code-gauge 10.2.0 → 10.2.1

package/dist/crossFileNearMiss.js.map

@@ -1 +1 @@
-{"version":3,"file":"crossFileNearMiss.js","names":[],"sources":["../src/crossFileNearMiss.ts"],"sourcesContent":["import {\n  createLcsLengthCounter,\n  lcsLength,\n  markLcs,\n  type CountedOccurrence,\n  type Token,\n  type TokenRange,\n  type TokenSegment,\n} from './duplication.js';\n\n/**\n * Cross-file near-miss (Type-3) clone detection, following the within-file detector's model\n * (native/src/near_miss.rs): candidate block pairs are filtered through an n-gram inverted index\n * (NIL, Nakagawa et al. 2021), then verified by token-level longest common subsequence against the\n * larger block (NiCad's per-fragment similarity), backed by an information-weighted content gate,\n * with a statement-order-insensitive fallback and a local match over the anchored cores of two\n * blocks. Only pairs of blocks in different files are compared: a same-file pair is the\n * within-file detector's concern.\n */\n\nexport interface NearMissSourceFile {\n  tokens?: Token[];\n  containerStatements?: TokenRange[][];\n  nearMissBlocks?: TokenRange[];\n}\n\n/** One copy (a block or its matched cores) in a near-miss group; anchors carry `spanCountedElsewhere` (see collectCrossFileNearMissGroups). */\nexport interface NearMissOccurrence extends CountedOccurrence {\n  fileIndex: number;\n}\n\n/** N-gram size of the candidate index and local-match anchors (NIL's default). */\nconst ngramSize = 5;\n/** Filtration threshold: shared distinct n-grams over the smaller block's (NIL's default). */\nconst filtrationPercent = 10;\n/**\n * Pairs whose longer block exceeds this multiple of the shorter are compared only when whole-block\n * similarity still allows their ratio (below a minSimilarityPercent of 34). The candidate scan\n * stops at this floor while walking length-ordered postings, so pairs of very different lengths are\n * neither counted nor verified; `maxNgramBlockFrequency` is what bounds the scan's total cost.\n */\nconst maxLengthRatio = 3;\n/**\n * A structural match must also share content: more than this percent of the larger side's\n * information-weighted content-bearing tokens (names and literal values), so blocks of the same\n * shape that call different APIs on different data are not clones.\n */\nconst minContentSimilarityPercent = 50;\n/**\n * Caps content weights so only names and values spread over more than a quarter of the blocks are\n * discounted: a family of copies shares its content across several blocks, and uncapped rarity\n * weighting would let each copy's few unique edits outweigh everything the family shares.\n */\nconst maxContentWeight = 3;\n/**\n * Anchors must cover at least this percent of the shorter core: sparser chains are coincidental\n * runs of common n-grams in merely similar-looking code, and the cheap bound spares their content\n * and LCS checks. Not the similarity threshold itself, since n-grams repeated within a block\n * (repetitive statements) never anchor.\n */\nconst minAnchorCoveragePercent = 50;\n/** Anchors farther apart than this (in either block) split a local match into separate chains. */\nconst maxAnchorGapTokens = 30;\n/** Statement-order-insensitive comparison needs this many top-level statements per block. */\nconst minReorderStatementCount = 2;\n/**\n * N-grams occurring in more blocks than this are stop n-grams (syntax boilerplate such as a chain\n * of closing braces), left out of the index and of each block's n-gram count. Counting shared\n * n-grams costs the square of an n-gram's block frequency, so without the cap a project's most\n * common n-grams make filtration quadratic in the block count, while they discriminate nothing.\n */\nconst maxNgramBlockFrequency = 1000;\n\ninterface NormalizedBlock {\n  fileIndex: number;\n  range: TokenRange;\n  /** Interned non-identifier symbols (>= 0) and identifiers as -(file-level id + 1). */\n  symbols: Int32Array;\n  isContent: Uint8Array;\n  /**\n   * Per token of the block's file, a key of its content for a non-identifier: unlike a symbol,\n   * numbered by first occurrence, it does not depend on the order of the files.\n   */\n  fileTokenKeys: Float64Array;\n  /** Identifiers anonymized by first occurrence within the block. */\n  sequence: Int32Array;\n  /**\n   * The content of the block as written: per token its content key, or the number of its\n   * identifier. Equal for equal content wherever it lies, which symbols, numbered by first\n   * occurrence, are not; ordered like `written_content` in native/src/near_miss.rs, so that both\n   * detectors take the sides of a pair in the same order.\n   */\n  writtenContent: Float64Array;\n  /** The sequence sorted, for the token-bag upper bound on the LCS. */\n  sortedSequence: Int32Array;\n  /** Distinct non-stop n-gram hashes. */\n  ngrams: Int32Array;\n  /**\n   * The n-grams occurring exactly once in the block, sorted, with their offsets in the parallel\n   * array: two blocks' local-match anchors intersect by merging.\n   */\n  uniqueNgrams: Int32Array;\n  uniqueNgramOffsets: Int32Array;\n  contentCounts: Map<number, number>;\n  /**\n   * The sequence with its top-level statements in canonical order and the block offset each of its\n   * tokens comes from, when the block has enough statements for statement-order-insensitive\n   * comparison.\n   */\n  canonical: { sequence: Int32Array; offsets: Int32Array } | undefined;\n}\n\n/** A verified core in each block of a pair, as file token ranges. */\ntype CorePair = [[number, number], [number, number]];\n\n/** What alignPair needs to recover the tokens a verified match pairs. */\ninterface Alignment {\n  /** Whether the blocks matched only with their statements in canonical order. */\n  reordered: boolean;\n}\n\n/** How a verified pair matched: whole blocks, or every anchored core pair (one per gap-split chain segment). */\ntype PairMatch = (Alignment & { cores?: undefined }) | { cores: (Alignment & { pair: CorePair })[] };\n\n/** A verified near-miss pair of blocks; without `cores`, the blocks match as wholes. */\ninterface MatchEdge extends Alignment {\n  left: number;\n  right: number;\n  cores?: CorePair;\n}\n\n/**\n * Clusters verified cross-file near-miss pairs into groups. A node (a whole block or a matched\n * core) overlapping an occurrence of `reportedSpansByFile` (the exact cross-file groups) is an\n * anchor: it links near-miss copies to the content an exact group already reports, and appears in\n * the near-miss group marked `spanCountedElsewhere` so block counting does not count its span\n * twice. A pair of two spans that both overlap reported spans is dropped before clustering (blocks\n * wholly covered by reported spans are not even compared), so it neither collapses the blocks' core\n * nodes nor widens a node beyond what the pairs clustering it matched, and a group needs at least\n * one non-anchor node. A block that matched only locally is reported as its matched cores\n * (overlapping cores merged), each clustered with its own partners, so code no verified pair\n * matched never counts as duplicated.\n */\nexport function collectCrossFileNearMissGroups(\n  files: NearMissSourceFile[],\n  reportedSpansByFile: { startTokenIndex: number; endTokenIndex: number }[][],\n  minTokens: number,\n  minSimilarityPercent: number\n): NearMissOccurrence[][] {\n  if (minSimilarityPercent >= 100) {\n    return [];\n  }\n  const blocks = normalizeBlocks(files);\n  const matcher = createMatcher(blocks, minTokens, minSimilarityPercent);\n  const overlapsReportedSpan = reportedSpansByFile.map(createOverlapTest);\n  const coveredByReportedSpans = reportedSpansByFile.map(createCoverageTest);\n  const fullyReported = blocks.map(({ fileIndex, range }) => coveredByReportedSpans[fileIndex]?.(range) ?? false);\n  const touchesReported = blocks.map(({ fileIndex, range }) => overlapsReportedSpan[fileIndex]?.(range) ?? false);\n  const touchesCore = ({ fileIndex }: NormalizedBlock, [startTokenIndex, endTokenIndex]: [number, number]): boolean =>\n    overlapsReportedSpan[fileIndex]?.({ startTokenIndex, endTokenIndex }) ?? false;\n  const edges: MatchEdge[] = [];\n  forEachCandidatePair(blocks, fullyReported, minSimilarityPercent, (left, right) => {\n    const leftBlock = blocks[left];\n    const rightBlock = blocks[right];\n    const match = leftBlock && rightBlock && matcher(leftBlock, rightBlock, right);\n    if (!match) {\n      return;\n    }\n    if (!match.cores) {\n      if (!(touchesReported[left] && touchesReported[right])) {\n        edges.push({ left, right, reordered: match.reordered });\n      }\n      return;\n    }\n    for (const { pair, reordered } of match.cores) {\n      if (!(touchesCore(leftBlock, pair[0]) && touchesCore(rightBlock, pair[1]))) {\n        edges.push({ left, right, cores: pair, reordered });\n      }\n    }\n  });\n\n  // Clustering runs over (block, core) nodes: a block with a recorded whole match is one node,\n  // and otherwise each union of its overlapping local cores is its own node, so disjoint cores\n  // matched with different partners fall into separate groups.\n  const matchedWhole = blocks.map(() => false);\n  const localCores = blocks.map((): [number, number][] => []);\n  for (const edge of edges) {\n    for (const [index, core] of sidesOf(edge)) {\n      if (core) {\n        localCores[index]?.push(core);\n      } else {\n        matchedWhole[index] = true;\n      }\n    }\n  }\n  const nodes: { blockIndex: number; core: [number, number] | undefined }[] = [];\n  const firstNodeByBlock: number[] = [];\n  for (const blockIndex of blocks.keys()) {\n    firstNodeByBlock.push(nodes.length);\n    const cores = matchedWhole[blockIndex] ? [] : mergeOverlappingCores(localCores[blockIndex] ?? []);\n    if (cores.length === 0) {\n      nodes.push({ blockIndex, core: undefined });\n    }\n    for (const core of cores) {\n      nodes.push({ blockIndex, core });\n    }\n  }\n  const nodeOf = (blockIndex: number, core: [number, number] | undefined): number => {\n    const first = firstNodeByBlock[blockIndex] ?? 0;\n    if (!core || matchedWhole[blockIndex]) {\n      return first;\n    }\n    for (let node = first; nodes[node]?.blockIndex === blockIndex; node += 1) {\n      const span = nodes[node]?.core;\n      if (span && span[0] <= core[0] && core[1] <= span[1]) {\n        return node;\n      }\n    }\n    throw new Error(\"every local core lies in one of its block's merged cores\");\n  };\n\n  const parent = nodes.map((_, index) => index);\n  const find = (index: number): number => {\n    let root = index;\n    while (parent[root] !== root) {\n      root = parent[root] ?? root;\n    }\n    for (let current = index; parent[current] !== root;) {\n      const next = parent[current] ?? root;\n      parent[current] = root;\n      current = next;\n    }\n    return root;\n  };\n  // Anchoring is judged per node: a core is an anchor only when a reported span overlaps the core\n  // itself, not merely elsewhere in its block.\n  const anchored = nodes.map(({ blockIndex, core }) => {\n    const block = blocks[blockIndex];\n    const [startTokenIndex, endTokenIndex] = core ?? [\n      block?.range.startTokenIndex ?? 0,\n      block?.range.endTokenIndex ?? 0,\n    ];\n    return overlapsReportedSpan[block?.fileIndex ?? 0]?.({ startTokenIndex, endTokenIndex }) ?? false;\n  });\n  const edgeNodes = edges.map((edge) => sidesOf(edge).map(([index, core]) => nodeOf(index, core)));\n  for (const [leftNode = 0, rightNode = 0] of edgeNodes) {\n    const leftRoot = find(leftNode);\n    const rightRoot = find(rightNode);\n    parent[Math.max(leftRoot, rightRoot)] = Math.min(leftRoot, rightRoot);\n  }\n  const matchedRunsByNode = alignPairs(blocks, edges, edgeNodes, nodes.length);\n\n  const membersByRoot = new Map<number, number[]>();\n  for (const node of nodes.keys()) {\n    const root = find(node);\n    const members = membersByRoot.get(root) ?? [];\n    members.push(node);\n    membersByRoot.set(root, members);\n  }\n  const groups: NearMissOccurrence[][] = [];\n  for (const members of membersByRoot.values()) {\n    // Components form only through cross-file pairs, so two members always span two files.\n    if (members.length < 2 || members.every((node) => anchored[node])) {\n      continue;\n    }\n    // A group's nodes from one block become ONE occurrence whose segments are its cores, so the\n    // fragment-weighted count charges the block as one copy (as for gapped clones), not once per core.\n    const coresByBlock = new Map<number, BlockCopy>();\n    for (const node of members) {\n      const { blockIndex = 0, core } = nodes[node] ?? {};\n      const entry = coresByBlock.get(blockIndex) ?? { cores: [], anchor: false, matchedRuns: [] };\n      entry.cores.push(core);\n      entry.anchor ||= anchored[node] ?? false;\n      entry.matchedRuns.push(...(matchedRunsByNode[node] ?? []));\n      coresByBlock.set(blockIndex, entry);\n    }\n    groups.push(\n      [...coresByBlock].flatMap(([blockIndex, copy]) => {\n        const block = blocks[blockIndex];\n        return block ? [toOccurrence(block, files, copy)] : [];\n      })\n    );\n  }\n  return groups;\n}\n\n/** The nodes of one block in a group: its cores (`undefined` for the whole block) and what its partners match of them. */\ninterface BlockCopy {\n  cores: ([number, number] | undefined)[];\n  anchor: boolean;\n  matchedRuns: [number, number][];\n}\n\n/** The (block, core) on each side of a pair. */\nfunction sidesOf({ left, right, cores }: MatchEdge): [number, [number, number] | undefined][] {\n  return [\n    [left, cores?.[0]],\n    [right, cores?.[1]],\n  ];\n}\n\n/**\n * Per node, the token runs its partners match. The lines of a near-miss copy that are duplicated\n * are those a verified partner matches, which takes the pairing itself. Every pair is aligned:\n * verifying it already cost a longest common subsequence of the same sequences, and any choice\n * among a copy's partners would let their number or their order decide what the copy counts.\n */\nfunction alignPairs(\n  blocks: NormalizedBlock[],\n  edges: MatchEdge[],\n  edgeNodes: number[][],\n  nodeCount: number\n): [number, number][][] {\n  const matchedRunsByNode = Array.from({ length: nodeCount }, (): [number, number][] => []);\n  for (const [edgeIndex, edge] of edges.entries()) {\n    const left = blocks[edge.left];\n    const right = blocks[edge.right];\n    if (left && right) {\n      const [leftRuns, rightRuns] = alignPair(left, right, edge);\n      matchedRunsByNode[edgeNodes[edgeIndex]?.[0] ?? 0]?.push(...leftRuns);\n      matchedRunsByNode[edgeNodes[edgeIndex]?.[1] ?? 0]?.push(...rightRuns);\n    }\n  }\n  return matchedRunsByNode;\n}\n\n/**\n * The token runs a verified match pairs, per side as file token ranges: those of a longest common\n * subsequence of what the match compared, its cores or else the whole blocks.\n */\nfunction alignPair(\n  left: NormalizedBlock,\n  right: NormalizedBlock,\n  { cores, reordered }: MatchEdge\n): [[number, number][], [number, number][]] {\n  const leftCompared = comparedSequence(left, cores?.[0], reordered);\n  const rightCompared = comparedSequence(right, cores?.[1], reordered);\n  // Of the equally long subsequences, which one is marked depends on which sequence comes first,\n  // so the sides are taken in the order of their content rather than of their files.\n  if (compareSequences(contentOf(left, cores?.[0]), contentOf(right, cores?.[1])) <= 0) {\n    const [leftMatched, rightMatched] = markLcs(leftCompared.sequence, rightCompared.sequence);\n    return [toRuns(leftCompared.positions, leftMatched), toRuns(rightCompared.positions, rightMatched)];\n  }\n  const [rightMatched, leftMatched] = markLcs(rightCompared.sequence, leftCompared.sequence);\n  return [toRuns(leftCompared.positions, leftMatched), toRuns(rightCompared.positions, rightMatched)];\n}\n\n/**\n * The content of a block or of a core of it, as written (see `writtenContent`), with the\n * identifiers of a core numbered within it.\n */\nfunction contentOf(block: NormalizedBlock, core: [number, number] | undefined): Float64Array {\n  if (!core) {\n    return block.writtenContent;\n  }\n  const { sequence, positions } = comparedSequence(block, core, false);\n  return Float64Array.from(sequence, (symbol, index) =>\n    symbol < 0 ? symbol : (block.fileTokenKeys[positions[index] ?? 0] ?? 0)\n  );\n}\n\n/** The maximal runs of the matched positions, as half-open ranges. */\nfunction toRuns(positions: Int32Array, matched: Uint8Array): [number, number][] {\n  const runs: [number, number][] = [];\n  for (const position of positions.filter((_, index) => matched[index] === 1).toSorted()) {\n    const last = runs.at(-1);\n    if (last?.[1] === position) {\n      last[1] = position + 1;\n    } else {\n      runs.push([position, position + 1]);\n    }\n  }\n  return runs;\n}\n\n/**\n * The sequence a match compared, with the file token index of each of its symbols: a core, the\n * block with its statements in canonical order, or the block as written.