@clazic/kordoc 2.3.0 → 2.3.1

package/dist/mcp.js

@@ -10,13 +10,13 @@ import {
   markdownToHwpx,
   markdownToXlsx,
   parse
-} from "./chunk-2ZGLFZCN.js";
+} from "./chunk-ZOEUKD77.js";
 import {
   KordocError,
   VERSION,
   sanitizeError,
   toArrayBuffer
-} from "./chunk-WWILSVMJ.js";
+} from "./chunk-W5KUC23B.js";
 import "./chunk-MOL7MDBG.js";
 import "./chunk-ZWE3DS7E.js";
 

package/dist/{provider-OBY3XFSZ.js → provider-WYHC4NHI.js}

@@ -150,7 +150,7 @@ function isBatchProvider(p) {
 async function ocrPages(doc, provider, pageFilter, effectivePageCount, warnings, concurrency = 1, onProgress) {
   const blocks = [];
   if (isBatchProvider(provider)) {
-    return ocrPagesBatch(doc, provider, pageFilter, effectivePageCount, warnings, onProgress);
+    return ocrPagesBatch(doc, provider, pageFilter, effectivePageCount, warnings, concurrency, onProgress);
   }
   if (concurrency <= 1) {
     for (let i = 1; i <= effectivePageCount; i++) {
@@ -197,8 +197,7 @@ async function ocrPages(doc, provider, pageFilter, effectivePageCount, warnings,
   }
   return blocks;
 }
-async function ocrPagesBatch(doc, provider, pageFilter, effectivePageCount, warnings, onProgress) {
-  const blocks = [];
+async function ocrPagesBatch(doc, provider, pageFilter, effectivePageCount, warnings, concurrency = 1, onProgress) {
   const pageNumbers = [];
   for (let i = 1; i <= effectivePageCount; i++) {
     if (pageFilter && !pageFilter.has(i)) continue;
@@ -215,16 +214,16 @@ async function ocrPagesBatch(doc, provider, pageFilter, effectivePageCount, warn
     batches.push(pageImages.slice(i, i + provider.batchSize));
   }
   let processed = 0;
-  for (const batch of batches) {
+  const batchTasks = batches.map((batch, batchIdx) => async () => {
+    const pageBlocks = [];
     try {
       const results = await provider.processBatch(batch);
       for (const { pageNum } of batch) {
         const result = results.get(pageNum);
-        if (result) {
-          for (const b of ocrResultToBlocks(result, pageNum)) blocks.push(b);
-        }
-        processed++;
-        onProgress?.(processed, pageNumbers.length);
+        pageBlocks.push({
+          pageNum,
+          blocks: result ? ocrResultToBlocks(result, pageNum) : []
+        });
       }
     } catch (err) {
       const range = `${batch[0].pageNum}-${batch[batch.length - 1].pageNum}`;
@@ -232,8 +231,20 @@ async function ocrPagesBatch(doc, provider, pageFilter, effectivePageCount, warn
         message: `\uBC30\uCE58 OCR \uC2E4\uD328 (\uD398\uC774\uC9C0 ${range}): ${err instanceof Error ? err.message : "\uC54C \uC218 \uC5C6\uB294 \uC624\uB958"}`,
         code: "OCR_PAGE_FAILED"
       });
-      processed += batch.length;
-      onProgress?.(processed, pageNumbers.length);
+      for (const { pageNum } of batch) {
+        pageBlocks.push({ pageNum, blocks: [] });
+      }
+    }
+    processed += batch.length;
+    onProgress?.(processed, pageNumbers.length);
+    return { batchIdx, pageBlocks };
+  });
+  const effectiveConcurrency = Math.max(1, concurrency);
+  const batchResults = await runWithConcurrency(batchTasks, effectiveConcurrency);
+  const blocks = [];
+  for (const result of batchResults) {
+    for (const { blocks: pageBlks } of result.pageBlocks) {
+      for (const b of pageBlks) blocks.push(b);
     }
   }
   return blocks;
@@ -250,4 +261,4 @@ async function renderPageToPng(page) {
 export {
   ocrPages
 };
-//# sourceMappingURL=provider-OBY3XFSZ.js.map
+//# sourceMappingURL=provider-WYHC4NHI.js.map

package/dist/provider-WYHC4NHI.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/ocr/markdown-to-blocks.ts","../src/ocr/provider.ts"],"sourcesContent":["/**\n * Markdown → IRBlock[] 역파싱\n *\n * Vision LLM(gemini/claude/codex 등)이 반환한 Markdown 문자열을\n * kordoc의 IRBlock[] 중간 표현으로 변환.\n * 기존 blocksToMarkdown()의 역방향 처리.\n */\n\nimport type { IRBlock, IRTable, IRCell } from \"../types.js\"\n\n/**\n * Markdown 문자열을 IRBlock[] 배열로 변환.\n *\n * 지원 요소:\n * - 헤딩: # ~ ######\n * - 테이블: | col1 | col2 | (파이프 구분, |---|---| 구분선 포함)\n * - 순서/비순서 리스트: - / 1.\n * - 구분선: ---, ***, ___\n * - 일반 텍스트 (paragraph)\n */\nexport function markdownToBlocks(markdown: string, pageNumber: number): IRBlock[] {\n  const blocks: IRBlock[] = []\n  const lines = markdown.split(\"\\n\")\n  let i = 0\n\n  while (i < lines.length) {\n    const line = lines[i]\n\n    // 빈 줄 스킵\n    if (line.trim() === \"\") {\n      i++\n      continue\n    }\n\n    // 1. 헤딩: # ~ ######\n    const headingMatch = line.match(/^(#{1,6})\\s+(.+)$/)\n    if (headingMatch) {\n      blocks.push({\n        type: \"heading\",\n        level: headingMatch[1].length,\n        text: headingMatch[2].trim(),\n        pageNumber,\n      })\n      i++\n      continue\n    }\n\n    // 2. 구분선: ---, ***, ___\n    if (/^[-*_]{3,}\\s*$/.test(line.trim())) {\n      blocks.push({ type: \"separator\", pageNumber })\n      i++\n      continue\n    }\n\n    // 3. 테이블: | 로 시작하는 연속 행 수집\n    if (line.trim().startsWith(\"|\")) {\n      const tableLines: string[] = []\n      while (i < lines.length && lines[i].trim().startsWith(\"|\")) {\n        tableLines.push(lines[i])\n        i++\n      }\n      const table = parseMarkdownTable(tableLines)\n      if (table) {\n        blocks.push({ type: \"table\", table, pageNumber })\n      }\n      continue\n    }\n\n    // 4. 비순서 리스트: -, *, +\n    const ulMatch = line.match(/^(\\s*)[-*+]\\s+(.+)$/)\n    if (ulMatch) {\n      blocks.push({\n        type: \"list\",\n        listType: \"unordered\",\n        text: ulMatch[2].trim(),\n        pageNumber,\n      })\n      i++\n      continue\n    }\n\n    // 5. 순서 리스트: 1.\n    const olMatch = line.match(/^(\\s*)\\d+\\.\\s+(.+)$/)\n    if (olMatch) {\n      blocks.push({\n        type: \"list\",\n        listType: \"ordered\",\n        text: olMatch[2].trim(),\n        pageNumber,\n      })\n      i++\n      continue\n    }\n\n    // 6. 일반 텍스트 — 구조적 행이 나올 때까지 병합\n    const paraLines: string[] = []\n    while (i < lines.length && lines[i].trim() !== \"\" && !isStructuralLine(lines[i])) {\n      paraLines.push(lines[i].trim())\n      i++\n    }\n    if (paraLines.length > 0) {\n      blocks.push({\n        type: \"paragraph\",\n        text: paraLines.join(\"\\n\"),\n        pageNumber,\n      })\n    }\n  }\n\n  return blocks\n}\n\n/**\n * 구조적 행 판별 — paragraph 병합 중단 트리거.\n */\nfunction isStructuralLine(line: string): boolean {\n  if (/^#{1,6}\\s+/.test(line)) return true\n  if (line.trim().startsWith(\"|\")) return true\n  if (/^[-*_]{3,}\\s*$/.test(line.trim())) return true\n  if (/^\\s*[-*+]\\s+/.test(line)) return true\n  if (/^\\s*\\d+\\.