIndex - startTokenIndex}`;\n}\n\n/**\n * A single 32-bit summary of a range for the coarse rolling-hash phase. Deliberately\n * density-agnostic: density is a property of the final candidate REGION, and folding literal\n * values into per-statement hashes would make a dense statement inside a logic-heavy window\n * (`const weights = [1, 2, 3];`) block the window from ever being enumerated. The coarse phase\n * over-approximates on shape alone; the exact region fingerprint still applies the density rule.\n */\nfunction fingerprintHash(tokens: Token[], startTokenIndex: number, endTokenIndex: number): number {\n  const [primary, secondary] = fingerprintHashPair(tokens, startTokenIndex, endTokenIndex, false);\n  // XOR already coerces to int32, matching the native side's i32 arithmetic.\n  return primary ^ Math.imul(secondary, 31);\n}\n\nfunction fingerprintHashPair(\n  tokens: Token[],\n  startTokenIndex: number,\n  endTokenIndex: number,\n  foldLiteralValues: boolean\n): [number, number] {\n  const indexByIdentifier = new Map<string, number>();\n  let primary = 5381;\n  let secondary = 52_711;\n  for (let index = startTokenIndex; index < endTokenIndex; index += 1) {\n    const token = tokens[index];\n    if (!token) {\n      continue;\n    }\n    // Each accumulator consumes its own independent per-token hash: sharing one would collapse\n    // the key to 32 effective bits (a single djb2 collision would equate whole regions).\n    let part: number;\n    let part2: number;\n    if (token.kind === 'id') {\n      let identifierIndex = indexByIdentifier.get(token.text);\n      if (identifierIndex === undefined) {\n        identifierIndex = indexByIdentifier.size;\n        indexByIdentifier.set(token.text, identifierIndex);\n      }\n      part = anonymizedIndexHash(identifierIndex);\n      part2 = anonymizedIndexHash2(identifierIndex);\n    } else {\n      part = token.textHash;\n      part2 = token.textHash2;\n    }\n    // oxlint-disable-next-line unicorn/prefer-math-trunc -- `| 0` wraps the sum to int32 (Math.trunc does not), which must match the native side's wrapping i32 arithmetic.\n    primary = (Math.imul(primary, 31) + part) | 0;\n    secondary = Math.imul(secondary, 37) ^ part2;\n    if (foldLiteralValues && token.literalHash !== undefined && token.literalHash2 !== undefined) {\n      // oxlint-disable-next-line unicorn/prefer-math-trunc -- `| 0` wraps the sum to int32 (Math.trunc does not), which must match the native side's wrapping i32 arithmetic.\n      primary = (Math.imul(primary, 31) + token.literalHash) | 0;\n      secondary = Math.imul(secondary, 37) ^ token.literalHash2;\n    }\n  }\n  return [primary, secondary];\n}\n\n/** djb2-style hash; XOR keeps the value in signed 32-bit range, which is fine for a grouping key. */\nfunction hashText(text: string): number {\n  let hash = 5381;\n  for (let index = 0; index < text.length; index += 1) {\n    // oxlint-disable-next-line unicorn/prefer-code-point -- djb2 hashes UTF-16 code units; codePointAt would hash surrogate pairs twice (full code point, then the lone low surrogate).\n    hash = Math.imul(hash, 33) ^ text.charCodeAt(index);\n  }\n  return hash;\n}\n\n/** FNV-1a over UTF-16 code units: independent of hashText so the two accumulators never share input. */\nfunction hashText2(text: string): number {\n  let hash = -2_128_831_035; // 2166136261 as int32 (the FNV-1a offset basis)\n  for (let index = 0; index < text.length; index += 1) {\n    // oxlint-disable-next-line unicorn/prefer-code-point -- hashes UTF-16 code units like hashText.\n    hash = Math.imul(hash ^ text.charCodeAt(index), 16_777_619);\n  }\n  return hash;\n}\n\nfunction combineHashes(hash: number, value: number): number {\n  return Math.imul(hash, 31) + value;\n}\n\n/**\n * Merges duplicate groups separated by a small token gap into one gapped (Type-3) clone group: a\n * copy edited in one spot splits into two exact groups whose occurrences sit side by side in the\n * same order. Occurrences are paired greedily in source order; a merge happens when the pairing\n * fully pairs at least one group with at least two pairs. Equal-cardinality groups whose\n * occurrences all pair merge into one group as before. When cardinalities differ (a fragment also\n * occurs standalone: prefix ×3, suffix ×2), the fully-paired group is subsumed into the merged\n * gapped group while the other group is RETAINED with ALL its occurrences: dropping the leftover\n * would lose duplicated-line coverage, and reporting it alone would make a single-occurrence group\n * (contradicting duplicateBlockGroupCount's \"appears more than once\" meaning). Line coverage\n * unions ranges, so the overlap between the retained exact group and the merged group is harmless.\n * Merging repeats to a fixpoint so a clone edited in several spots still reassembles; it\n * terminates because a merge either removes the input groups it replaces or marks their paired\n * occurrences as counted elsewhere, and only unmarked occurrences of remaining groups pair, so the\n * number of pairable occurrences strictly decreases with every merge. Gap tokens are not matched\n * content: line coverage and sizes count only the matched segments. Generic so cross-file merging\n * can thread file identity through occurrences.\n */\nexport function mergeAdjacentGroups<T extends CountedOccurrence>(\n  groups: T[][],\n  maxGapTokens: number,\n  isReportableGroup: (group: T[]) => boolean = () => true\n): T[][] {\n  if (maxGapTokens <= 0 || groups.length < 2) {\n    return groups;\n  }\n  // Deterministic processing order (mirrored by the native side): by first occurrence position.\n  groups.sort(compareGroups);\n  for (let restart = true; restart;) {\n    restart = false;\n    for (let leftIndex = 0; leftIndex < groups.length && !restart; leftIndex += 1) {\n      for (let rightIndex = leftIndex + 1; rightIndex < groups.length; rightIndex += 1) {\n        const left = groups[leftIndex];\n        const right = groups[rightIndex];\n        if (!left || !right) {\n          continue;\n        }\n        const forward = mergeGroups(left, right, maxGapTokens, isReportableGroup);\n        const result = forward ?? mergeGroups(right, left, maxGapTokens, isReportableGroup);\n        if (!result) {\n          continue;\n        }\n        const leftReplaced = forward ? result.firstReplaced : result.secondReplaced;\n        const rightReplaced = forward ? result.secondReplaced : result.firstReplaced;\n        if (leftReplaced && rightReplaced) {\n          groups[leftIndex] = result.merged;\n          groups.splice(rightIndex, 1);\n        } else if (rightReplaced) {\n          groups[rightIndex] = result.merged;\n        } else if (leftReplaced) {\n          groups[leftIndex] = result.merged;\n        } else {\n          // Both groups stay (each is only partly paired, or reports nested copies of its own), so\n          // the merged group joins them instead of taking a place.\n          groups.push(result.merged);\n        }\n        // A group that stays keeps ALL its occurrences (line coverage must not shrink, and a\n        // reported group must keep >= 2 occurrences), so its paired occurrences now also live\n        // inside the merged group's occurrences: mark them so duplicateBlockCount counts each\n        // token span once, and so the same pair cannot merge again.\n        for (const occurrence of result.pairedRetained) {\n          occurrence.spanCountedElsewhere = true;\n        }\n        groups.sort(compareGroups);\n        restart = true;\n        break;\n      }\n    }\n  }\n  return groups;\n}\n\nfunction compareGroups(left: CountedOccurrence[], right: CountedOccurrence[]): number {\n  const leftFirst = left[0];\n  const rightFirst = right[0];\n  return (\n    (leftFirst?.startTokenIndex ?? 0) - (rightFirst?.startTokenIndex ?? 0) ||\n    (leftFirst?.endTokenIndex ?? 0) - (rightFirst?.endTokenIndex ?? 0)\n  );\n}\n\ninterface MergeResult<T> {\n  merged: T[];\n  /**\n   * Whether the merged group takes the respective input group's place: its standalone occurrences\n   * were all paired and it holds no nested copies that only it can report.\n   */\n  firstReplaced: boolean;\n  secondReplaced: boolean;\n  /** The occurrences of groups that stay, which the merged group's spans now also cover. */\n  pairedRetained: T[];\n}\n\n/**\n * Pairs `second` occurrences with gap-preceding `first` occurrences, greedily in source order:\n * each trailing occurrence takes the earliest unused leading occurrence within the gap, and a\n * pair's leading must start at or after the previous pair's trailing end so merged spans never\n * overlap. A merge needs at least two pairs (a merged group must still mean \"appears more than\n * once\") and must fully consume at least one group; for equal cardinalities this reduces to the\n * strict all-pairs merge, so pre-partial-merge behavior is unchanged there.\n */\nfunction mergeGroups<T extends CountedOccurrence>(\n  first: T[],\n  second: T[],\n  maxGapTokens: number,\n  isReportableGroup: (group: T[]) => boolean\n): MergeResult<T> | undefined {\n  // Occurrences a previous partial merge already paired into a merged group must not pair again:\n  // their spans already live inside that merged group, so re-pairing them would assemble a second,\n  // competing merged group instead of letting the existing merged group extend (and would count\n  // the same span twice). Consumption is still judged against the FULL group, so a group holding\n  // shared occurrences is never subsumed away; nested copies, which the merged span would not\n  // cover anyway, are left out of that judgment so they cannot veto a merge of the standalone\n  // copies.\n  const [leadings, firstLength] = pairableOccurrences(first);\n  const [trailings, secondLength] = pairableOccurrences(second);\n  const pairs: [T, T][] = [];\n  let leadingIndex = 0;\n  let previousTrailingEnd = -1;\n  for (const trailing of trailings) {\n    // Leadings ending too far before this trailing can never pair a later (even farther) one.\n    while (leadingIndex < leadings.length) {\n      const leading = leadings[leadingIndex];\n      if (leading && leading.endTokenIndex + maxGapTokens < trailing.startTokenIndex) {\n        leadingIndex += 1;\n      } else {\n        break;\n      }\n    }\n    const leading = leadings[leadingIndex];\n    if (\n      leading &&\n      leading.endTokenIndex <= trailing.startTokenIndex &&\n      leading.startTokenIndex >= previousTrailingEnd\n    ) {\n      pairs.push([leading, trailing]);\n      previousTrailingEnd = trailing.endTokenIndex;\n      leadingIndex += 1;\n    }\n  }\n  const firstFullyPaired = pairs.length === firstLength;\n  const secondFullyPaired = pairs.length === secondLength;\n  if (pairs.length < 2 || (!firstFullyPaired && !secondFullyPaired)) {\n    return undefined;\n  }\n  // A group holding nested copies stays even when all its standalone copies pair: the merged\n  // group's content is larger than what those copies matched, so only the original group can\n  // report which files share the matched fragment.\n  const firstReplaced = firstFullyPaired && !first.some((occurrence) => occurrence.nestedInLargerGroup);\n  const secondReplaced = secondFullyPaired && !second.some((occurrence) => occurrence.nestedInLargerGroup);\n  const merged = pairs.map(([leading, trailing]) => ({\n    ...leading,\n    // A merged occurrence is a fresh span combination; it never inherits shared-span marks.\n    spanCountedElsewhere: undefined,\n    nestedInLargerGroup: undefined,\n    segments: [...leading.segments, ...trailing.segments],\n    tokenCount: leading.tokenCount + trailing.tokenCount,\n    endTokenIndex: trailing.endTokenIndex,\n    endIndex: trailing.endIndex,\n    endLine: trailing.endLine,\n  }));\n  // Only occurrences of one file pair (file offsets exceed the gap), so a merged group spans the\n  // files its pairs sit in: with nested copies left out of pairing, that can be fewer files than\n  // the input groups covered, and a merged group that is no longer reportable must not form.\n  if (!isReportableGroup(merged)) {\n    return undefined;\n  }\n  const pairedRetained = [\n    ...