\n */\nfunction comparedSequence(\n  block: NormalizedBlock,\n  core: [number, number] | undefined,\n  reordered: boolean\n): { sequence: Int32Array; positions: Int32Array } {\n  const blockStart = block.range.startTokenIndex;\n  if (core) {\n    return {\n      sequence: anonymize(block.symbols.subarray(core[0] - blockStart, core[1] - blockStart)),\n      positions: Int32Array.from({ length: core[1] - core[0] }, (_, index) => core[0] + index),\n    };\n  }\n  if (reordered && block.canonical) {\n    return {\n      sequence: block.canonical.sequence,\n      positions: block.canonical.offsets.map((offset) => blockStart + offset),\n    };\n  }\n  return { sequence: block.sequence, positions: block.sequence.map((_, index) => blockStart + index) };\n}\n\n/** The unions of overlapping cores, in position order. */\nfunction mergeOverlappingCores(cores: [number, number][]): [number, number][] {\n  const merged: [number, number][] = [];\n  for (const [start, end] of cores.toSorted((left, right) => left[0] - right[0])) {\n    const last = merged.at(-1);\n    if (last && start < last[1]) {\n      last[1] = Math.max(last[1], end);\n    } else {\n      merged.push([start, end]);\n    }\n  }\n  return merged;\n}\n\n/** Whether the spans, merged, cover every token of a range. */\nfunction createCoverageTest(\n  spans: { startTokenIndex: number; endTokenIndex: number }[]\n): (range: { startTokenIndex: number; endTokenIndex: number }) => boolean {\n  // Touching spans merge too: together they cover a range across their boundary.\n  const merged: [number, number][] = [];\n  for (const { startTokenIndex, endTokenIndex } of spans.toSorted(\n    (left, right) => left.startTokenIndex - right.startTokenIndex\n  )) {\n    const last = merged.at(-1);\n    if (last && startTokenIndex <= last[1]) {\n      last[1] = Math.max(last[1], endTokenIndex);\n    } else {\n      merged.push([startTokenIndex, endTokenIndex]);\n    }\n  }\n  return (range) => {\n    let low = 0;\n    let high = merged.length;\n    while (low < high) {\n      const middle = (low + high) >>> 1;\n      if ((merged[middle]?.[0] ?? 0) <= range.startTokenIndex) {\n        low = middle + 1;\n      } else {\n        high = middle;\n      }\n    }\n    return (merged[low - 1]?.[1] ?? -1) >= range.endTokenIndex;\n  };\n}\n\n/**\n * Whether a range overlaps any of the spans: among the spans starting before the range ends\n * (binary search over sorted starts), the furthest end reaches past the range's start.\n */\nfunction createOverlapTest(\n  spans: { startTokenIndex: number; endTokenIndex: number }[]\n): (range: { startTokenIndex: number; endTokenIndex: number }) => boolean {\n  const sorted = spans.toSorted((left, right) => left.startTokenIndex - right.startTokenIndex);\n  const maxEndPrefix = new Int32Array(sorted.length);\n  let maxEnd = -1;\n  for (const [index, span] of sorted.entries()) {\n    maxEnd = Math.max(maxEnd, span.endTokenIndex);\n    maxEndPrefix[index] = maxEnd;\n  }\n  return (range) => {\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) < range.endTokenIndex) {\n        low = middle + 1;\n      } else {\n        high = middle;\n      }\n    }\n    return low > 0 && (maxEndPrefix[low - 1] ?? -1) > range.startTokenIndex;\n  };\n}\n\n/**\n * The block's occurrence with one segment per entry of `cores` (the whole block for `undefined`,\n * a whole match). Source offsets stay the block's: tokens carry none, and near-miss occurrences\n * report lines only.\n */\nfunction toOccurrence(\n  { fileIndex, range }: NormalizedBlock,\n  files: NearMissSourceFile[],\n  { cores, anchor, matchedRuns }: BlockCopy\n): NearMissOccurrence {\n  const whole = cores.includes(undefined);\n  const segments = cores\n    .map((core): [number, number] => core ?? [range.startTokenIndex, range.endTokenIndex])\n    .toSorted((left, right) => left[0] - right[0])\n    .map(([startTokenIndex, endTokenIndex]) => ({ startTokenIndex, endTokenIndex }));\n  const start = segments[0]?.startTokenIndex ?? range.startTokenIndex;\n  const end = segments.at(-1)?.endTokenIndex ?? range.endTokenIndex;\n  const tokens = files[fileIndex]?.tokens;\n  return {\n    fileIndex,\n    spanCountedElsewhere: anchor || undefined,\n    segments,\n    matchedRuns: mergeOverlappingCores(matchedRuns).map(([startTokenIndex, endTokenIndex]): TokenSegment => ({\n      startTokenIndex,\n      endTokenIndex,\n    })),\n    tokenCount: segments.reduce((sum, segment) => sum + segment.endTokenIndex - segment.startTokenIndex, 0),\n    startTokenIndex: start,\n    endTokenIndex: end,\n    startIndex: range.startIndex,\n    endIndex: range.endIndex,\n    startLine: whole ? range.startLine : (tokens?.[start]?.startRow ?? 0) + 1,\n    endLine: whole ? range.endLine : (tokens?.[end - 1]?.endRow ?? 0) + 1,\n  };\n}\n\n/**\n * Visits every cross-file block pair sharing at least `filtrationPercent` of the smaller block's\n * non-stop n-grams, except pairs of two blocks wholly covered by reported spans and pairs whose\n * length ratio rules out both\n * whole-block similarity and `maxLengthRatio`. Blocks are\n * indexed in ascending length, so each posting list is scanned backwards only while its blocks\n * are long enough; shared counts accumulate in a dense counter, so no pair map is materialized.\n */\nfunction forEachCandidatePair(\n  blocks: NormalizedBlock[],\n  fullyReported: boolean[],\n  minSimilarityPercent: number,\n  visit: (left: number, right: number) => void\n): void {\n  const blockFrequency = new Map<number, number>();\n  for (const block of blocks) {\n    for (const ngram of block.ngrams) {\n      blockFrequency.set(ngram, (blockFrequency.get(ngram) ?? 0) + 1);\n    }\n  }\n  for (const block of blocks) {\n    block.ngrams = block.ngrams.filter((ngram) => (blockFrequency.get(ngram) ?? 0) <= maxNgramBlockFrequency);\n  }\n\n  // Typed copies keep the posting loop, which dominates this phase, free of object dereferences.\n  const fileIndexes = Int32Array.from(blocks, (block) => block.fileIndex);\n  const reportedFlags = Uint8Array.from(fullyReported, Number);\n  const lengths = Int32Array.from(blocks, (block) => block.sequence.length);\n  const ngramCounts = Int32Array.from(blocks, (block) => block.ngrams.length);\n  const order = [...blocks.keys()].toSorted((left, right) => (lengths[left] ?? 0) - (lengths[right] ?? 0));\n  const postings = new Map<number, number[]>();\n  const sharedCounts = new Int32Array(blocks.length);\n  const touched: number[] = [];\n  for (const right of order) {\n    const fileIndex = fileIndexes[right];\n    const rightReported = reportedFlags[right] === 1;\n    const minLeftLength = Math.min(\n      Math.ceil((lengths[right] ?? 0) / maxLengthRatio),\n      Math.ceil((minSimilarityPercent * (lengths[right] ?? 0)) / 100)\n    );\n    const ngrams = blocks[right]?.ngrams ?? [];\n    for (const ngram of ngrams) {\n      const posting = postings.get(ngram);\n      if (!posting) {\n        postings.set(ngram, [right]);\n        continue;\n      }\n      for (let position = posting.length - 1; position >= 0; position -= 1) {\n        const left = posting[position] ?? 0;\n        if ((lengths[left] ?? 0) < minLeftLength) {\n          break;\n        }\n        if (fileIndexes[left] === fileIndex || (rightReported && reportedFlags[left] === 1)) {\n          continue;\n        }\n        if (sharedCounts[left] === 0) {\n          touched.push(left);\n        }\n        sharedCounts[left] = (sharedCounts[left] ?? 0) + 1;\n      }\n      posting.push(right);\n    }\n    for (const left of touched) {\n      const shared = sharedCounts[left] ?? 0;\n      sharedCounts[left] = 0;\n      if (shared * 100 >= filtrationPercent * Math.min(ngramCounts[left] ?? 0, ngrams.length)) {\n        visit(left, right);\n      }\n    }\n    touched.length = 0;\n  }\n}\n\n/**\n * Returns the pair verifier. Content symbols are weighted by integer self-information,\n * 1 + floor(log2((N + 1) / df)) over N blocks capped at `maxContentWeight`, so rare names and\n * values (the logic a copy preserves) outweigh ubiquitous ones, following the\n * information-theoretic weighting of ECScan's essence-clone detection (2025).\n */\nfunction createMatcher(\n  blocks: NormalizedBlock[],\n  minTokens: number,\n  minSimilarityPercent: number\n): (left: NormalizedBlock, right: NormalizedBlock, rightIndex: number) => PairMatch | undefined {\n  const documentFrequencies = new Map<number, number>();\n  for (const block of blocks) {\n    for (const symbol of block.contentCounts.keys()) {\n      documentFrequencies.set(symbol, (documentFrequencies.get(symbol) ?? 0) + 1);\n    }\n  }\n  const selfInformation = (documentFrequency: number): number =>\n    Math.min(31 - Math.clz32(Math.floor((blocks.length + 1) / documentFrequency)) + 1, maxContentWeight);\n  const weights = new Map<number, number>();\n  for (const [symbol, frequency] of documentFrequencies) {\n    weights.set(symbol, selfInformation(frequency));\n  }\n  const weigh = (counts: Map<number, number>): WeightedContent => {\n    const symbols = Int32Array.from(counts.keys()).toSorted();\n    // Span content comes from blocks, so every symbol has a weight.\n    const weightedCounts = Int32Array.from(symbols, (symbol) => (counts.get(symbol) ?? 0) * (weights.get(symbol) ?? 0));\n    let total = 0;\n    for (const count of weightedCounts) {\n      total += count;\n    }\n    return { symbols, weightedCounts, total };\n  };\n  const blockContents = new Map(blocks.map((block) => [block, weigh(block.contentCounts)]));\n\n  // Every candidate pair of one `right` block is visited consecutively, so one LCS counter (its\n  // position masks built once) serves them all.\n  let counterBlock = -1;\n  let counter: ((sequence: Int32Array) => number) | undefined;\n  const lcsLengthWithRight = (right: NormalizedBlock, rightIndex: number, sequence: Int32Array): number => {\n    if (counterBlock !== rightIndex || !counter) {\n      counterBlock = rightIndex;\n      counter = createLcsLengthCounter(right.sequence);\n    }\n    return counter(sequence);\n  };\n\n  /**\n   * Matches the cores two blocks share inside different surroundings (a copy wrapped in added\n   * code, or two copies embedded in different code), which whole-block similarity misses\n   * (CCAligner's large-gap and LVMapper's large-variance clones). N-grams unique to each block\n   * anchor the alignment; their longest chain increasing in both blocks (a run filter keeps only\n   * anchors continuing a diagonal, but the chain may shift diagonals at small insertions), split at\n   * gaps, delimits the cores, and every core pair that is a near-miss clone in its own right is\n   * returned.\n   */\n  const matchLocally = (left: NormalizedBlock, right: NormalizedBlock): PairMatch | undefined => {\n    const anchors: [number, number][] = [];\n    for (\n      let leftIndex = 0, rightIndex = 0;\n      leftIndex < left.uniqueNgrams.length && rightIndex < right.uniqueNgrams.length;\n    ) {\n      const leftHash = left.uniqueNgrams[leftIndex] ?? 0;\n      const rightHash = right.uniqueNgrams[rightIndex] ?? 0;\n      if (leftHash === rightHash) {\n        anchors.push([left.uniqueNgramOffsets[leftIndex] ?? 0, right.uniqueNgramOffsets[rightIndex] ?? 0]);\n      }\n      if (leftHash <= rightHash) {\n        leftIndex += 1;\n      }\n      if (rightHash <= leftHash) {\n        rightIndex += 1;\n      }\n    }\n    anchors.sort((first, second) => first[0] - second[0]);\n    // An isolated 5-gram match is often coincidental (n-grams are identifier-blind); a copied core\n    // yields runs of consecutive anchors, so only anchors continuing a diagonal run are chained.\n    const runAnchors = anchors.filter(([leftOffset, rightOffset], index) => {\n      const previous = anchors[index - 1];\n      const next = anchors[index + 1];\n      return (\n        (previous?.[0] === leftOffset - 1 && previous[1] === rightOffset - 1) ||\n        (next?.[0] === leftOffset + 1 && next[1] === rightOffset + 1)\n      );\n    });\n    const leftOffset = left.range.startTokenIndex;\n    const rightOffset = right.range.startTokenIndex;\n    const cores: (Alignment & { pair: CorePair })[] = [];\n    for (const segment of chainSegments(longestIncreasingChain(runAnchors))) {\n      const [leftStart, rightStart] = segment[0] ?? [0, 0];\n      const [leftLast, rightLast] = segment.at(-1) ?? [0, 0];\n      const leftEnd = leftLast + ngramSize;\n      const rightEnd = rightLast + ngramSize;\n      const leftLength = leftEnd - leftStart;\n      const rightLength = rightEnd - rightStart;\n      const shorter = Math.min(leftLength, rightLength);\n      const required = minSimilarityPercent * Math.max(leftLength, rightLength);\n      if (\n        shorter < minTokens ||\n        shorter * 100 < required ||\n        anchoredTokenCount(segment) * 100 < minAnchorCoveragePercent * shorter ||\n        !sharesContent(\n          weigh(countContent(left.symbols, left.isContent, leftStart, leftEnd)),\n          weigh(countContent(right.symbols, right.isContent, rightStart, rightEnd))\n        )\n      ) {\n        continue;\n      }\n      const coreLcsLength = lcsLength(\n        anonymize(left.symbols.subarray(leftStart, leftEnd)),\n        anonymize(right.symbols.subarray(rightStart, rightEnd))\n      );\n      if (coreLcsLength * 100 >= required) {\n        cores.push({\n          pair: [\n            [leftOffset + leftStart, leftOffset + leftEnd],\n            [rightOffset + rightStart, rightOffset + rightEnd],\n          ],\n          reordered: false,\n        });\n      }\n    }\n    return cores.length > 0 ? { cores } : undefined;\n  };\n\n  /** Cheapest bounds first: the LCS cannot exceed the shorter block's length nor the bag overlap. */\n  return (left, right, rightIndex) => {\n    const required = minSimilarityPercent * Math.max(left.sequence.length, right.sequence.length);\n    if (\n      Math.min(left.sequence.length, right.sequence.length) * 100 >= required &&\n      sharesContent(blockContents.get(left), blockContents.get(right))\n    ) {\n      if (sortedOverlap(left.sortedSequence, right.sortedSequence) * 100 >= required) {\n        const inOrder = lcsLengthWithRight(right, rightIndex, left.sequence);\n        if (inOrder * 100 >= required) {\n          return { reordered: false };\n        }\n      }\n      // With their top-level statements (each anonymized on its own) in a canonical order, a copy\n      // whose independent statements were swapped still matches.\n      if (left.canonical && right.canonical) {\n        const reordered = lcsLength(left.canonical.sequence, right.canonical.sequence);\n        if (reordered * 100 >= required) {\n          return { reordered: true };\n        }\n      }\n    }\n    // Of the equally long chains of anchors, the one kept depends on which block comes first, so\n    // the blocks are taken in the order of their content rather than of their files.\n    if (compareSequences(right.writtenContent, left.writtenContent) < 0) {\n      const swapped = matchLocally(right, left);\n      return (\n        swapped?.cores && {\n          cores: swapped.cores.map(({ pair: [rightCore, leftCore], ...alignment }) => ({\n            ...alignment,\n            pair: [leftCore, rightCore],\n          })),\n        }\n      );\n    }\n    return matchLocally(left, right);\n  };\n}\n\n/** Content-bearing symbols, sorted, with their information-weighted counts. */\ninterface WeightedContent {\n  symbols: Int32Array;\n  weightedCounts: Int32Array;\n  total: number;\n}\n\n/**\n * A structural match must be backed by shared content: more than `minContentSimilarityPercent` of\n * the larger side's information-weighted names and literal values. Two sides without content never\n * pass.\n */\nfunction sharesContent(left: WeightedContent | undefined, right: WeightedContent | undefined): boolean {\n  if (!left || !right) {\n    return false;\n  }\n  let overlap = 0;\n  for (let leftIndex = 0, rightIndex = 0; leftIndex < left.symbols.length && rightIndex < right.symbols.length;) {\n    const leftSymbol = left.symbols[leftIndex] ?? 0;\n    const rightSymbol = right.symbols[rightIndex] ?? 0;\n    if (leftSymbol === rightSymbol) {\n      overlap += Math.min(left.weightedCounts[leftIndex] ?? 0, right.weightedCounts[rightIndex] ?? 0);\n    }\n    if (leftSymbol <= rightSymbol) {\n      leftIndex += 1;\n    }\n    if (rightSymbol <= leftSymbol) {\n      rightIndex += 1;\n    }\n  }\n  return overlap * 100 > minContentSimilarityPercent * Math.max(left.total, right.total);\n}\n\n/**\n * The block's units (its top-level statements, given in file token indexes, and the token runs\n * between them), each anonymized on its own and sorted, concatenated, with the block offset of\n * every token; undefined with too few statements. `tokenKeys` holds the content key of each\n * non-identifier token of the block.