\\s+/.test(line)) return true\n  return false\n}\n\n/**\n * Markdown 테이블 행 배열을 IRTable로 변환.\n *\n * 구분선 행(|---|---|)은 제거 후 데이터 행만 파싱.\n * hasHeader: 구분선이 있었으면 true.\n */\nfunction parseMarkdownTable(lines: string[]): IRTable | null {\n  const hasSeparator = lines.some(line => /^\\|[\\s:|-]+\\|$/.test(line.trim()))\n\n  const rows: IRCell[][] = []\n  let maxCols = 0\n\n  for (const line of lines) {\n    // 구분선 행 스킵: |---|---| 패턴\n    if (/^\\|\\s*:?-+:?\\s*(\\|\\s*:?-+:?\\s*)+\\|?\\s*$/.test(line.trim())) continue\n\n    const parts = line.split(\"|\")\n    // 앞뒤 빈 요소 제거 (| 로 시작/종료하는 행)\n    const cells: IRCell[] = parts\n      .slice(1, parts[parts.length - 1].trim() === \"\" ? -1 : undefined)\n      .map(cell => ({\n        text: cell.trim(),\n        colSpan: 1,\n        rowSpan: 1,\n      }))\n\n    if (cells.length > 0) {\n      rows.push(cells)\n      maxCols = Math.max(maxCols, cells.length)\n    }\n  }\n\n  if (rows.length === 0) return null\n\n  // 열 수 통일 (부족한 셀은 빈 셀로 채움)\n  for (const row of rows) {\n    while (row.length < maxCols) {\n      row.push({ text: \"\", colSpan: 1, rowSpan: 1 })\n    }\n  }\n\n  return {\n    rows: rows.length,\n    cols: maxCols,\n    cells: rows,\n    hasHeader: hasSeparator && rows.length > 1,\n  }\n}\n","/**\n * OCR 프로바이더 브릿지 — PDF 페이지를 이미지로 렌더링하여 OCR 호출\n *\n * kordoc은 OCR 라이브러리를 번들하지 않음.\n * 사용자가 OcrProvider 함수를 제공하면 이미지 기반 PDF도 텍스트 추출 가능.\n *\n * @example\n * ```ts\n * import { parse } from \"kordoc\"\n *\n * const result = await parse(buffer, {\n *   ocr: async (pageImage, pageNumber, mimeType) => {\n *     // Tesseract, Claude Vision, Google Vision 등 사용\n *     return await myOcrService.recognize(pageImage)\n *   }\n * })\n * ```\n */\n\nimport type { OcrProvider, IRBlock, ParseWarning, StructuredOcrResult, BatchOcrProvider } from \"../types.js\"\nimport { markdownToBlocks } from \"./markdown-to-blocks.js\"\n\n/**\n * 동시 실행 수를 제한한 병렬 태스크 실행 헬퍼.\n *\n * limit개의 워커를 만들어 tasks 배열을 순서대로 처리.\n * 각 워커는 완료되는 즉시 다음 태스크를 가져가므로 순서가 보존됨.\n *\n * @param tasks - 실행할 비동기 함수 배열\n * @param limit - 최대 동시 실행 수\n * @returns 입력 순서와 동일한 결과 배열\n */\nasync function runWithConcurrency<T>(\n  tasks: (() => Promise<T>)[],\n  limit: number\n): Promise<T[]> {\n  const results: T[] = new Array(tasks.length)\n  let nextIndex = 0\n\n  // 각 워커는 처리할 태스크가 없을 때까지 반복\n  async function worker() {\n    while (nextIndex < tasks.length) {\n      const idx = nextIndex++\n      results[idx] = await tasks[idx]()\n    }\n  }\n\n  // limit개 워커를 동시 실행 (tasks가 limit보다 적으면 tasks 수만큼)\n  await Promise.all(Array.from({ length: Math.min(limit, tasks.length) }, () => worker()))\n  return results\n}\n\n/**\n * OCR 결과(string | StructuredOcrResult)를 IRBlock[]으로 변환.\n */\nfunction ocrResultToBlocks(result: string | StructuredOcrResult, pageNum: number): IRBlock[] {\n  const pageBlocks: IRBlock[] = []\n  if (typeof result === \"string\") {\n    // 순수 텍스트 → paragraph 블록\n    if (result.trim()) {\n      pageBlocks.push({ type: \"paragraph\", text: result.trim(), pageNumber: pageNum })\n    }\n  } else if (result && typeof result === \"object\" && \"markdown\" in result) {\n    // 구조화된 결과 → Markdown → IRBlock[]\n    const structured = result as StructuredOcrResult\n    if (structured.markdown.trim()) {\n      const converted = markdownToBlocks(structured.markdown, pageNum)\n      for (const b of converted) pageBlocks.push(b)\n    }\n  }\n  return pageBlocks\n}\n\n/** BatchOcrProvider 타입 가드 */\nfunction isBatchProvider(p: unknown): p is BatchOcrProvider {\n  return !!p && typeof p === \"object\" && \"__batch\" in p && (p as BatchOcrProvider).__batch === true\n}\n\n/**\n * 이미지 기반 PDF 페이지에 OCR을 적용하여 IRBlock[] 반환.\n *\n * pdfjs page 객체에서 viewport + render를 통해 PNG 생성 후\n * 사용자 제공 OcrProvider 호출.\n *\n * - string 반환: 단순 텍스트 → paragraph 블록\n * - StructuredOcrResult 반환: Markdown → markdownToBlocks()로 구조화\n * - concurrency > 1: 병렬 처리 (워커 풀 프로바이더 권장)\n *\n * canvas 미설치 시 pdfjs render 불가하므로 에러 반환.\n */\nexport async function ocrPages(\n  doc: { numPages: number; getPage(n: number): Promise<PdfPageProxy> },\n  provider: OcrProvider | BatchOcrProvider,\n  pageFilter: Set<number> | null,\n  effectivePageCount: number,\n  warnings?: ParseWarning[],\n  concurrency: number = 1,    // 기본값 1 = 순차 처리 (하위 호환)\n  onProgress?: (current: number, total: number) => void\n): Promise<IRBlock[]> {\n  const blocks: IRBlock[] = []\n\n  // ── 배치 처리 (BatchOcrProvider) ────────────────────\n  if (isBatchProvider(provider)) {\n    return ocrPagesBatch(doc, provider, pageFilter, effectivePageCount, warnings, concurrency, onProgress)\n  }\n\n  // ── 순차 처리 (concurrency === 1) ────────────────────\n  if (concurrency <= 1) {\n    for (let i = 1; i <= effectivePageCount; i++) {\n      if (pageFilter && !pageFilter.has(i)) continue\n      const page = await doc.getPage(i)\n      try {\n        const imageData = await renderPageToPng(page)\n        const result = await provider(imageData, i, \"image/png\")\n        for (const b of ocrResultToBlocks(result, i)) blocks.push(b)\n      } catch (err) {\n        // 개별 페이지 실패 시 경고 발행 후 계속 진행\n        warnings?.push({\n          page: i,\n          message: `페이지 ${i} OCR 실패: ${err instanceof Error ? err.message : \"알 수 없는 오류\"}`,\n          code: \"OCR_PAGE_FAILED\",\n        })\n      }\n    }\n    return blocks\n  }\n\n  // ── 병렬 처리 (concurrency > 1) ──────────────────────\n  // 처리 대상 페이지 번호 수집\n  const pageNumbers: number[] = []\n  for (let i = 1; i <= effectivePageCount; i++) {\n    if (pageFilter && !pageFilter.has(i)) continue\n    pageNumbers.push(i)\n  }\n\n  // 각 페이지에 대한 태스크 생성 (에러는 개별 캐치)\n  const tasks = pageNumbers.map(pageNum => async (): Promise<{ pageNum: number; pageBlocks: IRBlock[] } | null> => {\n    try {\n      const page = await doc.getPage(pageNum)\n      const imageData = await renderPageToPng(page)\n      const result = await provider(imageData, pageNum, \"image/png\")\n      return { pageNum, pageBlocks: ocrResultToBlocks(result, pageNum) }\n    } catch (err) {\n      // 개별 페이지 실패 시 경고 발행 후 null 반환\n      warnings?.push({\n        page: pageNum,\n        message: `페이지 ${pageNum} OCR 실패: ${err instanceof Error ? err.message : \"알 수 없는 오류\"}`,\n        code: \"OCR_PAGE_FAILED\",\n      })\n      return null\n    }\n  })\n\n  // 병렬 실행 — concurrency 수만큼 동시 처리\n  const taskResults = await runWithConcurrency(tasks, concurrency)\n\n  // 결과를 페이지 번호 순서대로 합산 (pageNumbers 순서 = 오름차순 보장)\n  for (const item of taskResults) {\n    if (!item) continue\n    for (const b of item.pageBlocks) blocks.push(b)\n  }\n\n  return blocks\n}\n\n/**\n * 배치 OCR 처리 — BatchOcrProvider를 사용하여 N페이지씩 묶어 처리.