(firstReplaced ? [] : pairs.map(([leading]) => leading)),\n    ...(secondReplaced ? [] : pairs.map(([, trailing]) => trailing)),\n  ];\n  return { merged, firstReplaced, secondReplaced, pairedRetained };\n}\n\n/** One pass over a group: its pairable occurrences and its non-nested occurrence count. */\nfunction pairableOccurrences<T extends CountedOccurrence>(group: T[]): [T[], number] {\n  const pairable: T[] = [];\n  let length = 0;\n  for (const occurrence of group) {\n    if (!occurrence.nestedInLargerGroup) {\n      length += 1;\n    }\n    // Either flag keeps an occurrence out of pairing: its span is already counted elsewhere, or it\n    // is a nested copy of content the merged span would not cover.\n    if (!occurrence.spanCountedElsewhere && !occurrence.nestedInLargerGroup) {\n      pairable.push(occurrence);\n    }\n  }\n  return [pairable, length];\n}\n\n/**\n * Redundant copies one group adds to duplicateBlockCount. Each redundant occurrence contributes\n * one count per matched fragment, so merging a gapped clone's fragments into one group does not\n * halve duplicateBlockCount: an edited two-fragment pair still counts 2, exactly as its unmerged\n * fragments did. Occurrence shapes can differ within one group (a\n * gap-merged exact pair plus an appended whole-block near-miss copy), so every occurrence's\n * fragments are summed and one representative — the largest — is deducted, keeping the count\n * independent of source order. Occurrences a partial gapped merge also paired into a merged group\n * are skipped: their spans are already counted there, and such a retained group deducts no\n * representative of its own — the merged group's representative already stands for the shared\n * content — so no token span contributes to the count twice.\n */\nexport function countRedundantFragments(group: CountedOccurrence[]): number {\n  let fragmentCount = 0;\n  let maxFragmentCount = 0;\n  let hasSharedOccurrence = false;\n  for (const occurrence of group) {\n    if (occurrence.spanCountedElsewhere) {\n      hasSharedOccurrence = true;\n      continue;\n    }\n    fragmentCount += occurrence.segments.length;\n    maxFragmentCount = Math.max(maxFragmentCount, occurrence.segments.length);\n  }\n  return hasSharedOccurrence ? fragmentCount : fragmentCount - maxFragmentCount;\n}\n\n/** Adds the 1-based code lines the segment's matched tokens cover; shared with cross-file coverage. */\nexport function collectSegmentLines(\n  segment: { startTokenIndex: number; endTokenIndex: number },\n  tokens: Token[],\n  codeLineNumbers: Set<number> | undefined,\n  duplicatedLines: Set<number>\n): void {\n  for (let index = segment.startTokenIndex; index < segment.endTokenIndex; index += 1) {\n    const token = tokens[index];\n    for (let row = token?.startRow ?? 0; row <= (token?.endRow ?? -1); row += 1) {\n      if (!codeLineNumbers || codeLineNumbers.has(row + 1)) {\n        duplicatedLines.add(row + 1);\n      }\n    }\n  }\n}\n"],"mappings":"AASA,MAAa,EAA0D,CACrE,UAAW,GACX,aAAc,GACd,qBAAsB,EACxB,EAQA,SAAgB,EAA0B,EAA4D,CACpG,MAAO,CACL,UAAW,EAAc,GAAS,UAAW,EAA0B,SAAS,EAChF,aAAc,EAAc,GAAS,aAAc,EAA0B,YAAY,EACzF,qBAAsB,EAAc,GAAS,qBAAsB,EAA0B,oBAAoB,CACnH,CACF,CAEA,SAAS,EAAc,EAA2B,EAA0B,CAC1E,OAAO,IAAU,IAAA,IAAa,OAAO,MAAM,CAAK,EAC5C,EACA,KAAK,IAAI,KAAK,IAAI,KAAK,MAAM,CAAK,EAAG,CAAC,EAAG,UAAa,CAC5D,CAiBA,SAAS,EAAe,EAAsB,EAA6B,CACzE,OAAO,EAAe,GAAK,CAC7B,CAiHA,SAAgB,EAAwB,EAA6B,CACnE,IAAM,EAAS,IAAI,WAAW,EAAO,OAAS,CAAC,EAC/C,IAAK,GAAM,CAAC,EAAO,KAAU,EAAO,QAAQ,EAC1C,EAAO,EAAQ,IAAM,EAAO,IAAU,IAAM,EAAM,cAAgB,IAAA,GAAY,EAAI,GAEpF,OAAO,CACT,CA0CA,SAAgB,EACd,EACA,EACA,EAC+B,CAC/B,IAAM,EAA4C,CAAC,EAC7C,EAAoC,CAAC,EACrC,EAA6B,CAAC,EACpC,IAAK,GAAM,CAAC,EAAc,KAAY,EAAS,QAAQ,EACrD,IAAK,IAAM,KAAc,EAAQ,WAC/B,EAAwB,KAAK,CAAY,EACzC,EAAW,KAAK,CAAU,EAG9B,IAAM,EAAa,GACjB,EAAS,EAAwB,IAAmB,GAChD,EAAyB,IAAI,IAC7B,EAAmB,EAAW,KAAK,EAAY,IACnD,EAA0B,EAAU,CAAc,CAAC,EAAE,QAAU,CAAC,EAAG,EAAY,CAAS,CAC1F,EACA,IAAK,GAAM,CAAC,EAAgB,KAAY,EAAiB,QAAQ,EAAG,CAClE,IAAM,EAAe,EAAwB,IAAmB,EAChE,IAAK,GAAM,CAAC,EAAO,KAAQ,EAAQ,kBAAkB,QAAQ,EAC3D,IAAK,IAAM,KAAa,EAAK,CAC3B,GAAI,IAAc,IAAA,GAChB,SAEF,IAAM,EAAc,EAAuB,IAAI,CAAS,EACpD,GACF,EAAY,OAAS,EACjB,EAAY,iBAAmB,IACjC,EAAY,eAAiB,IAE3B,EAAY,eAAiB,IAC/B,EAAY,aAAe,IAE7B,EAAY,SAAW,KAAK,IAAI,EAAY,SAAU,CAAK,EAC3D,EAAY,SAAW,KAAK,IAAI,EAAY,SAAU,CAAK,GAE3D,EAAuB,IAAI,EAAW,CACpC,MAAO,EACP,iBACA,eACA,SAAU,EACV,SAAU,CACZ,CAAC,CAEL,CAEJ,CAMA,IAAM,GAAW,EAA+B,IAA4B,CAC1E,GAAI,IAAc,IAAA,GAChB,MAAO,GAET,IAAM,EAAc,EAAuB,IAAI,CAAS,EAOxD,OANI,IAAgB,IAAA,IAAa,EAAY,MAAQ,EAC5C,GAEL,EACK,EAAY,eAAiB,GAE/B,EAAY,iBAAmB,IAAM,EAAY,SAAW,EAAY,UAAY,CAC7F,EAKM,EAAyB,GAAoC,CACjE,IAAM,EAAS,EAAiB,EAAO,eAAe,EAAE,iBAAmB,CAAC,EACtE,EAAY,EAAO,EAAO,OAChC,IAAK,IAAI,EAAQ,EAAO,MAAQ,EAAG,EAAQ,EAAO,MAAQ,EAAO,OAAQ,GAAS,EAChF,GAAI,EAAO,KAAW,EACpB,MAAO,GAGX,MAAO,EACT,EAEM,EAAmC,CAAC,EAC1C,IAAK,GAAM,CAAC,EAAgB,KAAY,EAAiB,QAAQ,EAC/D,IAAK,GAAM,CAAC,EAAO,KAAQ,EAAQ,kBAAkB,QAAQ,EAC3D,IAAK,GAAM,CAAC,EAAQ,KAAc,EAAI,QAAQ,EAAG,CAC/C,GAAI,CAAC,EAAQ,EAAW,CAAM,GAAK,CAAC,EAAsB,CAAE,iBAAgB,QAAO,QAAO,CAAC,EACzF,SAIF,IAAM,EAAgB,EAAQ,kBAAkB,EAAM,GAAG,EAAS,GAC5D,EAAe,EAAQ,kBAAkB,EAAQ,EAAE,GAAG,EAAS,GACjE,EAAQ,EAAe,EAAS,CAAC,GAAK,EAAQ,EAAc,EAAS,CAAC,GAG1E,EAAe,KAAK,CAAE,iBAAgB,QAAO,QAAO,CAAC,CACvD,CASJ,IAAM,EAAU,IAAI,IAAI,EAAe,IAAI,CAAQ,CAAC,EAChD,EAAW,EACf,KAAO,EAAS,OAAS,GAAG,CAC1B,IAAM,EAA4B,CAAC,EACnC,IAAK,IAAM,KAAU,EAAU,CAC7B,IAAM,EAAa,EAAW,EAAO,gBAC/B,EAAQ,IAAa,EAAO,OAC5B,EAAO,IAAa,EAAO,MAAQ,EAAO,OAAS,GACnD,EAAU,EAAU,EAAO,cAAc,EAC/C,GAAI,CAAC,GAAS,CAAC,GAAQ,CAAC,EACtB,SAEF,IAAM,EAAc,KAAK,EAAe,EAAQ,OAAQ,EAAQ,mBAAoB,EAAM,gBAAiB,EAAK,aAAa,IAC7H,EAAW,KAAK,CACd,UAAW,EAAY,EAAa,EAAM,gBAAiB,EAAK,cAAe,EAAO,CAAI,EAC1F,aAAc,EAAwB,EAAO,iBAAmB,CAClE,CAAC,EACD,EAAQ,KAAK,CAAM,CACrB,CACA,EAAW,CAAC,EACZ,IAAK,IAAM,KAAU,EACnB,IAAK,IAAM,IAAS,CAAC,EAAO,MAAO,EAAO,MAAQ,CAAC,EAAG,CACpD,IAAM,EAAY,CAAE,eAAgB,EAAO,eAAgB,QAAO,OAAQ,EAAO,OAAS,CAAE,EACtF,EAAe,EAAiB,EAAO,eAAe,EAAE,kBAAkB,EAAM,GAAG,EAAU,QAEjG,GAAQ,IAAI,EAAS,CAAS,CAAC,GAC9B,EAAQ,EAAc,EAAU,MAAM,GACtC,EAAsB,CAAS,IAIlC,EAAQ,IAAI,EAAS,CAAS,CAAC,EAC/B,EAAS,KAAK,CAAS,EACzB,CAEJ,CACA,OAAO,CACT,CAEA,SAAS,EAAS,EAAgC,CAChD,MAAO,GAAG,EAAO,eAAe,GAAG,EAAO,MAAM,GAAG,EAAO,QAC5D,CASA,SAAS,EAA0B,EAAiB,EAA0B,EAAqC,CACjH,IAAM,EAAkB,EAAW,IAAK,GACtC,EAAgB,EAAQ,EAAU,gBAAiB,EAAU,aAAa,CAC5E,EACM,EAA8C,CAAC,EACrD,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAW,OAAQ,GAAS,EAAG,CACzD,IAAM,EAA8B,CAAC,EACjC,EAAO,KACP,EAAa,EACX,EAAS,KAAK,IAAI,EAAW,OAAQ,EAAQ,GAAyB,EAC5E,IAAK,IAAI,EAAM,EAAO,EAAM,EAAQ,GAAO,EAAG,CAC5C,IAAM,EAAY,EAAW,GACvB,EAAgB,EAAgB,GACtC,GAAI,CAAC,GAAa,IAAkB,IAAA,GAClC,MAEF,EAAO,EAAc,EAAM,CAAa,EACxC,GAAc,EAAU,cAAgB,EAAU,gBAClD,IAAM,EAAiB,EAAM,EAAQ,EACrC,EAAI,GACF,GAAkB,GAA6B,GAAc,EACzD,EAAc,EAAM,CAAc,EAClC,IAAA,EACR,CACA,EAAkB,KAAK,CAAG,CAC5B,CACA,MAAO,CAAE,oBAAmB,iBAAgB,CAC9C,CAOA,SAAgB,EAAU,EAAe,EAAuB,CAC9D,IAAM,EAAa,EAAE,OAAS,KAAQ,EAChC,EAAgB,IAAI,IAC1B,IAAK,GAAM,CAAC,EAAO,KAAW,EAAE,QAAQ,EAAG,CACzC,IAAI,EAAO,EAAc,IAAI,CAAM,EAC9B,IACH,EAAO,IAAI,YAAY,CAAS,EAChC,EAAc,IAAI,EAAQ,CAAI,GAEhC,IAAM,EAAO,IAAU,EAIvB,EAAK,IAAS,EAAK,IAAS,GAAM,IAAM,EAAQ,GAClD,CAEA,IAAM,EAAI,IAAI,YAAY,CAAS,EACnC,IAAK,IAAM,KAAU,EAAG,CACtB,IAAM,EAAY,EAAc,IAAI,CAAM,EAEtC,EAAa,EACb,EAAS,EACb,IAAK,IAAI,EAAO,EAAG,EAAO,EAAW,GAAQ,EAAG,CAC9C,IAAM,EAAW,EAAE,IAAS,EAEtB,IAAM,IAAY,IAAS,GAAK,KAAc,EAE9C,GAAY,GAAY,EAAK,KAAgB,EACnD,EAAa,IAAa,GAC1B,IAAM,EAAa,EAAI,EAAU,EACjC,EAAS,IAAa,GAEtB,EAAE,GAAQ,EAAI,CAAC,CACjB,CACF,CAEA,IAAI,EAAS,EACb,IAAK,IAAM,KAAQ,EACjB,GAAU,EAAS,CAAI,EAEzB,OAAO,CACT,CAEA,SAAS,EAAS,EAAuB,CACvC,IAAI,EAAQ,GAAU,IAAU,EAAK,YAErC,MADA,IAAS,EAAQ,YAAmB,IAAU,EAAK,WAC3C,KAAK,KAAM,GAAS,IAAU,GAAM,UAAe,QAAa,IAAM,GAAM,GACtF,CAEA,SAAS,EACP,EACA,EACA,EACA,EACA,EACoB,CACpB,MAAO,CACL,cACA,WAAY,EAAgB,EAC5B,kBACA,gBACA,WAAY,EAAM,WAClB,SAAU,EAAK,SACf,UAAW,EAAM,UACjB,QAAS,EAAK,OAChB,CACF,CAGA,MAAM,EAAkC,CAAC,EACnC,EAAmC,CAAC,EAE1C,SAAS,EAAoB,EAAuB,CAClD,IAAI,EAAO,EAAsB,GAKjC,OAJI,IAAS,IAAA,KACX,EAAO,EAAS,IAAI,GAAO,EAC3B,EAAsB,GAAS,GAE1B,CACT,CAEA,SAAS,EAAqB,EAAuB,CACnD,IAAI,EAAO,EAAuB,GAKlC,OAJI,IAAS,IAAA,KACX,EAAO,EAAU,IAAI,GAAO,EAC5B,EAAuB,GAAS,GAE3B,CACT,CAUA,SAAS,EACP,EACA,EACA,EACA,EACQ,CAGR,GAAM,CAAC,EAAS,GAAa,EAAoB,EAAQ,EAAiB,EADrD,GADC,EAAmB,IAAkB,IAAM,EAAmB,IAAoB,GACtD,EAAgB,CACkC,CAAC,EACrG,MAAO,GAAG,EAAQ,GAAG,EAAU,GAAG,EAAgB,GACpD,CASA,SAAS,EAAgB,EAAiB,EAAyB,EAA+B,CAChG,GAAM,CAAC,EAAS,GAAa,EAAoB,EAAQ,EAAiB,EAAe,EAAK,EAE9F,OAAO,EAAU,KAAK,KAAK,EAAW,EAAE,CAC1C,CAEA,SAAS,EACP,EACA,EACA,EACA,EACkB,CAClB,IAAM,EAAoB,IAAI,IAC1B,EAAU,KACV,EAAY,MAChB,IAAK,IAAI,EAAQ,EAAiB,EAAQ,EAAe,GAAS,EAAG,CACnE,IAAM,EAAQ,EAAO,GACrB,GAAI,CAAC,EACH,SAIF,IAAI,EACA,EACJ,GAAI,EAAM,OAAS,KAAM,CACvB,IAAI,EAAkB,EAAkB,IAAI,EAAM,IAAI,EAClD,IAAoB,IAAA,KACtB,EAAkB,EAAkB,KACpC,EAAkB,IAAI,EAAM,KAAM,CAAe,GAEnD,EAAO,EAAoB,CAAe,EAC1C,EAAQ,EAAqB,CAAe,CAC9C,KACE,GAAO,EAAM,SACb,EAAQ,EAAM,UAGhB,EAAW,KAAK,KAAK,EAAS,EAAE,EAAI,EAAQ,EAC5C,EAAY,KAAK,KAAK,EAAW,EAAE,EAAI,EACnC,GAAqB,EAAM,cAAgB,IAAA,IAAa,EAAM,eAAiB,IAAA,KAEjF,EAAW,KAAK,KAAK,EAAS,EAAE,EAAI,EAAM,YAAe,EACzD,EAAY,KAAK,KAAK,EAAW,EAAE,EAAI,EAAM,aAEjD,CACA,MAAO,CAAC,EAAS,CAAS,CAC5B,CAGA,SAAS,EAAS,EAAsB,CACtC,IAAI,EAAO,KACX,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAK,OAAQ,GAAS,EAEhD,EAAO,KAAK,KAAK,EAAM,EAAE,EAAI,EAAK,WAAW,CAAK,EAEpD,OAAO,CACT,CAGA,SAAS,EAAU,EAAsB,CACvC,IAAI,EAAO,YACX,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAK,OAAQ,GAAS,EAEhD,EAAO,KAAK,KAAK,EAAO,EAAK,WAAW,CAAK,EAAG,QAAU,EAE5D,OAAO,CACT,CAEA,SAAS,EAAc,EAAc,EAAuB,CAC1D,OAAO,KAAK,KAAK,EAAM,EAAE,EAAI,CAC/B,CAoBA,SAAgB,EACd,EACA,EACA,MAAmD,GAC5C,CACP,GAAI,GAAgB,GAAK,EAAO,OAAS,EACvC,OAAO,EAGT,EAAO,KAAK,CAAa,EACzB,IAAK,IAAI,EAAU,GAAM,GAAU,CACjC,EAAU,GACV,IAAK,IAAI,EAAY,EAAG,EAAY,EAAO,QAAU,CAAC,EAAS,GAAa,EAC1E,IAAK,IAAI,EAAa,EAAY,EAAG,EAAa,EAAO,OAAQ,GAAc,EAAG,CAChF,IAAM,EAAO,EAAO,GACd,EAAQ,EAAO,GACrB,GAAI,CAAC,GAAQ,CAAC,EACZ,SAEF,IAAM,EAAU,EAAY,EAAM,EAAO,EAAc,CAAiB,EAClE,EAAS,GAAW,EAAY,EAAO,EAAM,EAAc,CAAiB,EAClF,GAAI,CAAC,EACH,SAEF,IAAM,EAAe,EAAU,EAAO,cAAgB,EAAO,eACvD,EAAgB,EAAU,EAAO,eAAiB,EAAO,cAC3D,GAAgB,GAClB,EAAO,GAAa,EAAO,OAC3B,EAAO,OAAO,EAAY,CAAC,GAClB,EACT,EAAO,GAAc,EAAO,OACnB,EACT,EAAO,GAAa,EAAO,OAI3B,EAAO,KAAK,EAAO,MAAM,EAM3B,IAAK,IAAM,KAAc,EAAO,eAC9B,EAAW,qBAAuB,GAEpC,EAAO,KAAK,CAAa,EACzB,EAAU,GACV,KACF,CAEJ,CACA,OAAO,CACT,CAEA,SAAS,EAAc,EAA2B,EAAoC,CACpF,IAAM,EAAY,EAAK,GACjB,EAAa,EAAM,GACzB,OACG,GAAW,iBAAmB,IAAM,GAAY,iBAAmB,KACnE,GAAW,eAAiB,IAAM,GAAY,eAAiB,EAEpE,CAsBA,SAAS,EACP,EACA,EACA,EACA,EAC4B,CAQ5B,GAAM,CAAC,EAAU,GAAe,EAAoB,CAAK,EACnD,CAAC,EAAW,GAAgB,EAAoB,CAAM,EACtD,EAAkB,CAAC,EACrB,EAAe,EACf,EAAsB,GAC1B,IAAK,IAAM,KAAY,EAAW,CAEhC,KAAO,EAAe,EAAS,QAAQ,CACrC,IAAM,EAAU,EAAS,GACzB,GAAI,GAAW,EAAQ,cAAgB,EAAe,EAAS,gBAC7D,GAAgB,OAEhB,KAEJ,CACA,IAAM,EAAU,EAAS,GAEvB,GACA,EAAQ,eAAiB,EAAS,iBAClC,EAAQ,iBAAmB,IAE3B,EAAM,KAAK,CAAC,EAAS,CAAQ,CAAC,EAC9B,EAAsB,EAAS,cAC/B,GAAgB,EAEpB,CACA,IAAM,EAAmB,EAAM,SAAW,EACpC,EAAoB,EAAM,SAAW,EAC3C,GAAI,EAAM,OAAS,GAAM,CAAC,GAAoB,CAAC,EAC7C,OAKF,IAAM,EAAgB,GAAoB,CAAC,EAAM,KAAM,GAAe,EAAW,mBAAmB,EAC9F,EAAiB,GAAqB,CAAC,EAAO,KAAM,GAAe,EAAW,mBAAmB,EACjG,EAAS,EAAM,KAAK,CAAC,EAAS,MAAe,CACjD,GAAG,EAEH,qBAAsB,IAAA,GACtB,oBAAqB,IAAA,GACrB,SAAU,CAAC,GAAG,EAAQ,SAAU,GAAG,EAAS,QAAQ,EACpD,WAAY,EAAQ,WAAa,EAAS,WAC1C,cAAe,EAAS,cACxB,SAAU,EAAS,SACnB,QAAS,EAAS,OACpB,EAAE,EAIG,KAAkB,CAAM,EAO7B,MAAO,CAAE,SAAQ,gBAAe,iBAAgB,eAAA,CAH9C,GAAI,EAAgB,CAAC,EAAI,EAAM,KAAK,CAAC,KAAa,CAAO,EACzD,GAAI,EAAiB,CAAC,EAAI,EAAM,KAAK,EAAG,KAAc,CAAQ,CAEH,CAAE,CACjE,CAGA,SAAS,EAAiD,EAA2B,CACnF,IAAM,EAAgB,CAAC,EACnB,EAAS,EACb,IAAK,IAAM,KAAc,EAClB,EAAW,sBACd,GAAU,GAIR,CAAC,EAAW,sBAAwB,CAAC,EAAW,qBAClD,EAAS,KAAK,CAAU,EAG5B,MAAO,CAAC,EAAU,CAAM,CAC1B,CAcA,SAAgB,EAAwB,EAAoC,CAC1E,IAAI,EAAgB,EAChB,EAAmB,EACnB,EAAsB,GAC1B,IAAK,IAAM,KAAc,EAAO,CAC9B,GAAI,EAAW,qBAAsB,CACnC,EAAsB,GACtB,QACF,CACA,GAAiB,EAAW,SAAS,OACrC,EAAmB,KAAK,IAAI,EAAkB,EAAW,SAAS,MAAM,CAC1E,CACA,OAAO,EAAsB,EAAgB,EAAgB,CAC/D,CAGA,SAAgB,EACd,EACA,EACA,EACA,EACM,CACN,IAAK,IAAI,EAAQ,EAAQ,gBAAiB,EAAQ,EAAQ,cAAe,GAAS,EAAG,CACnF,IAAM,EAAQ,EAAO,GACrB,IAAK,IAAI,EAAM,GAAO,UAAY,EAAG,IAAQ,GAAO,QAAU,IAAK,GAAO,GACpE,CAAC,GAAmB,EAAgB,IAAI,EAAM,CAAC,IACjD,EAAgB,IAAI,EAAM,CAAC,CAGjC,CACF"}