\n */\nfunction canonicalSequenceOf(\n  symbols: Int32Array,\n  tokenKeys: Float64Array,\n  statements: [number, number][],\n  blockStart: number\n): NormalizedBlock['canonical'] {\n  if (statements.length < minReorderStatementCount) {\n    return undefined;\n  }\n  const units: { unit: Int32Array; content: Float64Array; start: number }[] = [];\n  const pushUnit = (start: number, end: number): void => {\n    const unit = anonymize(symbols.subarray(start, end));\n    // Ordered by the content of their tokens: symbols, numbered by first occurrence, would order\n    // the units by where other code stands.\n    const content = Float64Array.from(unit, (symbol, index) => (symbol < 0 ? symbol : (tokenKeys[start + index] ?? 0)));\n    units.push({ unit, content, start });\n  };\n  let cursor = 0;\n  for (const [statementStart, statementEnd] of statements) {\n    const start = statementStart - blockStart;\n    if (cursor < start) {\n      pushUnit(cursor, start);\n    }\n    pushUnit(start, statementEnd - blockStart);\n    cursor = statementEnd - blockStart;\n  }\n  if (cursor < symbols.length) {\n    pushUnit(cursor, symbols.length);\n  }\n  units.sort((left, right) => compareSequences(left.content, right.content) || left.start - right.start);\n  const sequence = new Int32Array(symbols.length);\n  const offsets = new Int32Array(symbols.length);\n  let position = 0;\n  for (const { unit, start } of units) {\n    sequence.set(unit, position);\n    for (let index = 0; index < unit.length; index += 1) {\n      offsets[position + index] = start + index;\n    }\n    position += unit.length;\n  }\n  return { sequence, offsets };\n}\n\n/** Lexicographic order. */\nfunction compareSequences(left: Float64Array, right: Float64Array): number {\n  for (let index = 0; index < Math.min(left.length, right.length); index += 1) {\n    const difference = (left[index] ?? 0) - (right[index] ?? 0);\n    if (difference !== 0) {\n      return difference;\n    }\n  }\n  return left.length - right.length;\n}\n\n/**\n * The longest chain of anchors increasing in both blocks (anchors arrive sorted by left offset),\n * via patience sorting over right offsets.\n */\nfunction longestIncreasingChain(anchors: [number, number][]): [number, number][] {\n  const tailIndexes: number[] = [];\n  const predecessors: number[] = [];\n  for (const [index, [, rightOffset]] of anchors.entries()) {\n    let low = 0;\n    let high = tailIndexes.length;\n    while (low < high) {\n      const middle = (low + high) >>> 1;\n      if ((anchors[tailIndexes[middle] ?? 0]?.[1] ?? 0) < rightOffset) {\n        low = middle + 1;\n      } else {\n        high = middle;\n      }\n    }\n    predecessors.push(low > 0 ? (tailIndexes[low - 1] ?? -1) : -1);\n    tailIndexes[low] = index;\n  }\n  const chain: [number, number][] = [];\n  for (let cursor = tailIndexes.at(-1) ?? -1; cursor >= 0; cursor = predecessors[cursor] ?? -1) {\n    const anchor = anchors[cursor];\n    if (anchor) {\n      chain.push(anchor);\n    }\n  }\n  return chain.toReversed();\n}\n\n/** The chain's segments, split where consecutive anchors lie more than `maxAnchorGapTokens` apart in either block. */\nfunction chainSegments(chain: [number, number][]): [number, number][][] {\n  const segments: [number, number][][] = [];\n  for (const [index, anchor] of chain.entries()) {\n    const previous = chain[index - 1];\n    const continues =\n      previous !== undefined &&\n      anchor[0] - (previous[0] + ngramSize) <= maxAnchorGapTokens &&\n      anchor[1] - (previous[1] + ngramSize) <= maxAnchorGapTokens;\n    if (continues) {\n      segments.at(-1)?.push(anchor);\n    } else {\n      segments.push([anchor]);\n    }\n  }\n  return segments;\n}\n\n/** Left-block tokens the segment's anchors cover (overlapping anchors count once). */\nfunction anchoredTokenCount(segment: [number, number][]): number {\n  let count = ngramSize;\n  for (let index = 1; index < segment.length; index += 1) {\n    count += Math.min((segment[index]?.[0] ?? 0) - (segment[index - 1]?.[0] ?? 0), ngramSize);\n  }\n  return count;\n}\n\n/** Multiset intersection size of two ascending arrays. */\nfunction sortedOverlap(left: Int32Array, right: Int32Array): number {\n  let overlap = 0;\n  let leftIndex = 0;\n  let rightIndex = 0;\n  while (leftIndex < left.length && rightIndex < right.length) {\n    const leftValue = left[leftIndex] ?? 0;\n    const rightValue = right[rightIndex] ?? 0;\n    if (leftValue === rightValue) {\n      overlap += 1;\n      leftIndex += 1;\n      rightIndex += 1;\n    } else if (leftValue < rightValue) {\n      leftIndex += 1;\n    } else {\n      rightIndex += 1;\n    }\n  }\n  return overlap;\n}\n\n/**\n * Normalizes every block like the within-file detector. Non-identifier symbols are interned\n * project-wide from the tokens' hash pairs, so equal tokens compare equal across files, while\n * identifiers are interned per file and re-anonymized per compared range.\n */\nfunction normalizeBlocks(files: NearMissSourceFile[]): NormalizedBlock[] {\n  const symbolByTokenKey = new Map<number, number>();\n  const blocks: NormalizedBlock[] = [];\n  for (const [fileIndex, { tokens, containerStatements, nearMissBlocks }] of files.entries()) {\n    if (!tokens || !nearMissBlocks?.length) {\n      continue;\n    }\n    const symbols = new Int32Array(tokens.length);\n    const isContent = new Uint8Array(tokens.length);\n    const fileTokenKeys = new Float64Array(tokens.length);\n    const idByIdentifier = new Map<string, number>();\n    for (const [index, token] of tokens.entries()) {\n      if (token.kind === 'id') {\n        let id = idByIdentifier.get(token.text);\n        if (id === undefined) {\n          id = idByIdentifier.size;\n          idByIdentifier.set(token.text, id);\n        }\n        symbols[index] = -(id + 1);\n        continue;\n      }\n      const key = tokenKey(token);\n      fileTokenKeys[index] = key;\n      let symbol = symbolByTokenKey.get(key);\n      if (symbol === undefined) {\n        symbol = symbolByTokenKey.size;\n        symbolByTokenKey.set(key, symbol);\n      }\n      symbols[index] = symbol;\n      isContent[index] = token.isName || token.literalHash !== undefined ? 1 : 0;\n    }\n    const findStatements = createTopLevelStatementFinder(containerStatements ?? []);\n    for (const range of nearMissBlocks) {\n      const { startTokenIndex: start, endTokenIndex: end } = range;\n      const blockSymbols = symbols.subarray(start, end);\n      const blockIsContent = isContent.subarray(start, end);\n      const sequence = anonymize(blockSymbols);\n      const ngramHashes = collectNgramHashes(blockSymbols, fileTokenKeys.subarray(start, end));\n      const occurrenceCounts = new Map<number, number>();\n      for (const hash of ngramHashes) {\n        occurrenceCounts.set(hash, (occurrenceCounts.get(hash) ?? 0) + 1);\n      }\n      const uniqueOffsets = ngramHashes\n        .keys()\n        .filter((offset) => occurrenceCounts.get(ngramHashes[offset] ?? 0) === 1)\n        .toArray()\n        .toSorted((first, second) => (ngramHashes[first] ?? 0) - (ngramHashes[second] ?? 0));\n      blocks.push({\n        fileIndex,\n        range,\n        symbols: blockSymbols,\n        isContent: blockIsContent,\n        fileTokenKeys,\n        sequence,\n        writtenContent: Float64Array.from(sequence, (symbol, index) =>\n          symbol < 0 ? symbol : (fileTokenKeys[start + index] ?? 0)\n        ),\n        sortedSequence: sequence.toSorted(),\n        ngrams: Int32Array.from(occurrenceCounts.keys()),\n        uniqueNgrams: Int32Array.from(uniqueOffsets, (offset) => ngramHashes[offset] ?? 0),\n        uniqueNgramOffsets: Int32Array.from(uniqueOffsets),\n        contentCounts: countContent(blockSymbols, blockIsContent, 0, blockSymbols.length),\n        canonical: canonicalSequenceOf(\n          blockSymbols,\n          fileTokenKeys.subarray(start, end),\n          findStatements(start, end),\n          start\n        ),\n      });\n    }\n  }\n  return blocks;\n}\n\n/**\n * Returns a lookup of the outermost container statements inside a token range, excluding a\n * statement spanning the whole range (the block itself).\n */\nfunction createTopLevelStatementFinder(\n  containerStatements: TokenRange[][]\n): (start: number, end: number) => [number, number][] {\n  const statements = containerStatements\n    .flat()\n    .filter((statement) => statement.startTokenIndex < statement.endTokenIndex)\n    .map((statement): [number, number] => [statement.startTokenIndex, statement.endTokenIndex])\n    .toSorted((left, right) => left[0] - right[0] || right[1] - left[1]);\n  return (start, end) => {\n    let low = 0;\n    let high = statements.length;\n    while (low < high) {\n      const middle = (low + high) >>> 1;\n      if ((statements[middle]?.[0] ?? 0) < start) {\n        low = middle + 1;\n      } else {\n        high = middle;\n      }\n    }\n    const topLevel: [number, number][] = [];\n    for (let index = low; index < statements.length; index += 1) {\n      const statement = statements[index];\n      if (!statement || statement[0] >= end) {\n        break;\n      }\n      const last = topLevel.at(-1);\n      const nested = last !== undefined && statement[0] < last[1];\n      if (statement[1] <= end && !(statement[0] === start && statement[1] === end) && !nested) {\n        topLevel.push(statement);\n      }\n    }\n    return topLevel;\n  };\n}\n\nfunction countContent(symbols: Int32Array, isContent: Uint8Array, start: number, end: number): Map<number, number> {\n  const counts = new Map<number, number>();\n  for (let offset = start; offset < end; offset += 1) {\n    if (isContent[offset] === 1) {\n      const symbol = symbols[offset] ?? 0;\n      counts.set(symbol, (counts.get(symbol) ?? 0) + 1);\n    }\n  }\n  return counts;\n}\n\n/** Identifiers renumbered by first occurrence within `symbols`, so a range compares the same wherever it sits in its file. */\nfunction anonymize(symbols: Int32Array): Int32Array {\n  const indexByIdentifier = new Map<number, number>();\n  return symbols.map((symbol) => {\n    if (symbol >= 0) {\n      return symbol;\n    }\n    let index = indexByIdentifier.get(symbol);\n    if (index === undefined) {\n      index = indexByIdentifier.size;\n      indexByIdentifier.set(symbol, index);\n    }\n    return -(index + 1);\n  });\n}\n\n/**\n * A 53-bit key from the token's two independent text hashes, each mixed with the matching literal\n * value hash: exact in a JavaScript number, so interning never merges distinct tokens unless 53\n * hash bits collide.\n */\nfunction tokenKey(token: Token): number {\n  const primary = token.textHash ^ Math.imul(token.literalHash ?? 0, 0x9E_37_79_B1);\n  const secondary = token.textHash2 ^ Math.imul(token.literalHash2 ?? 0, 0x85_EB_CA_6B);\n  return (primary >>> 0) * 0x20_00_00 + (secondary >>> 11);\n}\n\n/**\n * N-gram hash per start offset, identifier-blind (every identifier hashes as -1) so a block copied\n * into different surroundings (renumbering its identifiers) or with reordered statements still\n * shares its n-grams. Hashed from the content keys of the tokens: with symbols, numbered by first\n * occurrence, which n-grams collide would depend on the order of the files.\n */\nfunction collectNgramHashes(symbols: Int32Array, tokenKeys: Float64Array): Int32Array {\n  const hashes = new Int32Array(Math.max(symbols.length - ngramSize + 1, 0));\n  for (let start = 0; start < hashes.length; start += 1) {\n    let hash = 5381;\n    for (let offset = 0; offset < ngramSize; offset += 1) {\n      const tokenKey = tokenKeys[start + offset] ?? 0;\n      // A token key holds 53 bits; both halves enter the hash.\n      const value = (symbols[start + offset] ?? 0) < 0 ? -1 : tokenKey ^ Math.floor(tokenKey / 0x1_00_00_00_00);\n      // oxlint-disable-next-line unicorn/prefer-math-trunc -- `| 0` wraps the sum to int32.\n      hash = (Math.imul(hash, 31) + value) | 0;\n    }\n    hashes[start] = hash;\n  }\n  return hashes;\n}\n"],"mappings":"sFA+IA,SAAgB,EACd,EACA,EACA,EACA,EACwB,CACxB,GAAI,GAAwB,IAC1B,MAAO,CAAC,EAEV,IAAM,EAAS,EAAgB,CAAK,EAC9B,EAAU,EAAc,EAAQ,EAAW,CAAoB,EAC/D,EAAuB,EAAoB,IAAI,CAAiB,EAChE,EAAyB,EAAoB,IAAI,CAAkB,EACnE,EAAgB,EAAO,KAAK,CAAE,YAAW,WAAY,EAAuB,EAAU,GAAG,CAAK,GAAK,EAAK,EACxG,EAAkB,EAAO,KAAK,CAAE,YAAW,WAAY,EAAqB,EAAU,GAAG,CAAK,GAAK,EAAK,EACxG,GAAe,CAAE,aAA8B,CAAC,EAAiB,KACrE,EAAqB,EAAU,GAAG,CAAE,kBAAiB,eAAc,CAAC,GAAK,GACrE,EAAqB,CAAC,EAC5B,EAAqB,EAAQ,EAAe,GAAuB,EAAM,IAAU,CACjF,IAAM,EAAY,EAAO,GACnB,EAAa,EAAO,GACpB,EAAQ,GAAa,GAAc,EAAQ,EAAW,EAAY,CAAK,EACxE,KAGL,IAAI,CAAC,EAAM,MAAO,CACV,EAAgB,IAAS,EAAgB,IAC7C,EAAM,KAAK,CAAE,OAAM,QAAO,UAAW,EAAM,SAAU,CAAC,EAExD,MACF,CACA,IAAK,GAAM,CAAE,OAAM,eAAe,EAAM,MAChC,EAAY,EAAW,EAAK,EAAE,GAAK,EAAY,EAAY,EAAK,EAAE,GACtE,EAAM,KAAK,CAAE,OAAM,QAAO,MAAO,EAAM,WAAU,CAAC,CAHtD,CAMF,CAAC,EAKD,IAAM,EAAe,EAAO,QAAU,EAAK,EACrC,EAAa,EAAO,QAA8B,CAAC,CAAC,EAC1D,IAAK,IAAM,KAAQ,EACjB,IAAK,GAAM,CAAC,EAAO,KAAS,EAAQ,CAAI,EAClC,EACF,EAAW,EAAM,EAAE,KAAK,CAAI,EAE5B,EAAa,GAAS,GAI5B,IAAM,EAAsE,CAAC,EACvE,EAA6B,CAAC,EACpC,IAAK,IAAM,KAAc,EAAO,KAAK,EAAG,CACtC,EAAiB,KAAK,EAAM,MAAM,EAClC,IAAM,EAAQ,EAAa,GAAc,CAAC,EAAI,EAAsB,EAAW,IAAe,CAAC,CAAC,EAC5F,EAAM,SAAW,GACnB,EAAM,KAAK,CAAE,aAAY,KAAM,IAAA,EAAU,CAAC,EAE5C,IAAK,IAAM,KAAQ,EACjB,EAAM,KAAK,CAAE,aAAY,MAAK,CAAC,CAEnC,CACA,IAAM,GAAU,EAAoB,IAA+C,CACjF,IAAM,EAAQ,EAAiB,IAAe,EAC9C,GAAI,CAAC,GAAQ,EAAa,GACxB,OAAO,EAET,IAAK,IAAI,EAAO,EAAO,EAAM,EAAK,EAAE,aAAe,EAAY,GAAQ,EAAG,CACxE,IAAM,EAAO,EAAM,EAAK,EAAE,KAC1B,GAAI,GAAQ,EAAK,IAAM,EAAK,IAAM,EAAK,IAAM,EAAK,GAChD,OAAO,CAEX,CACA,MAAU,MAAM,0DAA0D,CAC5E,EAEM,EAAS,EAAM,KAAK,EAAG,IAAU,CAAK,EACtC,EAAQ,GAA0B,CACtC,IAAI,EAAO,EACX,KAAO,EAAO,KAAU,GACtB,EAAO,EAAO,IAAS,EAEzB,IAAK,IAAI,EAAU,EAAO,EAAO,KAAa,GAAO,CACnD,IAAM,EAAO,EAAO,IAAY,EAChC,EAAO,GAAW,EAClB,EAAU,CACZ,CACA,OAAO,CACT,EAGM,EAAW,EAAM,KAAK,CAAE,aAAY,UAAW,CACnD,IAAM,EAAQ,EAAO,GACf,CAAC,EAAiB,GAAiB,GAAQ,CAC/C,GAAO,MAAM,iBAAmB,EAChC,GAAO,MAAM,eAAiB,CAChC,EACA,OAAO,EAAqB,GAAO,WAAa,EAAE,GAAG,CAAE,kBAAiB,eAAc,CAAC,GAAK,EAC9F,CAAC,EACK,EAAY,EAAM,IAAK,GAAS,EAAQ,CAAI,CAAC,CAAC,KAAK,CAAC,EAAO,KAAU,EAAO,EAAO,CAAI,CAAC,CAAC,EAC/F,IAAK,GAAM,CAAC,EAAW,EAAG,EAAY,KAAM,EAAW,CACrD,IAAM,EAAW,EAAK,CAAQ,EACxB,EAAY,EAAK,CAAS,EAChC,EAAO,KAAK,IAAI,EAAU,CAAS,GAAK,KAAK,IAAI,EAAU,CAAS,CACtE,CACA,IAAM,EAAoB,EAAW,EAAQ,EAAO,EAAW,EAAM,MAAM,EAErE,kBAAgB,IAAI,IAC1B,IAAK,IAAM,KAAQ,EAAM,KAAK,EAAG,CAC/B,IAAM,EAAO,EAAK,CAAI,EAChB,EAAU,EAAc,IAAI,CAAI,GAAK,CAAC,EAC5C,EAAQ,KAAK,CAAI,EACjB,EAAc,IAAI,EAAM,CAAO,CACjC,CACA,IAAM,EAAiC,CAAC,EACxC,IAAK,IAAM,KAAW,EAAc,OAAO,EAAG,CAE5C,GAAI,EAAQ,OAAS,GAAK,EAAQ,MAAO,GAAS,EAAS,EAAK,EAC9D,SAIF,IAAM,kBAAe,IAAI,IACzB,IAAK,IAAM,KAAQ,EAAS,CAC1B,GAAM,CAAE,aAAa,EAAG,QAAS,EAAM,IAAS,CAAC,EAC3C,EAAQ,EAAa,IAAI,CAAU,GAAK,CAAE,MAAO,CAAC,EAAG,OAAQ,GAAO,YAAa,CAAC,CAAE,EAC1F,EAAM,MAAM,KAAK,CAAI,EACrB,EAAM,SAAW,EAAS,IAAS,GACnC,EAAM,YAAY,KAAK,GAAI,EAAkB,IAAS,CAAC,CAAE,EACzD,EAAa,IAAI,EAAY,CAAK,CACpC,CACA,EAAO,KACL,CAAC,GAAG,CAAY,CAAC,CAAC,SAAS,CAAC,EAAY,KAAU,CAChD,IAAM,EAAQ,EAAO,GACrB,OAAO,EAAQ,CAAC,EAAa,EAAO,EAAO,CAAI,CAAC,EAAI,CAAC,CACvD,CAAC,CACH,CACF,CACA,OAAO,CACT,CAUA,SAAS,EAAQ,CAAE,OAAM,QAAO,SAA8D,CAC5F,MAAO,CACL,CAAC,EAAM,IAAQ,EAAE,EACjB,CAAC,EAAO,IAAQ,EAAE,CACpB,CACF,CAQA,SAAS,EACP,EACA,EACA,EACA,EACsB,CACtB,IAAM,EAAoB,MAAM,KAAK,CAAE,OAAQ,CAAU,MAA6B,CAAC,CAAC,EACxF,IAAK,GAAM,CAAC,EAAW,KAAS,EAAM,QAAQ,EAAG,CAC/C,IAAM,EAAO,EAAO,EAAK,MACnB,EAAQ,EAAO,EAAK,OAC1B,GAAI,GAAQ,EAAO,CACjB,GAAM,CAAC,EAAU,GAAa,EAAU,EAAM,EAAO,CAAI,EACzD,EAAkB,EAAU,EAAU,GAAG,IAAM,EAAE,EAAE,KAAK,GAAG,CAAQ,EACnE,EAAkB,EAAU,EAAU,GAAG,IAAM,EAAE,EAAE,KAAK,GAAG,CAAS,CACtE,CACF,CACA,OAAO,CACT,CAMA,SAAS,EACP,EACA,EACA,CAAE,QAAO,aACiC,CAC1C,IAAM,EAAe,EAAiB,EAAM,IAAQ,GAAI,CAAS,EAC3D,EAAgB,EAAiB,EAAO,IAAQ,GAAI,CAAS,EAGnE,GAAI,EAAiB,EAAU,EAAM,IAAQ,EAAE,EAAG,EAAU,EAAO,IAAQ,EAAE,CAAC,GAAK,EAAG,CACpF,GAAM,CAAC,EAAa,GAAgB,EAAQ,EAAa,SAAU,EAAc,QAAQ,EACzF,MAAO,CAAC,EAAO,EAAa,UAAW,CAAW,EAAG,EAAO,EAAc,UAAW,CAAY,CAAC,CACpG,CACA,GAAM,CAAC,EAAc,GAAe,EAAQ,EAAc,SAAU,EAAa,QAAQ,EACzF,MAAO,CAAC,EAAO,EAAa,UAAW,CAAW,EAAG,EAAO,EAAc,UAAW,CAAY,CAAC,CACpG,CAMA,SAAS,EAAU,EAAwB,EAAkD,CAC3F,GAAI,CAAC,EACH,OAAO,EAAM,eAEf,GAAM,CAAE,WAAU,aAAc,EAAiB,EAAO,EAAM,EAAK,EACnE,OAAO,aAAa,KAAK,GAAW,EAAQ,IAC1C,EAAS,EAAI,EAAU,EAAM,cAAc,EAAU,IAAU,IAAM,CACvE,CACF,CAGA,SAAS,EAAO,EAAuB,EAAyC,CAC9E,IAAM,EAA2B,CAAC,EAClC,IAAK,IAAM,KAAY,EAAU,QAAQ,EAAG,IAAU,EAAQ,KAAW,CAAC,CAAC,CAAC,SAAS,EAAG,CACtF,IAAM,EAAO,EAAK,GAAG,EAAE,EACnB,IAAO,KAAO,EAChB,EAAK,GAAK,EAAW,EAErB,EAAK,KAAK,CAAC,EAAU,EAAW,CAAC,CAAC,CAEtC,CACA,OAAO,CACT,CAMA,SAAS,EACP,EACA,EACA,EACiD,CACjD,IAAM,EAAa,EAAM,MAAM,gBAa/B,OAZI,EACK,CACL,SAAU,EAAU,EAAM,QAAQ,SAAS,EAAK,GAAK,EAAY,EAAK,GAAK,CAAU,CAAC,EACtF,UAAW,WAAW,KAAK,CAAE,OAAQ,EAAK,GAAK,EAAK,EAAG,GAAI,EAAG,IAAU,EAAK,GAAK,CAAK,CACzF,EAEE,GAAa,EAAM,UACd,CACL,SAAU,EAAM,UAAU,SAC1B,UAAW,EAAM,UAAU,QAAQ,IAAK,GAAW,EAAa,CAAM,CACxE,EAEK,CAAE,SAAU,EAAM,SAAU,UAAW,EAAM,SAAS,KAAK,EAAG,IAAU,EAAa,CAAK,CAAE,CACrG,CAGA,SAAS,EAAsB,EAA+C,CAC5E,IAAM,EAA6B,CAAC,EACpC,IAAK,GAAM,CAAC,EAAO,KAAQ,EAAM,UAAU,EAAM,IAAU,EAAK,GAAK,EAAM,EAAE,EAAG,CAC9E,IAAM,EAAO,EAAO,GAAG,EAAE,EACrB,GAAQ,EAAQ,EAAK,GACvB,EAAK,GAAK,KAAK,IAAI,EAAK,GAAI,CAAG,EAE/B,EAAO,KAAK,CAAC,EAAO,CAAG,CAAC,CAE5B,CACA,OAAO,CACT,CAGA,SAAS,EACP,EACwE,CAExE,IAAM,EAA6B,CAAC,EACpC,IAAK,GAAM,CAAE,