\n *\n * concurrency > 1이면 여러 배치를 동시에 실행하여 속도 향상.\n * 예: 5페이지/배치 × 4 동시 = 20페이지 동시 처리.\n */\nasync function ocrPagesBatch(\n  doc: { numPages: number; getPage(n: number): Promise<PdfPageProxy> },\n  provider: BatchOcrProvider,\n  pageFilter: Set<number> | null,\n  effectivePageCount: number,\n  warnings?: ParseWarning[],\n  concurrency: number = 1,\n  onProgress?: (current: number, total: number) => void\n): Promise<IRBlock[]> {\n  // 1. 대상 페이지 번호 수집\n  const pageNumbers: number[] = []\n  for (let i = 1; i <= effectivePageCount; i++) {\n    if (pageFilter && !pageFilter.has(i)) continue\n    pageNumbers.push(i)\n  }\n\n  // 2. 모든 페이지를 PNG로 렌더링\n  const pageImages: Array<{image: Uint8Array, pageNum: number}> = []\n  for (const pageNum of pageNumbers) {\n    const page = await doc.getPage(pageNum)\n    const image = await renderPageToPng(page)\n    pageImages.push({ image, pageNum })\n  }\n\n  // 3. batchSize개씩 그룹핑\n  const batches: Array<typeof pageImages> = []\n  for (let i = 0; i < pageImages.length; i += provider.batchSize) {\n    batches.push(pageImages.slice(i, i + provider.batchSize))\n  }\n\n  // 4. 배치 태스크 생성 — 각 배치의 결과를 {batchIdx, pageBlocks} 형태로 반환\n  let processed = 0\n  type BatchResult = { batchIdx: number; pageBlocks: Array<{pageNum: number; blocks: IRBlock[]}> }\n\n  const batchTasks = batches.map((batch, batchIdx) => async (): Promise<BatchResult> => {\n    const pageBlocks: Array<{pageNum: number; blocks: IRBlock[]}> = []\n    try {\n      const results = await provider.processBatch(batch)\n      for (const { pageNum } of batch) {\n        const result = results.get(pageNum)\n        pageBlocks.push({\n          pageNum,\n          blocks: result ? ocrResultToBlocks(result, pageNum) : [],\n        })\n      }\n    } catch (err) {\n      const range = `${batch[0].pageNum}-${batch[batch.length - 1].pageNum}`\n      warnings?.push({\n        message: `배치 OCR 실패 (페이지 ${range}): ${err instanceof Error ? err.message : \"알 수 없는 오류\"}`,\n        code: \"OCR_PAGE_FAILED\",\n      })\n      // 실패 배치는 빈 결과\n      for (const { pageNum } of batch) {\n        pageBlocks.push({ pageNum, blocks: [] })\n      }\n    }\n    // 진행률 갱신 (병렬 실행 중 atomic하지 않지만 표시용으로 충분)\n    processed += batch.length\n    onProgress?.(processed, pageNumbers.length)\n    return { batchIdx, pageBlocks }\n  })\n\n  // 5. 병렬 실행 — concurrency개 배치를 동시 처리\n  const effectiveConcurrency = Math.max(1, concurrency)\n  const batchResults = await runWithConcurrency(batchTasks, effectiveConcurrency)\n\n  // 6. 배치 순서대로 블록 합산 (페이지 순서 보존)\n  const blocks: IRBlock[] = []\n  for (const result of batchResults) {\n    for (const { blocks: pageBlks } of result.pageBlocks) {\n      for (const b of pageBlks) blocks.push(b)\n    }\n  }\n\n  return blocks\n}\n\ninterface PdfPageProxy {\n  getViewport(params: { scale: number }): { width: number; height: number }\n  render(params: { canvasContext: unknown; viewport: unknown }): { promise: Promise<void> }\n}\n\n/**\n * PDF 페이지를 PNG로 렌더링.\n * @napi-rs/canvas 사용 (kordoc 번들 의존성, 별도 설치 불필요)\n */\nasync function renderPageToPng(page: PdfPageProxy): Promise<Uint8Array> {\n  const { createCanvas } = await import(\"@napi-rs/canvas\")\n\n  const scale = 2.0 // 300 DPI 근사\n  const viewport = page.getViewport({ scale })\n  const canvas = createCanvas(Math.floor(viewport.width), Math.floor(viewport.height))\n  const ctx = canvas.getContext(\"2d\")\n\n  await page.render({ canvasContext: ctx as unknown, viewport }).promise\n  return new Uint8Array(canvas.toBuffer(\"image/png\"))\n}\n"],"mappings":";;;;AAoBO,SAAS,iBAAiB,UAAkB,YAA+B;AAChF,QAAM,SAAoB,CAAC;AAC3B,QAAM,QAAQ,SAAS,MAAM,IAAI;AACjC,MAAI,IAAI;AAER,SAAO,IAAI,MAAM,QAAQ;AACvB,UAAM,OAAO,MAAM,CAAC;AAGpB,QAAI,KAAK,KAAK,MAAM,IAAI;AACtB;AACA;AAAA,IACF;AAGA,UAAM,eAAe,KAAK,MAAM,mBAAmB;AACnD,QAAI,cAAc;AAChB,aAAO,KAAK;AAAA,QACV,MAAM;AAAA,QACN,OAAO,aAAa,CAAC,EAAE;AAAA,QACvB,MAAM,aAAa,CAAC,EAAE,KAAK;AAAA,QAC3B;AAAA,MACF,CAAC;AACD;AACA;AAAA,IACF;AAGA,QAAI,iBAAiB,KAAK,KAAK,KAAK,CAAC,GAAG;AACtC,aAAO,KAAK,EAAE,MAAM,aAAa,WAAW,CAAC;AAC7C;AACA;AAAA,IACF;AAGA,QAAI,KAAK,KAAK,EAAE,WAAW,GAAG,GAAG;AAC/B,YAAM,aAAuB,CAAC;AAC9B,aAAO,IAAI,MAAM,UAAU,MAAM,CAAC,EAAE,KAAK,EAAE,WAAW,GAAG,GAAG;AAC1D,mBAAW,KAAK,MAAM,CAAC,CAAC;AACxB;AAAA,MACF;AACA,YAAM,QAAQ,mBAAmB,UAAU;AAC3C,UAAI,OAAO;AACT,eAAO,KAAK,EAAE,MAAM,SAAS,OAAO,WAAW,CAAC;AAAA,MAClD;AACA;AAAA,IACF;AAGA,UAAM,UAAU,KAAK,MAAM,qBAAqB;AAChD,QAAI,SAAS;AACX,aAAO,KAAK;AAAA,QACV,MAAM;AAAA,QACN,UAAU;AAAA,QACV,MAAM,QAAQ,CAAC,EAAE,KAAK;AAAA,QACtB;AAAA,MACF,CAAC;AACD;AACA;AAAA,IACF;AAGA,UAAM,UAAU,KAAK,MAAM,qBAAqB;AAChD,QAAI,SAAS;AACX,aAAO,KAAK;AAAA,QACV,MAAM;AAAA,QACN,UAAU;AAAA,QACV,MAAM,QAAQ,CAAC,EAAE,KAAK;AAAA,QACtB;AAAA,MACF,CAAC;AACD;AACA;AAAA,IACF;AAGA,UAAM,YAAsB,CAAC;AAC7B,WAAO,IAAI,MAAM,UAAU,MAAM,CAAC,EAAE,KAAK,MAAM,MAAM,CAAC,iBAAiB,MAAM,CAAC,CAAC,GAAG;AAChF,gBAAU,KAAK,MAAM,CAAC,EAAE,KAAK,CAAC;AAC9B;AAAA,IACF;AACA,QAAI,UAAU,SAAS,GAAG;AACxB,aAAO,KAAK;AAAA,QACV,MAAM;AAAA,QACN,MAAM,UAAU,KAAK,IAAI;AAAA,QACzB;AAAA,MACF,CAAC;AAAA,IACH;AAAA,EACF;AAEA,SAAO;AACT;AAKA,SAAS,iBAAiB,MAAuB;AAC/C,MAAI,aAAa,KAAK,IAAI,EAAG,QAAO;AACpC,MAAI,KAAK,KAAK,EAAE,WAAW,GAAG,EAAG,QAAO;AACxC,MAAI,iBAAiB,KAAK,KAAK,KAAK,CAAC,EAAG,QAAO;AAC/C,MAAI,eAAe,KAAK,IAAI,EAAG,QAAO;AACtC,MAAI,eAAe,KAAK,IAAI,EAAG,QAAO;AACtC,SAAO;AACT;AAQA,SAAS,mBAAmB,OAAiC;AAC3D,QAAM,eAAe,MAAM,KAAK,UAAQ,iBAAiB,KAAK,KAAK,KAAK,CAAC,CAAC;AAE1E,QAAM,OAAmB,CAAC;AAC1B,MAAI,UAAU;AAEd,aAAW,QAAQ,OAAO;AAExB,QAAI,0CAA0C,KAAK,KAAK,KAAK,CAAC,EAAG;AAEjE,UAAM,QAAQ,KAAK,MAAM,GAAG;AAE5B,UAAM,QAAkB,MACrB,MAAM,GAAG,MAAM,MAAM,SAAS,CAAC,EAAE,KAAK,MAAM,KAAK,KAAK,MAAS,EAC/D,IAAI,WAAS;AAAA,MACZ,MAAM,KAAK,KAAK;AAAA,MAChB,SAAS;AAAA,MACT,SAAS;AAAA,IACX,EAAE;AAEJ,QAAI,MAAM,SAAS,GAAG;AACpB,WAAK,KAAK,KAAK;AACf,gBAAU,KAAK,IAAI,SAAS,MAAM,MAAM;AAAA,IAC1C;AAAA,EACF;AAEA,MAAI,KAAK,WAAW,EAAG,QAAO;AAG9B,aAAW,OAAO,MAAM;AACtB,WAAO,IAAI,SAAS,SAAS;AAC3B,UAAI,KAAK,EAAE,MAAM,IAAI,SAAS,GAAG,SAAS,EAAE,CAAC;AAAA,IAC/C;AAAA,EACF;AAEA,SAAO;AAAA,IACL,MAAM,KAAK;AAAA,IACX,MAAM;AAAA,IACN,OAAO;AAAA,IACP,WAAW,gBAAgB,KAAK,SAAS;AAAA,EAC3C;AACF;;;AC3IA,eAAe,mBACb,OACA,OACc;AACd,QAAM,UAAe,IAAI,MAAM,MAAM,MAAM;AAC3C,MAAI,YAAY;AAGhB,iBAAe,SAAS;AACtB,WAAO,YAAY,MAAM,QAAQ;AAC/B,YAAM,MAAM;AACZ,cAAQ,GAAG,IAAI,MAAM,MAAM,GAAG,EAAE;AAAA,IAClC;AAAA,EACF;AAGA,QAAM,QAAQ,IAAI,MAAM,KAAK,EAAE,QAAQ,KAAK,IAAI,OAAO,MAAM,MAAM,EAAE,GAAG,MAAM,OAAO,CAAC,CAAC;AACvF,SAAO;AACT;AAKA,SAAS,kBAAkB,QAAsC,SAA4B;AAC3F,QAAM,aAAwB,CAAC;AAC/B,MAAI