+{"version":3,"file":"duplication.js","names":[],"sources":["../src/duplication.ts"],"sourcesContent":["import type { DuplicationOptions } from './types.js';\n\n/**\n * Project-level duplication machinery operating on normalized token streams. Tokenization itself\n * (parsing, identifier anonymization, literal normalization) happens in the Rust addon, which\n * serializes each file's Token stream and statement structure; the helpers here match statement\n * windows across files, merge gap-adjacent groups, and count duplicated lines over that data.\n */\n\nexport const defaultDuplicationOptions: Required<DuplicationOptions> = {\n  minTokens: 40,\n  maxGapTokens: 30,\n  minSimilarityPercent: 70,\n};\n\n/**\n * Fills defaults for absent settings, applying the same normalization as the native boundary's\n * clampToU32 — NaN (e.g. `Number(unsetEnvVariable)`) counts as absent, and other values truncate\n * and clamp to [0, u32::MAX] — so the TypeScript half of cross-file matching cannot diverge from\n * the natively collected candidates on such input.\n */\nexport function resolveDuplicationOptions(options?: DuplicationOptions): Required<DuplicationOptions> {\n  return {\n    minTokens: resolveOption(options?.minTokens, defaultDuplicationOptions.minTokens),\n    maxGapTokens: resolveOption(options?.maxGapTokens, defaultDuplicationOptions.maxGapTokens),\n    minSimilarityPercent: resolveOption(options?.minSimilarityPercent, defaultDuplicationOptions.minSimilarityPercent),\n  };\n}\n\nfunction resolveOption(value: number | undefined, fallback: number): number {\n  return value === undefined || Number.isNaN(value)\n    ? fallback\n    : Math.min(Math.max(Math.trunc(value), 0), 0xFF_FF_FF_FF);\n}\n\n/** Minimum consecutive statements for a statement-sequence duplicate candidate. */\nconst minSequenceStatementCount = 2;\n/**\n * Caps the window length so statement-sequence enumeration stays linear in the statement count.\n * Heterogeneous clones longer than the cap are reported as capped windows (a deliberate\n * conservative undercount trading completeness for bounded discovery cost).\n */\nconst maxSequenceStatementCount = 100;\n\n/**\n * A region whose normalized tokens are at least 20% literal values is data-like (a lookup table, a\n * constant list, a value-mapping switch), not logic: literal values re-enter its fingerprint so\n * tables that merely share their shape stop counting as copy-paste. Compared in integer math\n * (5 * literals >= total) so the TypeScript and native sides cannot disagree on the boundary.\n */\nfunction isLiteralDense(literalCount: number, tokenCount: number): boolean {\n  return literalCount * 5 >= tokenCount;\n}\n\nexport interface Token {\n  /** Normalization target: identifiers to anonymize, literal kind tags, or the raw token text. */\n  kind: 'id' | 'text';\n  text: string;\n  /**\n   * Two INDEPENDENT hashes of `text` (djb2 and FNV-1a), precomputed so fingerprinting nested\n   * regions never re-hashes a token. Feeding the same per-token hash to both fingerprint\n   * accumulators would collapse the key to 32 effective bits: one djb2 collision between two\n   * token texts would then equate whole regions.\n   */\n  textHash: number;\n  textHash2: number;\n  /** Hash pair of a value-carrying literal's value, folded into data-like region fingerprints. */\n  literalHash?: number;\n  literalHash2?: number;\n  /**\n   * True for verbatim-kept NAMES (member/callee/type names, named grammar leaves): together with\n   * value-carrying literals these are the content-bearing tokens the near-miss content gate\n   * counts. Keywords, operators, and punctuation come from unnamed nodes and stay false.\n   */\n  isName?: boolean;\n  /** 0-based source rows the token occupies, so line coverage counts only matched-token lines. */\n  startRow: number;\n  endRow: number;\n}\n\n/** A token span with its source position; plain data so cross-file matching can retain it. */\nexport interface TokenRange {\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\n/** A contiguous run of matched tokens; gapped (merged) duplicates carry several per occurrence. */\ninterface TokenSegment {\n  startTokenIndex: number;\n  endTokenIndex: number;\n}\n\ninterface DuplicateCandidate {\n  fingerprint: string;\n  tokenCount: number;\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\nexport interface CountedOccurrence {\n  /** Matched token runs; more than one once gapped groups are merged. */\n  segments: TokenSegment[];\n  /**\n   * Set on occurrences whose span another reported group already counts: a retained group's\n   * occurrences that a partial gapped merge also paired into a merged group, and cross-file copies\n   * nested inside a larger group's region. Block counting must not count them again.\n   */\n  spanCountedElsewhere?: boolean;\n  /**\n   * Set on cross-file copies nested inside a larger group's region (they also set\n   * `spanCountedElsewhere`). They never pair in gapped merging, and they do not keep their group's\n   * standalone copies from merging, which they are not copies of. They do keep the merged group\n   * from taking their group's place, since only the original group reports the nesting.\n   */\n  nestedInLargerGroup?: boolean;\n  /** Sum of segment token counts (the gap tokens are not matched content). */\n  tokenCount: number;\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\n/** A duplicate region found in one file, exported for cross-file matching by fingerprint. */\nexport interface CrossFileDuplicateCandidate {\n  /** Content key: equal fingerprints mean equal normalized token sequences (up to hash collision). */\n  fingerprint: string;\n  tokenCount: number;\n  /** Token positions within the owning file, for cross-file gapped (Type-3) merging. */\n  startTokenIndex: number;\n  endTokenIndex: number;\n  startIndex: number;\n  endIndex: number;\n  startLine: number;\n  endLine: number;\n}\n\n/**\n * One file's contribution to cross-file clone detection: catalogued candidates plus the normalized\n * token stream and statement structure, so the project-level pass can match partial statement runs\n * (windows that a single file cannot know repeat elsewhere) and merge gap-adjacent groups.\n */\nexport interface CrossFileDuplicationFileData {\n  candidates: CrossFileDuplicateCandidate[];\n  tokens: Token[];\n  containerStatements: TokenRange[][];\n  /**\n   * The mutually disjoint blocks (at least `minTokens` long, not literal-dense) compared for\n   * cross-file near-miss (Type-3) clones. Optional for backward compatibility; without it, the\n   * file takes part in exact and gapped cross-file matching only.\n   */\n  nearMissBlocks?: TokenRange[];\n  /**\n   * 1-based lines that are neither blank nor comment-only, so cross-file line coverage counts only\n   * code lines (blank rows inside multi-row tokens such as template literals carry no content).\n   * Optional for backward compatibility; without it, every matched-token row counts.\n   */\n  codeLineNumbers?: Set<number>;\n}\n\n/** literalCountPrefix[i] = value-carrying literal tokens in tokens[0..i), for O(1) density checks. */\nexport function buildLiteralCountPrefix(tokens: Token[]): Int32Array {\n  const prefix = new Int32Array(tokens.length + 1);\n  for (const [index, token] of tokens.entries()) {\n    prefix[index + 1] = (prefix[index] ?? 0) + (token.literalHash === undefined ? 0 : 1);\n  }\n  return prefix;\n}\n\ninterface WindowOccurrences {\n  count: number;\n  /** -1 once occurrences span more than one container. */\n  containerIndex: number;\n  /** -1 once occurrences span more than one context (file). */\n  contextIndex: number;\n  minStart: number;\n  maxStart: number;\n}\n\ninterface SequenceWindow {\n  containerIndex: number;\n  start: number;\n  length: number;\n}\n\n/** One file's token stream and statement containers, as a window-matching context. */\nexport interface SequenceWindowContext {\n  tokens: Token[];\n  literalCountPrefix: Int32Array;\n  containers: TokenRange[][];\n}\n\nexport interface ContextualSequenceCandidate {\n  candidate: DuplicateCandidate;\n  contextIndex: number;\n}\n\n/**\n * Enumerates runs of consecutive sibling statements over one or more contexts (files). Every\n * container statement participates; only the window length is capped, so enumeration stays linear\n * in the statement count. Windows are grouped by a cheap rolling hash of per-statement\n * fingerprints, and only locally maximal repeated windows — those whose one-statement extensions\n * stop repeating — become candidates with an exact (window-consistent) fingerprint. Without the\n * maximality filter a degenerate file of near-identical statements would fingerprint every\n * sub-window of every repeated region. With `requireMultipleContexts` a window only counts as\n * repeated when its occurrences span at least two contexts (CPD-style cross-file matching): a\n * repeat confined to one file is that file's own concern, and emitting it here would flood the\n * project-level selection with unusable single-file groups.\n */\nexport function collectSequenceWindowCandidates(\n  contexts: SequenceWindowContext[],\n  minTokens: number,\n  requireMultipleContexts: boolean\n): ContextualSequenceCandidate[] {\n  const candidates: ContextualSequenceCandidate[] = [];\n  const contextIndexByContainer: number[] = [];\n  const containers: TokenRange[][] = [];\n  for (const [contextIndex, context] of contexts.entries()) {\n    for (const statements of context.containers) {\n      contextIndexByContainer.push(contextIndex);\n      containers.push(statements);\n    }\n  }\n  const contextAt = (containerIndex: number): SequenceWindowContext | undefined =>\n    contexts[contextIndexByContainer[containerIndex] ?? 0];\n  const occurrencesByWindowKey = new Map<number, WindowOccurrences>();\n  const containerWindows = containers.map((statements, containerIndex) =>\n    enumerateContainerWindows(contextAt(containerIndex)?.tokens ?? [], statements, minTokens)\n  );\n  for (const [containerIndex, windows] of containerWindows.entries()) {\n    const contextIndex = contextIndexByContainer[containerIndex] ?? 0;\n    for (const [start, row] of windows.windowKeysByStart.entries()) {\n      for (const windowKey of row) {\n        if (windowKey === undefined) {\n          continue;\n        }\n        const occurrences = occurrencesByWindowKey.get(windowKey);\n        if (occurrences) {\n          occurrences.count += 1;\n          if (occurrences.containerIndex !== containerIndex) {\n            occurrences.containerIndex = -1;\n          }\n          if (occurrences.contextIndex !== contextIndex) {\n            occurrences.contextIndex = -1;\n          }\n          occurrences.minStart = Math.min(occurrences.minStart, start);\n          occurrences.maxStart = Math.max(occurrences.maxStart, start);\n        } else {\n          occurrencesByWindowKey.set(windowKey, {\n            count: 1,\n            containerIndex,\n            contextIndex,\n            minStart: start,\n            maxStart: start,\n          });\n        }\n      }\n    }\n  }\n\n  // A window only \"repeats\" when two of its occurrences can coexist without overlapping: sliding\n  // matches inside a homogeneous run (start spread smaller than the window length) can never both\n  // be counted and must neither qualify a window nor dominate its sub-windows. Cross-context\n  // matching instead requires occurrences in two contexts, which coexist by construction.\n  const repeats = (windowKey: number | undefined, length: number): boolean => {\n    if (windowKey === undefined) {\n      return false;\n    }\n    const occurrences = occurrencesByWindowKey.get(windowKey);\n    if (occurrences === undefined || occurrences.count < 2) {\n      return false;\n    }\n    if (requireMultipleContexts) {\n      return occurrences.contextIndex === -1;\n    }\n    return occurrences.containerIndex === -1 || occurrences.maxStart - occurrences.minStart >= length;\n  };\n\n  // A window whose statements all share one normalized shape (sixteen `let x = 0;` declarations,\n  // a constant table) is a homogeneous preamble, not a copy-paste: requiring two distinct\n  // per-statement shapes keeps such runs out of duplicate groups and the duplication ratio.