kBAAiB,mBAAmB,EAAM,UACpD,EAAM,IAAU,EAAK,gBAAkB,EAAM,eAChD,EAAG,CACD,IAAM,EAAO,EAAO,GAAG,EAAE,EACrB,GAAQ,GAAmB,EAAK,GAClC,EAAK,GAAK,KAAK,IAAI,EAAK,GAAI,CAAa,EAEzC,EAAO,KAAK,CAAC,EAAiB,CAAa,CAAC,CAEhD,CACA,MAAQ,IAAU,CAChB,IAAI,EAAM,EACN,EAAO,EAAO,OAClB,KAAO,EAAM,GAAM,CACjB,IAAM,EAAU,EAAM,IAAU,GAC3B,EAAO,EAAO,GAAG,IAAM,IAAM,EAAM,gBACtC,EAAM,EAAS,EAEf,EAAO,CAEX,CACA,OAAQ,EAAO,EAAM,EAAE,GAAG,IAAM,KAAO,EAAM,aAC/C,CACF,CAMA,SAAS,EACP,EACwE,CACxE,IAAM,EAAS,EAAM,UAAU,EAAM,IAAU,EAAK,gBAAkB,EAAM,eAAe,EACrF,EAAe,IAAI,WAAW,EAAO,MAAM,EAC7C,EAAS,GACb,IAAK,GAAM,CAAC,EAAO,KAAS,EAAO,QAAQ,EACzC,EAAS,KAAK,IAAI,EAAQ,EAAK,aAAa,EAC5C,EAAa,GAAS,EAExB,MAAQ,IAAU,CAChB,IAAI,EAAM,EACN,EAAO,EAAO,OAClB,KAAO,EAAM,GAAM,CACjB,IAAM,EAAU,EAAM,IAAU,GAC3B,EAAO,EAAO,EAAE,iBAAmB,GAAK,EAAM,cACjD,EAAM,EAAS,EAEf,EAAO,CAEX,CACA,OAAO,EAAM,IAAM,EAAa,EAAM,IAAM,IAAM,EAAM,eAC1D,CACF,CAOA,SAAS,EACP,CAAE,YAAW,SACb,EACA,CAAE,QAAO,SAAQ,eACG,CACpB,IAAM,EAAQ,EAAM,SAAS,IAAA,EAAS,EAChC,EAAW,EACd,IAAK,GAA2B,GAAQ,CAAC,EAAM,gBAAiB,EAAM,aAAa,CAAC,CAAC,CACrF,UAAU,EAAM,IAAU,EAAK,GAAK,EAAM,EAAE,CAAC,CAC7C,KAAK,CAAC,EAAiB,MAAoB,CAAE,kBAAiB,eAAc,EAAE,EAC3E,EAAQ,EAAS,EAAE,EAAE,iBAAmB,EAAM,gBAC9C,EAAM,EAAS,GAAG,EAAE,CAAC,EAAE,eAAiB,EAAM,cAC9C,EAAS,EAAM,EAAU,EAAE,OACjC,MAAO,CACL,YACA,qBAAsB,GAAU,IAAA,GAChC,WACA,YAAa,EAAsB,CAAW,CAAC,CAAC,KAAK,CAAC,EAAiB,MAAkC,CACvG,kBACA,eACF,EAAE,EACF,WAAY,EAAS,QAAQ,EAAK,IAAY,EAAM,EAAQ,cAAgB,EAAQ,gBAAiB,CAAC,EACtG,gBAAiB,EACjB,cAAe,EACf,WAAY,EAAM,WAClB,SAAU,EAAM,SAChB,UAAW,EAAQ,EAAM,WAAa,IAAS,EAAM,EAAE,UAAY,GAAK,EACxE,QAAS,EAAQ,EAAM,SAAW,IAAS,EAAM,EAAE,EAAE,QAAU,GAAK,CACtE,CACF,CAUA,SAAS,EACP,EACA,EACA,EACA,EACM,CACN,IAAM,kBAAiB,IAAI,IAC3B,IAAK,IAAM,KAAS,EAClB,IAAK,IAAM,KAAS,EAAM,OACxB,EAAe,IAAI,GAAQ,EAAe,IAAI,CAAK,GAAK,GAAK,CAAC,EAGlE,IAAK,IAAM,KAAS,EAClB,EAAM,OAAS,EAAM,OAAO,OAAQ,IAAW,EAAe,IAAI,CAAK,GAAK,IAAM,GAAsB,EAI1G,IAAM,EAAc,WAAW,KAAK,EAAS,GAAU,EAAM,SAAS,EAChE,EAAgB,WAAW,KAAK,EAAe,MAAM,EACrD,EAAU,WAAW,KAAK,EAAS,GAAU,EAAM,SAAS,MAAM,EAClE,EAAc,WAAW,KAAK,EAAS,GAAU,EAAM,OAAO,MAAM,EACpE,EAAQ,CAAC,GAAG,EAAO,KAAK,CAAC,CAAC,CAAC,UAAU,EAAM,KAAW,EAAQ,IAAS,IAAM,EAAQ,IAAU,EAAE,EACjG,kBAAW,IAAI,IACf,EAAe,IAAI,WAAW,EAAO,MAAM,EAC3C,EAAoB,CAAC,EAC3B,IAAK,IAAM,KAAS,EAAO,CACzB,IAAM,EAAY,EAAY,GACxB,EAAgB,EAAc,KAAW,EACzC,EAAgB,KAAK,IACzB,KAAK,MAAM,EAAQ,IAAU,GAAK,CAAc,EAChD,KAAK,KAAM,GAAwB,EAAQ,IAAU,GAAM,GAAG,CAChE,EACM,EAAS,EAAO,EAAM,EAAE,QAAU,CAAC,EACzC,IAAK,IAAM,KAAS,EAAQ,CAC1B,IAAM,EAAU,EAAS,IAAI,CAAK,EAClC,GAAI,CAAC,EAAS,CACZ,EAAS,IAAI,EAAO,CAAC,CAAK,CAAC,EAC3B,QACF,CACA,IAAK,IAAI,EAAW,EAAQ,OAAS,EAAG,GAAY,EAAG,IAAe,CACpE,IAAM,EAAO,EAAQ,IAAa,EAClC,IAAK,EAAQ,IAAS,GAAK,EACzB,MAEE,EAAY,KAAU,GAAc,GAAiB,EAAc,KAAU,IAG7E,EAAa,KAAU,GACzB,EAAQ,KAAK,CAAI,EAEnB,EAAa,IAAS,EAAa,IAAS,GAAK,EACnD,CACA,EAAQ,KAAK,CAAK,CACpB,CACA,IAAK,IAAM,KAAQ,EAAS,CAC1B,IAAM,EAAS,EAAa,IAAS,EACrC,EAAa,GAAQ,EACjB,EAAS,KAAO,GAAoB,KAAK,IAAI,EAAY,IAAS,EAAG,EAAO,MAAM,GACpF,EAAM,EAAM,CAAK,CAErB,CACA,EAAQ,OAAS,CACnB,CACF,CAQA,SAAS,EACP,EACA,EACA,EAC8F,CAC9F,IAAM,kBAAsB,IAAI,IAChC,IAAK,IAAM,KAAS,EAClB,IAAK,IAAM,KAAU,EAAM,cAAc,KAAK,EAC5C,EAAoB,IAAI,GAAS,EAAoB,IAAI,CAAM,GAAK,GAAK,CAAC,EAG9E,IAAM,EAAmB,GACvB,KAAK,IAAI,GAAK,KAAK,MAAM,KAAK,OAAO,EAAO,OAAS,GAAK,CAAiB,CAAC,EAAI,EAAG,CAAgB,EAC/F,kBAAU,IAAI,IACpB,IAAK,GAAM,CAAC,EAAQ,KAAc,EAChC,EAAQ,IAAI,EAAQ,EAAgB,CAAS,CAAC,EAEhD,IAAM,EAAS,GAAiD,CAC9D,IAAM,EAAU,WAAW,KAAK,EAAO,KAAK,CAAC,CAAC,CAAC,SAAS,EAElD,EAAiB,WAAW,KAAK,EAAU,IAAY,EAAO,IAAI,CAAM,GAAK,IAAM,EAAQ,IAAI,CAAM,GAAK,EAAE,EAC9G,EAAQ,EACZ,IAAK,IAAM,KAAS,EAClB,GAAS,EAEX,MAAO,CAAE,UAAS,iBAAgB,OAAM,CAC1C,EACM,EAAgB,IAAI,IAAI,EAAO,IAAK,GAAU,CAAC,EAAO,EAAM,EAAM,aAAa,CAAC,CAAC,CAAC,EAIpF,EAAe,GACf,EACE,GAAsB,EAAwB,EAAoB,MAClE,IAAiB,GAAc,CAAC,KAClC,EAAe,EACf,EAAU,EAAuB,EAAM,QAAQ,GAE1C,EAAQ,CAAQ,GAYnB,GAAgB,EAAuB,IAAkD,CAC7F,IAAM,EAA8B,CAAC,EACrC,IACE,IAAI,EAAY,EAAG,EAAa,EAChC,EAAY,EAAK,aAAa,QAAU,EAAa,EAAM,aAAa,QACxE,CACA,IAAM,EAAW,EAAK,aAAa,IAAc,EAC3C,EAAY,EAAM,aAAa,IAAe,EAChD,IAAa,GACf,EAAQ,KAAK,CAAC,EAAK,mBAAmB,IAAc,EAAG,EAAM,mBAAmB,IAAe,CAAC,CAAC,EAE/F,GAAY,IACd,GAAa,GAEX,GAAa,IACf,GAAc,EAElB,CACA,EAAQ,MAAM,EAAO,IAAW,EAAM,GAAK,EAAO,EAAE,EAGpD,IAAM,EAAa,EAAQ,QAAQ,CAAC,EAAY,GAAc,IAAU,CACtE,IAAM,EAAW,EAAQ,EAAQ,GAC3B,EAAO,EAAQ,EAAQ,GAC7B,OACG,IAAW,KAAO,EAAa,GAAK,EAAS,KAAO,EAAc,GAClE,IAAO,KAAO,EAAa,GAAK,EAAK,KAAO,EAAc,CAE/D,CAAC,EACK,EAAa,EAAK,MAAM,gBACxB,EAAc,EAAM,MAAM,gBAC1B,EAA4C,CAAC,EACnD,IAAK,IAAM,KAAW,EAAc,EAAuB,CAAU,CAAC,EAAG,CACvE,GAAM,CAAC,EAAW,GAAc,EAAQ,IAAM,CAAC,EAAG,CAAC,EAC7C,CAAC,EAAU,GAAa,EAAQ,GAAG,EAAE,GAAK,CAAC,EAAG,CAAC,EAC/C,EAAU,EAAW,EACrB,EAAW,EAAY,EACvB,EAAa,EAAU,EACvB,EAAc,EAAW,EACzB,EAAU,KAAK,IAAI,EAAY,CAAW,EAC1C,EAAW,EAAuB,KAAK,IAAI,EAAY,CAAW,EAEtE,EAAU,GACV,EAAU,IAAM,GAChB,EAAmB,CAAO,EAAI,IAAM,GAA2B,GAC/D,CAAC,EACC,EAAM,EAAa,EAAK,QAAS,EAAK,UAAW,EAAW,CAAO,CAAC,EACpE,EAAM,EAAa,EAAM,QAAS,EAAM,UAAW,EAAY,CAAQ,CAAC,CAC1E,GAIoB,EACpB,EAAU,EAAK,QAAQ,SAAS,EAAW,CAAO,CAAC,EACnD,EAAU,EAAM,QAAQ,SAAS,EAAY,CAAQ,CAAC,CAExC,EAAI,KAAO,GACzB,EAAM,KAAK,CACT,KAAM,CACJ,CAAC,EAAa,EAAW,EAAa,CAAO,EAC7C,CAAC,EAAc,EAAY,EAAc,CAAQ,CACnD,EACA,UAAW,EACb,CAAC,CAEL,CACA,OAAO,EAAM,OAAS,EAAI,CAAE,OAAM,EAAI,IAAA,EACxC,EAGA,OAAQ,EAAM,EAAO,IAAe,CAClC,IAAM,EAAW,EAAuB,KAAK,IAAI,EAAK,SAAS,OAAQ,EAAM,SAAS,MAAM,EAC5F,GACE,KAAK,IAAI,EAAK,SAAS,OAAQ,EAAM,SAAS,MAAM,EAAI,KAAO,GAC/D,EAAc,EAAc,IAAI,CAAI,EAAG,EAAc,IAAI,CAAK,CAAC,EAC/D,CACA,GAAI,EAAc,EAAK,eAAgB,EAAM,cAAc,EAAI,KAAO,GACpD,EAAmB,EAAO,EAAY,EAAK,QACjD,EAAI,KAAO,EACnB,MAAO,CAAE,UAAW,EAAM,EAK9B,GAAI,EAAK,WAAa,EAAM,WACR,EAAU,EAAK,UAAU,SAAU,EAAM,UAAU,QACzD,EAAI,KAAO,EACrB,MAAO,CAAE,UAAW,EAAK,CAG/B,CAGA,GAAI,EAAiB,EAAM,eAAgB,EAAK,cAAc,EAAI,EAAG,CACnE,IAAM,EAAU,EAAa,EAAO,CAAI,EACxC,OACE,GAAS,OAAS,CAChB,MAAO,EAAQ,MAAM,KAAK,CAAE,KAAM,CAAC,EAAW,GAAW,GAAG,MAAiB,CAC3E,GAAG,EACH,KAAM,CAAC,EAAU,CAAS,CAC5B,EAAE,CACJ,CAEJ,CACA,OAAO,EAAa,EAAM,CAAK,CACjC,CACF,CAcA,SAAS,EAAc,EAAmC,EAA6C,CACrG,GAAI,CAAC,GAAQ,CAAC,EACZ,MAAO,GAET,IAAI,EAAU,EACd,IAAK,IAAI,EAAY,EAAG,EAAa,EAAG,EAAY,EAAK,QAAQ,QAAU,EAAa,EAAM,QAAQ,QAAS,CAC7G,IAAM,EAAa,EAAK,QAAQ,IAAc,EACxC,EAAc,EAAM,QAAQ,IAAe,EAC7C,IAAe,IACjB,GAAW,KAAK,IAAI,EAAK,eAAe,IAAc,EAAG,EAAM,eAAe,IAAe,CAAC,GAE5F,GAAc,IAChB,GAAa,GAEX,GAAe,IACjB,GAAc,EAElB,CACA,OAAO,EAAU,IAAM,GAA8B,KAAK,IAAI,EAAK,MAAO,EAAM,KAAK,CACvF,CAQA,SAAS,EACP,EACA,EACA,EACA,EAC8B,CAC9B,GAAI,EAAW,OAAS,EACtB,OAEF,IAAM,EAAsE,CAAC,EACvE,GAAY,EAAe,IAAsB,CACrD,IAAM,EAAO,EAAU,EAAQ,SAAS,EAAO,CAAG,CAAC,EAG7C,EAAU,aAAa,KAAK,GAAO,EAAQ,IAAW,EAAS,EAAI,EAAU,EAAU,EAAQ,IAAU,CAAG,EAClH,EAAM,KAAK,CAAE,OAAM,UAAS,OAAM,CAAC,CACrC,EACI,EAAS,EACb,IAAK,GAAM,CAAC,EAAgB,KAAiB,EAAY,CACvD,IAAM,EAAQ,EAAiB,EAC3B,EAAS,GACX,EAAS,EAAQ,CAAK,EAExB,EAAS,EAAO,EAAe,CAAU,EACzC,EAAS,EAAe,CAC1B,CACI,EAAS,EAAQ,QACnB,EAAS,EAAQ,EAAQ,MAAM,EAEjC,EAAM,MAAM,EAAM,IAAU,EAAiB,EAAK,QAAS,EAAM,OAAO,GAAK,EAAK,MAAQ,EAAM,KAAK,EACrG,IAAM,EAAW,IAAI,WAAW,EAAQ,MAAM,EACxC,EAAU,IAAI,WAAW,EAAQ,MAAM,EACzC,EAAW,EACf,IAAK,GAAM,CAAE,OAAM,WAAW,EAAO,CACnC,EAAS,IAAI,EAAM,CAAQ,EAC3B,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAK,OAAQ,GAAS,EAChD,EAAQ,EAAW,GAAS,EAAQ,EAEtC,GAAY,EAAK,MACnB,CACA,MAAO,CAAE,WAAU,SAAQ,CAC7B,CAGA,SAAS,EAAiB,EAAoB,EAA6B,CACzE,IAAK,IAAI,EAAQ,EAAG,EAAQ,KAAK,IAAI,EAAK,OAAQ,EAAM,MAAM,EAAG,GAAS,EAAG,CAC3E,IAAM,GAAc,EAAK,IAAU,IAAM,EAAM,IAAU,GACzD,GAAI,IAAe,EACjB,OAAO,CAEX,CACA,OAAO,EAAK,OAAS,EAAM,MAC7B,CAMA,SAAS,EAAuB,EAAiD,CAC/E,IAAM,EAAwB,CAAC,EACzB,EAAyB,CAAC,EAChC,IAAK,GAAM,CAAC,EAAO,EAAG,MAAiB,EAAQ,QAAQ,EAAG,CACxD,IAAI,EAAM,EACN,EAAO,EAAY,OACvB,KAAO,EAAM,GAAM,CACjB,IAAM,EAAU,EAAM,IAAU,GAC3B,EAAQ,EAAY,IAAW,EAAE,GAAG,IAAM,GAAK,EAClD,EAAM,EAAS,EAEf,EAAO,CAEX,CACA,EAAa,KAAK,EAAM,EAAK,EAAY,EAAM,IAAM,GAAM,EAAE,EAC7D,EAAY,GAAO,CACrB,CACA,IAAM,EAA4B,CAAC,EACnC,IAAK,IAAI,EAAS,EAAY,GAAG,EAAE,GAAK,GAAI,GAAU,EAAG,EAAS,EAAa,IAAW,GAAI,CAC5F,IAAM,EAAS,EAAQ,GACnB,GACF,EAAM,KAAK,CAAM,CAErB,CACA,OAAO,EAAM,WAAW,CAC1B,CAGA,SAAS,EAAc,EAAiD,CACtE,IAAM,EAAiC,CAAC,EACxC,IAAK,GAAM,CAAC,EAAO,KAAW,EAAM,QAAQ,EAAG,CAC7C,IAAM,EAAW,EAAM,EAAQ,GAE7B,IAAa,IAAA,IACb,EAAO,IAAM,EAAS,GAAK,IAAc,IACzC,EAAO,IAAM,EAAS,GAAK,IAAc,GAEzC,EAAS,GAAG,EAAE,CAAC,EAAE,KAAK,CAAM,EAE5B,EAAS,KAAK,CAAC,CAAM,CAAC,CAE1B,CACA,OAAO,CACT,CAGA,SAAS,EAAmB,EAAqC,CAC/D,IAAI,EAAQ,EACZ,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAQ,OAAQ,GAAS,EACnD,GAAS,KAAK,KAAK,EAAQ,EAAM,GAAG,IAAM,IAAM,EAAQ,EAAQ,EAAE,GAAG,IAAM,GAAI,CAAS,EAE1F,OAAO,CACT,CAGA,SAAS,EAAc,EAAkB,EAA2B,CAClE,IAAI,EAAU,EACV,EAAY,EACZ,EAAa,EACjB,KAAO,EAAY,EAAK,QAAU,EAAa,EAAM,QAAQ,CAC3D,IAAM,EAAY,EAAK,IAAc,EAC/B,EAAa,EAAM,IAAe,EACpC,IAAc,GAChB,GAAW,EACX,GAAa,EACb,GAAc,GACL,EAAY,EACrB,GAAa,EAEb,GAAc,CAElB,CACA,OAAO,CACT,CAOA,SAAS,EAAgB,EAAgD,CACvE,IAAM,kBAAmB,IAAI,IACvB,EAA4B,CAAC,EACnC,IAAK,GAAM,CAAC,EAAW,CAAE,SAAQ,sBAAqB,qBAAqB,EAAM,QAAQ,EAAG,CAC1F,GAAI,CAAC,GAAU,CAAC,GAAgB,OAC9B,SAEF,IAAM,EAAU,IAAI,WAAW,EAAO,MAAM,EACtC,EAAY,IAAI,WAAW,EAAO,MAAM,EACxC,EAAgB,IAAI,aAAa,EAAO,MAAM,EAC9C,kBAAiB,IAAI,IAC3B,IAAK,GAAM,CAAC,EAAO,KAAU,EAAO,QAAQ,EAAG,CAC7C,GAAI,EAAM,OAAS,KAAM,CACvB,IAAI,EAAK,EAAe,IAAI,EAAM,IAAI,EAClC,IAAO,IAAA,KACT,EAAK,EAAe,KACpB,EAAe,IAAI,EAAM,KAAM,CAAE,GAEnC,EAAQ,GAAS,EAAE,EAAK,GACxB,QACF,CACA,IAAM,EAAM,EAAS,CAAK,EAC1B,EAAc,GAAS,EACvB,IAAI,EAAS,EAAiB,IAAI,CAAG,EACjC,IAAW,IAAA,KACb,EAAS,EAAiB,KAC1B,EAAiB,IAAI,EAAK,CAAM,GAElC,EAAQ,GAAS,EACjB,EAAU,GAAS,EAAM,QAAU,EAAM,cAAgB,IAAA,GAAY,EAAI,CAC3E,CACA,IAAM,EAAiB,EAA8B,GAAuB,CAAC,CAAC,EAC9E,IAAK,IAAM,KAAS,EAAgB,CAClC,GAAM,CAAE,gBAAiB,EAAO,cAAe,GAAQ,EACjD,EAAe,EAAQ,SAAS,EAAO,CAAG,EAC1C,EAAiB,EAAU,SAAS,EAAO,CAAG,EAC9C,EAAW,EAAU,CAAY,EACjC,EAAc,EAAmB,EAAc,EAAc,SAAS,EAAO,CAAG,CAAC,EACjF,kBAAmB,IAAI,IAC7B,IAAK,IAAM,KAAQ,EACjB,EAAiB,IAAI,GAAO,EAAiB,IAAI,CAAI,GAAK,GAAK,CAAC,EAElE,IAAM,EAAgB,EACnB,KAAK,CAAC,CACN,OAAQ,GAAW,EAAiB,IAAI,EAAY,IAAW,CAAC,IAAM,CAAC,CAAC,CACxE,QAAQ,CAAC,CACT,UAAU,EAAO,KAAY,EAAY,IAAU,IAAM,EAAY,IAAW,EAAE,EACrF,EAAO,KAAK,CACV,YACA,QACA,QAAS,EACT,UAAW,EACX,gBACA,WACA,eAAgB,aAAa,KAAK,GAAW,EAAQ,IACnD,EAAS,EAAI,EAAU,EAAc,EAAQ,IAAU,CACzD,EACA,eAAgB,EAAS,SAAS,EAClC,OAAQ,WAAW,KAAK,EAAiB,KAAK,CAAC,EAC/C,aAAc,WAAW,KAAK,EAAgB,GAAW,EAAY,IAAW,CAAC,EACjF,mBAAoB,WAAW,KAAK,CAAa,EACjD,cAAe,EAAa,EAAc,EAAgB,EAAG,EAAa,MAAM,EAChF,UAAW,EACT,EACA,EAAc,SAAS,EAAO,CAAG,EACjC,EAAe,EAAO,CAAG,EACzB,CACF,CACF,CAAC,CACH,CACF,CACA,OAAO,CACT,CAMA,SAAS,EACP,EACoD,CACpD,IAAM,EAAa,EAChB,KAAK,CAAC,CACN,OAAQ,GAAc,EAAU,gBAAkB,EAAU,aAAa,CAAC,CAC1E,IAAK,GAAgC,CAAC,EAAU,gBAAiB,EAAU,aAAa,CAAC,CAAC,CAC1F,UAAU,EAAM,IAAU,EAAK,GAAK,EAAM,IAAM,EAAM,GAAK,EAAK,EAAE,EACrE,OAAQ,EAAO,IAAQ,CACrB,IAAI,EAAM,EACN,EAAO,EAAW,OACtB,KAAO,EAAM,GAAM,CACjB,IAAM,EAAU,EAAM,IAAU,GAC3B,EAAW,EAAO,GAAG,IAAM,GAAK,EACnC,EAAM,EAAS,EAEf,EAAO,CAEX,CACA,IAAM,EAA+B,CAAC,EACtC,IAAK,IAAI,EAAQ,EAAK,EAAQ,EAAW,OAAQ,GAAS,EAAG,CAC3D,IAAM,EAAY,EAAW,GAC7B,GAAI,CAAC,GAAa,EAAU,IAAM,EAChC,MAEF,IAAM,EAAO,EAAS,GAAG,EAAE,EACrB,EAAS,IAAS,IAAA,IAAa,EAAU,GAAK,EAAK,GACrD,EAAU,IAAM,IAAS,EAAU,KAAO,GAAS,EAAU,KAAO,IAAQ,CAAC,GAC/E,EAAS,KAAK,CAAS,CAE3B,CACA,OAAO,CACT,CACF,CAEA,SAAS,EAAa,EAAqB,EAAuB,EAAe,EAAkC,CACjH,IAAM,kBAAS,IAAI,IACnB,IAAK,IAAI,EAAS,EAAO,EAAS,EAAK,GAAU,EAC/C,GAAI,EAAU,KAAY,EAAG,CAC3B,IAAM,EAAS,EAAQ,IAAW,EAClC,EAAO,IAAI,GAAS,EAAO,IAAI,CAAM,GAAK,GAAK,CAAC,CAClD,CAEF,OAAO,CACT,CAGA,SAAS,EAAU,EAAiC,CAClD,IAAM,kBAAoB,IAAI,IAC9B,OAAO,EAAQ,IAAK,GAAW,CAC7B,GAAI,GAAU,EACZ,OAAO,EAET,IAAI,EAAQ,EAAkB,IAAI,CAAM,EAKxC,OAJI,IAAU,IAAA,KACZ,EAAQ,EAAkB,KAC1B,EAAkB,IAAI,EAAQ,CAAK,GAE9B,EAAE,EAAQ,EACnB,CAAC,CACH,CAOA,SAAS,EAAS,EAAsB,CACtC,IAAM,EAAU,EAAM,SAAW,KAAK,KAAK,EAAM,aAAe,EAAG,UAAa,EAC1E,EAAY,EAAM,UAAY,KAAK,KAAK,EAAM,cAAgB,EAAG,UAAa,EACpF,OAAQ,IAAY,GAAK,SAAc,IAAc,GACvD,CAQA,SAAS,EAAmB,EAAqB,EAAqC,CACpF,IAAM,EAAS,IAAI,WAAW,KAAK,IAAI,EAAQ,OAAS,EAAY,EAAG,CAAC,CAAC,EACzE,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAO,OAAQ,GAAS,EAAG,CACrD,IAAI,EAAO,KACX,IAAK,IAAI,EAAS,EAAG,EAAS,EAAW,GAAU,EAAG,CACpD,IAAM,EAAW,EAAU,EAAQ,IAAW,EAExC,GAAS,EAAQ,EAAQ,IAAW,GAAK,EAAI,GAAK,EAAW,KAAK,MAAM,EAAW,UAAe,EAExG,EAAQ,KAAK,KAAK,EAAM,EAAE,EAAI,EAAS,CACzC,CACA,EAAO,GAAS,CAClB,CACA,OAAO,CACT"}
+{"version":3,"file":"crossFileNearMiss.js","names":[],"sources":["../src/crossFileNearMiss.ts"],"sourcesContent":["import {\n  createLcsLengthCounter,\n  lcsLength,\n  markLcs,\n  type CountedOccurrence,\n  type Token,\n  type TokenRange,\n  type TokenSegment,\n} from './duplication.js';\n\n/**\n * Cross-file near-miss (Type-3) clone detection, following the within-file detector's model\n * (native/src/near_miss.rs): candidate block pairs are filtered through an n-gram inverted index\n * (NIL, Nakagawa et al. 2021), then verified by token-level longest common subsequence against the\n * larger block (NiCad's per-fragment similarity), backed by an information-weighted content gate,\n * with a statement-order-insensitive fallback and a local match over the anchored cores of two\n * blocks. Only pairs of blocks in different files are compared: a same-file pair is the\n * within-file detector's concern.\n */\n\nexport interface NearMissSourceFile {\n  tokens?: Token[];\n  containerStatements?: TokenRange[][];\n  nearMissBlocks?: TokenRange[];\n}\n\n/** One copy (a block or its matched cores) in a near-miss group; anchors carry `spanCountedElsewhere` (see collectCrossFileNearMissGroups). */\nexport interface NearMissOccurrence extends CountedOccurrence {\n  fileIndex: number;\n}\n\n/** N-gram size of the candidate index and local-match anchors (NIL's default). */\nconst ngramSize = 5;\n/** Filtration threshold: shared distinct n-grams over the smaller block's (NIL's default). */\nconst filtrationPercent = 10;\n/**\n * Pairs whose longer block exceeds this multiple of the shorter are compared only when whole-block\n * similarity still allows their ratio (below a minSimilarityPercent of 34). The candidate scan\n * stops at this floor while walking length-ordered postings, so pairs of very different lengths are\n * neither counted nor verified; `maxNgramBlockFrequency` is what bounds the scan's total cost.\n */\nconst maxLengthRatio = 3;\n/**\n * A structural match must also share content: more than this percent of the larger side's\n * information-weighted content-bearing tokens (names and literal values), so blocks of the same\n * shape that call different APIs on different data are not clones.\n */\nconst minContentSimilarityPercent = 50;\n/**\n * Caps content weights so only names and values spread over more than a quarter of the blocks are\n * discounted: a family of copies shares its content across several blocks, and uncapped rarity\n * weighting would let each copy's few unique edits outweigh everything the family shares.\n */\nconst maxContentWeight = 3;\n/**\n * Anchors must cover at least this percent of the shorter core: sparser chains are coincidental\n * runs of common n-grams in merely similar-looking code, and the cheap bound spares their content\n * and LCS checks. Not the similarity threshold itself, since n-grams repeated within a block\n * (repetitive statements) never anchor.\n */\nconst minAnchorCoveragePercent = 50;\n/** Anchors farther apart than this (in either block) split a local match into separate chains. */\nconst maxAnchorGapTokens = 30;\n/** Statement-order-insensitive comparison needs this many top-level statements per block. */\nconst minReorderStatementCount = 2;\n/**\n * N-grams occurring in more blocks than this are stop n-grams (syntax boilerplate such as a chain\n * of closing braces), left out of the index and of each block's n-gram count. Counting shared\n * n-grams costs the square of an n-gram's block frequency, so without the cap a project's most\n * common n-grams make filtration quadratic in the block count, while they discriminate nothing.\n */\nconst maxNgramBlockFrequency = 1000;\n\ninterface NormalizedBlock {\n  fileIndex: number;\n  range: TokenRange;\n  /** Interned non-identifier symbols (>= 0) and identifiers as -(file-level id + 1). */\n  symbols: Int32Array;\n  isContent: Uint8Array;\n  /**\n   * Per token of the block's file, a key of its content for a non-identifier: unlike a symbol,\n   * numbered by first occurrence, it does not depend on the order of the files.\n   */\n  fileTokenKeys: Float64Array;\n  /** Identifiers anonymized by first occurrence within the block. */\n  sequence: Int32Array;\n  /**\n   * The content of the block as written: per token its content key, or the number of its\n   * identifier. Equal for equal content wherever it lies, which symbols, numbered by first\n   * occurrence, are not; ordered like `written_content` in native/src/near_miss.rs, so that both\n   * detectors take the sides of a pair in the same order.\n   */\n  writtenContent: Float64Array;\n  /** The sequence sorted, for the token-bag upper bound on the LCS. */\n  sortedSequence: Int32Array;\n  /** Distinct non-stop n-gram hashes. */\n  ngrams: Int32Array;\n  /**\n   * The n-grams occurring exactly once in the block, sorted, with their offsets in the parallel\n   * array: two blocks' local-match anchors intersect by merging.\n   */\n  uniqueNgrams: Int32Array;\n  uniqueNgramOffsets: Int32Array;\n  contentCounts: Map<number, number>;\n  /**\n   * The sequence with its top-level statements in canonical order and the block offset each of its\n   * tokens comes from, when the block has enough statements for statement-order-insensitive\n   * comparison.