,OAAO,WAAW,UAAU;AAE9B,QAAI,OAAO,KAAK,GAAG;AACjB,iBAAW,KAAK,EAAE,MAAM,aAAa,MAAM,OAAO,KAAK,GAAG,YAAY,QAAQ,CAAC;AAAA,IACjF;AAAA,EACF,WAAW,UAAU,OAAO,WAAW,YAAY,cAAc,QAAQ;AAEvE,UAAM,aAAa;AACnB,QAAI,WAAW,SAAS,KAAK,GAAG;AAC9B,YAAM,YAAY,iBAAiB,WAAW,UAAU,OAAO;AAC/D,iBAAW,KAAK,UAAW,YAAW,KAAK,CAAC;AAAA,IAC9C;AAAA,EACF;AACA,SAAO;AACT;AAGA,SAAS,gBAAgB,GAAmC;AAC1D,SAAO,CAAC,CAAC,KAAK,OAAO,MAAM,YAAY,aAAa,KAAM,EAAuB,YAAY;AAC/F;AAcA,eAAsB,SACpB,KACA,UACA,YACA,oBACA,UACA,cAAsB,GACtB,YACoB;AACpB,QAAM,SAAoB,CAAC;AAG3B,MAAI,gBAAgB,QAAQ,GAAG;AAC7B,WAAO,cAAc,KAAK,UAAU,YAAY,oBAAoB,UAAU,aAAa,UAAU;AAAA,EACvG;AAGA,MAAI,eAAe,GAAG;AACpB,aAAS,IAAI,GAAG,KAAK,oBAAoB,KAAK;AAC5C,UAAI,cAAc,CAAC,WAAW,IAAI,CAAC,EAAG;AACtC,YAAM,OAAO,MAAM,IAAI,QAAQ,CAAC;AAChC,UAAI;AACF,cAAM,YAAY,MAAM,gBAAgB,IAAI;AAC5C,cAAM,SAAS,MAAM,SAAS,WAAW,GAAG,WAAW;AACvD,mBAAW,KAAK,kBAAkB,QAAQ,CAAC,EAAG,QAAO,KAAK,CAAC;AAAA,MAC7D,SAAS,KAAK;AAEZ,kBAAU,KAAK;AAAA,UACb,MAAM;AAAA,UACN,SAAS,sBAAO,CAAC,sBAAY,eAAe,QAAQ,IAAI,UAAU,yCAAW;AAAA,UAC7E,MAAM;AAAA,QACR,CAAC;AAAA,MACH;AAAA,IACF;AACA,WAAO;AAAA,EACT;AAIA,QAAM,cAAwB,CAAC;AAC/B,WAAS,IAAI,GAAG,KAAK,oBAAoB,KAAK;AAC5C,QAAI,cAAc,CAAC,WAAW,IAAI,CAAC,EAAG;AACtC,gBAAY,KAAK,CAAC;AAAA,EACpB;AAGA,QAAM,QAAQ,YAAY,IAAI,aAAW,YAAwE;AAC/G,QAAI;AACF,YAAM,OAAO,MAAM,IAAI,QAAQ,OAAO;AACtC,YAAM,YAAY,MAAM,gBAAgB,IAAI;AAC5C,YAAM,SAAS,MAAM,SAAS,WAAW,SAAS,WAAW;AAC7D,aAAO,EAAE,SAAS,YAAY,kBAAkB,QAAQ,OAAO,EAAE;AAAA,IACnE,SAAS,KAAK;AAEZ,gBAAU,KAAK;AAAA,QACb,MAAM;AAAA,QACN,SAAS,sBAAO,OAAO,sBAAY,eAAe,QAAQ,IAAI,UAAU,yCAAW;AAAA,QACnF,MAAM;AAAA,MACR,CAAC;AACD,aAAO;AAAA,IACT;AAAA,EACF,CAAC;AAGD,QAAM,cAAc,MAAM,mBAAmB,OAAO,WAAW;AAG/D,aAAW,QAAQ,aAAa;AAC9B,QAAI,CAAC,KAAM;AACX,eAAW,KAAK,KAAK,WAAY,QAAO,KAAK,CAAC;AAAA,EAChD;AAEA,SAAO;AACT;AAQA,eAAe,cACb,KACA,UACA,YACA,oBACA,UACA,cAAsB,GACtB,YACoB;AAEpB,QAAM,cAAwB,CAAC;AAC/B,WAAS,IAAI,GAAG,KAAK,oBAAoB,KAAK;AAC5C,QAAI,cAAc,CAAC,WAAW,IAAI,CAAC,EAAG;AACtC,gBAAY,KAAK,CAAC;AAAA,EACpB;AAGA,QAAM,aAA0D,CAAC;AACjE,aAAW,WAAW,aAAa;AACjC,UAAM,OAAO,MAAM,IAAI,QAAQ,OAAO;AACtC,UAAM,QAAQ,MAAM,gBAAgB,IAAI;AACxC,eAAW,KAAK,EAAE,OAAO,QAAQ,CAAC;AAAA,EACpC;AAGA,QAAM,UAAoC,CAAC;AAC3C,WAAS,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK,SAAS,WAAW;AAC9D,YAAQ,KAAK,WAAW,MAAM,GAAG,IAAI,SAAS,SAAS,CAAC;AAAA,EAC1D;AAGA,MAAI,YAAY;AAGhB,QAAM,aAAa,QAAQ,IAAI,CAAC,OAAO,aAAa,YAAkC;AACpF,UAAM,aAA0D,CAAC;AACjE,QAAI;AACF,YAAM,UAAU,MAAM,SAAS,aAAa,KAAK;AACjD,iBAAW,EAAE,QAAQ,KAAK,OAAO;AAC/B,cAAM,SAAS,QAAQ,IAAI,OAAO;AAClC,mBAAW,KAAK;AAAA,UACd;AAAA,UACA,QAAQ,SAAS,kBAAkB,QAAQ,OAAO,IAAI,CAAC;AAAA,QACzD,CAAC;AAAA,MACH;AAAA,IACF,SAAS,KAAK;AACZ,YAAM,QAAQ,GAAG,MAAM,CAAC,EAAE,OAAO,IAAI,MAAM,MAAM,SAAS,CAAC,EAAE,OAAO;AACpE,gBAAU,KAAK;AAAA,QACb,SAAS,qDAAkB,KAAK,MAAM,eAAe,QAAQ,IAAI,UAAU,yCAAW;AAAA,QACtF,MAAM;AAAA,MACR,CAAC;AAED,iBAAW,EAAE,QAAQ,KAAK,OAAO;AAC/B,mBAAW,KAAK,EAAE,SAAS,QAAQ,CAAC,EAAE,CAAC;AAAA,MACzC;AAAA,IACF;AAEA,iBAAa,MAAM;AACnB,iBAAa,WAAW,YAAY,MAAM;AAC1C,WAAO,EAAE,UAAU,WAAW;AAAA,EAChC,CAAC;AAGD,QAAM,uBAAuB,KAAK,IAAI,GAAG,WAAW;AACpD,QAAM,eAAe,MAAM,mBAAmB,YAAY,oBAAoB;AAG9E,QAAM,SAAoB,CAAC;AAC3B,aAAW,UAAU,cAAc;AACjC,eAAW,EAAE,QAAQ,SAAS,KAAK,OAAO,YAAY;AACpD,iBAAW,KAAK,SAAU,QAAO,KAAK,CAAC;AAAA,IACzC;AAAA,EACF;AAEA,SAAO;AACT;AAWA,eAAe,gBAAgB,MAAyC;AACtE,QAAM,EAAE,aAAa,IAAI,MAAM,OAAO,iBAAiB;AAEvD,QAAM,QAAQ;AACd,QAAM,WAAW,KAAK,YAAY,EAAE,MAAM,CAAC;AAC3C,QAAM,SAAS,aAAa,KAAK,MAAM,SAAS,KAAK,GAAG,KAAK,MAAM,SAAS,MAAM,CAAC;AACnF,QAAM,MAAM,OAAO,WAAW,IAAI;AAElC,QAAM,KAAK,OAAO,EAAE,eAAe,KAAgB,SAAS,CAAC,EAAE;AAC/D,SAAO,IAAI,WAAW,OAAO,SAAS,WAAW,CAAC;AACpD;","names":[]}

package/dist/{resolve-LBFYRHJI.js → resolve-4FSAQF2S.js}

@@ -201,7 +201,7 @@ async function resolveOcrProvider(mode, warnings, concurrency, batchSize) {
       return createTesseractProvider();
     }
     if (mode === "gemini" || mode === "claude" || mode === "codex") {
-      const { createBatchCliProvider, DEFAULT_BATCH_SIZES } = await import("./batch-provider-FUCIIS4M.js");
+      const { createBatchCliProvider, DEFAULT_BATCH_SIZES } = await import("./batch-provider-PCT4I4LK.js");
       const effectiveBatch = batchSize ?? DEFAULT_BATCH_SIZES[mode];
       if (effectiveBatch > 1) {
         return createBatchCliProvider(mode, effectiveBatch);
@@ -232,7 +232,7 @@ async function resolveOcrProvider(mode, warnings, concurrency, batchSize) {
     return createTesseractProvider();
   }
   if (detected === "gemini" || detected === "codex" || detected === "claude") {
-    const { createBatchCliProvider, DEFAULT_BATCH_SIZES } = await import("./batch-provider-FUCIIS4M.js");
+    const { createBatchCliProvider, DEFAULT_BATCH_SIZES } = await import("./batch-provider-PCT4I4LK.js");
     const effectiveBatch = batchSize ?? DEFAULT_BATCH_SIZES[detected];
     if (effectiveBatch > 1) {
       return createBatchCliProvider(detected, effectiveBatch);
@@ -244,4 +244,4 @@ async function resolveOcrProvider(mode, warnings, concurrency, batchSize) {
 export {
   resolveOcrProvider
 };
-//# sourceMappingURL=resolve-LBFYRHJI.js.map
+//# sourceMappingURL=resolve-4FSAQF2S.js.map

package/dist/{utils-QAK24RJS.js → utils-HSF5HI5T.js}

@@ -8,7 +8,7 @@ import {
   sanitizeError,
   sanitizeHref,
   toArrayBuffer
-} from "./chunk-WWILSVMJ.js";
+} from "./chunk-W5KUC23B.js";
 import "./chunk-ZWE3DS7E.js";
 export {
   KordocError,
@@ -20,4 +20,4 @@ export {
   sanitizeHref,
   toArrayBuffer
 };
-//# sourceMappingURL=utils-QAK24RJS.js.map
+//# sourceMappingURL=utils-HSF5HI5T.js.map

package/dist/{watch-MPHX3QIH.js → watch-R2JHXDGF.js}

@@ -2,10 +2,10 @@
 import {
   detectFormat,
   parse
-} from "./chunk-2ZGLFZCN.js";
+} from "./chunk-ZOEUKD77.js";
 import {
   toArrayBuffer
-} from "./chunk-WWILSVMJ.js";
+} from "./chunk-W5KUC23B.js";
 import "./chunk-MOL7MDBG.js";
 import "./chunk-ZWE3DS7E.js";
 
@@ -126,4 +126,4 @@ async function sendWebhook(url, payload) {
 export {
   watchDirectory
 };
-//# sourceMappingURL=watch-MPHX3QIH.js.map
+//# sourceMappingURL=watch-R2JHXDGF.js.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@clazic/kordoc",
-  "version": "2.3.0",
+  "version": "2.3.1",
   "description": "Parse Korean documents (HWP, HWPX, PDF, XLSX, DOCX) to Markdown",
   "type": "module",
   "exports": {