\n  const hasDistinctStatements = (window: SequenceWindow): boolean => {\n    const hashes = containerWindows[window.containerIndex]?.statementHashes ?? [];\n    const firstHash = hashes[window.start];\n    for (let index = window.start + 1; index < window.start + window.length; index += 1) {\n      if (hashes[index] !== firstHash) {\n        return true;\n      }\n    }\n    return false;\n  };\n\n  const maximalWindows: SequenceWindow[] = [];\n  for (const [containerIndex, windows] of containerWindows.entries()) {\n    for (const [start, row] of windows.windowKeysByStart.entries()) {\n      for (const [length, windowKey] of row.entries()) {\n        if (!repeats(windowKey, length) || !hasDistinctStatements({ containerIndex, start, length })) {\n          continue;\n        }\n        // Dominated windows are skipped: the one-statement extension also repeats, so a larger\n        // candidate covering this window exists.\n        const extendedRight = windows.windowKeysByStart[start]?.[length + 1];\n        const extendedLeft = windows.windowKeysByStart[start - 1]?.[length + 1];\n        if (repeats(extendedRight, length + 1) || repeats(extendedLeft, length + 1)) {\n          continue;\n        }\n        maximalWindows.push({ containerIndex, start, length });\n      }\n    }\n  }\n\n  // The rolling hash anonymizes identifiers per statement, so a window can look repeated coarsely\n  // while its exact (window-consistent) fingerprints differ, and a longer window's match can\n  // dominate sub-windows that other copies still need (three copies where only two extend one\n  // statement further). Every emitted window therefore exposes its repeating, unvisited\n  // sub-windows; `visited` bounds the worklist and lengths strictly decrease, so it terminates.\n  const visited = new Set(maximalWindows.map(windowId));\n  let frontier = maximalWindows;\n  while (frontier.length > 0) {\n    const emitted: SequenceWindow[] = [];\n    for (const window of frontier) {\n      const statements = containers[window.containerIndex];\n      const first = statements?.[window.start];\n      const last = statements?.[window.start + window.length - 1];\n      const context = contextAt(window.containerIndex);\n      if (!first || !last || !context) {\n        continue;\n      }\n      const fingerprint = `s:${fingerprintKey(context.tokens, context.literalCountPrefix, first.startTokenIndex, last.endTokenIndex)}`;\n      candidates.push({\n        candidate: toCandidate(fingerprint, first.startTokenIndex, last.endTokenIndex, first, last),\n        contextIndex: contextIndexByContainer[window.containerIndex] ?? 0,\n      });\n      emitted.push(window);\n    }\n    frontier = [];\n    for (const window of emitted) {\n      for (const start of [window.start, window.start + 1]) {\n        const subWindow = { containerIndex: window.containerIndex, start, length: window.length - 1 };\n        const subWindowKey = containerWindows[window.containerIndex]?.windowKeysByStart[start]?.[subWindow.length];\n        if (\n          visited.has(windowId(subWindow)) ||\n          !repeats(subWindowKey, subWindow.length) ||\n          !hasDistinctStatements(subWindow)\n        ) {\n          continue;\n        }\n        visited.add(windowId(subWindow));\n        frontier.push(subWindow);\n      }\n    }\n  }\n  return candidates;\n}\n\nfunction windowId(window: SequenceWindow): string {\n  return `${window.containerIndex}:${window.start}:${window.length}`;\n}\n\ninterface ContainerWindows {\n  /** windowKeysByStart[start][length] is the rolling-hash key of the window, or undefined if it is below the size thresholds. */\n  windowKeysByStart: (number | undefined)[][];\n  /** Per-statement fingerprint hashes, for the distinct-shape requirement on windows. */\n  statementHashes: number[];\n}\n\nfunction enumerateContainerWindows(tokens: Token[], statements: TokenRange[], minTokens: number): ContainerWindows {\n  const statementHashes = statements.map((statement) =>\n    fingerprintHash(tokens, statement.startTokenIndex, statement.endTokenIndex)\n  );\n  const windowKeysByStart: (number | undefined)[][] = [];\n  for (let start = 0; start < statements.length; start += 1) {\n    const row: (number | undefined)[] = [];\n    let hash = 5381;\n    let tokenCount = 0;\n    const maxEnd = Math.min(statements.length, start + maxSequenceStatementCount);\n    for (let end = start; end < maxEnd; end += 1) {\n      const statement = statements[end];\n      const statementHash = statementHashes[end];\n      if (!statement || statementHash === undefined) {\n        break;\n      }\n      hash = combineHashes(hash, statementHash);\n      tokenCount += statement.endTokenIndex - statement.startTokenIndex;\n      const statementCount = end - start + 1;\n      row[statementCount] =\n        statementCount >= minSequenceStatementCount && tokenCount >= minTokens\n          ? combineHashes(hash, statementCount)\n          : undefined;\n    }\n    windowKeysByStart.push(row);\n  }\n  return { windowKeysByStart, statementHashes };\n}\n\n/**\n * Longest-common-subsequence LENGTH of two symbol sequences via the Allison–Dix bit-parallel\n * recurrence (O(|a|/32 · |b|) words): per symbol of `b`, `x = match | v` and\n * `v = x & ~(x - ((v << 1) | 1))` over multi-word bit vectors; the set bits of `v` count the LCS.\n */\nexport function lcsLength(a: Int32Array, b: Int32Array): number {\n  return createLcsLengthCounter(a)(b);\n}\n\n/**\n * lcsLength against a fixed `a`, whose per-symbol position masks are built once and reused for\n * every `b` compared with it.\n */\nexport function createLcsLengthCounter(a: Int32Array): (b: Int32Array) => number {\n  const wordCount = (a.length + 31) >>> 5;\n  const positionMasks = new Map<number, Uint32Array>();\n  for (const [index, symbol] of a.entries()) {\n    let mask = positionMasks.get(symbol);\n    if (!mask) {\n      mask = new Uint32Array(wordCount);\n      positionMasks.set(symbol, mask);\n    }\n    const word = index >>> 5;\n    // The Uint32Array store wraps the signed int32 bit pattern to unsigned. The `?? 0` guards in\n    // this function are required by noUncheckedIndexedAccess (typed-array reads type as\n    // `number | undefined`), not redundancy: every index is in bounds.\n    mask[word] = (mask[word] ?? 0) | (1 << (index & 31));\n  }\n\n  const v = new Uint32Array(wordCount);\n  return (b) => {\n    v.fill(0);\n    for (const symbol of b) {\n      const matchMask = positionMasks.get(symbol);\n      // `(v << 1) | 1` shifts a carry bit across words; subtraction borrows across words.\n      let shiftCarry = 1;\n      let borrow = 0;\n      for (let word = 0; word < wordCount; word += 1) {\n        const previous = v[word] ?? 0;\n        // oxlint-disable-next-line unicorn/prefer-math-trunc -- `>>> 0` reinterprets the signed int32 bit pattern as unsigned so the borrow subtraction below compares magnitudes; Math.trunc would keep it negative.\n        const x = ((matchMask?.[word] ?? 0) | previous) >>> 0;\n        // oxlint-disable-next-line unicorn/prefer-math-trunc -- same unsigned reinterpretation as `x`.\n        const shifted = ((previous << 1) | shiftCarry) >>> 0;\n        shiftCarry = previous >>> 31;\n        const difference = x - shifted - borrow;\n        borrow = difference < 0 ? 1 : 0;\n        // The Uint32Array store wraps the signed int32 bit pattern to unsigned.\n        v[word] = x & ~difference;\n      }\n    }\n\n    let length = 0;\n    for (const word of v) {\n      length += popCount(word);\n    }\n    return length;\n  };\n}\n\nfunction popCount(value: number): number {\n  let count = value - ((value >>> 1) & 0x55_55_55_55);\n  count = (count & 0x33_33_33_33) + ((count >>> 2) & 0x33_33_33_33);\n  return (Math.imul((count + (count >>> 4)) & 0x0F_0F_0F_0F, 0x01_01_01_01) >>> 24) & 0xFF;\n}\n\nfunction toCandidate(\n  fingerprint: string,\n  startTokenIndex: number,\n  endTokenIndex: number,\n  first: TokenRange,\n  last: TokenRange\n): DuplicateCandidate {\n  return {\n    fingerprint,\n    tokenCount: endTokenIndex - startTokenIndex,\n    startTokenIndex,\n    endTokenIndex,\n    startIndex: first.startIndex,\n    endIndex: last.endIndex,\n    startLine: first.startLine,\n    endLine: last.endLine,\n  };\n}\n\n/** Caches of hashText/hashText2 over '$0', '$1', ... so anonymized identifiers hash without allocating. */\nconst anonymizedIndexHashes: number[] = [];\nconst anonymizedIndexHashes2: number[] = [];\n\nfunction anonymizedIndexHash(index: number): number {\n  let hash = anonymizedIndexHashes[index];\n  if (hash === undefined) {\n    hash = hashText(`$${index}`);\n    anonymizedIndexHashes[index] = hash;\n  }\n  return hash;\n}\n\nfunction anonymizedIndexHash2(index: number): number {\n  let hash = anonymizedIndexHashes2[index];\n  if (hash === undefined) {\n    hash = hashText2(`$${index}`);\n    anonymizedIndexHashes2[index] = hash;\n  }\n  return hash;\n}\n\n/**\n * Content key of a token range: two independent 32-bit hashes over the normalized token sequence\n * (identifiers anonymized consistently by first-occurrence order) plus the token count. Regions\n * with equal keys are treated as equal content; a collision would need both 32-bit hashes and the\n * length to coincide, which is negligible for a metrics report. The format and arithmetic match\n * fingerprint_key in native/src/duplication.rs exactly, so window candidates fingerprinted here\n * group together with the per-file candidates the addon catalogues.\n */\nfunction fingerprintKey(\n  tokens: Token[],\n  literalCountPrefix: Int32Array,\n  startTokenIndex: number,\n  endTokenIndex: number\n): string {\n  const literalCount = (literalCountPrefix[endTokenIndex] ?? 0) - (literalCountPrefix[startTokenIndex] ?? 0);\n  const literalDense = isLiteralDense(literalCount, endTokenIndex - startTokenIndex);\n  const [primary, secondary] = fingerprintHashPair(tokens, startTokenIndex, endTokenIndex, literalDense);\n  return `${primary}:${secondary}:${endTokenIndex - startTokenIndex}`;\n}\n\n/**\n * A single 32-bit summary of a range for the coarse rolling-hash phase. Deliberately\n * density-agnostic: density is a property of the final candidate REGION, and folding literal\n * values into per-statement hashes would make a dense statement inside a logic-heavy window\n * (`const weights = [1, 2, 3];`) block the window from ever being enumerated. The coarse phase\n * over-approximates on shape alone; the exact region fingerprint still applies the density rule.\n */\nfunction fingerprintHash(tokens: Token[], startTokenIndex: number, endTokenIndex: number): number {\n  const [primary, secondary] = fingerprintHashPair(tokens, startTokenIndex, endTokenIndex, false);\n  // XOR already coerces to int32, matching the native side's i32 arithmetic.\n  return primary ^ Math.imul(secondary, 31);\n}\n\nfunction fingerprintHashPair(\n  tokens: Token[],\n  startTokenIndex: number,\n  endTokenIndex: number,\n  foldLiteralValues: boolean\n): [number, number] {\n  const indexByIdentifier = new Map<string, number>();\n  let primary = 5381;\n  let secondary = 52_711;\n  for (let index = startTokenIndex; index < endTokenIndex; index += 1) {\n    const token = tokens[index];\n    if (!token) {\n      continue;\n    }\n    // Each accumulator consumes its own independent per-token hash: sharing one would collapse\n    // the key to 32 effective bits (a single djb2 collision would equate whole regions).\n    let part: number;\n    let part2: number;\n    if (token.kind === 'id') {\n      let identifierIndex = indexByIdentifier.get(token.text);\n      if (identifierIndex === undefined) {\n        identifierIndex = indexByIdentifier.size;\n        indexByIdentifier.set(token.text, identifierIndex);\n      }\n      part = anonymizedIndexHash(identifierIndex);\n      part2 = anonymizedIndexHash2(identifierIndex);\n    } else {\n      part = token.textHash;\n      part2 = token.textHash2;\n    }\n    // oxlint-disable-next-line unicorn/prefer-math-trunc -- `| 0` wraps the sum to int32 (Math.trunc does not), which must match the native side's wrapping i32 arithmetic.\n    primary = (Math.imul(primary, 31) + part) | 0;\n    secondary = Math.imul(secondary, 37) ^ part2;\n    if (foldLiteralValues && token.literalHash !