\n   */\n  canonical: { sequence: Int32Array; offsets: Int32Array } | undefined;\n}\n\n/** A verified core in each block of a pair, as file token ranges. */\ntype CorePair = [[number, number], [number, number]];\n\n/** What alignPair needs to recover the tokens a verified match pairs. */\ninterface Alignment {\n  /** Whether the blocks matched only with their statements in canonical order. */\n  reordered: boolean;\n}\n\n/** How a verified pair matched: whole blocks, or every anchored core pair (one per gap-split chain segment). */\ntype PairMatch = (Alignment & { cores?: undefined }) | { cores: (Alignment & { pair: CorePair })[] };\n\n/** A verified near-miss pair of blocks; without `cores`, the blocks match as wholes. */\ninterface MatchEdge {\n  left: number;\n  right: number;\n  cores?: CorePair;\n  /** The token runs the pair matches, on the left and on the right. */\n  matchedRuns: [[number, number][], [number, number][]];\n}\n\n/**\n * Clusters verified cross-file near-miss pairs into groups. A node (a whole block or a matched\n * core) overlapping an occurrence of `reportedSpansByFile` (the exact cross-file groups) is an\n * anchor: it links near-miss copies to the content an exact group already reports, and appears in\n * the near-miss group marked `spanCountedElsewhere` so block counting does not count its span\n * twice; it still counts the lines its partners match outside those spans. A pair matching\n * nothing outside reported spans has nothing to add to them and is dropped before clustering (two\n * blocks wholly covered by reported spans are not even compared), so it neither collapses the\n * blocks' core nodes nor widens a node beyond what the pairs clustering it matched. Any other\n * pair is kept however much of it is reported: what the exact pipeline reports of a copy depends\n * on where its copies stand, and the lines the copy counts must not. A group needs at least one\n * node reported spans do not cover entirely. A block that matched only locally is reported as its matched cores\n * (overlapping cores merged), each clustered with its own partners, so code no verified pair\n * matched never counts as duplicated.\n */\nexport function collectCrossFileNearMissGroups(\n  files: NearMissSourceFile[],\n  reportedSpansByFile: { startTokenIndex: number; endTokenIndex: number }[][],\n  minTokens: number,\n  minSimilarityPercent: number\n): NearMissOccurrence[][] {\n  if (minSimilarityPercent >= 100) {\n    return [];\n  }\n  const blocks = normalizeBlocks(files);\n  const matcher = createMatcher(blocks, minTokens, minSimilarityPercent);\n  const countReported = reportedSpansByFile.map(createReportedTokenCounter);\n  const reportedCount = (fileIndex: number, [start, end]: [number, number]): number =>\n    countReported[fileIndex]?.(start, end) ?? 0;\n  const unreportedCount = (fileIndex: number, runs: [number, number][]): number =>\n    runs.reduce((sum, run) => sum + run[1] - run[0] - reportedCount(fileIndex, run), 0);\n  const fullyReported = blocks.map((block) => {\n    const span = spanOf(block);\n    return reportedCount(block.fileIndex, span) === span[1] - span[0];\n  });\n  const edges: MatchEdge[] = [];\n  forEachCandidatePair(blocks, fullyReported, minSimilarityPercent, (left, right) => {\n    const leftBlock = blocks[left];\n    const rightBlock = blocks[right];\n    const match = leftBlock && rightBlock && matcher(leftBlock, rightBlock, right);\n    if (!match) {\n      return;\n    }\n    // The lines of a near-miss copy that are duplicated are those a verified partner matches,\n    // which takes the pairing itself. Every pair kept is aligned: verifying it already cost a longest\n    // common subsequence of the same sequences, and any choice among a copy's partners would let\n    // their number or their order decide what the copy counts.\n    for (const { pair: cores, reordered } of match.cores ?? [{ pair: undefined, reordered: match.reordered }]) {\n      const [leftSpan, rightSpan] = cores ?? [spanOf(leftBlock), spanOf(rightBlock)];\n      const addsLines = (leftRuns: [number, number][], rightRuns: [number, number][]): boolean =>\n        unreportedCount(leftBlock.fileIndex, leftRuns) + unreportedCount(rightBlock.fileIndex, rightRuns) > 0;\n      // Fully reported spans need no alignment to be dropped.\n      if (!addsLines([leftSpan], [rightSpan])) {\n        continue;\n      }\n      const matchedRuns = alignPair(leftBlock, rightBlock, { cores, reordered });\n      if (addsLines(...matchedRuns)) {\n        edges.push({ left, right, cores, matchedRuns });\n      }\n    }\n  });\n\n  // Clustering runs over (block, core) nodes: a block with a recorded whole match is one node,\n  // and otherwise each union of its overlapping local cores is its own node, so disjoint cores\n  // matched with different partners fall into separate groups.\n  const matchedWhole = blocks.map(() => false);\n  const localCores = blocks.map((): [number, number][] => []);\n  for (const edge of edges) {\n    for (const [index, core] of sidesOf(edge)) {\n      if (core) {\n        localCores[index]?.push(core);\n      } else {\n        matchedWhole[index] = true;\n      }\n    }\n  }\n  const nodes: { blockIndex: number; core: [number, number] | undefined }[] = [];\n  const firstNodeByBlock: number[] = [];\n  for (const blockIndex of blocks.keys()) {\n    firstNodeByBlock.push(nodes.length);\n    const cores = matchedWhole[blockIndex] ? [] : mergeOverlappingCores(localCores[blockIndex] ?? []);\n    if (cores.length === 0) {\n      nodes.push({ blockIndex, core: undefined });\n    }\n    for (const core of cores) {\n      nodes.push({ blockIndex, core });\n    }\n  }\n  const nodeOf = (blockIndex: number, core: [number, number] | undefined): number => {\n    const first = firstNodeByBlock[blockIndex] ?? 0;\n    if (!core || matchedWhole[blockIndex]) {\n      return first;\n    }\n    for (let node = first; nodes[node]?.blockIndex === blockIndex; node += 1) {\n      const span = nodes[node]?.core;\n      if (span && span[0] <= core[0] && core[1] <= span[1]) {\n        return node;\n      }\n    }\n    throw new Error(\"every local core lies in one of its block's merged cores\");\n  };\n\n  const parent = nodes.map((_, index) => index);\n  const find = (index: number): number => {\n    let root = index;\n    while (parent[root] !== root) {\n      root = parent[root] ?? root;\n    }\n    for (let current = index; parent[current] !== root;) {\n      const next = parent[current] ?? root;\n      parent[current] = root;\n      current = next;\n    }\n    return root;\n  };\n  // Anchoring is judged per node: a core is an anchor only when a reported span overlaps the core\n  // itself, not merely elsewhere in its block.\n  const nodeRanges = nodes.map(({ blockIndex, core }) => {\n    const block = blocks[blockIndex];\n    const [startTokenIndex, endTokenIndex] = core ?? [\n      block?.range.startTokenIndex ?? 0,\n      block?.range.endTokenIndex ?? 0,\n    ];\n    return {\n      length: endTokenIndex - startTokenIndex,\n      reported: reportedCount(block?.fileIndex ?? 0, [startTokenIndex, endTokenIndex]),\n    };\n  });\n  const anchored = nodeRanges.map(({ reported }) => reported > 0);\n  const reported = nodeRanges.map((range) => range.reported === range.length);\n  const edgeNodes = edges.map((edge) => sidesOf(edge).map(([index, core]) => nodeOf(index, core)));\n  for (const [leftNode = 0, rightNode = 0] of edgeNodes) {\n    const leftRoot = find(leftNode);\n    const rightRoot = find(rightNode);\n    parent[Math.max(leftRoot, rightRoot)] = Math.min(leftRoot, rightRoot);\n  }\n  const matchedRunsByNode = nodes.map((): [number, number][] => []);\n  for (const [edgeIndex, { matchedRuns }] of edges.entries()) {\n    const [leftNode = 0, rightNode = 0] = edgeNodes[edgeIndex] ?? [];\n    matchedRunsByNode[leftNode]?.push(...matchedRuns[0]);\n    matchedRunsByNode[rightNode]?.push(...matchedRuns[1]);\n  }\n\n  const membersByRoot = new Map<number, number[]>();\n  for (const node of nodes.keys()) {\n    const root = find(node);\n    const members = membersByRoot.get(root) ?? [];\n    members.push(node);\n    membersByRoot.set(root, members);\n  }\n  const groups: NearMissOccurrence[][] = [];\n  for (const members of membersByRoot.values()) {\n    // Components form only through cross-file pairs, so two members always span two files.\n    if (members.length < 2 || members.every((node) => reported[node])) {\n      continue;\n    }\n    // A group's nodes from one block become ONE occurrence whose segments are its cores, so the\n    // fragment-weighted count charges the block as one copy (as for gapped clones), not once per core.\n    const coresByBlock = new Map<number, BlockCopy>();\n    for (const node of members) {\n      const { blockIndex = 0, core } = nodes[node] ?? {};\n      const entry = coresByBlock.get(blockIndex) ?? { cores: [], anchor: false, matchedRuns: [] };\n      entry.cores.push(core);\n      entry.anchor ||= anchored[node] ?? false;\n      entry.matchedRuns.push(...(matchedRunsByNode[node] ?? []));\n      coresByBlock.set(blockIndex, entry);\n    }\n    groups.push(\n      [...coresByBlock].flatMap(([blockIndex, copy]) => {\n        const block = blocks[blockIndex];\n        return block ? [toOccurrence(block, files, copy)] : [];\n      })\n    );\n  }\n  return groups;\n}\n\n/** The nodes of one block in a group: its cores (`undefined` for the whole block) and what its partners match of them. */\ninterface BlockCopy {\n  cores: ([number, number] | undefined)[];\n  anchor: boolean;\n  matchedRuns: [number, number][];\n}\n\nfunction spanOf({ range }: NormalizedBlock): [number, number] {\n  return [range.startTokenIndex, range.endTokenIndex];\n}\n\n/** The (block, core) on each side of a pair. */\nfunction sidesOf({ left, right, cores }: MatchEdge): [number, [number, number] | undefined][] {\n  return [\n    [left, cores?.[0]],\n    [right, cores?.[1]],\n  ];\n}\n\n/**\n * The token runs a verified match pairs, per side as file token ranges: those of a longest common\n * subsequence of what the match compared, its cores or else the whole blocks.\n */\nfunction alignPair(\n  left: NormalizedBlock,\n  right: NormalizedBlock,\n  { cores, reordered }: Alignment & { cores?: CorePair }\n): [[number, number][], [number, number][]] {\n  const leftCompared = comparedSequence(left, cores?.[0], reordered);\n  const rightCompared = comparedSequence(right, cores?.[1], reordered);\n  // Of the equally long subsequences, which one is marked depends on which sequence comes first,\n  // so the sides are taken in the order of their content rather than of their files.\n  if (compareSequences(contentOf(left, cores?.[0]), contentOf(right, cores?.[1])) <= 0) {\n    const [leftMatched, rightMatched] = markLcs(leftCompared.sequence, rightCompared.sequence);\n    return [toRuns(leftCompared.positions, leftMatched), toRuns(rightCompared.positions, rightMatched)];\n  }\n  const [rightMatched, leftMatched] = markLcs(rightCompared.sequence, leftCompared.sequence);\n  return [toRuns(leftCompared.positions, leftMatched), toRuns(rightCompared.positions, rightMatched)];\n}\n\n/**\n * The content of a block or of a core of it, as written (see `writtenContent`), with the\n * identifiers of a core numbered within it.\n */\nfunction contentOf(block: NormalizedBlock, core: [number, number] | undefined): Float64Array {\n  if (!core) {\n    return block.writtenContent;\n  }\n  const { sequence, positions } = comparedSequence(block, core, false);\n  return Float64Array.from(sequence, (symbol, index) =>\n    symbol < 0 ? symbol : (block.fileTokenKeys[positions[index] ?? 0] ?? 0)\n  );\n}\n\n/** The maximal runs of the matched positions, as half-open ranges. */\nfunction toRuns(positions: Int32Array, matched: Uint8Array): [number, number][] {\n  const runs: [number, number][] = [];\n  for (const position of positions.filter((_, index) => matched[index] === 1).toSorted()) {\n    const last = runs.at(-1);\n    if (last?.[1] === position) {\n      last[1] = position + 1;\n    } else {\n      runs.push([position, position + 1]);\n    }\n  }\n  return runs;\n}\n\n/**\n * The sequence a match compared, with the file token index of each of its symbols: a core, the\n * block with its statements in canonical order, or the block as written.\n */\nfunction comparedSequence(\n  block: NormalizedBlock,\n  core: [number, number] | undefined,\n  reordered: boolean\n): { sequence: Int32Array; positions: Int32Array } {\n  const blockStart = block.range.startTokenIndex;\n  if (core) {\n    return {\n      sequence: anonymize(block.symbols.subarray(core[0] - blockStart, core[1] - blockStart)),\n      positions: Int32Array.from({ length: core[1] - core[0] }, (_, index) => core[0] + index),\n    };\n  }\n  if (reordered && block.canonical) {\n    return {\n      sequence: block.canonical.sequence,\n      positions: block.canonical.offsets.map((offset) => blockStart + offset),\n    };\n  }\n  return { sequence: block.sequence, positions: block.sequence.map((_, index) => blockStart + index) };\n}\n\n/** The unions of overlapping cores, in position order. */\nfunction mergeOverlappingCores(cores: [number, number][]): [number, number][] {\n  const merged: [number, number][] = [];\n  for (const [start, end] of cores.toSorted((left, right) => left[0] - right[0])) {\n    const last = merged.at(-1);\n    if (last && start < last[1]) {\n      last[1] = Math.max(last[1], end);\n    } else {\n      merged.push([start, end]);\n    }\n  }\n  return merged;\n}\n\n/** Counts the tokens of [start, end) that the spans cover. */\nfunction createReportedTokenCounter(\n  spans: { startTokenIndex: number; endTokenIndex: number }[]\n): (start: number, end: number) => number {\n  const merged = mergeOverlappingCores(spans.map((span) => [span.startTokenIndex, span.endTokenIndex]));\n  let covered = 0;\n  const coveredBefore = [0];\n  for (const [start, end] of merged) {\n    covered += end - start;\n    coveredBefore.push(covered);\n  }\n  const countBefore = (position: number): number => {\n    let low = 0;\n    let high = merged.length;\n    while (low < high) {\n      const middle = (low + high) >>> 1;\n      if ((merged[middle]?.[0] ?? 0) < position) {\n        low = middle + 1;\n      } else {\n        high = middle;\n      }\n    }\n    return (coveredBefore[low] ?? 0) - Math.max(0, (merged[low - 1]?.[1] ?? 0) - position);\n  };\n  return (start, end) => countBefore(end) - countBefore(start);\n}\n\n/**\n * The block's occurrence with one segment per entry of `cores` (the whole block for `undefined`,\n * a whole match). Source offsets stay the block's: tokens carry none, and near-miss occurrences\n * report lines only.\n */\nfunction toOccurrence(\n  { fileIndex, range }: NormalizedBlock,\n  files: NearMissSourceFile[],\n  { cores, anchor, matchedRuns }: BlockCopy\n): NearMissOccurrence {\n  const whole = cores.includes(undefined);\n  const segments = cores\n    .map((core): [number, number] => core ?? [range.startTokenIndex, range.endTokenIndex])\n    .toSorted((left, right) => left[0] - right[0])\n    .map(([startTokenIndex, endTokenIndex]) => ({ startTokenIndex, endTokenIndex }));\n  const start = segments[0]?.startTokenIndex ?? range.startTokenIndex;\n  const end = segments.at(-1)?.endTokenIndex ?? range.endTokenIndex;\n  const tokens = files[fileIndex]?.tokens;\n  return {\n    fileIndex,\n    spanCountedElsewhere: anchor || undefined,\n    segments,\n    matchedRuns: mergeOverlappingCores(matchedRuns).map(([startTokenIndex, endTokenIndex]): TokenSegment => ({\n      startTokenIndex,\n      endTokenIndex,\n    })),\n    tokenCount: segments.reduce((sum, segment) => sum + segment.endTokenIndex - segment.startTokenIndex, 0),\n    startTokenIndex: start,\n    endTokenIndex: end,\n    startIndex: range.startIndex,\n    endIndex: range.endIndex,\n    startLine: whole ? range.startLine : (tokens?.[start]?.startRow ?? 0) + 1,\n    endLine: whole ? range.endLine : (tokens?.[end - 1]?.endRow ?? 0) + 1,\n  };\n}\n\n/**\n * Visits every cross-file block pair sharing at least `filtrationPercent` of the smaller block's\n * non-stop n-grams, except pairs of two blocks wholly covered by reported spans and pairs whose\n * length ratio rules out both\n * whole-block similarity and `maxLengthRatio`. Blocks are\n * indexed in ascending length, so each posting list is scanned backwards only while its blocks\n * are long enough; shared counts accumulate in a dense counter, so no pair map is materialized.\n */\nfunction forEachCandidatePair(\n  blocks: NormalizedBlock[],\n  fullyReported: boolean[],\n  minSimilarityPercent: number,\n  visit: (left: number, right: number) => void\n): void {\n  const blockFrequency = new Map<number, number>();\n  for (const block of blocks) {\n    for (const ngram of block.ngrams) {\n      blockFrequency.set(ngram, (blockFrequency.get(ngram) ?? 0) + 1);\n    }\n  }\n  for (const block of blocks) {\n    block.ngrams = block.ngrams.filter((ngram) => (blockFrequency.get(ngram) ?? 0) <= maxNgramBlockFrequency);\n  }\n\n  // Typed copies keep the posting loop, which dominates this phase, free of object dereferences.