package/dist/batch-provider-FUCIIS4M.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/ocr/batch-provider.ts"],"sourcesContent":["/**\n * CLI 배치 OCR 프로바이더\n *\n * 여러 페이지 이미지를 단일 CLI 호출로 처리하여 API 호출 수를 대폭 감소.\n * gemini/claude: @file 멀티 참조, codex: --image 멀티 플래그\n *\n * 299페이지 기준:\n * - 기존: CLI 299회 호출 (~30분)\n * - 배치: CLI 3~6회 호출 (~3분)\n */\n\nimport { spawnSync } from \"child_process\"\nimport { writeFileSync, readFileSync, unlinkSync, mkdirSync } from \"fs\"\nimport { join } from \"path\"\nimport { tmpdir } from \"os\"\nimport type { StructuredOcrResult, BatchOcrProvider } from \"../types.js\"\n\n/** 배치 OCR 프롬프트 */\nconst BATCH_OCR_PROMPT =\n  \"다음 문서 페이지 이미지들을 OCR하여 순수 Markdown으로 변환하세요.\\n\\n\" +\n  \"규칙:\\n\" +\n  \"- 각 페이지 결과 사이에 반드시 이 구분자를 삽입: <!-- PAGE_BREAK -->\\n\" +\n  \"- 테이블은 Markdown 테이블 문법 사용 (| 구분, |---|---| 헤더 구분선 포함)\\n\" +\n  \"- 병합된 셀은 해당 위치에 내용 기재\\n\" +\n  \"- 헤딩은 글자 크기에 따라 ## ~ ###### 사용\\n\" +\n  \"- 리스트는 - 또는 1. 사용\\n\" +\n  \"- 이미지, 도형 등 비텍스트 요소는 무시\\n\" +\n  \"- 원문의 읽기 순서와 구조를 유지\\n\" +\n  \"- ```로 감싸지 말고 순수 Markdown만 출력\"\n\n/** 모드별 기본 배치 크기 (CLI 내부 타임아웃 + 실측 기반)\n *\n * gemini CLI: 10장 이상에서 AbortError 발생 (내부 타임아웃).\n * 5장 배치가 안정적으로 동작 확인 (35초/배치).\n * 299페이지 = 60배치 = 기존 299회 대비 80% 감소.\n */\nexport const DEFAULT_BATCH_SIZES: Record<string, number> = {\n  gemini: 5,\n  claude: 5,\n  codex: 10,\n}\n\n/** 임시 디렉토리 — gemini CLI는 cwd 하위 + gitignore 밖만 @참조 가능 */\nlet _batchTempDir: string | null = null\nfunction getBatchTempDir(): string {\n  if (!_batchTempDir) {\n    _batchTempDir = join(process.cwd(), \"_kordoc_ocr_tmp\")\n    mkdirSync(_batchTempDir, { recursive: true })\n  }\n  return _batchTempDir\n}\n\n/**\n * 배치 CLI 프로바이더 생성\n */\nexport function createBatchCliProvider(\n  mode: \"gemini\" | \"claude\" | \"codex\",\n  batchSize: number\n): BatchOcrProvider {\n  return {\n    __batch: true as const,\n    batchSize,\n    async processBatch(pages) {\n      const results = new Map<number, StructuredOcrResult>()\n      const tempDir = getBatchTempDir()\n      const tempFiles: string[] = []\n\n      try {\n        // 1. Write all page images to temp files\n        for (const { image, pageNum } of pages) {\n          const path = join(tempDir, `batch-p${pageNum}.png`)\n          writeFileSync(path, image)\n          tempFiles.push(path)\n        }\n\n        // 2. Call CLI with all file references\n        let output: string\n        if (mode === \"codex\") {\n          output = callBatchCodexCli(tempFiles)\n        } else {\n          output = callBatchCli(mode, tempFiles)\n        }\n\n        // 3. Parse response by PAGE_BREAK separator\n        const cleaned = stripCodeFence(output.trim())\n        const parts = cleaned.split(/<!--\\s*PAGE_BREAK\\s*-->/)\n          .map(p => p.trim())\n          .filter(p => p.length > 0)\n\n        // 4. Map results to page numbers (best-effort if count mismatch)\n        for (let i = 0; i < pages.length; i++) {\n          const pageNum = pages[i].pageNum\n          if (i < parts.length) {\n            results.set(pageNum, { markdown: parts[i] })\n          }\n          // If fewer parts than pages, remaining pages get no result\n        }\n      } finally {\n        // 5. Clean up temp files\n        for (const f of tempFiles) {\n          try { unlinkSync(f) } catch { /* ignore */ }\n        }\n      }\n\n      return results\n    },\n  }\n}\n\n/** gemini/claude 배치 호출 */\nfunction callBatchCli(mode: \"gemini\" | \"claude\", imagePaths: string[]): string {\n  const fileRefs = imagePaths.map(p => `@${p}`).join(\"\\n\")\n  const prompt = `${BATCH_OCR_PROMPT}\\n\\n${fileRefs}`\n\n  let args: string[]\n  if (mode === \"gemini\") {\n    args = [\"--prompt\", prompt, \"--yolo\"]\n    const model = process.env.KORDOC_GEMINI_MODEL\n    if (model) args.push(\"--model\", model)\n  } else {\n    // claude\n    args = [\"--print\", prompt]\n    const model = process.env.KORDOC_CLAUDE_MODEL\n    if (model) args.push(\"--model\", model)\n  }\n\n  const timeoutMs = 60_000 + imagePaths.length * 20_000\n  const result = spawnSync(mode, args, {\n    encoding: \"utf-8\",\n    timeout: timeoutMs,\n    maxBuffer: 50 * 1024 * 1024, // 50MB (large batch output)\n    ...(mode === \"claude\" ? { cwd: tmpdir() } : {}),\n  })\n\n  if (result.error) {\n    throw new Error(`${mode} 배치 OCR 실패: ${result.error.message}`)\n  }\n  if (result.status !== 0) {\n    const errMsg = result.stderr?.trim() || `exit code ${result.status}`\n    throw new Error(`${mode} 배치 OCR 실패: ${errMsg}`)\n  }\n\n  return result.stdout || \"\"\n}\n\n/** codex 배치 호출 — --image를 여러 번 지정 */\nfunction callBatchCodexCli(imagePaths: string[]): string {\n  const outPath = join(tmpdir(), `kordoc-codex-batch-${Date.now()}.txt`)\n  try {\n    const args = [\"exec\", BATCH_OCR_PROMPT]\n    for (const p of imagePaths) {\n      args.push(\"--image\", p)\n    }\n    args.push(\"--output-last-message\", outPath)\n    const model = process.env.KORDOC_CODEX_MODEL\n    if (model) args.push(\"--model\", model)\n\n    const timeoutMs = 60_000 + imagePaths.length * 20_000\n    const result = spawnSync(\"codex\", args, {\n      encoding: \"utf-8\",\n      timeout: timeoutMs,\n      maxBuffer: 50 * 1024 * 1024,\n      input: \"\",\n    })\n\n    if (result.error) {\n      throw new Error(`codex 배치 OCR 실패: ${result.error.message}`)\n    }\n    if (result.status !== 0) {\n      const errMsg = result.stderr?.trim() || `exit code ${result.status}`\n      throw new Error(`codex 배치 OCR 실패: ${errMsg}`)\n    }\n\n    try {\n      return readFileSync(outPath, \"utf-8\")\n    } catch {\n      return result.stdout || \"\"\n    }\n  } finally {\n    try { unlinkSync(outPath) } catch { /* ignore */ }\n  }\n}\n\n/** LLM 출력에서 코드 펜스 제거 (cli-provider.ts와 동일 로직) */\nfunction stripCodeFence(text: string): string {\n  const match = text.match(/^```(?:markdown|md)?\\s*\\n([\\s\\S]*?)\\n```\\s*$/m)\n  return match ? match[1].trim() : text\n}\n"],"mappings":";;;;AAWA,SAAS,iBAAiB;AAC1B,SAAS,eAAe,cAAc,YAAY,iBAAiB;AACnE,SAAS,YAAY;AACrB,SAAS,cAAc;AAIvB,IAAM,mBACJ;AAiBK,IAAM,sBAA8C;AAAA,EACzD,QAAQ;AAAA,EACR,QAAQ;AAAA,EACR,OAAO;AACT;AAGA,IAAI,gBAA+B;AACnC,SAAS,kBAA0B;AACjC,MAAI,CAAC,eAAe;AAClB,oBAAgB,KAAK,QAAQ,IAAI,GAAG,iBAAiB;AACrD,cAAU,eAAe,EAAE,WAAW,KAAK,CAAC;AAAA,EAC9C;AACA,SAAO;AACT;AAKO,SAAS,uBACd,MACA,WACkB;AAClB,SAAO;AAAA,IACL,SAAS;AAAA,IACT;AAAA,IACA,MAAM,aAAa,OAAO;AACxB,YAAM,UAAU,oBAAI,IAAiC;AACrD,YAAM,UAAU,gBAAgB;AAChC,YAAM,YAAsB,CAAC;AAE7B,UAAI;AAEF,mBAAW,EAAE,OAAO,QAAQ,KAAK,OAAO;AACtC,gBAAM,OAAO,KAAK,SAAS,UAAU,OAAO,MAAM;AAClD,wBAAc,MAAM,KAAK;AACzB,oBAAU,KAAK,IAAI;AAAA,QACrB;AAGA,YAAI;AACJ,YAAI,SAAS,SAAS;AACpB,mBAAS,kBAAkB,SAAS;AAAA,QACtC,OAAO;AACL,mBAAS,aAAa,MAAM,SAAS;AAAA,QACvC;AAGA,cAAM,UAAU,eAAe,OAAO,KAAK,CAAC;AAC5C,cAAM,QAAQ,QAAQ,MAAM,yBAAyB,EAClD,IAAI,OAAK,EAAE,KAAK,CAAC,EACjB,OAAO,OAAK,EAAE,SAAS,CAAC;AAG3B,iBAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACrC,gBAAM,UAAU,MAAM,CAAC,EAAE;AACzB,cAAI,IAAI,MAAM,QAAQ;AACpB,oBAAQ,IAAI,SAAS,EAAE,UAAU,MAAM,CAAC,EAAE,CAAC