== undefined && token.literalHash2 !== undefined) {\n      // oxlint-disable-next-line unicorn/prefer-math-trunc -- `| 0` wraps the sum to int32 (Math.trunc does not), which must match the native side's wrapping i32 arithmetic.\n      primary = (Math.imul(primary, 31) + token.literalHash) | 0;\n      secondary = Math.imul(secondary, 37) ^ token.literalHash2;\n    }\n  }\n  return [primary, secondary];\n}\n\n/** djb2-style hash; XOR keeps the value in signed 32-bit range, which is fine for a grouping key. */\nfunction hashText(text: string): number {\n  let hash = 5381;\n  for (let index = 0; index < text.length; index += 1) {\n    // oxlint-disable-next-line unicorn/prefer-code-point -- djb2 hashes UTF-16 code units; codePointAt would hash surrogate pairs twice (full code point, then the lone low surrogate).\n    hash = Math.imul(hash, 33) ^ text.charCodeAt(index);\n  }\n  return hash;\n}\n\n/** FNV-1a over UTF-16 code units: independent of hashText so the two accumulators never share input. */\nfunction hashText2(text: string): number {\n  let hash = -2_128_831_035; // 2166136261 as int32 (the FNV-1a offset basis)\n  for (let index = 0; index < text.length; index += 1) {\n    // oxlint-disable-next-line unicorn/prefer-code-point -- hashes UTF-16 code units like hashText.\n    hash = Math.imul(hash ^ text.charCodeAt(index), 16_777_619);\n  }\n  return hash;\n}\n\nfunction combineHashes(hash: number, value: number): number {\n  return Math.imul(hash, 31) + value;\n}\n\n/**\n * Merges duplicate groups separated by a small token gap into one gapped (Type-3) clone group: a\n * copy edited in one spot splits into two exact groups whose occurrences sit side by side in the\n * same order. Occurrences are paired greedily in source order; a merge happens when the pairing\n * fully pairs at least one group with at least two pairs. Equal-cardinality groups whose\n * occurrences all pair merge into one group as before. When cardinalities differ (a fragment also\n * occurs standalone: prefix ×3, suffix ×2), the fully-paired group is subsumed into the merged\n * gapped group while the other group is RETAINED with ALL its occurrences: dropping the leftover\n * would lose duplicated-line coverage, and reporting it alone would make a single-occurrence group\n * (contradicting duplicateBlockGroupCount's \"appears more than once\" meaning). Line coverage\n * unions ranges, so the overlap between the retained exact group and the merged group is harmless.\n * Merging repeats to a fixpoint so a clone edited in several spots still reassembles; it\n * terminates because a merge either removes the input groups it replaces or marks their paired\n * occurrences as counted elsewhere, and only unmarked occurrences of remaining groups pair, so the\n * number of pairable occurrences strictly decreases with every merge. Gap tokens are not matched\n * content: line coverage and sizes count only the matched segments. Generic so cross-file merging\n * can thread file identity through occurrences.\n */\nexport function mergeAdjacentGroups<T extends CountedOccurrence>(\n  groups: T[][],\n  maxGapTokens: number,\n  isReportableGroup: (group: T[]) => boolean = () => true\n): T[][] {\n  if (maxGapTokens <= 0 || groups.length < 2) {\n    return groups;\n  }\n  // Deterministic processing order (mirrored by the native side): by first occurrence position.\n  groups.sort(compareGroups);\n  for (let restart = true; restart;) {\n    restart = false;\n    const partnersByGroup = collectGapAdjacentPartners(groups, maxGapTokens);\n    for (let leftIndex = 0; leftIndex < groups.length && !restart; leftIndex += 1) {\n      for (const rightIndex of partnersByGroup[leftIndex] ?? []) {\n        const left = groups[leftIndex];\n        const right = groups[rightIndex];\n        if (!left || !right) {\n          continue;\n        }\n        const forward = mergeGroups(left, right, maxGapTokens, isReportableGroup);\n        const result = forward ?? mergeGroups(right, left, maxGapTokens, isReportableGroup);\n        if (!result) {\n          continue;\n        }\n        const leftReplaced = forward ? result.firstReplaced : result.secondReplaced;\n        const rightReplaced = forward ? result.secondReplaced : result.firstReplaced;\n        if (leftReplaced && rightReplaced) {\n          groups[leftIndex] = result.merged;\n          groups.splice(rightIndex, 1);\n        } else if (rightReplaced) {\n          groups[rightIndex] = result.merged;\n        } else if (leftReplaced) {\n          groups[leftIndex] = result.merged;\n        } else {\n          // Both groups stay (each is only partly paired, or reports nested copies of its own), so\n          // the merged group joins them instead of taking a place.\n          groups.push(result.merged);\n        }\n        // A group that stays keeps ALL its occurrences (line coverage must not shrink, and a\n        // reported group must keep >= 2 occurrences), so its paired occurrences now also live\n        // inside the merged group's occurrences: mark them so duplicateBlockCount counts each\n        // token span once, and so the same pair cannot merge again.\n        for (const occurrence of result.pairedRetained) {\n          occurrence.spanCountedElsewhere = true;\n        }\n        groups.sort(compareGroups);\n        restart = true;\n        break;\n      }\n    }\n  }\n  return groups;\n}\n\n/**\n * Per group index, the ascending indexes of later groups that mergeGroups can pair with it in\n * either direction: an occurrence of one starts within `maxGapTokens` after an occurrence of the\n * other ends. Every other pair fails mergeGroups, so skipping it leaves the fixpoint unchanged\n * while each restart stops costing a comparison per pair of groups.\n */\nfunction collectGapAdjacentPartners(groups: CountedOccurrence[][], maxGapTokens: number): number[][] {\n  const starts: { startTokenIndex: number; groupIndex: number }[] = [];\n  for (const [groupIndex, group] of groups.entries()) {\n    for (const occurrence of group) {\n      starts.push({ startTokenIndex: occurrence.startTokenIndex, groupIndex });\n    }\n  }\n  starts.sort((left, right) => left.startTokenIndex - right.startTokenIndex);\n  const partners = groups.map(() => new Set<number>());\n  for (const [groupIndex, group] of groups.entries()) {\n    for (const occurrence of group) {\n      for (\n        let index = lowerBoundByStart(starts, occurrence.endTokenIndex);\n        index < starts.length && (starts[index]?.startTokenIndex ?? 0) <= occurrence.endTokenIndex + maxGapTokens;\n        index += 1\n      ) {\n        const other = starts[index]?.groupIndex ?? groupIndex;\n        if (other !== groupIndex) {\n          partners[Math.min(other, groupIndex)]?.add(Math.max(other, groupIndex));\n        }\n      }\n    }\n  }\n  return partners.map((set) => [...set].toSorted((left, right) => left - right));\n}\n\nfunction lowerBoundByStart(sorted: { startTokenIndex: number }[], target: number): number {\n  let low = 0;\n  let high = sorted.length;\n  while (low < high) {\n    const middle = (low + high) >>> 1;\n    if ((sorted[middle]?.startTokenIndex ?? 0) < target) {\n      low = middle + 1;\n    } else {\n      high = middle;\n    }\n  }\n  return low;\n}\n\nfunction compareGroups(left: CountedOccurrence[], right: CountedOccurrence[]): number {\n  const leftFirst = left[0];\n  const rightFirst = right[0];\n  return (\n    (leftFirst?.startTokenIndex ?? 0) - (rightFirst?.startTokenIndex ?? 0) ||\n    (leftFirst?.endTokenIndex ?? 0) - (rightFirst?.endTokenIndex ?? 0)\n  );\n}\n\ninterface MergeResult<T> {\n  merged: T[];\n  /**\n   * Whether the merged group takes the respective input group's place: its standalone occurrences\n   * were all paired and it holds no nested copies that only it can report.\n   */\n  firstReplaced: boolean;\n  secondReplaced: boolean;\n  /** The occurrences of groups that stay, which the merged group's spans now also cover. */\n  pairedRetained: T[];\n}\n\n/**\n * Pairs `second` occurrences with gap-preceding `first` occurrences, greedily in source order:\n * each trailing occurrence takes the earliest unused leading occurrence within the gap, and a\n * pair's leading must start at or after the previous pair's trailing end so merged spans never\n * overlap. A merge needs at least two pairs (a merged group must still mean \"appears more than\n * once\") and must fully consume at least one group; for equal cardinalities this reduces to the\n * strict all-pairs merge, so pre-partial-merge behavior is unchanged there.\n */\nfunction mergeGroups<T extends CountedOccurrence>(\n  first: T[],\n  second: T[],\n  maxGapTokens: number,\n  isReportableGroup: (group: T[]) => boolean\n): MergeResult<T> | undefined {\n  // Occurrences a previous partial merge already paired into a merged group must not pair again:\n  // their spans already live inside that merged group, so re-pairing them would assemble a second,\n  // competing merged group instead of letting the existing merged group extend (and would count\n  // the same span twice). Consumption is still judged against the FULL group, so a group holding\n  // shared occurrences is never subsumed away; nested copies, which the merged span would not\n  // cover anyway, are left out of that judgment so they cannot veto a merge of the standalone\n  // copies.\n  const [leadings, firstLength] = pairableOccurrences(first);\n  const [trailings, secondLength] = pairableOccurrences(second);\n  const pairs: [T, T][] = [];\n  let leadingIndex = 0;\n  let previousTrailingEnd = -1;\n  for (const trailing of trailings) {\n    // Leadings ending too far before this trailing can never pair a later (even farther) one.\n    while (leadingIndex < leadings.length) {\n      const leading = leadings[leadingIndex];\n      if (leading && leading.endTokenIndex + maxGapTokens < trailing.startTokenIndex) {\n        leadingIndex += 1;\n      } else {\n        break;\n      }\n    }\n    const leading = leadings[leadingIndex];\n    if (\n      leading &&\n      leading.endTokenIndex <= trailing.startTokenIndex &&\n      leading.startTokenIndex >= previousTrailingEnd\n    ) {\n      pairs.push([leading, trailing]);\n      previousTrailingEnd = trailing.endTokenIndex;\n      leadingIndex += 1;\n    }\n  }\n  const firstFullyPaired = pairs.length === firstLength;\n  const secondFullyPaired = pairs.length === secondLength;\n  if (pairs.length < 2 || (!firstFullyPaired && !secondFullyPaired)) {\n    return undefined;\n  }\n  // A group holding nested copies stays even when all its standalone copies pair: the merged\n  // group's content is larger than what those copies matched, so only the original group can\n  // report which files share the matched fragment.\n  const firstReplaced = firstFullyPaired && !first.some((occurrence) => occurrence.nestedInLargerGroup);\n  const secondReplaced = secondFullyPaired && !second.some((occurrence) => occurrence.nestedInLargerGroup);\n  const merged = pairs.map(([leading, trailing]) => ({\n    ...leading,\n    // A merged occurrence is a fresh span combination; it never inherits shared-span marks.\n    spanCountedElsewhere: undefined,\n    nestedInLargerGroup: undefined,\n    segments: [...leading.segments, ...trailing.segments],\n    tokenCount: leading.tokenCount + trailing.tokenCount,\n    endTokenIndex: trailing.endTokenIndex,\n    endIndex: trailing.endIndex,\n    endLine: trailing.endLine,\n  }));\n  // Only occurrences of one file pair (file offsets exceed the gap), so a merged group spans the\n  // files its pairs sit in: with nested copies left out of pairing, that can be fewer files than\n  // the input groups covered, and a merged group that is no longer reportable must not form.\n  if (!isReportableGroup(merged)) {\n    return undefined;\n  }\n  const pairedRetained = [\n    ...(firstReplaced ? [] : pairs.map(([leading]) => leading)),\n    ...