\n  const fileIndexes = Int32Array.from(blocks, (block) => block.fileIndex);\n  const reportedFlags = Uint8Array.from(fullyReported, Number);\n  const lengths = Int32Array.from(blocks, (block) => block.sequence.length);\n  const ngramCounts = Int32Array.from(blocks, (block) => block.ngrams.length);\n  const order = [...blocks.keys()].toSorted((left, right) => (lengths[left] ?? 0) - (lengths[right] ?? 0));\n  const postings = new Map<number, number[]>();\n  const sharedCounts = new Int32Array(blocks.length);\n  const touched: number[] = [];\n  for (const right of order) {\n    const fileIndex = fileIndexes[right];\n    const rightReported = reportedFlags[right] === 1;\n    const minLeftLength = Math.min(\n      Math.ceil((lengths[right] ?? 0) / maxLengthRatio),\n      Math.ceil((minSimilarityPercent * (lengths[right] ?? 0)) / 100)\n    );\n    const ngrams = blocks[right]?.ngrams ?? [];\n    for (const ngram of ngrams) {\n      const posting = postings.get(ngram);\n      if (!posting) {\n        postings.set(ngram, [right]);\n        continue;\n      }\n      for (let position = posting.length - 1; position >= 0; position -= 1) {\n        const left = posting[position] ?? 0;\n        if ((lengths[left] ?? 0) < minLeftLength) {\n          break;\n        }\n        if (fileIndexes[left] === fileIndex || (rightReported && reportedFlags[left] === 1)) {\n          continue;\n        }\n        if (sharedCounts[left] === 0) {\n          touched.push(left);\n        }\n        sharedCounts[left] = (sharedCounts[left] ?? 0) + 1;\n      }\n      posting.push(right);\n    }\n    for (const left of touched) {\n      const shared = sharedCounts[left] ?? 0;\n      sharedCounts[left] = 0;\n      if (shared * 100 >= filtrationPercent * Math.min(ngramCounts[left] ?? 0, ngrams.length)) {\n        visit(left, right);\n      }\n    }\n    touched.length = 0;\n  }\n}\n\n/**\n * Returns the pair verifier. Content symbols are weighted by integer self-information,\n * 1 + floor(log2((N + 1) / df)) over N blocks capped at `maxContentWeight`, so rare names and\n * values (the logic a copy preserves) outweigh ubiquitous ones, following the\n * information-theoretic weighting of ECScan's essence-clone detection (2025).\n */\nfunction createMatcher(\n  blocks: NormalizedBlock[],\n  minTokens: number,\n  minSimilarityPercent: number\n): (left: NormalizedBlock, right: NormalizedBlock, rightIndex: number) => PairMatch | undefined {\n  const documentFrequencies = new Map<number, number>();\n  for (const block of blocks) {\n    for (const symbol of block.contentCounts.keys()) {\n      documentFrequencies.set(symbol, (documentFrequencies.get(symbol) ?? 0) + 1);\n    }\n  }\n  const selfInformation = (documentFrequency: number): number =>\n    Math.min(31 - Math.clz32(Math.floor((blocks.length + 1) / documentFrequency)) + 1, maxContentWeight);\n  const weights = new Map<number, number>();\n  for (const [symbol, frequency] of documentFrequencies) {\n    weights.set(symbol, selfInformation(frequency));\n  }\n  const weigh = (counts: Map<number, number>): WeightedContent => {\n    const symbols = Int32Array.from(counts.keys()).toSorted();\n    // Span content comes from blocks, so every symbol has a weight.\n    const weightedCounts = Int32Array.from(symbols, (symbol) => (counts.get(symbol) ?? 0) * (weights.get(symbol) ?? 0));\n    let total = 0;\n    for (const count of weightedCounts) {\n      total += count;\n    }\n    return { symbols, weightedCounts, total };\n  };\n  const blockContents = new Map(blocks.map((block) => [block, weigh(block.contentCounts)]));\n\n  // Every candidate pair of one `right` block is visited consecutively, so one LCS counter (its\n  // position masks built once) serves them all.\n  let counterBlock = -1;\n  let counter: ((sequence: Int32Array) => number) | undefined;\n  const lcsLengthWithRight = (right: NormalizedBlock, rightIndex: number, sequence: Int32Array): number => {\n    if (counterBlock !== rightIndex || !counter) {\n      counterBlock = rightIndex;\n      counter = createLcsLengthCounter(right.sequence);\n    }\n    return counter(sequence);\n  };\n\n  /**\n   * Matches the cores two blocks share inside different surroundings (a copy wrapped in added\n   * code, or two copies embedded in different code), which whole-block similarity misses\n   * (CCAligner's large-gap and LVMapper's large-variance clones). N-grams unique to each block\n   * anchor the alignment; their longest chain increasing in both blocks (a run filter keeps only\n   * anchors continuing a diagonal, but the chain may shift diagonals at small insertions), split at\n   * gaps, delimits the cores, and every core pair that is a near-miss clone in its own right is\n   * returned.\n   */\n  const matchLocally = (left: NormalizedBlock, right: NormalizedBlock): PairMatch | undefined => {\n    const anchors: [number, number][] = [];\n    for (\n      let leftIndex = 0, rightIndex = 0;\n      leftIndex < left.uniqueNgrams.length && rightIndex < right.uniqueNgrams.length;\n    ) {\n      const leftHash = left.uniqueNgrams[leftIndex] ?? 0;\n      const rightHash = right.uniqueNgrams[rightIndex] ?? 0;\n      if (leftHash === rightHash) {\n        anchors.push([left.uniqueNgramOffsets[leftIndex] ?? 0, right.uniqueNgramOffsets[rightIndex] ?? 0]);\n      }\n      if (leftHash <= rightHash) {\n        leftIndex += 1;\n      }\n      if (rightHash <= leftHash) {\n        rightIndex += 1;\n      }\n    }\n    anchors.sort((first, second) => first[0] - second[0]);\n    // An isolated 5-gram match is often coincidental (n-grams are identifier-blind); a copied core\n    // yields runs of consecutive anchors, so only anchors continuing a diagonal run are chained.\n    const runAnchors = anchors.filter(([leftOffset, rightOffset], index) => {\n      const previous = anchors[index - 1];\n      const next = anchors[index + 1];\n      return (\n        (previous?.[0] === leftOffset - 1 && previous[1] === rightOffset - 1) ||\n        (next?.[0] === leftOffset + 1 && next[1] === rightOffset + 1)\n      );\n    });\n    const leftOffset = left.range.startTokenIndex;\n    const rightOffset = right.range.startTokenIndex;\n    const cores: (Alignment & { pair: CorePair })[] = [];\n    for (const segment of chainSegments(longestIncreasingChain(runAnchors))) {\n      const [leftStart, rightStart] = segment[0] ?? [0, 0];\n      const [leftLast, rightLast] = segment.at(-1) ?? [0, 0];\n      const leftEnd = leftLast + ngramSize;\n      const rightEnd = rightLast + ngramSize;\n      const leftLength = leftEnd - leftStart;\n      const rightLength = rightEnd - rightStart;\n      const shorter = Math.min(leftLength, rightLength);\n      const required = minSimilarityPercent * Math.max(leftLength, rightLength);\n      if (\n        shorter < minTokens ||\n        shorter * 100 < required ||\n        anchoredTokenCount(segment) * 100 < minAnchorCoveragePercent * shorter ||\n        !sharesContent(\n          weigh(countContent(left.symbols, left.isContent, leftStart, leftEnd)),\n          weigh(countContent(right.symbols, right.isContent, rightStart, rightEnd))\n        )\n      ) {\n        continue;\n      }\n      const coreLcsLength = lcsLength(\n        anonymize(left.symbols.subarray(leftStart, leftEnd)),\n        anonymize(right.symbols.subarray(rightStart, rightEnd))\n      );\n      if (coreLcsLength * 100 >= required) {\n        cores.push({\n          pair: [\n            [leftOffset + leftStart, leftOffset + leftEnd],\n            [rightOffset + rightStart, rightOffset + rightEnd],\n          ],\n          reordered: false,\n        });\n      }\n    }\n    return cores.length > 0 ? { cores } : undefined;\n  };\n\n  /** Cheapest bounds first: the LCS cannot exceed the shorter block's length nor the bag overlap. */\n  return (left, right, rightIndex) => {\n    const required = minSimilarityPercent * Math.max(left.sequence.length, right.sequence.length);\n    if (\n      Math.min(left.sequence.length, right.sequence.length) * 100 >= required &&\n      sharesContent(blockContents.get(left), blockContents.get(right))\n    ) {\n      if (sortedOverlap(left.sortedSequence, right.sortedSequence) * 100 >= required) {\n        const inOrder = lcsLengthWithRight(right, rightIndex, left.sequence);\n        if (inOrder * 100 >= required) {\n          return { reordered: false };\n        }\n      }\n      // With their top-level statements (each anonymized on its own) in a canonical order, a copy\n      // whose independent statements were swapped still matches.\n      if (left.canonical && right.canonical) {\n        const reordered = lcsLength(left.canonical.sequence, right.canonical.sequence);\n        if (reordered * 100 >= required) {\n          return { reordered: true };\n        }\n      }\n    }\n    // Of the equally long chains of anchors, the one kept depends on which block comes first, so\n    // the blocks are taken in the order of their content rather than of their files.\n    if (compareSequences(right.writtenContent, left.writtenContent) < 0) {\n      const swapped = matchLocally(right, left);\n      return (\n        swapped?.cores && {\n          cores: swapped.cores.map(({ pair: [rightCore, leftCore], ...alignment }) => ({\n            ...alignment,\n            pair: [leftCore, rightCore],\n          })),\n        }\n      );\n    }\n    return matchLocally(left, right);\n  };\n}\n\n/** Content-bearing symbols, sorted, with their information-weighted counts. */\ninterface WeightedContent {\n  symbols: Int32Array;\n  weightedCounts: Int32Array;\n  total: number;\n}\n\n/**\n * A structural match must be backed by shared content: more than `minContentSimilarityPercent` of\n * the larger side's information-weighted names and literal values. Two sides without content never\n * pass.\n */\nfunction sharesContent(left: WeightedContent | undefined, right: WeightedContent | undefined): boolean {\n  if (!left || !right) {\n    return false;\n  }\n  let overlap = 0;\n  for (let leftIndex = 0, rightIndex = 0; leftIndex < left.symbols.length && rightIndex < right.symbols.length;) {\n    const leftSymbol = left.symbols[leftIndex] ?? 0;\n    const rightSymbol = right.symbols[rightIndex] ?? 0;\n    if (leftSymbol === rightSymbol) {\n      overlap += Math.min(left.weightedCounts[leftIndex] ?? 0, right.weightedCounts[rightIndex] ?? 0);\n    }\n    if (leftSymbol <= rightSymbol) {\n      leftIndex += 1;\n    }\n    if (rightSymbol <= leftSymbol) {\n      rightIndex += 1;\n    }\n  }\n  return overlap * 100 > minContentSimilarityPercent * Math.max(left.total, right.total);\n}\n\n/**\n * The block's units (its top-level statements, given in file token indexes, and the token runs\n * between them), each anonymized on its own and sorted, concatenated, with the block offset of\n * every token; undefined with too few statements. `tokenKeys` holds the content key of each\n * non-identifier token of the block.\n */\nfunction canonicalSequenceOf(\n  symbols: Int32Array,\n  tokenKeys: Float64Array,\n  statements: [number, number][],\n  blockStart: number\n): NormalizedBlock['canonical'] {\n  if (statements.length < minReorderStatementCount) {\n    return undefined;\n  }\n  const units: { unit: Int32Array; content: Float64Array; start: number }[] = [];\n  const pushUnit = (start: number, end: number): void => {\n    const unit = anonymize(symbols.subarray(start, end));\n    // Ordered by the content of their tokens: symbols, numbered by first occurrence, would order\n    // the units by where other code stands.\n    const content = Float64Array.from(unit, (symbol, index) => (symbol < 0 ? symbol : (tokenKeys[start + index] ?? 0)));\n    units.push({ unit, content, start });\n  };\n  let cursor = 0;\n  for (const [statementStart, statementEnd] of statements) {\n    const start = statementStart - blockStart;\n    if (cursor < start) {\n      pushUnit(cursor, start);\n    }\n    pushUnit(start, statementEnd - blockStart);\n    cursor = statementEnd - blockStart;\n  }\n  if (cursor < symbols.length) {\n    pushUnit(cursor, symbols.length);\n  }\n  units.sort((left, right) => compareSequences(left.content, right.content) || left.start - right.start);\n  const sequence = new Int32Array(symbols.length);\n  const offsets = new Int32Array(symbols.length);\n  let position = 0;\n  for (const { unit, start } of units) {\n    sequence.set(unit, position);\n    for (let index = 0; index < unit.length; index += 1) {\n      offsets[position + index] = start + index;\n    }\n    position += unit.length;\n  }\n  return { sequence, offsets };\n}\n\n/** Lexicographic order. */\nfunction compareSequences(left: Float64Array, right: Float64Array): number {\n  for (let index = 0; index < Math.min(left.length, right.length); index += 1) {\n    const difference = (left[index] ?? 0) - (right[index] ?? 0);\n    if (difference !== 0) {\n      return difference;\n    }\n  }\n  return left.length - right.length;\n}\n\n/**\n * The longest chain of anchors increasing in both blocks (anchors arrive sorted by left offset),\n * via patience sorting over right offsets.\n */\nfunction longestIncreasingChain(anchors: [number, number][]): [number, number][] {\n  const tailIndexes: number[] = [];\n  const predecessors: number[] = [];\n  for (const [index, [, rightOffset]] of anchors.entries()) {\n    let low = 0;\n    let high = tailIndexes.length;\n    while (low < high) {\n      const middle = (low + high) >>> 1;\n      if ((anchors[tailIndexes[middle] ?? 0]?.[1] ?? 0) < rightOffset) {\n        low = middle + 1;\n      } else {\n        high = middle;\n      }\n    }\n    predecessors.push(low > 0 ? (tailIndexes[low - 1] ?? -1) : -1);\n    tailIndexes[low] = index;\n  }\n  const chain: [number, number][] = [];\n  for (let cursor = tailIndexes.at(-1) ?? -1; cursor >= 0; cursor = predecessors[cursor] ?? -1) {\n    const anchor = anchors[cursor];\n    if (anchor) {\n      chain.push(anchor);\n    }\n  }\n  return chain.toReversed();\n}\n\n/** The chain's segments, split where consecutive anchors lie more than `maxAnchorGapTokens` apart in either block. */\nfunction chainSegments(chain: [number, number][]): [number, number][][] {\n  const segments: [number, number][][] = [];\n  for (const [index, anchor] of chain.entries()) {\n    const previous = chain[index - 1];\n    const continues =\n      previous !== undefined &&\n      anchor[0] - (previous[0] + ngramSize) <= maxAnchorGapTokens &&\n      anchor[1] - (previous[1] + ngramSize) <= maxAnchorGapTokens;\n    if (continues) {\n      segments.at(-1)?.push(anchor);\n    } else {\n      segments.push([anchor]);\n    }\n  }\n  return segments;\n}\n\n/** Left-block tokens the segment's anchors cover (overlapping anchors count once). */\nfunction anchoredTokenCount(segment: [number, number][]): number {\n  let count = ngramSize;\n  for (let index = 1; index < segment.length; index += 1) {\n    count += Math.min((segment[index]?.[0] ?? 0) - (segment[index - 1]?.[0] ?? 0), ngramSize);\n  }\n  return count;\n}\n\n/** Multiset intersection size of two ascending arrays. */\nfunction sortedOverlap(left: Int32Array, right: Int32Array): number {\n  let overlap = 0;\n  let leftIndex = 0;\n  let rightIndex = 0;\n  while (leftIndex < left.length && rightIndex < right.length) {\n    const leftValue = left[leftIndex] ?? 0;\n    const rightValue = right[rightIndex] ?? 0;\n    if (leftValue === rightValue) {\n      overlap += 1;\n      leftIndex += 1;\n      rightIndex += 1;\n    } else if (leftValue < rightValue) {\n      leftIndex += 1;\n    } else {\n      rightIndex += 1;\n    }\n  }\n  return overlap;\n}\n\n/**\n * Normalizes every block like the within-file detector. Non-identifier symbols are interned\n * project-wide from the tokens' hash pairs, so equal tokens compare equal across files, while\n * identifiers are interned per file and re-anonymized per compared range.\n */\nfunction normalizeBlocks(files: NearMissSourceFile[]): NormalizedBlock[] {\n  const symbolByTokenKey = new Map<number, number>();\n  const blocks: NormalizedBlock[] = [];\n  for (const [fileIndex, { tokens, containerStatements, nearMissBlocks }] of files.entries()) {\n    if (!tokens || !nearMissBlocks?.length) {\n      continue;\n    }\n    const symbols = new Int32Array(tokens.length);\n    const isContent = new Uint8Array(tokens.length);\n    const fileTokenKeys = new Float64Array(tokens.length);\n    const idByIdentifier = new Map<string, number>();\n    for (const [index, token] of tokens.entries()) {\n      if (token.kind === 'id') {\n        let id = idByIdentifier.get(token.text);\n        if (id === undefined) {\n          id = idByIdentifier.size;\n          idByIdentifier.set(token.text, id);\n        }\n        symbols[index] = -(id + 1);\n        continue;\n      }\n      const key = tokenKey(token);\n      fileTokenKeys[index] = key;\n      let symbol = symbolByTokenKey.get(key);\n      if (symbol === undefined) {\n        symbol = symbolByTokenKey.size;\n        symbolByTokenKey.set(key, symbol);\n      }\n      symbols[index] = symbol;\n      isContent[index] = token.isName || token.literalHash !