;AAAA,UAC7C;AAAA,QAEF;AAAA,MACF,UAAE;AAEA,mBAAW,KAAK,WAAW;AACzB,cAAI;AAAE,uBAAW,CAAC;AAAA,UAAE,QAAQ;AAAA,UAAe;AAAA,QAC7C;AAAA,MACF;AAEA,aAAO;AAAA,IACT;AAAA,EACF;AACF;AAGA,SAAS,aAAa,MAA2B,YAA8B;AAC7E,QAAM,WAAW,WAAW,IAAI,OAAK,IAAI,CAAC,EAAE,EAAE,KAAK,IAAI;AACvD,QAAM,SAAS,GAAG,gBAAgB;AAAA;AAAA,EAAO,QAAQ;AAEjD,MAAI;AACJ,MAAI,SAAS,UAAU;AACrB,WAAO,CAAC,YAAY,QAAQ,QAAQ;AACpC,UAAM,QAAQ,QAAQ,IAAI;AAC1B,QAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AAAA,EACvC,OAAO;AAEL,WAAO,CAAC,WAAW,MAAM;AACzB,UAAM,QAAQ,QAAQ,IAAI;AAC1B,QAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AAAA,EACvC;AAEA,QAAM,YAAY,MAAS,WAAW,SAAS;AAC/C,QAAM,SAAS,UAAU,MAAM,MAAM;AAAA,IACnC,UAAU;AAAA,IACV,SAAS;AAAA,IACT,WAAW,KAAK,OAAO;AAAA;AAAA,IACvB,GAAI,SAAS,WAAW,EAAE,KAAK,OAAO,EAAE,IAAI,CAAC;AAAA,EAC/C,CAAC;AAED,MAAI,OAAO,OAAO;AAChB,UAAM,IAAI,MAAM,GAAG,IAAI,mCAAe,OAAO,MAAM,OAAO,EAAE;AAAA,EAC9D;AACA,MAAI,OAAO,WAAW,GAAG;AACvB,UAAM,SAAS,OAAO,QAAQ,KAAK,KAAK,aAAa,OAAO,MAAM;AAClE,UAAM,IAAI,MAAM,GAAG,IAAI,mCAAe,MAAM,EAAE;AAAA,EAChD;AAEA,SAAO,OAAO,UAAU;AAC1B;AAGA,SAAS,kBAAkB,YAA8B;AACvD,QAAM,UAAU,KAAK,OAAO,GAAG,sBAAsB,KAAK,IAAI,CAAC,MAAM;AACrE,MAAI;AACF,UAAM,OAAO,CAAC,QAAQ,gBAAgB;AACtC,eAAW,KAAK,YAAY;AAC1B,WAAK,KAAK,WAAW,CAAC;AAAA,IACxB;AACA,SAAK,KAAK,yBAAyB,OAAO;AAC1C,UAAM,QAAQ,QAAQ,IAAI;AAC1B,QAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AAErC,UAAM,YAAY,MAAS,WAAW,SAAS;AAC/C,UAAM,SAAS,UAAU,SAAS,MAAM;AAAA,MACtC,UAAU;AAAA,MACV,SAAS;AAAA,MACT,WAAW,KAAK,OAAO;AAAA,MACvB,OAAO;AAAA,IACT,CAAC;AAED,QAAI,OAAO,OAAO;AAChB,YAAM,IAAI,MAAM,wCAAoB,OAAO,MAAM,OAAO,EAAE;AAAA,IAC5D;AACA,QAAI,OAAO,WAAW,GAAG;AACvB,YAAM,SAAS,OAAO,QAAQ,KAAK,KAAK,aAAa,OAAO,MAAM;AAClE,YAAM,IAAI,MAAM,wCAAoB,MAAM,EAAE;AAAA,IAC9C;AAEA,QAAI;AACF,aAAO,aAAa,SAAS,OAAO;AAAA,IACtC,QAAQ;AACN,aAAO,OAAO,UAAU;AAAA,IAC1B;AAAA,EACF,UAAE;AACA,QAAI;AAAE,iBAAW,OAAO;AAAA,IAAE,QAAQ;AAAA,IAAe;AAAA,EACnD;AACF;AAGA,SAAS,eAAe,MAAsB;AAC5C,QAAM,QAAQ,KAAK,MAAM,+CAA+C;AACxE,SAAO,QAAQ,MAAM,CAAC,EAAE,KAAK,IAAI;AACnC;","names":[]}

package/dist/provider-OBY3XFSZ.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/ocr/markdown-to-blocks.ts","../src/ocr/provider.ts"],"sourcesContent":["/**\n * Markdown → IRBlock[] 역파싱\n *\n * Vision LLM(gemini/claude/codex 등)이 반환한 Markdown 문자열을\n * kordoc의 IRBlock[] 중간 표현으로 변환.\n * 기존 blocksToMarkdown()의 역방향 처리.\n */\n\nimport type { IRBlock, IRTable, IRCell } from \"../types.js\"\n\n/**\n * Markdown 문자열을 IRBlock[] 배열로 변환.\n *\n * 지원 요소:\n * - 헤딩: # ~ ######\n * - 테이블: | col1 | col2 | (파이프 구분, |---|---| 구분선 포함)\n * - 순서/비순서 리스트: - / 1.\n * - 구분선: ---, ***, ___\n * - 일반 텍스트 (paragraph)\n */\nexport function markdownToBlocks(markdown: string, pageNumber: number): IRBlock[] {\n  const blocks: IRBlock[] = []\n  const lines = markdown.split(\"\\n\")\n  let i = 0\n\n  while (i < lines.length) {\n    const line = lines[i]\n\n    // 빈 줄 스킵\n    if (line.trim() === \"\") {\n      i++\n      continue\n    }\n\n    // 1. 헤딩: # ~ ######\n    const headingMatch = line.match(/^(#{1,6})\\s+(.+)$/)\n    if (headingMatch) {\n      blocks.push({\n        type: \"heading\",\n        level: headingMatch[1].length,\n        text: headingMatch[2].trim(),\n        pageNumber,\n      })\n      i++\n      continue\n    }\n\n    // 2. 구분선: ---, ***, ___\n    if (/^[-*_]{3,}\\s*$/.test(line.trim())) {\n      blocks.push({ type: \"separator\", pageNumber })\n      i++\n      continue\n    }\n\n    // 3. 테이블: | 로 시작하는 연속 행 수집\n    if (line.trim().startsWith(\"|\")) {\n      const tableLines: string[] = []\n      while (i < lines.length && lines[i].trim().startsWith(\"|\")) {\n        tableLines.push(lines[i])\n        i++\n      }\n      const table = parseMarkdownTable(tableLines)\n      if (table) {\n        blocks.push({ type: \"table\", table, pageNumber })\n      }\n      continue\n    }\n\n    // 4. 비순서 리스트: -, *, +\n    const ulMatch = line.match(/^(\\s*)[-*+]\\s+(.+)$/)\n    if (ulMatch) {\n      blocks.push({\n        type: \"list\",\n        listType: \"unordered\",\n        text: ulMatch[2].trim(),\n        pageNumber,\n      })\n      i++\n      continue\n    }\n\n    // 5. 순서 리스트: 1.\n    const olMatch = line.match(/^(\\s*)\\d+\\.\\s+(.+)$/)\n    if (olMatch) {\n      blocks.push({\n        type: \"list\",\n        listType: \"ordered\",\n        text: olMatch[2].trim(),\n        pageNumber,\n      })\n      i++\n      continue\n    }\n\n    // 6. 일반 텍스트 — 구조적 행이 나올 때까지 병합\n    const paraLines: string[] = []\n    while (i < lines.length && lines[i].trim() !== \"\" && !isStructuralLine(lines[i])) {\n      paraLines.push(lines[i].trim())\n      i++\n    }\n    if (paraLines.length > 0) {\n      blocks.push({\n        type: \"paragraph\",\n        text: paraLines.join(\"\\n\"),\n        pageNumber,\n      })\n    }\n  }\n\n  return blocks\n}\n\n/**\n * 구조적 행 판별 — paragraph 병합 중단 트리거.\n */\nfunction isStructuralLine(line: string): boolean {\n  if (/^#{1,6}\\s+/.test(line)) return true\n  if (line.trim().startsWith(\"|\")) return true\n  if (/^[-*_]{3,}\\s*$/.test(line.trim())) return true\n  if (/^\\s*[-*+]\\s+/.test(line)) return true\n  if (/^\\s*\\d+\\.\\s+/.test(line)) return true\n  return false\n}\n\n/**\n * Markdown 테이블 행 배열을 IRTable로 변환.\n *\n * 구분선 행(|---|---|)은 제거 후 데이터 행만 파싱.\n * hasHeader: 구분선이 있었으면 true.\n */\nfunction parseMarkdownTable(lines: string[]): IRTable | null {\n  const hasSeparator = lines.some(line => /^\\|[\\s:|-]+\\|$/.test(line.trim()))\n\n  const rows: IRCell[][] = []\n  let maxCols = 0\n\n  for (const line of lines) {\n    // 구분선 행 스킵: |---|---| 패턴\n    if (/^\\|\\s*:?-+:?\\s*(\\|\\s*:?-+:?\\s*)+\\|?\\s*$/.test(line.trim())) continue\n\n    const parts = line.split(\"|\")\n    // 앞뒤 빈 요소 제거 (| 로 시작/종료하는 행)\n    const cells: IRCell[] = parts\n      .slice(1, parts[parts.length - 1].trim() === \"\" ? -1 : undefined)\n      .map(cell => ({\n        text: cell.trim(),\n        colSpan: 1,\n        rowSpan: 1,\n      }))\n\n    if (cells.length > 0) {\n      rows.push(cells)\n      maxCols = Math.max(maxCols, cells.length)\n    }\n  }\n\n  if (rows.length === 0) return null\n\n  // 열 수 통일 (부족한 셀은 빈 셀로 채움)\n  for (const row of rows) {\n    while (row.length < maxCols) {\n      row.push({ text: \"\", colSpan: 1, rowSpan: 1 })\n    }\n  }\n\n  return {\n    rows: rows.length,\n    cols: maxCols,\n    cells: rows,\n    hasHeader: hasSeparator && rows.length > 1,\n  }\n}\n","/**\n * OCR 프로바이더 브릿지 — PDF 페이지를 이미지로 렌더링하여 OCR 호출\n *\n * kordoc은 OCR 라이브러리를 번들하지 않음.\n * 사용자가 OcrProvider 함수를 제공하면 이미지 기반 PDF도 텍스트 추출 가능.