(secondReplaced ? [] : pairs.map(([, trailing]) => trailing)),\n  ];\n  return { merged, firstReplaced, secondReplaced, pairedRetained };\n}\n\n/** One pass over a group: its pairable occurrences and its non-nested occurrence count. */\nfunction pairableOccurrences<T extends CountedOccurrence>(group: T[]): [T[], number] {\n  const pairable: T[] = [];\n  let length = 0;\n  for (const occurrence of group) {\n    if (!occurrence.nestedInLargerGroup) {\n      length += 1;\n    }\n    // Either flag keeps an occurrence out of pairing: its span is already counted elsewhere, or it\n    // is a nested copy of content the merged span would not cover.\n    if (!occurrence.spanCountedElsewhere && !occurrence.nestedInLargerGroup) {\n      pairable.push(occurrence);\n    }\n  }\n  return [pairable, length];\n}\n\n/**\n * Redundant copies one group adds to duplicateBlockCount. Each redundant occurrence contributes\n * one count per matched fragment, so merging a gapped clone's fragments into one group does not\n * halve duplicateBlockCount: an edited two-fragment pair still counts 2, exactly as its unmerged\n * fragments did. Occurrence shapes can differ within one group (a\n * gap-merged exact pair plus an appended whole-block near-miss copy), so every occurrence's\n * fragments are summed and one representative — the largest — is deducted, keeping the count\n * independent of source order. Occurrences a partial gapped merge also paired into a merged group\n * are skipped: their spans are already counted there, and such a retained group deducts no\n * representative of its own — the merged group's representative already stands for the shared\n * content — so no token span contributes to the count twice.\n */\nexport function countRedundantFragments(group: CountedOccurrence[]): number {\n  let fragmentCount = 0;\n  let maxFragmentCount = 0;\n  let hasSharedOccurrence = false;\n  for (const occurrence of group) {\n    if (occurrence.spanCountedElsewhere) {\n      hasSharedOccurrence = true;\n      continue;\n    }\n    fragmentCount += occurrence.segments.length;\n    maxFragmentCount = Math.max(maxFragmentCount, occurrence.segments.length);\n  }\n  return hasSharedOccurrence ? fragmentCount : fragmentCount - maxFragmentCount;\n}\n\n/** Adds the 1-based code lines the segment's matched tokens cover; shared with cross-file coverage. */\nexport function collectSegmentLines(\n  segment: { startTokenIndex: number; endTokenIndex: number },\n  tokens: Token[],\n  codeLineNumbers: Set<number> | undefined,\n  duplicatedLines: Set<number>\n): void {\n  for (let index = segment.startTokenIndex; index < segment.endTokenIndex; index += 1) {\n    const token = tokens[index];\n    for (let row = token?.startRow ?? 0; row <= (token?.endRow ?? -1); row += 1) {\n      if (!codeLineNumbers || codeLineNumbers.has(row + 1)) {\n        duplicatedLines.add(row + 1);\n      }\n    }\n  }\n}\n"],"mappings":"AASA,MAAa,EAA0D,CACrE,UAAW,GACX,aAAc,GACd,qBAAsB,EACxB,EAQA,SAAgB,EAA0B,EAA4D,CACpG,MAAO,CACL,UAAW,EAAc,GAAS,UAAW,EAA0B,SAAS,EAChF,aAAc,EAAc,GAAS,aAAc,EAA0B,YAAY,EACzF,qBAAsB,EAAc,GAAS,qBAAsB,EAA0B,oBAAoB,CACnH,CACF,CAEA,SAAS,EAAc,EAA2B,EAA0B,CAC1E,OAAO,IAAU,IAAA,IAAa,OAAO,MAAM,CAAK,EAC5C,EACA,KAAK,IAAI,KAAK,IAAI,KAAK,MAAM,CAAK,EAAG,CAAC,EAAG,UAAa,CAC5D,CAiBA,SAAS,EAAe,EAAsB,EAA6B,CACzE,OAAO,EAAe,GAAK,CAC7B,CAuHA,SAAgB,EAAwB,EAA6B,CACnE,IAAM,EAAS,IAAI,WAAW,EAAO,OAAS,CAAC,EAC/C,IAAK,GAAM,CAAC,EAAO,KAAU,EAAO,QAAQ,EAC1C,EAAO,EAAQ,IAAM,EAAO,IAAU,IAAM,EAAM,cAAgB,IAAA,GAAY,EAAI,GAEpF,OAAO,CACT,CA0CA,SAAgB,EACd,EACA,EACA,EAC+B,CAC/B,IAAM,EAA4C,CAAC,EAC7C,EAAoC,CAAC,EACrC,EAA6B,CAAC,EACpC,IAAK,GAAM,CAAC,EAAc,KAAY,EAAS,QAAQ,EACrD,IAAK,IAAM,KAAc,EAAQ,WAC/B,EAAwB,KAAK,CAAY,EACzC,EAAW,KAAK,CAAU,EAG9B,IAAM,EAAa,GACjB,EAAS,EAAwB,IAAmB,GAChD,EAAyB,IAAI,IAC7B,EAAmB,EAAW,KAAK,EAAY,IACnD,EAA0B,EAAU,CAAc,CAAC,EAAE,QAAU,CAAC,EAAG,EAAY,CAAS,CAC1F,EACA,IAAK,GAAM,CAAC,EAAgB,KAAY,EAAiB,QAAQ,EAAG,CAClE,IAAM,EAAe,EAAwB,IAAmB,EAChE,IAAK,GAAM,CAAC,EAAO,KAAQ,EAAQ,kBAAkB,QAAQ,EAC3D,IAAK,IAAM,KAAa,EAAK,CAC3B,GAAI,IAAc,IAAA,GAChB,SAEF,IAAM,EAAc,EAAuB,IAAI,CAAS,EACpD,GACF,EAAY,OAAS,EACjB,EAAY,iBAAmB,IACjC,EAAY,eAAiB,IAE3B,EAAY,eAAiB,IAC/B,EAAY,aAAe,IAE7B,EAAY,SAAW,KAAK,IAAI,EAAY,SAAU,CAAK,EAC3D,EAAY,SAAW,KAAK,IAAI,EAAY,SAAU,CAAK,GAE3D,EAAuB,IAAI,EAAW,CACpC,MAAO,EACP,iBACA,eACA,SAAU,EACV,SAAU,CACZ,CAAC,CAEL,CAEJ,CAMA,IAAM,GAAW,EAA+B,IAA4B,CAC1E,GAAI,IAAc,IAAA,GAChB,MAAO,GAET,IAAM,EAAc,EAAuB,IAAI,CAAS,EAOxD,OANI,IAAgB,IAAA,IAAa,EAAY,MAAQ,EAC5C,GAEL,EACK,EAAY,eAAiB,GAE/B,EAAY,iBAAmB,IAAM,EAAY,SAAW,EAAY,UAAY,CAC7F,EAKM,EAAyB,GAAoC,CACjE,IAAM,EAAS,EAAiB,EAAO,eAAe,EAAE,iBAAmB,CAAC,EACtE,EAAY,EAAO,EAAO,OAChC,IAAK,IAAI,EAAQ,EAAO,MAAQ,EAAG,EAAQ,EAAO,MAAQ,EAAO,OAAQ,GAAS,EAChF,GAAI,EAAO,KAAW,EACpB,MAAO,GAGX,MAAO,EACT,EAEM,EAAmC,CAAC,EAC1C,IAAK,GAAM,CAAC,EAAgB,KAAY,EAAiB,QAAQ,EAC/D,IAAK,GAAM,CAAC,EAAO,KAAQ,EAAQ,kBAAkB,QAAQ,EAC3D,IAAK,GAAM,CAAC,EAAQ,KAAc,EAAI,QAAQ,EAAG,CAC/C,GAAI,CAAC,EAAQ,EAAW,CAAM,GAAK,CAAC,EAAsB,CAAE,iBAAgB,QAAO,QAAO,CAAC,EACzF,SAIF,IAAM,EAAgB,EAAQ,kBAAkB,EAAM,GAAG,EAAS,GAC5D,EAAe,EAAQ,kBAAkB,EAAQ,EAAE,GAAG,EAAS,GACjE,EAAQ,EAAe,EAAS,CAAC,GAAK,EAAQ,EAAc,EAAS,CAAC,GAG1E,EAAe,KAAK,CAAE,iBAAgB,QAAO,QAAO,CAAC,CACvD,CASJ,IAAM,EAAU,IAAI,IAAI,EAAe,IAAI,CAAQ,CAAC,EAChD,EAAW,EACf,KAAO,EAAS,OAAS,GAAG,CAC1B,IAAM,EAA4B,CAAC,EACnC,IAAK,IAAM,KAAU,EAAU,CAC7B,IAAM,EAAa,EAAW,EAAO,gBAC/B,EAAQ,IAAa,EAAO,OAC5B,EAAO,IAAa,EAAO,MAAQ,EAAO,OAAS,GACnD,EAAU,EAAU,EAAO,cAAc,EAC/C,GAAI,CAAC,GAAS,CAAC,GAAQ,CAAC,EACtB,SAEF,IAAM,EAAc,KAAK,EAAe,EAAQ,OAAQ,EAAQ,mBAAoB,EAAM,gBAAiB,EAAK,aAAa,IAC7H,EAAW,KAAK,CACd,UAAW,EAAY,EAAa,EAAM,gBAAiB,EAAK,cAAe,EAAO,CAAI,EAC1F,aAAc,EAAwB,EAAO,iBAAmB,CAClE,CAAC,EACD,EAAQ,KAAK,CAAM,CACrB,CACA,EAAW,CAAC,EACZ,IAAK,IAAM,KAAU,EACnB,IAAK,IAAM,IAAS,CAAC,EAAO,MAAO,EAAO,MAAQ,CAAC,EAAG,CACpD,IAAM,EAAY,CAAE,eAAgB,EAAO,eAAgB,QAAO,OAAQ,EAAO,OAAS,CAAE,EACtF,EAAe,EAAiB,EAAO,eAAe,EAAE,kBAAkB,EAAM,GAAG,EAAU,QAEjG,GAAQ,IAAI,EAAS,CAAS,CAAC,GAC9B,EAAQ,EAAc,EAAU,MAAM,GACtC,EAAsB,CAAS,IAIlC,EAAQ,IAAI,EAAS,CAAS,CAAC,EAC/B,EAAS,KAAK,CAAS,EACzB,CAEJ,CACA,OAAO,CACT,CAEA,SAAS,EAAS,EAAgC,CAChD,MAAO,GAAG,EAAO,eAAe,GAAG,EAAO,MAAM,GAAG,EAAO,QAC5D,CASA,SAAS,EAA0B,EAAiB,EAA0B,EAAqC,CACjH,IAAM,EAAkB,EAAW,IAAK,GACtC,EAAgB,EAAQ,EAAU,gBAAiB,EAAU,aAAa,CAC5E,EACM,EAA8C,CAAC,EACrD,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAW,OAAQ,GAAS,EAAG,CACzD,IAAM,EAA8B,CAAC,EACjC,EAAO,KACP,EAAa,EACX,EAAS,KAAK,IAAI,EAAW,OAAQ,EAAQ,GAAyB,EAC5E,IAAK,IAAI,EAAM,EAAO,EAAM,EAAQ,GAAO,EAAG,CAC5C,IAAM,EAAY,EAAW,GACvB,EAAgB,EAAgB,GACtC,GAAI,CAAC,GAAa,IAAkB,IAAA,GAClC,MAEF,EAAO,EAAc,EAAM,CAAa,EACxC,GAAc,EAAU,cAAgB,EAAU,gBAClD,IAAM,EAAiB,EAAM,EAAQ,EACrC,EAAI,GACF,GAAkB,GAA6B,GAAc,EACzD,EAAc,EAAM,CAAc,EAClC,IAAA,EACR,CACA,EAAkB,KAAK,CAAG,CAC5B,CACA,MAAO,CAAE,oBAAmB,iBAAgB,CAC9C,CAOA,SAAgB,EAAU,EAAe,EAAuB,CAC9D,OAAO,EAAuB,CAAC,CAAC,CAAC,CAAC,CACpC,CAMA,SAAgB,EAAuB,EAA0C,CAC/E,IAAM,EAAa,EAAE,OAAS,KAAQ,EAChC,EAAgB,IAAI,IAC1B,IAAK,GAAM,CAAC,EAAO,KAAW,EAAE,QAAQ,EAAG,CACzC,IAAI,EAAO,EAAc,IAAI,CAAM,EAC9B,IACH,EAAO,IAAI,YAAY,CAAS,EAChC,EAAc,IAAI,EAAQ,CAAI,GAEhC,IAAM,EAAO,IAAU,EAIvB,EAAK,IAAS,EAAK,IAAS,GAAM,IAAM,EAAQ,GAClD,CAEA,IAAM,EAAI,IAAI,YAAY,CAAS,EACnC,MAAQ,IAAM,CACZ,EAAE,KAAK,CAAC,EACR,IAAK,IAAM,KAAU,EAAG,CACtB,IAAM,EAAY,EAAc,IAAI,CAAM,EAEtC,EAAa,EACb,EAAS,EACb,IAAK,IAAI,EAAO,EAAG,EAAO,EAAW,GAAQ,EAAG,CAC9C,IAAM,EAAW,EAAE,IAAS,EAEtB,IAAM,IAAY,IAAS,GAAK,KAAc,EAE9C,GAAY,GAAY,EAAK,KAAgB,EACnD,EAAa,IAAa,GAC1B,IAAM,EAAa,EAAI,EAAU,EACjC,EAAS,IAAa,GAEtB,EAAE,GAAQ,EAAI,CAAC,CACjB,CACF,CAEA,IAAI,EAAS,EACb,IAAK,IAAM,KAAQ,EACjB,GAAU,EAAS,CAAI,EAEzB,OAAO,CACT,CACF,CAEA,SAAS,EAAS,EAAuB,CACvC,IAAI,EAAQ,GAAU,IAAU,EAAK,YAErC,MADA,IAAS,EAAQ,YAAmB,IAAU,EAAK,WAC3C,KAAK,KAAM,GAAS,IAAU,GAAM,UAAe,QAAa,IAAM,GAAM,GACtF,CAEA,SAAS,EACP,EACA,EACA,EACA,EACA,EACoB,CACpB,MAAO,CACL,cACA,WAAY,EAAgB,EAC5B,kBACA,gBACA,WAAY,EAAM,WAClB,SAAU,EAAK,SACf,UAAW,EAAM,UACjB,QAAS,EAAK,OAChB,CACF,CAGA,MAAM,EAAkC,CAAC,EACnC,EAAmC,CAAC,EAE1C,SAAS,EAAoB,EAAuB,CAClD,IAAI,EAAO,EAAsB,GAKjC,OAJI,IAAS,IAAA,KACX,EAAO,EAAS,IAAI,GAAO,EAC3B,EAAsB,GAAS,GAE1B,CACT,CAEA,SAAS,EAAqB,EAAuB,CACnD,IAAI,EAAO,EAAuB,GAKlC,OAJI,IAAS,IAAA,KACX,EAAO,EAAU,IAAI,GAAO,EAC5B,EAAuB,GAAS,GAE3B,CACT,CAUA,SAAS,EACP,EACA,EACA,EACA,EACQ,CAGR,GAAM,CAAC,EAAS,GAAa,EAAoB,EAAQ,EAAiB,EADrD,GADC,EAAmB,IAAkB,IAAM,EAAmB,IAAoB,GACtD,EAAgB,CACkC,CAAC,EACrG,MAAO,GAAG,EAAQ,GAAG,EAAU,GAAG,EAAgB,GACpD,CASA,SAAS,EAAgB,EAAiB,EAAyB,EAA+B,CAChG,GAAM,CAAC,EAAS,GAAa,EAAoB,EAAQ,EAAiB,EAAe,EAAK,EAE9F,OAAO,EAAU,KAAK,KAAK,EAAW,EAAE,CAC1C,CAEA,SAAS,EACP,EACA,EACA,EACA,EACkB,CAClB,IAAM,EAAoB,IAAI,IAC1B,EAAU,KACV,EAAY,MAChB,IAAK,IAAI,EAAQ,EAAiB,EAAQ,EAAe,GAAS,EAAG,CACnE,IAAM,EAAQ,EAAO,GACrB,GAAI,CAAC,EACH,SAIF,IAAI,EACA,EACJ,GAAI,EAAM,OAAS,KAAM,CACvB,IAAI,EAAkB,EAAkB,IAAI,EAAM,IAAI,EAClD,IAAoB,IAAA,KACtB,EAAkB,EAAkB,KACpC,EAAkB,IAAI,EAAM,KAAM,CAAe,GAEnD,EAAO,EAAoB,CAAe,EAC1C,EAAQ,EAAqB,CAAe,CAC9C,KACE,GAAO,EAAM,SACb,EAAQ,EAAM,UAGhB,EAAW,KAAK,KAAK,EAAS,EAAE,EAAI,EAAQ,EAC5C,EAAY,KAAK,KAAK,EAAW,EAAE,EAAI,EACnC,GAAqB,EAAM,cAAgB,IAAA,IAAa,EAAM,eAAiB,IAAA,KAEjF,EAAW,KAAK,KAAK,EAAS,EAAE,EAAI,EAAM,YAAe,EACzD,EAAY,KAAK,KAAK,EAAW,EAAE,EAAI,EAAM,aAEjD,CACA,MAAO,CAAC,EAAS,CAAS,CAC5B,CAGA,SAAS,EAAS,EAAsB,CACtC,IAAI,EAAO,KACX,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAK,OAAQ,GAAS,EAEhD,EAAO,KAAK,KAAK,EAAM,EAAE,EAAI,EAAK,WAAW,CAAK,EAEpD,OAAO,CACT,CAGA,SAAS,EAAU,EAAsB,CACvC,IAAI,EAAO,YACX,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAK,OAAQ,GAAS,EAEhD,EAAO,KAAK,KAAK,EAAO,EAAK,WAAW,CAAK,EAAG,QAAU,EAE5D,OAAO,CACT,CAEA,SAAS,EAAc,EAAc,EAAuB,CAC1D,OAAO,KAAK,KAAK,EAAM,EAAE,EAAI,CAC/B,CAoBA,SAAgB,EACd,EACA,EACA,MAAmD,GAC5C,CACP,GAAI,GAAgB,GAAK,EAAO,OAAS,EACvC,OAAO,EAGT,EAAO,KAAK,CAAa,EACzB,IAAK,IAAI,EAAU,GAAM,GAAU,CACjC,EAAU,GACV,IAAM,EAAkB,EAA2B,EAAQ,CAAY,EACvE,IAAK,IAAI,EAAY,EAAG,EAAY,EAAO,QAAU,CAAC,EAAS,GAAa,EAC1E,IAAK,IAAM,KAAc,EAAgB,IAAc,CAAC,EAAG,CACzD,IAAM,EAAO,EAAO,GACd,EAAQ,EAAO,GACrB,GAAI,CAAC,GAAQ,CAAC,EACZ,SAEF,IAAM,EAAU,EAAY,EAAM,EAAO,EAAc,CAAiB,EAClE,EAAS,GAAW,EAAY,EAAO,EAAM,EAAc,CAAiB,EAClF,GAAI,CAAC,EACH,SAEF,IAAM,EAAe,EAAU,EAAO,cAAgB,EAAO,eACvD,EAAgB,EAAU,EAAO,eAAiB,EAAO,cAC3D,GAAgB,GAClB,EAAO,GAAa,EAAO,OAC3B,EAAO,OAAO,EAAY,CAAC,GAClB,EACT,EAAO,GAAc,EAAO,OACnB,EACT,EAAO,GAAa,EAAO,OAI3B,EAAO,KAAK,EAAO,MAAM,EAM3B,IAAK,IAAM,KAAc,EAAO,eAC9B,EAAW,qBAAuB,GAEpC,EAAO,KAAK,CAAa,EACzB,EAAU,GACV,KACF,CAEJ,CACA,OAAO,CACT,CAQA,SAAS,EAA2B,EAA+B,EAAkC,CACnG,IAAM,EAA4D,CAAC,EACnE,IAAK,GAAM,CAAC,EAAY,KAAU,EAAO,QAAQ,EAC/C,IAAK,IAAM,KAAc,EACvB,EAAO,KAAK,CAAE,gBAAiB,EAAW,gBAAiB,YAAW,CAAC,EAG3E,EAAO,MAAM,EAAM,IAAU,EAAK,gBAAkB,EAAM,eAAe,EACzE,IAAM,EAAW,EAAO,QAAU,IAAI,GAAa,EACnD,IAAK,GAAM,CAAC,EAAY,KAAU,EAAO,QAAQ,EAC/C,IAAK,IAAM,KAAc,EACvB,IACE,IAAI,EAAQ,EAAkB,EAAQ,EAAW,aAAa,EAC9D,EAAQ,EAAO,SAAW,EAAO,EAAM,EAAE,iBAAmB,IAAM,EAAW,cAAgB,EAC7F,GAAS,EACT,CACA,IAAM,EAAQ,EAAO,EAAM,EAAE,YAAc,EACvC,IAAU,GACZ,EAAS,KAAK,IAAI,EAAO,CAAU,EAAE,EAAE,IAAI,KAAK,IAAI,EAAO,CAAU,CAAC,CAE1E,CAGJ,OAAO,EAAS,IAAK,GAAQ,CAAC,GAAG,CAAG,CAAC,CAAC,UAAU,EAAM,IAAU,EAAO,CAAK,CAAC,CAC/E,CAEA,SAAS,EAAkB,EAAuC,EAAwB,CACxF,IAAI,EAAM,EACN,EAAO,EAAO,OAClB,KAAO,EAAM,GAAM,CACjB,IAAM,EAAU,EAAM,IAAU,GAC3B,EAAO,EAAO,EAAE,iBAAmB,GAAK,EAC3C,EAAM,EAAS,EAEf,EAAO,CAEX,CACA,OAAO,CACT,CAEA,SAAS,EAAc,EAA2B,EAAoC,CACpF,IAAM,EAAY,EAAK,GACjB,EAAa,EAAM,GACzB,OACG,GAAW,iBAAmB,IAAM,GAAY,iBAAmB,KACnE,GAAW,eAAiB,IAAM,GAAY,eAAiB,EAEpE,CAsBA,SAAS,EACP,EACA,EACA,EACA,EAC4B,CAQ5B,GAAM,CAAC,EAAU,GAAe,EAAoB,CAAK,EACnD,CAAC,EAAW,GAAgB,EAAoB,CAAM,EACtD,EAAkB,CAAC,EACrB,EAAe,EACf,EAAsB,GAC1B,IAAK,IAAM,KAAY,EAAW,CAEhC,KAAO,EAAe,EAAS,QAAQ,CACrC,IAAM,EAAU,EAAS,GACzB,GAAI,GAAW,EAAQ,cAAgB,EAAe,EAAS,gBAC7D,GAAgB,OAEhB,KAEJ,CACA,IAAM,EAAU,EAAS,GAEvB,GACA,EAAQ,eAAiB,EAAS,iBAClC,EAAQ,iBAAmB,IAE3B,EAAM,KAAK,CAAC,EAAS,CAAQ,CAAC,EAC9B,EAAsB,EAAS,cAC/B,GAAgB,EAEpB,CACA,IAAM,EAAmB,EAAM,SAAW,EACpC,EAAoB,EAAM,SAAW,EAC3C,GAAI,EAAM,OAAS,GAAM,CAAC,GAAoB,CAAC,EAC7C,OAKF,IAAM,EAAgB,GAAoB,CAAC,EAAM,KAAM,GAAe,EAAW,mBAAmB,EAC9F,EAAiB,GAAqB,CAAC,EAAO,KAAM,GAAe,EAAW,mBAAmB,EACjG,EAAS,EAAM,KAAK,CAAC,EAAS,MAAe,CACjD,GAAG,EAEH,qBAAsB,IAAA,GACtB,oBAAqB,IAAA,GACrB,SAAU,CAAC,GAAG,EAAQ,SAAU,GAAG,EAAS,QAAQ,EACpD,WAAY,EAAQ,WAAa,EAAS,WAC1C,cAAe,EAAS,cACxB,SAAU,EAAS,SACnB,QAAS,EAAS,OACpB,EAAE,EAIG,KAAkB,CAAM,EAO7B,MAAO,CAAE,SAAQ,gBAAe,iBAAgB,eAAA,CAH9C,GAAI,EAAgB,CAAC,EAAI,EAAM,KAAK,CAAC,KAAa,CAAO,EACzD,GAAI,EAAiB,CAAC,EAAI,EAAM,KAAK,EAAG,KAAc,CAAQ,CAEH,CAAE,CACjE,CAGA,SAAS,EAAiD,EAA2B,CACnF,IAAM,EAAgB,CAAC,EACnB,EAAS,EACb,IAAK,IAAM,KAAc,EAClB,EAAW,sBACd,GAAU,GAIR,CAAC,EAAW,sBAAwB,CAAC,EAAW,qBAClD,EAAS,KAAK,CAAU,EAG5B,MAAO,CAAC,EAAU,CAAM,CAC1B,CAcA,SAAgB,EAAwB,EAAoC,CAC1E,IAAI,EAAgB,EAChB,EAAmB,EACnB,EAAsB,GAC1B,IAAK,IAAM,KAAc,EAAO,CAC9B,GAAI,EAAW,qBAAsB,CACnC,EAAsB,GACtB,QACF,CACA,GAAiB,EAAW,SAAS,OACrC,EAAmB,KAAK,IAAI,EAAkB,EAAW,SAAS,MAAM,CAC1E,CACA,OAAO,EAAsB,EAAgB,EAAgB,CAC/D,CAGA,SAAgB,EACd,EACA,EACA,EACA,EACM,CACN,IAAK,IAAI,EAAQ,EAAQ,gBAAiB,EAAQ,EAAQ,cAAe,GAAS,EAAG,CACnF,IAAM,EAAQ,EAAO,GACrB,IAAK,IAAI,EAAM,GAAO,UAAY,EAAG,IAAQ,GAAO,QAAU,IAAK,GAAO,GACpE,CAAC,GAAmB,EAAgB,IAAI,EAAM,CAAC,IACjD,EAAgB,IAAI,EAAM,CAAC,CAGjC,CACF"}