== undefined ? 1 : 0;\n    }\n    const findStatements = createTopLevelStatementFinder(containerStatements ?? []);\n    for (const range of nearMissBlocks) {\n      const { startTokenIndex: start, endTokenIndex: end } = range;\n      const blockSymbols = symbols.subarray(start, end);\n      const blockIsContent = isContent.subarray(start, end);\n      const sequence = anonymize(blockSymbols);\n      const ngramHashes = collectNgramHashes(blockSymbols, fileTokenKeys.subarray(start, end));\n      const occurrenceCounts = new Map<number, number>();\n      for (const hash of ngramHashes) {\n        occurrenceCounts.set(hash, (occurrenceCounts.get(hash) ?? 0) + 1);\n      }\n      const uniqueOffsets = ngramHashes\n        .keys()\n        .filter((offset) => occurrenceCounts.get(ngramHashes[offset] ?? 0) === 1)\n        .toArray()\n        .toSorted((first, second) => (ngramHashes[first] ?? 0) - (ngramHashes[second] ?? 0));\n      blocks.push({\n        fileIndex,\n        range,\n        symbols: blockSymbols,\n        isContent: blockIsContent,\n        fileTokenKeys,\n        sequence,\n        writtenContent: Float64Array.from(sequence, (symbol, index) =>\n          symbol < 0 ? symbol : (fileTokenKeys[start + index] ?? 0)\n        ),\n        sortedSequence: sequence.toSorted(),\n        ngrams: Int32Array.from(occurrenceCounts.keys()),\n        uniqueNgrams: Int32Array.from(uniqueOffsets, (offset) => ngramHashes[offset] ?? 0),\n        uniqueNgramOffsets: Int32Array.from(uniqueOffsets),\n        contentCounts: countContent(blockSymbols, blockIsContent, 0, blockSymbols.length),\n        canonical: canonicalSequenceOf(\n          blockSymbols,\n          fileTokenKeys.subarray(start, end),\n          findStatements(start, end),\n          start\n        ),\n      });\n    }\n  }\n  return blocks;\n}\n\n/**\n * Returns a lookup of the outermost container statements inside a token range, excluding a\n * statement spanning the whole range (the block itself).\n */\nfunction createTopLevelStatementFinder(\n  containerStatements: TokenRange[][]\n): (start: number, end: number) => [number, number][] {\n  const statements = containerStatements\n    .flat()\n    .filter((statement) => statement.startTokenIndex < statement.endTokenIndex)\n    .map((statement): [number, number] => [statement.startTokenIndex, statement.endTokenIndex])\n    .toSorted((left, right) => left[0] - right[0] || right[1] - left[1]);\n  return (start, end) => {\n    let low = 0;\n    let high = statements.length;\n    while (low < high) {\n      const middle = (low + high) >>> 1;\n      if ((statements[middle]?.[0] ?? 0) < start) {\n        low = middle + 1;\n      } else {\n        high = middle;\n      }\n    }\n    const topLevel: [number, number][] = [];\n    for (let index = low; index < statements.length; index += 1) {\n      const statement = statements[index];\n      if (!statement || statement[0] >= end) {\n        break;\n      }\n      const last = topLevel.at(-1);\n      const nested = last !== undefined && statement[0] < last[1];\n      if (statement[1] <= end && !(statement[0] === start && statement[1] === end) && !nested) {\n        topLevel.push(statement);\n      }\n    }\n    return topLevel;\n  };\n}\n\nfunction countContent(symbols: Int32Array, isContent: Uint8Array, start: number, end: number): Map<number, number> {\n  const counts = new Map<number, number>();\n  for (let offset = start; offset < end; offset += 1) {\n    if (isContent[offset] === 1) {\n      const symbol = symbols[offset] ?? 0;\n      counts.set(symbol, (counts.get(symbol) ?? 0) + 1);\n    }\n  }\n  return counts;\n}\n\n/** Identifiers renumbered by first occurrence within `symbols`, so a range compares the same wherever it sits in its file. */\nfunction anonymize(symbols: Int32Array): Int32Array {\n  const indexByIdentifier = new Map<number, number>();\n  return symbols.map((symbol) => {\n    if (symbol >= 0) {\n      return symbol;\n    }\n    let index = indexByIdentifier.get(symbol);\n    if (index === undefined) {\n      index = indexByIdentifier.size;\n      indexByIdentifier.set(symbol, index);\n    }\n    return -(index + 1);\n  });\n}\n\n/**\n * A 53-bit key from the token's two independent text hashes, each mixed with the matching literal\n * value hash: exact in a JavaScript number, so interning never merges distinct tokens unless 53\n * hash bits collide.\n */\nfunction tokenKey(token: Token): number {\n  const primary = token.textHash ^ Math.imul(token.literalHash ?? 0, 0x9E_37_79_B1);\n  const secondary = token.textHash2 ^ Math.imul(token.literalHash2 ?? 0, 0x85_EB_CA_6B);\n  return (primary >>> 0) * 0x20_00_00 + (secondary >>> 11);\n}\n\n/**\n * N-gram hash per start offset, identifier-blind (every identifier hashes as -1) so a block copied\n * into different surroundings (renumbering its identifiers) or with reordered statements still\n * shares its n-grams. Hashed from the content keys of the tokens: with symbols, numbered by first\n * occurrence, which n-grams collide would depend on the order of the files.\n */\nfunction collectNgramHashes(symbols: Int32Array, tokenKeys: Float64Array): Int32Array {\n  const hashes = new Int32Array(Math.max(symbols.length - ngramSize + 1, 0));\n  for (let start = 0; start < hashes.length; start += 1) {\n    let hash = 5381;\n    for (let offset = 0; offset < ngramSize; offset += 1) {\n      const tokenKey = tokenKeys[start + offset] ?? 0;\n      // A token key holds 53 bits; both halves enter the hash.\n      const value = (symbols[start + offset] ?? 0) < 0 ? -1 : tokenKey ^ Math.floor(tokenKey / 0x1_00_00_00_00);\n      // oxlint-disable-next-line unicorn/prefer-math-trunc -- `| 0` wraps the sum to int32.\n      hash = (Math.imul(hash, 31) + value) | 0;\n    }\n    hashes[start] = hash;\n  }\n  return hashes;\n}\n"],"mappings":"sFAoJA,SAAgB,EACd,EACA,EACA,EACA,EACwB,CACxB,GAAI,GAAwB,IAC1B,MAAO,CAAC,EAEV,IAAM,EAAS,EAAgB,CAAK,EAC9B,EAAU,EAAc,EAAQ,EAAW,CAAoB,EAC/D,EAAgB,EAAoB,IAAI,CAA0B,EAClE,GAAiB,EAAmB,CAAC,EAAO,KAChD,EAAc,EAAU,GAAG,EAAO,CAAG,GAAK,EACtC,GAAmB,EAAmB,IAC1C,EAAK,QAAQ,EAAK,IAAQ,EAAM,EAAI,GAAK,EAAI,GAAK,EAAc,EAAW,CAAG,EAAG,CAAC,EAC9E,EAAgB,EAAO,IAAK,GAAU,CAC1C,IAAM,EAAO,EAAO,CAAK,EACzB,OAAO,EAAc,EAAM,UAAW,CAAI,IAAM,EAAK,GAAK,EAAK,EACjE,CAAC,EACK,EAAqB,CAAC,EAC5B,EAAqB,EAAQ,EAAe,GAAuB,EAAM,IAAU,CACjF,IAAM,EAAY,EAAO,GACnB,EAAa,EAAO,GACpB,EAAQ,GAAa,GAAc,EAAQ,EAAW,EAAY,CAAK,EACxE,KAOL,IAAK,GAAM,CAAE,KAAM,EAAO,eAAe,EAAM,OAAS,CAAC,CAAE,KAAM,IAAA,GAAW,UAAW,EAAM,SAAU,CAAC,EAAG,CACzG,GAAM,CAAC,EAAU,GAAa,GAAS,CAAC,EAAO,CAAS,EAAG,EAAO,CAAU,CAAC,EACvE,GAAa,EAA8B,IAC/C,EAAgB,EAAU,UAAW,CAAQ,EAAI,EAAgB,EAAW,UAAW,CAAS,EAAI,EAEtG,GAAI,CAAC,EAAU,CAAC,CAAQ,EAAG,CAAC,CAAS,CAAC,EACpC,SAEF,IAAM,EAAc,EAAU,EAAW,EAAY,CAAE,QAAO,WAAU,CAAC,EACrE,EAAU,GAAG,CAAW,GAC1B,EAAM,KAAK,CAAE,OAAM,QAAO,QAAO,aAAY,CAAC,CAElD,CACF,CAAC,EAKD,IAAM,EAAe,EAAO,QAAU,EAAK,EACrC,EAAa,EAAO,QAA8B,CAAC,CAAC,EAC1D,IAAK,IAAM,KAAQ,EACjB,IAAK,GAAM,CAAC,EAAO,KAAS,EAAQ,CAAI,EAClC,EACF,EAAW,EAAM,EAAE,KAAK,CAAI,EAE5B,EAAa,GAAS,GAI5B,IAAM,EAAsE,CAAC,EACvE,EAA6B,CAAC,EACpC,IAAK,IAAM,KAAc,EAAO,KAAK,EAAG,CACtC,EAAiB,KAAK,EAAM,MAAM,EAClC,IAAM,EAAQ,EAAa,GAAc,CAAC,EAAI,EAAsB,EAAW,IAAe,CAAC,CAAC,EAC5F,EAAM,SAAW,GACnB,EAAM,KAAK,CAAE,aAAY,KAAM,IAAA,EAAU,CAAC,EAE5C,IAAK,IAAM,KAAQ,EACjB,EAAM,KAAK,CAAE,aAAY,MAAK,CAAC,CAEnC,CACA,IAAM,GAAU,EAAoB,IAA+C,CACjF,IAAM,EAAQ,EAAiB,IAAe,EAC9C,GAAI,CAAC,GAAQ,EAAa,GACxB,OAAO,EAET,IAAK,IAAI,EAAO,EAAO,EAAM,EAAK,EAAE,aAAe,EAAY,GAAQ,EAAG,CACxE,IAAM,EAAO,EAAM,EAAK,EAAE,KAC1B,GAAI,GAAQ,EAAK,IAAM,EAAK,IAAM,EAAK,IAAM,EAAK,GAChD,OAAO,CAEX,CACA,MAAU,MAAM,0DAA0D,CAC5E,EAEM,EAAS,EAAM,KAAK,EAAG,IAAU,CAAK,EACtC,EAAQ,GAA0B,CACtC,IAAI,EAAO,EACX,KAAO,EAAO,KAAU,GACtB,EAAO,EAAO,IAAS,EAEzB,IAAK,IAAI,EAAU,EAAO,EAAO,KAAa,GAAO,CACnD,IAAM,EAAO,EAAO,IAAY,EAChC,EAAO,GAAW,EAClB,EAAU,CACZ,CACA,OAAO,CACT,EAGM,EAAa,EAAM,KAAK,CAAE,aAAY,UAAW,CACrD,IAAM,EAAQ,EAAO,GACf,CAAC,EAAiB,GAAiB,GAAQ,CAC/C,GAAO,MAAM,iBAAmB,EAChC,GAAO,MAAM,eAAiB,CAChC,EACA,MAAO,CACL,OAAQ,EAAgB,EACxB,SAAU,EAAc,GAAO,WAAa,EAAG,CAAC,EAAiB,CAAa,CAAC,CACjF,CACF,CAAC,EACK,EAAW,EAAW,KAAK,CAAE,cAAe,EAAW,CAAC,EACxD,EAAW,EAAW,IAAK,GAAU,EAAM,WAAa,EAAM,MAAM,EACpE,EAAY,EAAM,IAAK,GAAS,EAAQ,CAAI,CAAC,CAAC,KAAK,CAAC,EAAO,KAAU,EAAO,EAAO,CAAI,CAAC,CAAC,EAC/F,IAAK,GAAM,CAAC,EAAW,EAAG,EAAY,KAAM,EAAW,CACrD,IAAM,EAAW,EAAK,CAAQ,EACxB,EAAY,EAAK,CAAS,EAChC,EAAO,KAAK,IAAI,EAAU,CAAS,GAAK,KAAK,IAAI,EAAU,CAAS,CACtE,CACA,IAAM,EAAoB,EAAM,QAA8B,CAAC,CAAC,EAChE,IAAK,GAAM,CAAC,EAAW,CAAE,kBAAkB,EAAM,QAAQ,EAAG,CAC1D,GAAM,CAAC,EAAW,EAAG,EAAY,GAAK,EAAU,IAAc,CAAC,EAC/D,EAAkB,EAAS,EAAE,KAAK,GAAG,EAAY,EAAE,EACnD,EAAkB,EAAU,EAAE,KAAK,GAAG,EAAY,EAAE,CACtD,CAEA,IAAM,kBAAgB,IAAI,IAC1B,IAAK,IAAM,KAAQ,EAAM,KAAK,EAAG,CAC/B,IAAM,EAAO,EAAK,CAAI,EAChB,EAAU,EAAc,IAAI,CAAI,GAAK,CAAC,EAC5C,EAAQ,KAAK,CAAI,EACjB,EAAc,IAAI,EAAM,CAAO,CACjC,CACA,IAAM,EAAiC,CAAC,EACxC,IAAK,IAAM,KAAW,EAAc,OAAO,EAAG,CAE5C,GAAI,EAAQ,OAAS,GAAK,EAAQ,MAAO,GAAS,EAAS,EAAK,EAC9D,SAIF,IAAM,kBAAe,IAAI,IACzB,IAAK,IAAM,KAAQ,EAAS,CAC1B,GAAM,CAAE,aAAa,EAAG,QAAS,EAAM,IAAS,CAAC,EAC3C,EAAQ,EAAa,IAAI,CAAU,GAAK,CAAE,MAAO,CAAC,EAAG,OAAQ,GAAO,YAAa,CAAC,CAAE,EAC1F,EAAM,MAAM,KAAK,CAAI,EACrB,EAAM,SAAW,EAAS,IAAS,GACnC,EAAM,YAAY,KAAK,GAAI,EAAkB,IAAS,CAAC,CAAE,EACzD,EAAa,IAAI,EAAY,CAAK,CACpC,CACA,EAAO,KACL,CAAC,GAAG,CAAY,CAAC,CAAC,SAAS,CAAC,EAAY,KAAU,CAChD,IAAM,EAAQ,EAAO,GACrB,OAAO,EAAQ,CAAC,EAAa,EAAO,EAAO,CAAI,CAAC,EAAI,CAAC,CACvD,CAAC,CACH,CACF,CACA,OAAO,CACT,CASA,SAAS,EAAO,CAAE,SAA4C,CAC5D,MAAO,CAAC,EAAM,gBAAiB,EAAM,aAAa,CACpD,CAGA,SAAS,EAAQ,CAAE,OAAM,QAAO,SAA8D,CAC5F,MAAO,CACL,CAAC,EAAM,IAAQ,EAAE,EACjB,CAAC,EAAO,IAAQ,EAAE,CACpB,CACF,CAMA,SAAS,EACP,EACA,EACA,CAAE,QAAO,aACiC,CAC1C,IAAM,EAAe,EAAiB,EAAM,IAAQ,GAAI,CAAS,EAC3D,EAAgB,EAAiB,EAAO,IAAQ,GAAI,CAAS,EAGnE,GAAI,EAAiB,EAAU,EAAM,IAAQ,EAAE,EAAG,EAAU,EAAO,IAAQ,EAAE,CAAC,GAAK,EAAG,CACpF,GAAM,CAAC,EAAa,GAAgB,EAAQ,EAAa,SAAU,EAAc,QAAQ,EACzF,MAAO,CAAC,EAAO,EAAa,UAAW,CAAW,EAAG,EAAO,EAAc,UAAW,CAAY,CAAC,CACpG,CACA,GAAM,CAAC,EAAc,GAAe,EAAQ,EAAc,SAAU,EAAa,QAAQ,EACzF,MAAO,CAAC,EAAO,EAAa,UAAW,CAAW,EAAG,EAAO,EAAc,UAAW,CAAY,CAAC,CACpG,CAMA,SAAS,EAAU,EAAwB,EAAkD,CAC3F,GAAI,CAAC,EACH,OAAO,EAAM,eAEf,GAAM,CAAE,WAAU,aAAc,EAAiB,EAAO,EAAM,EAAK,EACnE,OAAO,aAAa,KAAK,GAAW,EAAQ,IAC1C,EAAS,EAAI,EAAU,EAAM,cAAc,EAAU,IAAU,IAAM,CACvE,CACF,CAGA,SAAS,EAAO,EAAuB,EAAyC,CAC9E,IAAM,EAA2B,CAAC,EAClC,IAAK,IAAM,KAAY,EAAU,QAAQ,EAAG,IAAU,EAAQ,KAAW,CAAC,CAAC,CAAC,SAAS,EAAG,CACtF,IAAM,EAAO,EAAK,GAAG,EAAE,EACnB,IAAO,KAAO,EAChB,EAAK,GAAK,EAAW,EAErB,EAAK,KAAK,CAAC,EAAU,EAAW,CAAC,CAAC,CAEtC,CACA,OAAO,CACT,CAMA,SAAS,EACP,EACA,EACA,EACiD,CACjD,IAAM,EAAa,EAAM,MAAM,gBAa/B,OAZI,EACK,CACL,SAAU,EAAU,EAAM,QAAQ,SAAS,EAAK,GAAK,EAAY,EAAK,GAAK,CAAU,CAAC,EACtF,UAAW,WAAW,KAAK,CAAE,OAAQ,EAAK,GAAK,EAAK,EAAG,GAAI,EAAG,IAAU,EAAK,GAAK,CAAK,CACzF,EAEE,GAAa,EAAM,UACd,CACL,SAAU,EAAM,UAAU,SAC1B,UAAW,EAAM,UAAU,QAAQ,IAAK,GAAW,EAAa,CAAM,CACxE,EAEK,CAAE,SAAU,EAAM,SAAU,UAAW,EAAM,SAAS,KAAK,EAAG,IAAU,EAAa,CAAK,CAAE,CACrG,CAGA,SAAS,EAAsB,EAA+C,CAC5E,IAAM,EAA6B,CAAC,EACpC,IAAK,GAAM,CAAC,EAAO,KAAQ,EAAM,UAAU,EAAM,IAAU,EAAK,GAAK,EAAM,EAAE,EAAG,CAC9E,IAAM,EAAO,EAAO,GAAG,EAAE,EACrB,GAAQ,EAAQ,EAAK,GACvB,EAAK,GAAK,KAAK,IAAI,EAAK,GAAI,CAAG,EAE/B,EAAO,KAAK,CAAC,EAAO,CAAG,CAAC,CAE5B,CACA,OAAO,CACT,CAGA,SAAS,EACP,EACwC,CACxC,IAAM,EAAS,EAAsB,EAAM,IAAK,GAAS,CAAC,EAAK,gBAAiB,EAAK,aAAa,CAAC,CAAC,EAChG,EAAU,EACR,EAAgB,CAAC,CAAC,EACxB,IAAK,GAAM,CAAC,EAAO,KAAQ,EACzB,GAAW,EAAM,EACjB,EAAc,KAAK,CAAO,EAE5B,IAAM,EAAe,GAA6B,CAChD,IAAI,EAAM,EACN,EAAO,EAAO,OAClB,KAAO,EAAM,GAAM,CACjB,IAAM,EAAU,EAAM,IAAU,GAC3B,EAAO,EAAO,GAAG,IAAM,GAAK,EAC/B,EAAM,EAAS,EAEf,EAAO,CAEX,CACA,OAAQ,EAAc,IAAQ,GAAK,KAAK,IAAI,GAAI,EAAO,EAAM,EAAE,GAAG,IAAM,GAAK,CAAQ,CACvF,EACA,OAAQ,EAAO,IAAQ,EAAY,CAAG,EAAI,EAAY,CAAK,CAC7D,CAOA,SAAS,EACP,CAAE,YAAW,SACb,EACA,CAAE,QAAO,SAAQ,eACG,CACpB,IAAM,EAAQ,EAAM,SAAS,IAAA,EAAS,EAChC,EAAW,EACd,IAAK,GAA2B,GAAQ,CAAC,EAAM,gBAAiB,EAAM,aAAa,CAAC,CAAC,CACrF,UAAU,EAAM,IAAU,EAAK,GAAK,EAAM,EAAE,CAAC,CAC7C,KAAK,CAAC,EAAiB,MAAoB,CAAE,kBAAiB,eAAc,EAAE,EAC3E,EAAQ,EAAS,EAAE,EAAE,iBAAmB,EAAM,gBAC9C,EAAM,EAAS,GAAG,EAAE,CAAC,EAAE,eAAiB,EAAM,cAC9C,EAAS,EAAM,EAAU,EAAE,OACjC,MAAO,CACL,YACA,qBAAsB,GAAU,IAAA,GAChC,WACA,YAAa,EAAsB,CAAW,CAAC,CAAC,KAAK,CAAC,EAAiB,MAAkC,CACvG,kBACA,eACF,EAAE,EACF,WAAY,EAAS,QAAQ,EAAK,IAAY,EAAM,EAAQ,cAAgB,EAAQ,gBAAiB,CAAC,EACtG,gBAAiB,EACjB,cAAe,EACf,WAAY,EAAM,WAClB,SAAU,EAAM,SAChB,UAAW,EAAQ,EAAM,WAAa,IAAS,EAAM,EAAE,UAAY,GAAK,EACxE,QAAS,EAAQ,EAAM,SAAW,IAAS,EAAM,EAAE,EAAE,QAAU,GAAK,CACtE,CACF,CAUA,SAAS,EACP,EACA,EACA,EACA,EACM,CACN,IAAM,kBAAiB,IAAI,IAC3B,IAAK,IAAM,KAAS,EAClB,IAAK,IAAM,KAAS,EAAM,OACxB,EAAe,IAAI,GAAQ,EAAe,IAAI,CAAK,GAAK,GAAK,CAAC,EAGlE,IAAK,IAAM,KAAS,EAClB,EAAM,OAAS,EAAM,OAAO,OAAQ,IAAW,EAAe,IAAI,CAAK,GAAK,IAAM,GAAsB,EAI1G,IAAM,EAAc,WAAW,KAAK,EAAS,GAAU,EAAM,SAAS,EAChE,EAAgB,WAAW,KAAK,EAAe,MAAM,EACrD,EAAU,WAAW,KAAK,EAAS,GAAU,EAAM,SAAS,MAAM,EAClE,EAAc,WAAW,KAAK,EAAS,GAAU,EAAM,OAAO,MAAM,EACpE,EAAQ,CAAC,GAAG,EAAO,KAAK,CAAC,CAAC,CAAC,UAAU,EAAM,KAAW,EAAQ,IAAS,IAAM,EAAQ,IAAU,EAAE,EACjG,kBAAW,IAAI,IACf,EAAe,IAAI,WAAW,EAAO,MAAM,EAC3C,EAAoB,CAAC,EAC3B,IAAK,IAAM,KAAS,EAAO,CACzB,IAAM,EAAY,EAAY,GACxB,EAAgB,EAAc,KAAW,EACzC,EAAgB,KAAK,IACzB,KAAK,MAAM,EAAQ,IAAU,GAAK,CAAc,EAChD,KAAK,KAAM,GAAwB,EAAQ,IAAU,GAAM,GAAG,CAChE,EACM,EAAS,EAAO,EAAM,EAAE,QAAU,CAAC,EACzC,IAAK,IAAM,KAAS,EAAQ,CAC1B,IAAM,EAAU,EAAS,IAAI,CAAK,EAClC,GAAI,CAAC,EAAS,CACZ,EAAS,IAAI,EAAO,CAAC,CAAK,CAAC,EAC3B,QACF,CACA,IAAK,IAAI,EAAW,EAAQ,OAAS,EAAG,GAAY,EAAG,IAAe,CACpE,IAAM,EAAO,EAAQ,IAAa,EAClC,IAAK,EAAQ,IAAS,GAAK,EACzB,MAEE,EAAY,KAAU,GAAc,GAAiB,EAAc,KAAU,IAG7E,EAAa,KAAU,GACzB,EAAQ,KAAK,CAAI,EAEnB,EAAa,IAAS,EAAa,IAAS,GAAK,EACnD,CACA,EAAQ,KAAK,CAAK,CACpB,CACA,IAAK,IAAM,KAAQ,EAAS,CAC1B,IAAM,EAAS,EAAa,IAAS,EACrC,EAAa,GAAQ,EACjB,EAAS,KAAO,GAAoB,KAAK,IAAI,EAAY,IAAS,EAAG,EAAO,MAAM,GACpF,EAAM,EAAM,CAAK,CAErB,CACA,EAAQ,OAAS,CACnB,CACF,CAQA,SAAS,EACP,EACA,EACA,EAC8F,CAC9F,IAAM,kBAAsB,IAAI,IAChC,IAAK,IAAM,KAAS,EAClB,IAAK,IAAM,KAAU,EAAM,cAAc,KAAK,EAC5C,EAAoB,IAAI,GAAS,EAAoB,IAAI,CAAM,GAAK,GAAK,CAAC,EAG9E,IAAM,EAAmB,GACvB,KAAK,IAAI,GAAK,KAAK,MAAM,KAAK,OAAO,EAAO,OAAS,GAAK,CAAiB,CAAC,EAAI,EAAG,CAAgB,EAC/F,kBAAU,IAAI,IACpB,IAAK,GAAM,CAAC,EAAQ,KAAc,EAChC,EAAQ,IAAI,EAAQ,EAAgB,CAAS,CAAC,EAEhD,IAAM,EAAS,GAAiD,CAC9D,IAAM,EAAU,WAAW,KAAK,EAAO,KAAK,CAAC,CAAC,CAAC,SAAS,EAElD,EAAiB,WAAW,KAAK,EAAU,IAAY,EAAO,IAAI,CAAM,GAAK,IAAM,EAAQ,IAAI,CAAM,GAAK,EAAE,EAC9G,EAAQ,EACZ,IAAK,IAAM,KAAS,EAClB,GAAS,EAEX,MAAO,CAAE,UAAS,iBAAgB,OAAM,CAC1C,EACM,EAAgB,IAAI,IAAI,EAAO,IAAK,GAAU,CAAC,EAAO,EAAM,EAAM,aAAa,CAAC,CAAC,CAAC,EAIpF,EAAe,GACf,EACE,GAAsB,EAAwB,EAAoB,MAClE,IAAiB,GAAc,CAAC,KAClC,EAAe,EACf,EAAU,EAAuB,EAAM,QAAQ,GAE1C,EAAQ,CAAQ,GAYnB,GAAgB,EAAuB,IAAkD,CAC7F,IAAM,EAA8B,CAAC,EACrC,IACE,IAAI,EAAY,EAAG,EAAa,EAChC,EAAY,EAAK,aAAa,QAAU,EAAa,EAAM,aAAa,QACxE,CACA,IAAM,EAAW,EAAK,aAAa,IAAc,EAC3C,EAAY,EAAM,aAAa,IAAe,EAChD,IAAa,GACf,EAAQ,KAAK,CAAC,EAAK,mBAAmB,IAAc,EAAG,EAAM,mBAAmB,IAAe,CAAC,CAAC,EAE/F,GAAY,IACd,GAAa,GAEX,GAAa,IACf,GAAc,EAElB,CACA,EAAQ,MAAM,EAAO,IAAW,EAAM,GAAK,EAAO,EAAE,EAGpD,IAAM,EAAa,EAAQ,QAAQ,CAAC,EAAY,GAAc,IAAU,CACtE,IAAM,EAAW,EAAQ,EAAQ,GAC3B,EAAO,EAAQ,EAAQ,GAC7B,OACG,IAAW,KAAO,EAAa,GAAK,EAAS,KAAO,EAAc,GAClE,IAAO,KAAO,EAAa,GAAK,EAAK,KAAO,EAAc,CAE/D,CAAC,EACK,EAAa,EAAK,MAAM,gBACxB,EAAc,EAAM,MAAM,gBAC1B,EAA4C,CAAC,EACnD,IAAK,IAAM,KAAW,EAAc,EAAuB,CAAU,CAAC,EAAG,CACvE,GAAM,CAAC,EAAW,GAAc,EAAQ,IAAM,CAAC,EAAG,CAAC,EAC7C,CAAC,EAAU,GAAa,EAAQ,GAAG,EAAE,GAAK,CAAC,EAAG,CAAC,EAC/C,EAAU,EAAW,EACrB,EAAW,EAAY,EACvB,EAAa,EAAU,EACvB,EAAc,EAAW,EACzB,EAAU,KAAK,IAAI,EAAY,CAAW,EAC1C,EAAW,EAAuB,KAAK,IAAI,EAAY,CAAW,EAEtE,EAAU,GACV,EAAU,IAAM,GAChB,EAAmB,CAAO,EAAI,IAAM,GAA2B,GAC/D,CAAC,EACC,EAAM,EAAa,EAAK,QAAS,EAAK,UAAW,EAAW,CAAO,CAAC,EACpE,EAAM,EAAa,EAAM,QAAS,EAAM,UAAW,EAAY,CAAQ,CAAC,CAC1E,GAIoB,EACpB,EAAU,EAAK,QAAQ,SAAS,EAAW,CAAO,CAAC,EACnD,EAAU,EAAM,QAAQ,SAAS,EAAY,CAAQ,CAAC,CAExC,EAAI,KAAO,GACzB,EAAM,KAAK,CACT,KAAM,CACJ,CAAC,EAAa,EAAW,EAAa,CAAO,EAC7C,CAAC,EAAc,EAAY,EAAc,CAAQ,CACnD,EACA,UAAW,EACb,CAAC,CAEL,CACA,OAAO,EAAM,OAAS,EAAI,CAAE,OAAM,EAAI,IAAA,EACxC,EAGA,OAAQ,EAAM,EAAO,IAAe,CAClC,IAAM,EAAW,EAAuB,KAAK,IAAI,EAAK,SAAS,OAAQ,EAAM,SAAS,MAAM,EAC5F,GACE,KAAK,IAAI,EAAK,SAAS,OAAQ,EAAM,SAAS,MAAM,EAAI,KAAO,GAC/D,EAAc,EAAc,IAAI,CAAI,EAAG,EAAc,IAAI,CAAK,CAAC,EAC/D,CACA,GAAI,EAAc,EAAK,eAAgB,EAAM,cAAc,EAAI,KAAO,GACpD,EAAmB,EAAO,EAAY,EAAK,QACjD,EAAI,KAAO,EACnB,MAAO,CAAE,UAAW,EAAM,EAK9B,GAAI,EAAK,WAAa,EAAM,WACR,EAAU,EAAK,UAAU,SAAU,EAAM,UAAU,QACzD,EAAI,KAAO,EACrB,MAAO,CAAE,UAAW,EAAK,CAG/B,CAGA,GAAI,EAAiB,EAAM,eAAgB,EAAK,cAAc,EAAI,EAAG,CACnE,IAAM,EAAU,EAAa,EAAO,CAAI,EACxC,OACE,GAAS,OAAS,CAChB,MAAO,EAAQ,MAAM,KAAK,CAAE,KAAM,CAAC,EAAW,GAAW,GAAG,MAAiB,CAC3E,GAAG,EACH,KAAM,CAAC,EAAU,CAAS,CAC5B,EAAE,CACJ,CAEJ,CACA,OAAO,EAAa,EAAM,CAAK,CACjC,CACF,CAcA,SAAS,EAAc,EAAmC,EAA6C,CACrG,GAAI,CAAC,GAAQ,CAAC,EACZ,MAAO,GAET,IAAI,EAAU,EACd,IAAK,IAAI,EAAY,EAAG,EAAa,EAAG,EAAY,EAAK,QAAQ,QAAU,EAAa,EAAM,QAAQ,QAAS,CAC7G,IAAM,EAAa,EAAK,QAAQ,IAAc,EACxC,EAAc,EAAM,QAAQ,IAAe,EAC7C,IAAe,IACjB,GAAW,KAAK,IAAI,EAAK,eAAe,IAAc,EAAG,EAAM,eAAe,IAAe,CAAC,GAE5F,GAAc,IAChB,GAAa,GAEX,GAAe,IACjB,GAAc,EAElB,CACA,OAAO,EAAU,IAAM,GAA8B,KAAK,IAAI,EAAK,MAAO,EAAM,KAAK,CACvF,CAQA,SAAS,EACP,EACA,EACA,EACA,EAC8B,CAC9B,GAAI,EAAW,OAAS,EACtB,OAEF,IAAM,EAAsE,CAAC,EACvE,GAAY,EAAe,IAAsB