\n *\n * @example\n * ```ts\n * import { parse } from \"kordoc\"\n *\n * const result = await parse(buffer, {\n *   ocr: async (pageImage, pageNumber, mimeType) => {\n *     // Tesseract, Claude Vision, Google Vision 등 사용\n *     return await myOcrService.recognize(pageImage)\n *   }\n * })\n * ```\n */\n\nimport type { OcrProvider, IRBlock, ParseWarning, StructuredOcrResult, BatchOcrProvider } from \"../types.js\"\nimport { markdownToBlocks } from \"./markdown-to-blocks.js\"\n\n/**\n * 동시 실행 수를 제한한 병렬 태스크 실행 헬퍼.\n *\n * limit개의 워커를 만들어 tasks 배열을 순서대로 처리.\n * 각 워커는 완료되는 즉시 다음 태스크를 가져가므로 순서가 보존됨.\n *\n * @param tasks - 실행할 비동기 함수 배열\n * @param limit - 최대 동시 실행 수\n * @returns 입력 순서와 동일한 결과 배열\n */\nasync function runWithConcurrency<T>(\n  tasks: (() => Promise<T>)[],\n  limit: number\n): Promise<T[]> {\n  const results: T[] = new Array(tasks.length)\n  let nextIndex = 0\n\n  // 각 워커는 처리할 태스크가 없을 때까지 반복\n  async function worker() {\n    while (nextIndex < tasks.length) {\n      const idx = nextIndex++\n      results[idx] = await tasks[idx]()\n    }\n  }\n\n  // limit개 워커를 동시 실행 (tasks가 limit보다 적으면 tasks 수만큼)\n  await Promise.all(Array.from({ length: Math.min(limit, tasks.length) }, () => worker()))\n  return results\n}\n\n/**\n * OCR 결과(string | StructuredOcrResult)를 IRBlock[]으로 변환.\n */\nfunction ocrResultToBlocks(result: string | StructuredOcrResult, pageNum: number): IRBlock[] {\n  const pageBlocks: IRBlock[] = []\n  if (typeof result === \"string\") {\n    // 순수 텍스트 → paragraph 블록\n    if (result.trim()) {\n      pageBlocks.push({ type: \"paragraph\", text: result.trim(), pageNumber: pageNum })\n    }\n  } else if (result && typeof result === \"object\" && \"markdown\" in result) {\n    // 구조화된 결과 → Markdown → IRBlock[]\n    const structured = result as StructuredOcrResult\n    if (structured.markdown.trim()) {\n      const converted = markdownToBlocks(structured.markdown, pageNum)\n      for (const b of converted) pageBlocks.push(b)\n    }\n  }\n  return pageBlocks\n}\n\n/** BatchOcrProvider 타입 가드 */\nfunction isBatchProvider(p: unknown): p is BatchOcrProvider {\n  return !!p && typeof p === \"object\" && \"__batch\" in p && (p as BatchOcrProvider).__batch === true\n}\n\n/**\n * 이미지 기반 PDF 페이지에 OCR을 적용하여 IRBlock[] 반환.\n *\n * pdfjs page 객체에서 viewport + render를 통해 PNG 생성 후\n * 사용자 제공 OcrProvider 호출.\n *\n * - string 반환: 단순 텍스트 → paragraph 블록\n * - StructuredOcrResult 반환: Markdown → markdownToBlocks()로 구조화\n * - concurrency > 1: 병렬 처리 (워커 풀 프로바이더 권장)\n *\n * canvas 미설치 시 pdfjs render 불가하므로 에러 반환.\n */\nexport async function ocrPages(\n  doc: { numPages: number; getPage(n: number): Promise<PdfPageProxy> },\n  provider: OcrProvider | BatchOcrProvider,\n  pageFilter: Set<number> | null,\n  effectivePageCount: number,\n  warnings?: ParseWarning[],\n  concurrency: number = 1,    // 기본값 1 = 순차 처리 (하위 호환)\n  onProgress?: (current: number, total: number) => void\n): Promise<IRBlock[]> {\n  const blocks: IRBlock[] = []\n\n  // ── 배치 처리 (BatchOcrProvider) ────────────────────\n  if (isBatchProvider(provider)) {\n    return ocrPagesBatch(doc, provider, pageFilter, effectivePageCount, warnings, onProgress)\n  }\n\n  // ── 순차 처리 (concurrency === 1) ────────────────────\n  if (concurrency <= 1) {\n    for (let i = 1; i <= effectivePageCount; i++) {\n      if (pageFilter && !pageFilter.has(i)) continue\n      const page = await doc.getPage(i)\n      try {\n        const imageData = await renderPageToPng(page)\n        const result = await provider(imageData, i, \"image/png\")\n        for (const b of ocrResultToBlocks(result, i)) blocks.push(b)\n      } catch (err) {\n        // 개별 페이지 실패 시 경고 발행 후 계속 진행\n        warnings?.push({\n          page: i,\n          message: `페이지 ${i} OCR 실패: ${err instanceof Error ? err.message : \"알 수 없는 오류\"}`,\n          code: \"OCR_PAGE_FAILED\",\n        })\n      }\n    }\n    return blocks\n  }\n\n  // ── 병렬 처리 (concurrency > 1) ──────────────────────\n  // 처리 대상 페이지 번호 수집\n  const pageNumbers: number[] = []\n  for (let i = 1; i <= effectivePageCount; i++) {\n    if (pageFilter && !pageFilter.has(i)) continue\n    pageNumbers.push(i)\n  }\n\n  // 각 페이지에 대한 태스크 생성 (에러는 개별 캐치)\n  const tasks = pageNumbers.map(pageNum => async (): Promise<{ pageNum: number; pageBlocks: IRBlock[] } | null> => {\n    try {\n      const page = await doc.getPage(pageNum)\n      const imageData = await renderPageToPng(page)\n      const result = await provider(imageData, pageNum, \"image/png\")\n      return { pageNum, pageBlocks: ocrResultToBlocks(result, pageNum) }\n    } catch (err) {\n      // 개별 페이지 실패 시 경고 발행 후 null 반환\n      warnings?.push({\n        page: pageNum,\n        message: `페이지 ${pageNum} OCR 실패: ${err instanceof Error ? err.message : \"알 수 없는 오류\"}`,\n        code: \"OCR_PAGE_FAILED\",\n      })\n      return null\n    }\n  })\n\n  // 병렬 실행 — concurrency 수만큼 동시 처리\n  const taskResults = await runWithConcurrency(tasks, concurrency)\n\n  // 결과를 페이지 번호 순서대로 합산 (pageNumbers 순서 = 오름차순 보장)\n  for (const item of taskResults) {\n    if (!item) continue\n    for (const b of item.pageBlocks) blocks.push(b)\n  }\n\n  return blocks\n}\n\n/**\n * 배치 OCR 처리 — BatchOcrProvider를 사용하여 N페이지씩 묶어 처리.\n */\nasync function ocrPagesBatch(\n  doc: { numPages: number; getPage(n: number): Promise<PdfPageProxy> },\n  provider: BatchOcrProvider,\n  pageFilter: Set<number> | null,\n  effectivePageCount: number,\n  warnings?: ParseWarning[],\n  onProgress?: (current: number, total: number) => void\n): Promise<IRBlock[]> {\n  const blocks: IRBlock[] = []\n\n  // 1. 대상 페이지 번호 수집\n  const pageNumbers: number[] = []\n  for (let i = 1; i <= effectivePageCount; i++) {\n    if (pageFilter && !pageFilter.has(i)) continue\n    pageNumbers.push(i)\n  }\n\n  // 2. 모든 페이지를 PNG로 렌더링\n  const pageImages: Array<{image: Uint8Array, pageNum: number}> = []\n  for (const pageNum of pageNumbers) {\n    const page = await doc.getPage(pageNum)\n    const image = await renderPageToPng(page)\n    pageImages.push({ image, pageNum })\n  }\n\n  // 3. batchSize개씩 그룹핑\n  const batches: Array<typeof pageImages> = []\n  for (let i = 0; i < pageImages.length; i += provider.batchSize) {\n    batches.push(pageImages.slice(i, i + provider.batchSize))\n  }\n\n  // 4. 각 배치 처리\n  let processed = 0\n  for (const batch of batches) {\n    try {\n      const results = await provider.processBatch(batch)\n      for (const { pageNum } of batch) {\n        const result = results.get(pageNum)\n        if (result) {\n          for (const b of ocrResultToBlocks(result, pageNum)) blocks.push(b)\n        }\n        processed++\n        onProgress?.