package/dist/metrics.cjs

@@ -1,2 +1,2 @@
-"use strict";const e=require("./languages.cjs"),t=require("./nativeMetrics.cjs");var n=class{registry=e.createLanguageRegistry();getSupportedLanguages(){return[...new Set([...this.registry.values()].map(e=>e.name))]}measure(e,n){let i=this.resolveLanguage(n.language),a=n.includeSyntaxTree??!1;return r(t.measureCodeNative(e,i.name,a,n.duplication),a)}collectDuplicationCandidates(e,t){return this.collectCrossFileDuplicationFileData(e,t).candidates}collectCrossFileDuplicationFileData(e,n){let r=this.resolveLanguage(n.language),i=t.collectCrossFileDataNative(e,r.name,n.duplication?.minTokens);return{candidates:i.candidates,tokens:i.tokens,containerStatements:i.containerStatements,codeLineNumbers:new Set(i.codeLineNumbers)}}collectFunctionTokenSequences(e,n){let r=this.resolveLanguage(n.language);return t.collectFunctionTokenSequencesNative(e,r.name)}resolveLanguage(e){let t=this.registry.get(e);if(!t)throw Error(`Unsupported language: ${e}`);return t}};function r(e,t){return{language:e.language,bytes:e.bytes,lines:e.lines,functions:e.functions.map(e=>({name:e.name,nodeType:e.nodeType,startLine:e.startLine,startColumn:e.startColumn,endLine:e.endLine,endColumn:e.endColumn,cyclomaticComplexity:e.cyclomaticComplexity,cognitiveComplexity:e.cognitiveComplexity,nestingDepth:e.nestingDepth,ncss:e.ncss,parameterCount:e.parameterCount,halstead:i(e.halsteadCounts),depDegree:e.depDegree})),cyclomaticComplexity:e.cyclomaticComplexity,cognitiveComplexity:e.cognitiveComplexity,maxCognitiveComplexity:e.maxCognitiveComplexity,nestingDepth:e.nestingDepth,ncssCount:e.ncssCount,duplication:e.duplication,halstead:i(e.halsteadCounts),syntaxTree:t?e.syntaxTree:void 0}}function i(e){let{distinctOperators:t,distinctOperands:n,totalOperators:r,totalOperands:i}=e,a=t+n,o=r+i,s=a===0?0:o*Math.log2(a);return{distinctOperators:t,distinctOperands:n,totalOperators:r,totalOperands:i,vocabulary:a,length:o,volume:s,effort:(n===0?0:t/2*(i/n))*s}}const a=new n;function o(e,t){return a.measure(e,t)}function s(e,t){return a.collectDuplicationCandidates(e,t)}function c(e,t){return a.collectFunctionTokenSequences(e,t)}function l(e,t){return a.collectCrossFileDuplicationFileData(e,t)}exports.TreeMeasurer=n,exports.collectCrossFileDuplicationFileData=l,exports.collectDuplicationCandidates=s,exports.collectFunctionTokenSequences=c,exports.defaultMeasurer=a,exports.measureCode=o;
+"use strict";const e=require("./languages.cjs"),t=require("./nativeMetrics.cjs");var n=class{registry=e.createLanguageRegistry();getSupportedLanguages(){return[...new Set([...this.registry.values()].map(e=>e.name))]}measure(e,n){let r=this.resolveLanguage(n.language),a=n.includeSyntaxTree??!1;return i(t.measureCodeNative(e,r.name,a,n.duplication),a)}measureWithCrossFileData(e,n){let a=this.resolveLanguage(n.language),o=n.includeSyntaxTree??!1,s=t.measureCodeNative(e,a.name,o,n.duplication,!0);if(!s.crossFileData)throw Error(`The native addon returned no cross-file duplication data`);return{metrics:i(s,o),crossFileData:r(s.crossFileData)}}collectDuplicationCandidates(e,t){return this.collectCrossFileDuplicationFileData(e,t).candidates}collectCrossFileDuplicationFileData(e,n){let i=this.resolveLanguage(n.language);return r(t.collectCrossFileDataNative(e,i.name,n.duplication?.minTokens))}collectFunctionTokenSequences(e,n){let r=this.resolveLanguage(n.language);return t.collectFunctionTokenSequencesNative(e,r.name)}resolveLanguage(e){let t=this.registry.get(e);if(!t)throw Error(`Unsupported language: ${e}`);return t}};function r(e){return{candidates:e.candidates,tokens:e.tokens,containerStatements:e.containerStatements,nearMissBlocks:e.nearMissBlocks,codeLineNumbers:new Set(e.codeLineNumbers)}}function i(e,t){return{language:e.language,bytes:e.bytes,lines:e.lines,functions:e.functions.map(e=>({name:e.name,nodeType:e.nodeType,startLine:e.startLine,startColumn:e.startColumn,endLine:e.endLine,endColumn:e.endColumn,cyclomaticComplexity:e.cyclomaticComplexity,cognitiveComplexity:e.cognitiveComplexity,nestingDepth:e.nestingDepth,ncss:e.ncss,parameterCount:e.parameterCount,halstead:a(e.halsteadCounts),depDegree:e.depDegree})),cyclomaticComplexity:e.cyclomaticComplexity,cognitiveComplexity:e.cognitiveComplexity,maxCognitiveComplexity:e.maxCognitiveComplexity,nestingDepth:e.nestingDepth,ncssCount:e.ncssCount,duplication:e.duplication,halstead:a(e.halsteadCounts),syntaxTree:t?e.syntaxTree:void 0}}function a(e){let{distinctOperators:t,distinctOperands:n,totalOperators:r,totalOperands:i}=e,a=t+n,o=r+i,s=a===0?0:o*Math.log2(a);return{distinctOperators:t,distinctOperands:n,totalOperators:r,totalOperands:i,vocabulary:a,length:o,volume:s,effort:(n===0?0:t/2*(i/n))*s}}const o=new n;function s(e,t){return o.measure(e,t)}function c(e,t){return o.measureWithCrossFileData(e,t)}function l(e,t){return o.collectDuplicationCandidates(e,t)}function u(e,t){return o.collectFunctionTokenSequences(e,t)}function d(e,t){return o.collectCrossFileDuplicationFileData(e,t)}exports.TreeMeasurer=n,exports.collectCrossFileDuplicationFileData=d,exports.collectDuplicationCandidates=l,exports.collectFunctionTokenSequences=u,exports.defaultMeasurer=o,exports.measureCode=s,exports.measureCodeWithCrossFileData=c;
 //# sourceMappingURL=metrics.cjs.map

package/dist/metrics.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"metrics.cjs","names":["createLanguageRegistry","measureCodeNative","collectCrossFileDataNative","collectFunctionTokenSequencesNative"],"sources":["../src/metrics.ts"],"sourcesContent":["import type { CrossFileDuplicateCandidate, CrossFileDuplicationFileData } from './duplication.js';\nimport { createLanguageRegistry } from './languages.js';\nimport {\n  collectCrossFileDataNative,\n  collectFunctionTokenSequencesNative,\n  measureCodeNative,\n  type NativeHalsteadCounts,\n  type NativeMetricsPayload,\n} from './nativeMetrics.js';\nimport type { CodeMetrics, HalsteadMetrics, LanguageDefinition, LanguageName, MeasureOptions } from './types.js';\n\n/**\n * Measures code metrics for the built-in languages. Parsing and metric passes run in the bundled\n * Rust addon (tree-sitter); this class resolves language aliases, crosses the N-API boundary, and\n * derives the Halstead float metrics from the natively measured counts.\n */\nexport class TreeMeasurer {\n  private readonly registry = createLanguageRegistry();\n\n  getSupportedLanguages(): LanguageName[] {\n    return [...new Set([...this.registry.values()].map((language) => language.name))];\n  }\n\n  measure(code: string, options: MeasureOptions): CodeMetrics {\n    const language = this.resolveLanguage(options.language);\n    const includeSyntaxTree = options.includeSyntaxTree ?? false;\n    const payload = measureCodeNative(code, language.name, includeSyntaxTree, options.duplication);\n    return assembleNativeMetrics(payload, includeSyntaxTree);\n  }\n\n  /** Collects one file's duplicate candidates for cross-file clone detection. */\n  collectDuplicationCandidates(code: string, options: MeasureOptions): CrossFileDuplicateCandidate[] {\n    return this.collectCrossFileDuplicationFileData(code, options).candidates;\n  }\n\n  /**\n   * Collects one file's duplicate candidates, normalized tokens, and statement structure for\n   * cross-file clone detection with measureCrossFileDuplication.\n   */\n  collectCrossFileDuplicationFileData(code: string, options: MeasureOptions): CrossFileDuplicationFileData {\n    const language = this.resolveLanguage(options.language);\n    const payload = collectCrossFileDataNative(code, language.name, options.duplication?.minTokens);\n    return {\n      candidates: payload.candidates,\n      tokens: payload.tokens,\n      containerStatements: payload.containerStatements,\n      // Lets cross-file line coverage count only code lines, like within-file coverage.\n      codeLineNumbers: new Set(payload.codeLineNumbers),\n    };\n  }\n\n  /**\n   * Normalized token hash sequences of every function, index-parallel to the functions array of\n   * measure(): identifiers are anonymized by first occurrence within the function, literals by\n   * kind, and keywords/operators kept verbatim, so the regression gate can re-match renamed or\n   * moved functions across two revisions by token-LCS similarity.\n   */\n  collectFunctionTokenSequences(code: string, options: MeasureOptions): Int32Array[] {\n    const language = this.resolveLanguage(options.language);\n    return collectFunctionTokenSequencesNative(code, language.name);\n  }\n\n  private resolveLanguage(name: LanguageName): LanguageDefinition {\n    const language = this.registry.get(name);\n    if (!language) {\n      throw new Error(`Unsupported language: ${name}`);\n    }\n    return language;\n  }\n}\n\n/**\n * Completes a native measurement into CodeMetrics. The object is rebuilt field by field (rather\n * than spread from the parsed JSON) so the result has a stable shape, including\n * explicitly-undefined optional keys.\n */\nfunction assembleNativeMetrics(payload: NativeMetricsPayload, includeSyntaxTree: boolean): CodeMetrics {\n  return {\n    language: payload.language,\n    bytes: payload.bytes,\n    lines: payload.lines,\n    functions: payload.functions.map((fn) => ({\n      name: fn.name,\n      nodeType: fn.nodeType,\n      startLine: fn.startLine,\n      startColumn: fn.startColumn,\n      endLine: fn.endLine,\n      endColumn: fn.endColumn,\n      cyclomaticComplexity: fn.cyclomaticComplexity,\n      cognitiveComplexity: fn.cognitiveComplexity,\n      nestingDepth: fn.nestingDepth,\n      ncss: fn.ncss,\n      parameterCount: fn.parameterCount,\n      halstead: deriveHalsteadMetrics(fn.halsteadCounts),\n      depDegree: fn.depDegree,\n    })),\n    cyclomaticComplexity: payload.cyclomaticComplexity,\n    cognitiveComplexity: payload.cognitiveComplexity,\n    maxCognitiveComplexity: payload.maxCognitiveComplexity,\n    nestingDepth: payload.nestingDepth,\n    ncssCount: payload.ncssCount,\n    duplication: payload.duplication,\n    halstead: deriveHalsteadMetrics(payload.halsteadCounts),\n    syntaxTree: includeSyntaxTree ? payload.syntaxTree : undefined,\n  };\n}\n\nfunction deriveHalsteadMetrics(counts: NativeHalsteadCounts): HalsteadMetrics {\n  const { distinctOperators, distinctOperands, totalOperators, totalOperands } = counts;\n  const vocabulary = distinctOperators + distinctOperands;\n  const length = totalOperators + totalOperands;\n  const volume = vocabulary === 0 ? 0 : length * Math.log2(vocabulary);\n  const difficulty = distinctOperands === 0 ? 