,CACrD,IAAM,EAAO,EAAU,EAAQ,SAAS,EAAO,CAAG,CAAC,EAG7C,EAAU,aAAa,KAAK,GAAO,EAAQ,IAAW,EAAS,EAAI,EAAU,EAAU,EAAQ,IAAU,CAAG,EAClH,EAAM,KAAK,CAAE,OAAM,UAAS,OAAM,CAAC,CACrC,EACI,EAAS,EACb,IAAK,GAAM,CAAC,EAAgB,KAAiB,EAAY,CACvD,IAAM,EAAQ,EAAiB,EAC3B,EAAS,GACX,EAAS,EAAQ,CAAK,EAExB,EAAS,EAAO,EAAe,CAAU,EACzC,EAAS,EAAe,CAC1B,CACI,EAAS,EAAQ,QACnB,EAAS,EAAQ,EAAQ,MAAM,EAEjC,EAAM,MAAM,EAAM,IAAU,EAAiB,EAAK,QAAS,EAAM,OAAO,GAAK,EAAK,MAAQ,EAAM,KAAK,EACrG,IAAM,EAAW,IAAI,WAAW,EAAQ,MAAM,EACxC,EAAU,IAAI,WAAW,EAAQ,MAAM,EACzC,EAAW,EACf,IAAK,GAAM,CAAE,OAAM,WAAW,EAAO,CACnC,EAAS,IAAI,EAAM,CAAQ,EAC3B,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAK,OAAQ,GAAS,EAChD,EAAQ,EAAW,GAAS,EAAQ,EAEtC,GAAY,EAAK,MACnB,CACA,MAAO,CAAE,WAAU,SAAQ,CAC7B,CAGA,SAAS,EAAiB,EAAoB,EAA6B,CACzE,IAAK,IAAI,EAAQ,EAAG,EAAQ,KAAK,IAAI,EAAK,OAAQ,EAAM,MAAM,EAAG,GAAS,EAAG,CAC3E,IAAM,GAAc,EAAK,IAAU,IAAM,EAAM,IAAU,GACzD,GAAI,IAAe,EACjB,OAAO,CAEX,CACA,OAAO,EAAK,OAAS,EAAM,MAC7B,CAMA,SAAS,EAAuB,EAAiD,CAC/E,IAAM,EAAwB,CAAC,EACzB,EAAyB,CAAC,EAChC,IAAK,GAAM,CAAC,EAAO,EAAG,MAAiB,EAAQ,QAAQ,EAAG,CACxD,IAAI,EAAM,EACN,EAAO,EAAY,OACvB,KAAO,EAAM,GAAM,CACjB,IAAM,EAAU,EAAM,IAAU,GAC3B,EAAQ,EAAY,IAAW,EAAE,GAAG,IAAM,GAAK,EAClD,EAAM,EAAS,EAEf,EAAO,CAEX,CACA,EAAa,KAAK,EAAM,EAAK,EAAY,EAAM,IAAM,GAAM,EAAE,EAC7D,EAAY,GAAO,CACrB,CACA,IAAM,EAA4B,CAAC,EACnC,IAAK,IAAI,EAAS,EAAY,GAAG,EAAE,GAAK,GAAI,GAAU,EAAG,EAAS,EAAa,IAAW,GAAI,CAC5F,IAAM,EAAS,EAAQ,GACnB,GACF,EAAM,KAAK,CAAM,CAErB,CACA,OAAO,EAAM,WAAW,CAC1B,CAGA,SAAS,EAAc,EAAiD,CACtE,IAAM,EAAiC,CAAC,EACxC,IAAK,GAAM,CAAC,EAAO,KAAW,EAAM,QAAQ,EAAG,CAC7C,IAAM,EAAW,EAAM,EAAQ,GAE7B,IAAa,IAAA,IACb,EAAO,IAAM,EAAS,GAAK,IAAc,IACzC,EAAO,IAAM,EAAS,GAAK,IAAc,GAEzC,EAAS,GAAG,EAAE,CAAC,EAAE,KAAK,CAAM,EAE5B,EAAS,KAAK,CAAC,CAAM,CAAC,CAE1B,CACA,OAAO,CACT,CAGA,SAAS,EAAmB,EAAqC,CAC/D,IAAI,EAAQ,EACZ,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAQ,OAAQ,GAAS,EACnD,GAAS,KAAK,KAAK,EAAQ,EAAM,GAAG,IAAM,IAAM,EAAQ,EAAQ,EAAE,GAAG,IAAM,GAAI,CAAS,EAE1F,OAAO,CACT,CAGA,SAAS,EAAc,EAAkB,EAA2B,CAClE,IAAI,EAAU,EACV,EAAY,EACZ,EAAa,EACjB,KAAO,EAAY,EAAK,QAAU,EAAa,EAAM,QAAQ,CAC3D,IAAM,EAAY,EAAK,IAAc,EAC/B,EAAa,EAAM,IAAe,EACpC,IAAc,GAChB,GAAW,EACX,GAAa,EACb,GAAc,GACL,EAAY,EACrB,GAAa,EAEb,GAAc,CAElB,CACA,OAAO,CACT,CAOA,SAAS,EAAgB,EAAgD,CACvE,IAAM,kBAAmB,IAAI,IACvB,EAA4B,CAAC,EACnC,IAAK,GAAM,CAAC,EAAW,CAAE,SAAQ,sBAAqB,qBAAqB,EAAM,QAAQ,EAAG,CAC1F,GAAI,CAAC,GAAU,CAAC,GAAgB,OAC9B,SAEF,IAAM,EAAU,IAAI,WAAW,EAAO,MAAM,EACtC,EAAY,IAAI,WAAW,EAAO,MAAM,EACxC,EAAgB,IAAI,aAAa,EAAO,MAAM,EAC9C,kBAAiB,IAAI,IAC3B,IAAK,GAAM,CAAC,EAAO,KAAU,EAAO,QAAQ,EAAG,CAC7C,GAAI,EAAM,OAAS,KAAM,CACvB,IAAI,EAAK,EAAe,IAAI,EAAM,IAAI,EAClC,IAAO,IAAA,KACT,EAAK,EAAe,KACpB,EAAe,IAAI,EAAM,KAAM,CAAE,GAEnC,EAAQ,GAAS,EAAE,EAAK,GACxB,QACF,CACA,IAAM,EAAM,EAAS,CAAK,EAC1B,EAAc,GAAS,EACvB,IAAI,EAAS,EAAiB,IAAI,CAAG,EACjC,IAAW,IAAA,KACb,EAAS,EAAiB,KAC1B,EAAiB,IAAI,EAAK,CAAM,GAElC,EAAQ,GAAS,EACjB,EAAU,GAAS,EAAM,QAAU,EAAM,cAAgB,IAAA,GAAY,EAAI,CAC3E,CACA,IAAM,EAAiB,EAA8B,GAAuB,CAAC,CAAC,EAC9E,IAAK,IAAM,KAAS,EAAgB,CAClC,GAAM,CAAE,gBAAiB,EAAO,cAAe,GAAQ,EACjD,EAAe,EAAQ,SAAS,EAAO,CAAG,EAC1C,EAAiB,EAAU,SAAS,EAAO,CAAG,EAC9C,EAAW,EAAU,CAAY,EACjC,EAAc,EAAmB,EAAc,EAAc,SAAS,EAAO,CAAG,CAAC,EACjF,kBAAmB,IAAI,IAC7B,IAAK,IAAM,KAAQ,EACjB,EAAiB,IAAI,GAAO,EAAiB,IAAI,CAAI,GAAK,GAAK,CAAC,EAElE,IAAM,EAAgB,EACnB,KAAK,CAAC,CACN,OAAQ,GAAW,EAAiB,IAAI,EAAY,IAAW,CAAC,IAAM,CAAC,CAAC,CACxE,QAAQ,CAAC,CACT,UAAU,EAAO,KAAY,EAAY,IAAU,IAAM,EAAY,IAAW,EAAE,EACrF,EAAO,KAAK,CACV,YACA,QACA,QAAS,EACT,UAAW,EACX,gBACA,WACA,eAAgB,aAAa,KAAK,GAAW,EAAQ,IACnD,EAAS,EAAI,EAAU,EAAc,EAAQ,IAAU,CACzD,EACA,eAAgB,EAAS,SAAS,EAClC,OAAQ,WAAW,KAAK,EAAiB,KAAK,CAAC,EAC/C,aAAc,WAAW,KAAK,EAAgB,GAAW,EAAY,IAAW,CAAC,EACjF,mBAAoB,WAAW,KAAK,CAAa,EACjD,cAAe,EAAa,EAAc,EAAgB,EAAG,EAAa,MAAM,EAChF,UAAW,EACT,EACA,EAAc,SAAS,EAAO,CAAG,EACjC,EAAe,EAAO,CAAG,EACzB,CACF,CACF,CAAC,CACH,CACF,CACA,OAAO,CACT,CAMA,SAAS,EACP,EACoD,CACpD,IAAM,EAAa,EAChB,KAAK,CAAC,CACN,OAAQ,GAAc,EAAU,gBAAkB,EAAU,aAAa,CAAC,CAC1E,IAAK,GAAgC,CAAC,EAAU,gBAAiB,EAAU,aAAa,CAAC,CAAC,CAC1F,UAAU,EAAM,IAAU,EAAK,GAAK,EAAM,IAAM,EAAM,GAAK,EAAK,EAAE,EACrE,OAAQ,EAAO,IAAQ,CACrB,IAAI,EAAM,EACN,EAAO,EAAW,OACtB,KAAO,EAAM,GAAM,CACjB,IAAM,EAAU,EAAM,IAAU,GAC3B,EAAW,EAAO,GAAG,IAAM,GAAK,EACnC,EAAM,EAAS,EAEf,EAAO,CAEX,CACA,IAAM,EAA+B,CAAC,EACtC,IAAK,IAAI,EAAQ,EAAK,EAAQ,EAAW,OAAQ,GAAS,EAAG,CAC3D,IAAM,EAAY,EAAW,GAC7B,GAAI,CAAC,GAAa,EAAU,IAAM,EAChC,MAEF,IAAM,EAAO,EAAS,GAAG,EAAE,EACrB,EAAS,IAAS,IAAA,IAAa,EAAU,GAAK,EAAK,GACrD,EAAU,IAAM,IAAS,EAAU,KAAO,GAAS,EAAU,KAAO,IAAQ,CAAC,GAC/E,EAAS,KAAK,CAAS,CAE3B,CACA,OAAO,CACT,CACF,CAEA,SAAS,EAAa,EAAqB,EAAuB,EAAe,EAAkC,CACjH,IAAM,kBAAS,IAAI,IACnB,IAAK,IAAI,EAAS,EAAO,EAAS,EAAK,GAAU,EAC/C,GAAI,EAAU,KAAY,EAAG,CAC3B,IAAM,EAAS,EAAQ,IAAW,EAClC,EAAO,IAAI,GAAS,EAAO,IAAI,CAAM,GAAK,GAAK,CAAC,CAClD,CAEF,OAAO,CACT,CAGA,SAAS,EAAU,EAAiC,CAClD,IAAM,kBAAoB,IAAI,IAC9B,OAAO,EAAQ,IAAK,GAAW,CAC7B,GAAI,GAAU,EACZ,OAAO,EAET,IAAI,EAAQ,EAAkB,IAAI,CAAM,EAKxC,OAJI,IAAU,IAAA,KACZ,EAAQ,EAAkB,KAC1B,EAAkB,IAAI,EAAQ,CAAK,GAE9B,EAAE,EAAQ,EACnB,CAAC,CACH,CAOA,SAAS,EAAS,EAAsB,CACtC,IAAM,EAAU,EAAM,SAAW,KAAK,KAAK,EAAM,aAAe,EAAG,UAAa,EAC1E,EAAY,EAAM,UAAY,KAAK,KAAK,EAAM,cAAgB,EAAG,UAAa,EACpF,OAAQ,IAAY,GAAK,SAAc,IAAc,GACvD,CAQA,SAAS,EAAmB,EAAqB,EAAqC,CACpF,IAAM,EAAS,IAAI,WAAW,KAAK,IAAI,EAAQ,OAAS,EAAY,EAAG,CAAC,CAAC,EACzE,IAAK,IAAI,EAAQ,EAAG,EAAQ,EAAO,OAAQ,GAAS,EAAG,CACrD,IAAI,EAAO,KACX,IAAK,IAAI,EAAS,EAAG,EAAS,EAAW,GAAU,EAAG,CACpD,IAAM,EAAW,EAAU,EAAQ,IAAW,EAExC,GAAS,EAAQ,EAAQ,IAAW,GAAK,EAAI,GAAK,EAAW,KAAK,MAAM,EAAW,UAAe,EAExG,EAAQ,KAAK,KAAK,EAAM,EAAE,EAAI,EAAS,CACzC,CACA,EAAO,GAAS,CAClB,CACA,OAAO,CACT"}

package/dist/duplication.cjs

@@ -1,2 +1,2 @@
-"use strict";let e=require("@willbooster/shared-lib");const t={minTokens:40,maxGapTokens:30,minSimilarityPercent:70};function n(e){return{minTokens:r(e?.minTokens,t.minTokens),maxGapTokens:r(e?.maxGapTokens,t.maxGapTokens),minSimilarityPercent:r(e?.minSimilarityPercent,t.minSimilarityPercent)}}function r(t,n){return t===void 0||Number.isNaN(t)?n:(0,e.clamp)(Math.trunc(t),0,4294967295)}function i(e,t){return e*5>=t}function a(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 o(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 o=t=>e[i[t]??0],l=/* @__PURE__ */ new Map,u=a.map((e,n)=>c(o(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 d=(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},f=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},p=[];for(let[e,t]of u.entries())for(let[n,r]of t.windowKeysByStart.entries())for(let[i,a]of r.entries()){if(!d(a,i)||!f({containerIndex:e,start:n,length:i}))continue;let r=t.windowKeysByStart[n]?.[i+1],o=t.windowKeysByStart[n-1]?.[i+1];d(r,i+1)||d(o,i+1)||p.push({containerIndex:e,start:n,length:i})}let m=new Set(p.map(s)),h=p;for(;h.length>0;){let e=[];for(let t of h){let n=a[t.containerIndex],s=n?.[t.start],c=n?.[t.start+t.length-1],l=o(t.containerIndex);if(!s||!c||!l)continue;let u=`s:${x(l.tokens,l.literalCountPrefix,s.startTokenIndex,c.endTokenIndex)}`;r.push({candidate:g(u,s.startTokenIndex,c.endTokenIndex,s,c),contextIndex:i[t.containerIndex]??0}),e.push(t)}h=[];for(let t of e)for(let e of[t.start,t.start+1]){let n={containerIndex:t.containerIndex,start:e,length:t.length-1},r=u[t.containerIndex]?.windowKeysByStart[e]?.[n.length];!m.has(s(n))&&d(r,n.length)&&f(n)&&(m.add(s(n)),h.push(n))}}return r}function s(e){return`${e.containerIndex}:${e.start}:${e.length}`}function c(e,t,n){let r=t.map(t=>S(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=E(o,l),s+=c.endTokenIndex-c.startTokenIndex;let u=i-e+1;a[u]=u>=2&&s>=n?E(o,u):void 0}i.push(a)}return{windowKeysByStart:i,statementHashes:r}}function l(e,t){return u(e)(t)}function u(e){let t=d(e);return e=>{let n=0;for(let r of t(e))n+=h(r);return n}}function d(e){let t=e.length+31>>>5,n=/* @__PURE__ */ 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}}return r}}function f(e,t){let n=new Uint8Array(e.length),r=new Uint8Array(t.length);return p(e,t,n,r),[n,r]}function p(e,t,n,r){let i=0;for(;i<e.length&&i<t.length&&e[i]===t[i];)i+=1;let a=0;for(;a<e.length-i&&a<t.length-i&&e[e.length-1-a]===t[t.length-1-a];)a+=1;let o=e.length-a,s=t.length-a;n.fill(1,0,i).fill(1,o),r.fill(1,0,i).fill(1,s);let c=e.subarray(i,o),l=t.subarray(i,s),u=n.subarray(i,o),d=r.subarray(i,s);if(c.length===0||l.length===0)return;if(c.length===1){let e=l.indexOf(c[0]??0);e!==-1&&(u[0]=1,d[e]=1);return}let f=c.length>>>1,h=m(c.subarray(0,f),l),g=m(c.subarray(f).toReversed(),l.toReversed()),_=0,v=-1;for(let e=0;e<=l.length;e+=1){let t=(h[e]??0)+(g[l.length-e]??0);t>v&&(v=t,_=e)}p(c.subarray(0,f),l.subarray(0,_),u.subarray(0,f),d.subarray(0,_)),p(c.subarray(f),l.subarray(_),u.subarray(f),d.subarray(_))}function m(e,t){let n=d(t)(e),r=new Int32Array(t.length+1);for(let e=0;e<t.length;e+=1)r[e+1]=(r[e]??0)+((n[e>>>5]??0)>>>(e&31)&1);return r}function h(e){let t=e-(e>>>1&1431655765);return t=(t&858993459)+(t>>>2&858993459),Math.imul(t+(t>>>4)&252645135,16843009)>>>24&255}function g(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 _=[],v=[];function y(e){let t=_[e];return t===void 0&&(t=w(`$${e}`),_[e]=t),t}function b(e){let t=v[e];return t===void 0&&(t=T(`$${e}`),v[e]=t),t}function x(e,t,n,r){let[a,o]=C(e,n,r,i((t[r]??0)-(t[n]??0),r-n));return`${a}:${o}:${r-n}`}function S(e,t,n){let[r,i]=C(e,t,n,!1);return r^Math.imul(i,31)}function C(e,t,n,r){let i=/* @__PURE__ */ 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=y(e),c=b(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 w(e){let t=5381;for(let n=0;n<e.length;n+=1)t=Math.imul(t,33)^e.charCodeAt(n);return t}function T(e){let t=-2128831035;for(let n=0;n<e.length;n+=1)t=Math.imul(t^e.charCodeAt(n),16777619);return t}function E(e,t){return Math.imul(e,31)+t}function D(e,t,n=()=>!0){if(t<=0||e.length<2)return e;e.sort(A);for(let r=!0;r;){r=!1;let i=O(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=j(i,s,t,n),l=c??j(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(A),r=!0;break}}return e}function O(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(()=>/* @__PURE__ */ new Set);for(let[i,a]of e.entries())for(let e of a)for(let a=k(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 k(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 A(e,t){let n=e[0],r=t[0];return(n?.startTokenIndex??0)-(r?.startTokenIndex??0)||(n?.endTokenIndex??0)-(r?.endTokenIndex??0)}function j(e,t,n,r){let[i,a]=M(e),[o,s]=M(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,matchedRuns: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 M(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 N(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 P(e,t,n,r){let i=/* @__PURE__ */ new Map,a=/* @__PURE__ */ new Map,o=0;for(let r of e)for(let e=r.startTokenIndex;e<r.endTokenIndex;e+=1){for(;t&&(t[o]?.endTokenIndex??1/0)<=e;)o+=1;let r=!t||(t[o]?.startTokenIndex??1/0)<=e,s=n[e];for(let e=s?.startRow??0;e<=(s?.endRow??-1);e+=1)a.set(e+1,(a.get(e+1)??0)+1),i.set(e+1,(i.get(e+1)??0)+ +!!r)}let s=/* @__PURE__ */ new Set;for(let[e,t]of a)(!r||r.has(e))&&(i.get(e)??0)*2>t&&s.add(e);return s}exports.buildLiteralCountPrefix=a,exports.collectMatchedLines=P,exports.collectSequenceWindowCandidates=o,exports.countRedundantFragments=N,exports.createLcsLengthCounter=u,exports.defaultDuplicationOptions=t,exports.lcsLength=l,exports.markLcs=f,exports.mergeAdjacentGroups=D,exports.resolveDuplicationOptions=n;
+"use strict";let e=require("@willbooster/shared-lib");const t={minTokens:40,maxGapTokens:30,minSimilarityPercent:70};function n(e){return{minTokens:r(e?.minTokens,t.minTokens),maxGapTokens:r(e?.maxGapTokens,t.maxGapTokens),minSimilarityPercent:r(e?.minSimilarityPercent,t.minSimilarityPercent)}}function r(t,n){return t===void 0||Number.isNaN(t)?n:(0,e.clamp)(Math.trunc(t),0,4294967295)}function i(e,t){return e*5>=t}function a(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 o(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 o=t=>e[i[t]??0],l=/* @__PURE__ */ new Map,u=a.map((e,n)=>c(o(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 d=(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},f=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},p=[];for(let[e,t]of u.entries())for(let[n,r]of t.windowKeysByStart.entries())for(let[i,a]of r.entries()){if(!d(a,i)||!f({containerIndex:e,start:n,length:i}))continue;let r=t.windowKeysByStart[n]?.[i+1],o=t.windowKeysByStart[n-1]?.[i+1];d(r,i+1)||d(o,i+1)||p.push({containerIndex:e,start:n,length:i})}let m=new Set(p.map(s)),h=p;for(;h.length>0;){let e=[];for(let t of h){let n=a[t.containerIndex],s=n?.[t.start],c=n?.[t.start+t.length-1],l=o(t.containerIndex);if(!s||!c||!l)continue;let u=`s:${x(l.tokens,l.literalCountPrefix,s.startTokenIndex,c.endTokenIndex)}`;r.push({candidate:g(u,s.startTokenIndex,c.endTokenIndex,s,c),contextIndex:i[t.containerIndex]??0}),e.push(t)}h=[];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];!m.has(s(n))&&d(r,n.length)&&f(n)&&(m.add(s(n)),h.push(n))}}return r}function s(e){return`${e.containerIndex}:${e.start}:${e.length}`}function c(e,t,n){let r=t.map(t=>S(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=E(o,l),s+=c.endTokenIndex-c.startTokenIndex;let u=i-e+1;a[u]=u>=2&&s>=n?E(o,u):void 0}i.push(a)}return{windowKeysByStart:i,statementHashes:r}}function l(e,t){return u(e)(t)}function u(e){let t=d(e);return e=>{let n=0;for(let r of t(e))n+=h(r);return n}}function d(e){let t=e.length+31>>>5,n=/* @__PURE__ */ 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}}return r}}function f(e,t){let n=new Uint8Array(e.length),r=new Uint8Array(t.length);return p(e,t,n,r),[n,r]}function p(e,t,n,r){let i=0;for(;i<e.length&&i<t.length&&e[i]===t[i];)i+=1;let a=0;for(;a<e.length-i&&a<t.length-i&&e[e.length-1-a]===t[t.length-1-a];)a+=1;let o=e.length-a,s=t.length-a;n.fill(1,0,i).fill(1,o),r.fill(1,0,i).fill(1,s);let c=e.subarray(i,o),l=t.subarray(i,s),u=n.subarray(i,o),d=r.subarray(i,s);if(c.length===0||l.length===0)return;if(c.length===1){let e=l.indexOf(c[0]??0);e!==-1&&(u[0]=1,d[e]=1);return}let f=c.length>>>1,h=m(c.subarray(0,f),l),g=m(c.subarray(f).toReversed(),l.toReversed()),_=0,v=-1;for(let e=0;e<=l.length;e+=1){let t=(h[e]??0)+(g[l.length-e]??0);t>v&&(v=t,_=e)}p(c.subarray(0,f),l.subarray(0,_),u.subarray(0,f),d.subarray(0,_)),p(c.subarray(f),l.subarray(_),u.subarray(f),d.subarray(_))}function m(e,t){let n=d(t)(e),r=new Int32Array(t.length+1);for(let e=0;e<t.length;e+=1)r[e+1]=(r[e]??0)+((n[e>>>5]??0)>>>(e&31)&1);return r}function h(e){let t=e-(e>>>1&1431655765);return t=(t&858993459)+(t>>>2&858993459),Math.imul(t+(t>>>4)&252645135,16843009)>>>24&255}function g(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 _=[],v=[];function y(e){let t=_[e];return t===void 0&&(t=w(`$${e}`),_[e]=t),t}function b(e){let t=v[e];return t===void 0&&(t=T(`$${e}`),v[e]=t),t}function x(e,t,n,r){let[a,o]=C(e,n,r,i((t[r]??0)-(t[n]??0),r-n));return`${a}:${o}:${r-n}`}function S(e,t,n){let[r,i]=C(e,t,n,!1);return r^Math.imul(i,31)}function C(e,t,n,r){let i=/* @__PURE__ */ 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=y(e),c=b(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 w(e){let t=5381;for(let n=0;n<e.length;n+=1)t=Math.imul(t,33)^e.charCodeAt(n);return t}function T(e){let t=-2128831035;for(let n=0;n<e.length;n+=1)t=Math.imul(t^e.charCodeAt(n),16777619);return t}function E(e,t){return Math.imul(e,31)+t}function D(e,t,n=()=>!0){if(t<=0||e.length<2)return e;e.sort(A);for(let r=!0;r;){r=!1;let i=O(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=j(i,s,t,n),l=c??j(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(A),r=!0;break}}return e}function O(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(()=>/* @__PURE__ */ new Set);for(let[i,a]of e.entries())for(let e of a)for(let a=k(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 k(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 A(e,t){let n=e[0],r=t[0];return(n?.startTokenIndex??0)-(r?.startTokenIndex??0)||(n?.endTokenIndex??0)-(r?.endTokenIndex??0)}function j(e,t,n,r){let[i,a]=M(e),[o,s]=M(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,matchedRuns: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 M(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 N(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 P(e,t,n,r,i=[]){let a=/* @__PURE__ */ new Set;for(let e of i)for(let t=e.startTokenIndex;t<e.endTokenIndex;t+=1){let e=n[t];for(let t=e?.startRow??0;t<=(e?.endRow??-1);t+=1)a.add(t+1)}let o=/* @__PURE__ */ new Map,s=/* @__PURE__ */ new Map,c=0;for(let r of e)for(let e=r.startTokenIndex;e<r.endTokenIndex;e+=1){for(;t&&(t[c]?.endTokenIndex??1/0)<=e;)c+=1;let r=!t||(t[c]?.startTokenIndex??1/0)<=e,i=n[e];for(let e=i?.startRow??0;e<=(i?.endRow??-1);e+=1)s.set(e+1,(s.get(e+1)??0)+1),o.set(e+1,(o.get(e+1)??0)+ +!!r)}let l=/* @__PURE__ */ new Set;for(let[e,t]of s)(!r||r.has(e))&&((o.get(e)??0)*2>t||a.has(e))&&l.add(e);return l}exports.buildLiteralCountPrefix=a,exports.collectMatchedLines=P,exports.collectSequenceWindowCandidates=o,exports.countRedundantFragments=N,exports.createLcsLengthCounter=u,exports.defaultDuplicationOptions=t,exports.lcsLength=l,exports.markLcs=f,exports.mergeAdjacentGroups=D,exports.resolveDuplicationOptions=n;
 //# sourceMappingURL=duplication.cjs.map