0 : (distinctOperators / 2) * (totalOperands / distinctOperands);\n  const effort = difficulty * volume;\n\n  return {\n    distinctOperators,\n    distinctOperands,\n    totalOperators,\n    totalOperands,\n    vocabulary,\n    length,\n    volume,\n    effort,\n  };\n}\n\nexport const defaultMeasurer = new TreeMeasurer();\n\nexport function measureCode(code: string, options: MeasureOptions): CodeMetrics {\n  return defaultMeasurer.measure(code, options);\n}\n\n/** Standalone helper mirroring measureCode for the default measurer. */\nexport function collectDuplicationCandidates(code: string, options: MeasureOptions): CrossFileDuplicateCandidate[] {\n  return defaultMeasurer.collectDuplicationCandidates(code, options);\n}\n\n/** Standalone helper mirroring measureCode for the default measurer. */\nexport function collectFunctionTokenSequences(code: string, options: MeasureOptions): Int32Array[] {\n  return defaultMeasurer.collectFunctionTokenSequences(code, options);\n}\n\n/** Standalone helper mirroring measureCode for the default measurer. */\nexport function collectCrossFileDuplicationFileData(\n  code: string,\n  options: MeasureOptions\n): CrossFileDuplicationFileData {\n  return defaultMeasurer.collectCrossFileDuplicationFileData(code, options);\n}\n"],"mappings":"iFAgBA,IAAa,EAAb,KAA0B,CACxB,SAA4BA,EAAAA,uBAAuB,EAEnD,uBAAwC,CACtC,MAAO,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,KAAK,SAAS,OAAO,CAAC,CAAC,CAAC,IAAK,GAAa,EAAS,IAAI,CAAC,CAAC,CAClF,CAEA,QAAQ,EAAc,EAAsC,CAC1D,IAAM,EAAW,KAAK,gBAAgB,EAAQ,QAAQ,EAChD,EAAoB,EAAQ,mBAAqB,GAEvD,OAAO,EADSC,EAAAA,kBAAkB,EAAM,EAAS,KAAM,EAAmB,EAAQ,WAC/C,EAAG,CAAiB,CACzD,CAGA,6BAA6B,EAAc,EAAwD,CACjG,OAAO,KAAK,oCAAoC,EAAM,CAAO,CAAC,CAAC,UACjE,CAMA,oCAAoC,EAAc,EAAuD,CACvG,IAAM,EAAW,KAAK,gBAAgB,EAAQ,QAAQ,EAChD,EAAUC,EAAAA,2BAA2B,EAAM,EAAS,KAAM,EAAQ,aAAa,SAAS,EAC9F,MAAO,CACL,WAAY,EAAQ,WACpB,OAAQ,EAAQ,OAChB,oBAAqB,EAAQ,oBAE7B,gBAAiB,IAAI,IAAI,EAAQ,eAAe,CAClD,CACF,CAQA,8BAA8B,EAAc,EAAuC,CACjF,IAAM,EAAW,KAAK,gBAAgB,EAAQ,QAAQ,EACtD,OAAOC,EAAAA,oCAAoC,EAAM,EAAS,IAAI,CAChE,CAEA,gBAAwB,EAAwC,CAC9D,IAAM,EAAW,KAAK,SAAS,IAAI,CAAI,EACvC,GAAI,CAAC,EACH,MAAU,MAAM,yBAAyB,GAAM,EAEjD,OAAO,CACT,CACF,EAOA,SAAS,EAAsB,EAA+B,EAAyC,CACrG,MAAO,CACL,SAAU,EAAQ,SAClB,MAAO,EAAQ,MACf,MAAO,EAAQ,MACf,UAAW,EAAQ,UAAU,IAAK,IAAQ,CACxC,KAAM,EAAG,KACT,SAAU,EAAG,SACb,UAAW,EAAG,UACd,YAAa,EAAG,YAChB,QAAS,EAAG,QACZ,UAAW,EAAG,UACd,qBAAsB,EAAG,qBACzB,oBAAqB,EAAG,oBACxB,aAAc,EAAG,aACjB,KAAM,EAAG,KACT,eAAgB,EAAG,eACnB,SAAU,EAAsB,EAAG,cAAc,EACjD,UAAW,EAAG,SAChB,EAAE,EACF,qBAAsB,EAAQ,qBAC9B,oBAAqB,EAAQ,oBAC7B,uBAAwB,EAAQ,uBAChC,aAAc,EAAQ,aACtB,UAAW,EAAQ,UACnB,YAAa,EAAQ,YACrB,SAAU,EAAsB,EAAQ,cAAc,EACtD,WAAY,EAAoB,EAAQ,WAAa,IAAA,EACvD,CACF,CAEA,SAAS,EAAsB,EAA+C,CAC5E,GAAM,CAAE,oBAAmB,mBAAkB,iBAAgB,iBAAkB,EACzE,EAAa,EAAoB,EACjC,EAAS,EAAiB,EAC1B,EAAS,IAAe,EAAI,EAAI,EAAS,KAAK,KAAK,CAAU,EAInE,MAAO,CACL,oBACA,mBACA,iBACA,gBACA,aACA,SACA,SACA,QAXiB,IAAqB,EAAI,EAAK,EAAoB,GAAM,EAAgB,IAC/D,CAW5B,CACF,CAEA,MAAa,EAAkB,IAAI,EAEnC,SAAgB,EAAY,EAAc,EAAsC,CAC9E,OAAO,EAAgB,QAAQ,EAAM,CAAO,CAC9C,CAGA,SAAgB,EAA6B,EAAc,EAAwD,CACjH,OAAO,EAAgB,6BAA6B,EAAM,CAAO,CACnE,CAGA,SAAgB,EAA8B,EAAc,EAAuC,CACjG,OAAO,EAAgB,8BAA8B,EAAM,CAAO,CACpE,CAGA,SAAgB,EACd,EACA,EAC8B,CAC9B,OAAO,EAAgB,oCAAoC,EAAM,CAAO,CAC1E"}
+{"version":3,"file":"metrics.cjs","names":["createLanguageRegistry","measureCodeNative","collectCrossFileDataNative","collectFunctionTokenSequencesNative"],"sources":["../src/metrics.ts"],"sourcesContent":["import type { CrossFileDuplicateCandidate, CrossFileDuplicationFileData } from './duplication.js';\nimport { createLanguageRegistry } from './languages.js';\nimport {\n  collectCrossFileDataNative,\n  collectFunctionTokenSequencesNative,\n  measureCodeNative,\n  type NativeCrossFileDataPayload,\n  type NativeHalsteadCounts,\n  type NativeMetricsPayload,\n} from './nativeMetrics.js';\nimport type { CodeMetrics, HalsteadMetrics, LanguageDefinition, LanguageName, MeasureOptions } from './types.js';\n\n/**\n * Measures code metrics for the built-in languages. Parsing and metric passes run in the bundled\n * Rust addon (tree-sitter); this class resolves language aliases, crosses the N-API boundary, and\n * derives the Halstead float metrics from the natively measured counts.\n */\nexport class TreeMeasurer {\n  private readonly registry = createLanguageRegistry();\n\n  getSupportedLanguages(): LanguageName[] {\n    return [...new Set([...this.registry.values()].map((language) => language.name))];\n  }\n\n  measure(code: string, options: MeasureOptions): CodeMetrics {\n    const language = this.resolveLanguage(options.language);\n    const includeSyntaxTree = options.includeSyntaxTree ?? false;\n    const payload = measureCodeNative(code, language.name, includeSyntaxTree, options.duplication);\n    return assembleNativeMetrics(payload, includeSyntaxTree);\n  }\n\n  /** Measures a file and collects its cross-file clone-detection contribution from one parse. */\n  measureWithCrossFileData(\n    code: string,\n    options: MeasureOptions\n  ): { metrics: CodeMetrics; crossFileData: CrossFileDuplicationFileData } {\n    const language = this.resolveLanguage(options.language);\n    const includeSyntaxTree = options.includeSyntaxTree ?? false;\n    const payload = measureCodeNative(code, language.name, includeSyntaxTree, options.duplication, true);\n    if (!payload.crossFileData) {\n      throw new Error('The native addon returned no cross-file duplication data');\n    }\n    return {\n      metrics: assembleNativeMetrics(payload, includeSyntaxTree),\n      crossFileData: toCrossFileDuplicationFileData(payload.crossFileData),\n    };\n  }\n\n  /** Collects one file's duplicate candidates for cross-file clone detection. */\n  collectDuplicationCandidates(code: string, options: MeasureOptions): CrossFileDuplicateCandidate[] {\n    return this.collectCrossFileDuplicationFileData(code, options).candidates;\n  }\n\n  /**\n   * Collects one file's duplicate candidates, normalized tokens, and statement structure for\n   * cross-file clone detection with measureCrossFileDuplication.\n   */\n  collectCrossFileDuplicationFileData(code: string, options: MeasureOptions): CrossFileDuplicationFileData {\n    const language = this.resolveLanguage(options.language);\n    return toCrossFileDuplicationFileData(\n      collectCrossFileDataNative(code, language.name, options.duplication?.minTokens)\n    );\n  }\n\n  /**\n   * Normalized token hash sequences of every function, index-parallel to the functions array of\n   * measure(): identifiers are anonymized by first occurrence within the function, literals by\n   * kind, and keywords/operators kept verbatim, so the regression gate can re-match renamed or\n   * moved functions across two revisions by token-LCS similarity.\n   */\n  collectFunctionTokenSequences(code: string, options: MeasureOptions): Int32Array[] {\n    const language = this.resolveLanguage(options.language);\n    return collectFunctionTokenSequencesNative(code, language.name);\n  }\n\n  private resolveLanguage(name: LanguageName): LanguageDefinition {\n    const language = this.registry.get(name);\n    if (!language) {\n      throw new Error(`Unsupported language: ${name}`);\n    }\n    return language;\n  }\n}\n\nfunction toCrossFileDuplicationFileData(payload: NativeCrossFileDataPayload): CrossFileDuplicationFileData {\n  return {\n    candidates: payload.candidates,\n    tokens: payload.tokens,\n    containerStatements: payload.containerStatements,\n    nearMissBlocks: payload.nearMissBlocks,\n    // Lets cross-file line coverage count only code lines, like within-file coverage.\n    codeLineNumbers: new Set(payload.codeLineNumbers),\n  };\n}\n\n/**\n * Completes a native measurement into CodeMetrics. The object is rebuilt field by field (rather\n * than spread from the parsed JSON) so the result has a stable shape, including\n * explicitly-undefined optional keys.\n */\nfunction assembleNativeMetrics(payload: NativeMetricsPayload, includeSyntaxTree: boolean): CodeMetrics {\n  return {\n    language: payload.language,\n    bytes: payload.bytes,\n    lines: payload.lines,\n    functions: payload.functions.map((fn) => ({\n      name: fn.name,\n      nodeType: fn.nodeType,\n      startLine: fn.startLine,\n      startColumn: fn.startColumn,\n      endLine: fn.endLine,\n      endColumn: fn.endColumn,\n      cyclomaticComplexity: fn.cyclomaticComplexity,\n      cognitiveComplexity: fn.cognitiveComplexity,\n      nestingDepth: fn.nestingDepth,\n      ncss: fn.ncss,\n      parameterCount: fn.parameterCount,\n      halstead: deriveHalsteadMetrics(fn.halsteadCounts),\n      depDegree: fn.depDegree,\n    })),\n    cyclomaticComplexity: payload.cyclomaticComplexity,\n    cognitiveComplexity: payload.cognitiveComplexity,\n    maxCognitiveComplexity: payload.maxCognitiveComplexity,\n    nestingDepth: payload.nestingDepth,\n    ncssCount: payload.ncssCount,\n    duplication: payload.duplication,\n    halstead: deriveHalsteadMetrics(payload.halsteadCounts),\n    syntaxTree: includeSyntaxTree ? payload.syntaxTree : undefined,\n  };\n}\n\nfunction deriveHalsteadMetrics(counts: NativeHalsteadCounts): HalsteadMetrics {\n  const { distinctOperators, distinctOperands, totalOperators, totalOperands } = counts;\n  const vocabulary = distinctOperators + distinctOperands;\n  const length = totalOperators + totalOperands;\n  const volume = vocabulary === 0 ? 0 : length * Math.log2(vocabulary);\n  const difficulty = distinctOperands === 0 ? 0 : (distinctOperators / 2) * (totalOperands / distinctOperands);\n  const effort = difficulty * volume;\n\n  return {\n    distinctOperators,\n    distinctOperands,\n    totalOperators,\n    totalOperands,\n    vocabulary,\n    length,\n    volume,\n    effort,\n  };\n}\n\nexport const defaultMeasurer = new TreeMeasurer();\n\nexport function measureCode(code: string, options: MeasureOptions): CodeMetrics {\n  return defaultMeasurer.measure(code, options);\n}\n\n/** Standalone helper mirroring measureCode for the default measurer. */\nexport function measureCodeWithCrossFileData(\n  code: string,\n  options: MeasureOptions\n): { metrics: CodeMetrics; crossFileData: CrossFileDuplicationFileData } {\n  return defaultMeasurer.measureWithCrossFileData(code, options);\n}\n\n/** Standalone helper mirroring measureCode for the default measurer. */\nexport function collectDuplicationCandidates(code: string, options: MeasureOptions): CrossFileDuplicateCandidate[] {\n  return defaultMeasurer.collectDuplicationCandidates(code, options);\n}\n\n/** Standalone helper mirroring measureCode for the default measurer. */\nexport function collectFunctionTokenSequences(code: string, options: MeasureOptions): Int32Array[] {\n  return defaultMeasurer.collectFunctionTokenSequences(code, options);\n}\n\n/** Standalone helper mirroring measureCode for the default measurer. */\nexport function collectCrossFileDuplicationFileData(\n  code: string,\n  options: MeasureOptions\n): CrossFileDuplicationFileData {\n  return defaultMeasurer.collectCrossFileDuplicationFileData(code, options);\n}\n"],"mappings":"iFAiBA,IAAa,EAAb,KAA0B,CACxB,SAA4BA,EAAAA,uBAAuB,EAEnD,uBAAwC,CACtC,MAAO,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,KAAK,SAAS,OAAO,CAAC,CAAC,CAAC,IAAK,GAAa,EAAS,IAAI,CAAC,CAAC,CAClF,CAEA,QAAQ,EAAc,EAAsC,CAC1D,IAAM,EAAW,KAAK,gBAAgB,EAAQ,QAAQ,EAChD,EAAoB,EAAQ,mBAAqB,GAEvD,OAAO,EADSC,EAAAA,kBAAkB,EAAM,EAAS,KAAM,EAAmB,EAAQ,WAC/C,EAAG,CAAiB,CACzD,CAGA,yBACE,EACA,EACuE,CACvE,IAAM,EAAW,KAAK,gBAAgB,EAAQ,QAAQ,EAChD,EAAoB,EAAQ,mBAAqB,GACjD,EAAUA,EAAAA,kBAAkB,EAAM,EAAS,KAAM,EAAmB,EAAQ,YAAa,EAAI,EACnG,GAAI,CAAC,EAAQ,cACX,MAAU,MAAM,0DAA0D,EAE5E,MAAO,CACL,QAAS,EAAsB,EAAS,CAAiB,EACzD,cAAe,EAA+B,EAAQ,aAAa,CACrE,CACF,CAGA,6BAA6B,EAAc,EAAwD,CACjG,OAAO,KAAK,oCAAoC,EAAM,CAAO,CAAC,CAAC,UACjE,CAMA,oCAAoC,EAAc,EAAuD,CACvG,IAAM,EAAW,KAAK,gBAAgB,EAAQ,QAAQ,EACtD,OAAO,EACLC,EAAAA,2BAA2B,EAAM,EAAS,KAAM,EAAQ,aAAa,SAAS,CAChF,CACF,CAQA,8BAA8B,EAAc,EAAuC,CACjF,IAAM,EAAW,KAAK,gBAAgB,EAAQ,QAAQ,EACtD,OAAOC,EAAAA,oCAAoC,EAAM,EAAS,IAAI,CAChE,CAEA,gBAAwB,EAAwC,CAC9D,IAAM,EAAW,KAAK,SAAS,IAAI,CAAI,EACvC,GAAI,CAAC,EACH,MAAU,MAAM,yBAAyB,GAAM,EAEjD,OAAO,CACT,CACF,EAEA,SAAS,EAA+B,EAAmE,CACzG,MAAO,CACL,WAAY,EAAQ,WACpB,OAAQ,EAAQ,OAChB,oBAAqB,EAAQ,oBAC7B,eAAgB,EAAQ,eAExB,gBAAiB,IAAI,IAAI,EAAQ,eAAe,CAClD,CACF,CAOA,SAAS,EAAsB,EAA+B,EAAyC,CACrG,MAAO,CACL,SAAU,EAAQ,SAClB,MAAO,EAAQ,MACf,MAAO,EAAQ,MACf,UAAW,EAAQ,UAAU,IAAK,IAAQ,CACxC,KAAM,EAAG,KACT,SAAU,EAAG,SACb,UAAW,EAAG,UACd,YAAa,EAAG,YAChB,QAAS,EAAG,QACZ,UAAW,EAAG,UACd,qBAAsB,EAAG,qBACzB,oBAAqB,EAAG,oBACxB,aAAc,EAAG,aACjB,KAAM,EAAG,KACT,eAAgB,EAAG,eACnB,SAAU,EAAsB,EAAG,cAAc,EACjD,UAAW,EAAG,SAChB,EAAE,EACF,qBAAsB,EAAQ,qBAC9B,oBAAqB,EAAQ,oBAC7B,uBAAwB,EAAQ,uBAChC,aAAc,EAAQ,aACtB,UAAW,EAAQ,UACnB,YAAa,EAAQ,YACrB,SAAU,EAAsB,EAAQ,cAAc,EACtD,WAAY,EAAoB,EAAQ,WAAa,IAAA,EACvD,CACF,CAEA,SAAS,EAAsB,EAA+C,CAC5E,GAAM,CAAE,oBAAmB,mBAAkB,iBAAgB,iBAAkB,EACzE,EAAa,EAAoB,EACjC,EAAS,EAAiB,EAC1B,EAAS,IAAe,EAAI,EAAI,EAAS,KAAK,KAAK,CAAU,EAInE,MAAO,CACL,oBACA,mBACA,iBACA,gBACA,aACA,SACA,SACA,QAXiB,IAAqB,EAAI,EAAK,EAAoB,GAAM,EAAgB,IAC/D,CAW5B,CACF,CAEA,MAAa,EAAkB,IAAI,EAEnC,SAAgB,EAAY,EAAc,EAAsC,CAC9E,OAAO,EAAgB,QAAQ,EAAM,CAAO,CAC9C,CAGA,SAAgB,EACd,EACA,EACuE,CACvE,OAAO,EAAgB,yBAAyB,EAAM,CAAO,CAC/D,CAGA,SAAgB,EAA6B,EAAc,EAAwD,CACjH,OAAO,EAAgB,6BAA6B,EAAM,CAAO,CACnE,CAGA,SAAgB,EAA8B,EAAc,EAAuC,CACjG,OAAO,EAAgB,8BAA8B,EAAM,CAAO,CACpE,CAGA,SAAgB,EACd,EACA,EAC8B,CAC9B,OAAO,EAAgB,oCAAoC,EAAM,CAAO,CAC1E"}