(processed, pageNumbers.length)\n      }\n    } catch (err) {\n      // 배치 실패 — 해당 배치 전체 페이지에 경고 추가\n      const range = `${batch[0].pageNum}-${batch[batch.length - 1].pageNum}`\n      warnings?.push({\n        message: `배치 OCR 실패 (페이지 ${range}): ${err instanceof Error ? err.message : \"알 수 없는 오류\"}`,\n        code: \"OCR_PAGE_FAILED\",\n      })\n      processed += batch.length\n      onProgress?.(processed, pageNumbers.length)\n    }\n  }\n\n  return blocks\n}\n\ninterface PdfPageProxy {\n  getViewport(params: { scale: number }): { width: number; height: number }\n  render(params: { canvasContext: unknown; viewport: unknown }): { promise: Promise<void> }\n}\n\n/**\n * PDF 페이지를 PNG로 렌더링.\n * @napi-rs/canvas 사용 (kordoc 번들 의존성, 별도 설치 불필요)\n */\nasync function renderPageToPng(page: PdfPageProxy): Promise<Uint8Array> {\n  const { createCanvas } = await import(\"@napi-rs/canvas\")\n\n  const scale = 2.0 // 300 DPI 근사\n  const viewport = page.getViewport({ scale })\n  const canvas = createCanvas(Math.floor(viewport.width), Math.floor(viewport.height))\n  const ctx = canvas.getContext(\"2d\")\n\n  await page.render({ canvasContext: ctx as unknown, viewport }).promise\n  return new Uint8Array(canvas.toBuffer(\"image/png\"))\n}\n"],"mappings":";;;;AAoBO,SAAS,iBAAiB,UAAkB,YAA+B;AAChF,QAAM,SAAoB,CAAC;AAC3B,QAAM,QAAQ,SAAS,MAAM,IAAI;AACjC,MAAI,IAAI;AAER,SAAO,IAAI,MAAM,QAAQ;AACvB,UAAM,OAAO,MAAM,CAAC;AAGpB,QAAI,KAAK,KAAK,MAAM,IAAI;AACtB;AACA;AAAA,IACF;AAGA,UAAM,eAAe,KAAK,MAAM,mBAAmB;AACnD,QAAI,cAAc;AAChB,aAAO,KAAK;AAAA,QACV,MAAM;AAAA,QACN,OAAO,aAAa,CAAC,EAAE;AAAA,QACvB,MAAM,aAAa,CAAC,EAAE,KAAK;AAAA,QAC3B;AAAA,MACF,CAAC;AACD;AACA;AAAA,IACF;AAGA,QAAI,iBAAiB,KAAK,KAAK,KAAK,CAAC,GAAG;AACtC,aAAO,KAAK,EAAE,MAAM,aAAa,WAAW,CAAC;AAC7C;AACA;AAAA,IACF;AAGA,QAAI,KAAK,KAAK,EAAE,WAAW,GAAG,GAAG;AAC/B,YAAM,aAAuB,CAAC;AAC9B,aAAO,IAAI,MAAM,UAAU,MAAM,CAAC,EAAE,KAAK,EAAE,WAAW,GAAG,GAAG;AAC1D,mBAAW,KAAK,MAAM,CAAC,CAAC;AACxB;AAAA,MACF;AACA,YAAM,QAAQ,mBAAmB,UAAU;AAC3C,UAAI,OAAO;AACT,eAAO,KAAK,EAAE,MAAM,SAAS,OAAO,WAAW,CAAC;AAAA,MAClD;AACA;AAAA,IACF;AAGA,UAAM,UAAU,KAAK,MAAM,qBAAqB;AAChD,QAAI,SAAS;AACX,aAAO,KAAK;AAAA,QACV,MAAM;AAAA,QACN,UAAU;AAAA,QACV,MAAM,QAAQ,CAAC,EAAE,KAAK;AAAA,QACtB;AAAA,MACF,CAAC;AACD;AACA;AAAA,IACF;AAGA,UAAM,UAAU,KAAK,MAAM,qBAAqB;AAChD,QAAI,SAAS;AACX,aAAO,KAAK;AAAA,QACV,MAAM;AAAA,QACN,UAAU;AAAA,QACV,MAAM,QAAQ,CAAC,EAAE,KAAK;AAAA,QACtB;AAAA,MACF,CAAC;AACD;AACA;AAAA,IACF;AAGA,UAAM,YAAsB,CAAC;AAC7B,WAAO,IAAI,MAAM,UAAU,MAAM,CAAC,EAAE,KAAK,MAAM,MAAM,CAAC,iBAAiB,MAAM,CAAC,CAAC,GAAG;AAChF,gBAAU,KAAK,MAAM,CAAC,EAAE,KAAK,CAAC;AAC9B;AAAA,IACF;AACA,QAAI,UAAU,SAAS,GAAG;AACxB,aAAO,KAAK;AAAA,QACV,MAAM;AAAA,QACN,MAAM,UAAU,KAAK,IAAI;AAAA,QACzB;AAAA,MACF,CAAC;AAAA,IACH;AAAA,EACF;AAEA,SAAO;AACT;AAKA,SAAS,iBAAiB,MAAuB;AAC/C,MAAI,aAAa,KAAK,IAAI,EAAG,QAAO;AACpC,MAAI,KAAK,KAAK,EAAE,WAAW,GAAG,EAAG,QAAO;AACxC,MAAI,iBAAiB,KAAK,KAAK,KAAK,CAAC,EAAG,QAAO;AAC/C,MAAI,eAAe,KAAK,IAAI,EAAG,QAAO;AACtC,MAAI,eAAe,KAAK,IAAI,EAAG,QAAO;AACtC,SAAO;AACT;AAQA,SAAS,mBAAmB,OAAiC;AAC3D,QAAM,eAAe,MAAM,KAAK,UAAQ,iBAAiB,KAAK,KAAK,KAAK,CAAC,CAAC;AAE1E,QAAM,OAAmB,CAAC;AAC1B,MAAI,UAAU;AAEd,aAAW,QAAQ,OAAO;AAExB,QAAI,0CAA0C,KAAK,KAAK,KAAK,CAAC,EAAG;AAEjE,UAAM,QAAQ,KAAK,MAAM,GAAG;AAE5B,UAAM,QAAkB,MACrB,MAAM,GAAG,MAAM,MAAM,SAAS,CAAC,EAAE,KAAK,MAAM,KAAK,KAAK,MAAS,EAC/D,IAAI,WAAS;AAAA,MACZ,MAAM,KAAK,KAAK;AAAA,MAChB,SAAS;AAAA,MACT,SAAS;AAAA,IACX,EAAE;AAEJ,QAAI,MAAM,SAAS,GAAG;AACpB,WAAK,KAAK,KAAK;AACf,gBAAU,KAAK,IAAI,SAAS,MAAM,MAAM;AAAA,IAC1C;AAAA,EACF;AAEA,MAAI,KAAK,WAAW,EAAG,QAAO;AAG9B,aAAW,OAAO,MAAM;AACtB,WAAO,IAAI,SAAS,SAAS;AAC3B,UAAI,KAAK,EAAE,MAAM,IAAI,SAAS,GAAG,SAAS,EAAE,CAAC;AAAA,IAC/C;AAAA,EACF;AAEA,SAAO;AAAA,IACL,MAAM,KAAK;AAAA,IACX,MAAM;AAAA,IACN,OAAO;AAAA,IACP,WAAW,gBAAgB,KAAK,SAAS;AAAA,EAC3C;AACF;;;AC3IA,eAAe,mBACb,OACA,OACc;AACd,QAAM,UAAe,IAAI,MAAM,MAAM,MAAM;AAC3C,MAAI,YAAY;AAGhB,iBAAe,SAAS;AACtB,WAAO,YAAY,MAAM,QAAQ;AAC/B,YAAM,MAAM;AACZ,cAAQ,GAAG,IAAI,MAAM,MAAM,GAAG,EAAE;AAAA,IAClC;AAAA,EACF;AAGA,QAAM,QAAQ,IAAI,MAAM,KAAK,EAAE,QAAQ,KAAK,IAAI,OAAO,MAAM,MAAM,EAAE,GAAG,MAAM,OAAO,CAAC,CAAC;AACvF,SAAO;AACT;AAKA,SAAS,kBAAkB,QAAsC,SAA4B;AAC3F,QAAM,aAAwB,CAAC;AAC/B,MAAI,OAAO,WAAW,UAAU;AAE9B,QAAI,OAAO,KAAK,GAAG;AACjB,iBAAW,KAAK,EAAE,MAAM,aAAa,MAAM,OAAO,KAAK,GAAG,YAAY,QAAQ,CAAC;AAAA,IACjF;AAAA,EACF,WAAW,UAAU,OAAO,WAAW,YAAY,cAAc,QAAQ;AAEvE,UAAM,aAAa;AACnB,QAAI,WAAW,SAAS,KAAK,GAAG;AAC9B,YAAM,YAAY,iBAAiB,WAAW,UAAU,OAAO;AAC/D,iBAAW,KAAK,UAAW,YAAW,KAAK,CAAC;AAAA,IAC9C;AAAA,EACF;AACA,SAAO;AACT;AAGA,SAAS,gBAAgB,GAAmC;AAC1D,SAAO,CAAC,CAAC,KAAK,OAAO,MAAM,YAAY,aAAa,KAAM,EAAuB,YAAY;AAC/F;AAcA,eAAsB,SACpB,KACA,UACA,YACA,oBACA,UACA,cAAsB,GACtB,YACoB;AACpB,QAAM,SAAoB,CAAC;AAG3B,MAAI,gBAAgB,QAAQ,GAAG;AAC7B,WAAO,cAAc,KAAK,UAAU,YAAY,oBAAoB,UAAU,UAAU;AAAA,EAC1F;AAGA,MAAI,eAAe,GAAG;AACpB,aAAS,IAAI,GAAG,KAAK,oBAAoB,KAAK;AAC5C,UAAI,cAAc,CAAC,WAAW,IAAI,CAAC,EAAG;AACtC,YAAM,OAAO,MAAM,IAAI,QAAQ,CAAC;AAChC,UAAI;AACF,cAAM,YAAY,MAAM,gBAAgB,IAAI;AAC5C,cAAM,SAAS,MAAM,SAAS,WAAW,GAAG,WAAW;AACvD,mBAAW,KAAK,kBAAkB,QAAQ,CAAC,EAAG,QAAO,KAAK,CAAC;AAAA,MAC7D,SAAS,KAAK;AAEZ,kBAAU,KAAK;AAAA,UACb,MAAM;AAAA,UACN,SAAS,sBAAO,CAAC,sBAAY,eAAe,QAAQ,IAAI,UAAU,yCAAW;AAAA,UAC7E,MAAM;AAAA,QACR,CAAC;AAAA,MACH;AAAA,IACF;AACA,WAAO;AAAA,EACT;AAIA,QAAM,cAAwB,CAAC;AAC/B,WAAS,IAAI,GAAG,KAAK,oBAAoB,KAAK;AAC5C,QAAI,cAAc,CAAC,WAAW,IAAI,CAAC,EAAG;AACtC,gBAAY,KAAK,CAAC;AAAA,EACpB;AAGA,QAAM,QAAQ,YAAY,IAAI,aAAW,YAAwE;AAC/G,QAAI;AACF,YAAM,OAAO,MAAM,IAAI,QAAQ,OAAO;AACtC,YAAM,YAAY,MAAM,gBAAgB,IAAI;AAC5C,YAAM,SAAS,MAAM,SAAS,WAAW,SAAS,WAAW;AAC7D,aAAO,EAAE,SAAS,YAAY,kBAAkB,QAAQ,OAAO,EAAE;AAAA,IACnE,SAAS,KAAK;AAEZ,gBAAU,KAAK;AAAA,QACb,MAAM;AAAA,QACN,SAAS,sBAAO,OAAO,sBAAY,eAAe,QAAQ,IAAI,UAAU,yCAAW;AAAA,QACnF,MAAM;AAAA,MACR,CAAC;AACD,aAAO;AAAA,IACT;AAAA,EACF,CAAC;AAGD,QAAM,cAAc,MAAM,mBAAmB,OAAO,WAAW;AAG/D,aAAW,QAAQ,aAAa;AAC9B,QAAI,CAAC,KAAM;AACX,eAAW,KAAK,KAAK,WAAY,QAAO,KAAK,CAAC;AAAA,EAChD;AAEA,SAAO;AACT;AAKA,eAAe,cACb,KACA,UACA,YACA,oBACA,UACA,YACoB;AACpB,QAAM,SAAoB,CAAC;AAG3B,QAAM,cAAwB,CAAC;AAC/B,WAAS,IAAI,GAAG,KAAK,oBAAoB,KAAK;AAC5C,QAAI,cAAc,CAAC,WAAW,IAAI,CAAC,EAAG;AACtC,gBAAY,KAAK,CAAC;AAAA,EACpB;AAGA,QAAM,aAA0D,CAAC;AACjE,aAAW,WAAW,aAAa;AACjC,UAAM,OAAO,MAAM,IAAI,QAAQ,OAAO;AACtC,UAAM,QAAQ,MAAM,gBAAgB,IAAI;AACxC,eAAW,KAAK,EAAE,OAAO,QAAQ,CAAC;AAAA,EACpC;AAGA,QAAM,UAAoC,CAAC;AAC3C,WAAS,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK,SAAS,WAAW;AAC9D,YAAQ,KAAK,WAAW,MAAM,GAAG,IAAI,SAAS,SAAS,CAAC;AAAA,EAC1D;AAGA,MAAI,YAAY;AAChB,aAAW,SAAS,SAAS;AAC3B,QAAI;AACF,YAAM,UAAU,MAAM,SAAS,aAAa,KAAK;AACjD,iBAAW,EAAE,QAAQ,KAAK,OAAO;AAC/B,cAAM,SAAS,QAAQ,IAAI,OAAO;AAClC,YAAI,QAAQ;AACV,qBAAW,KAAK,kBAAkB,QAAQ,OAAO,EAAG,QAAO,KAAK,CAAC;AAAA,QACnE;AACA;AACA,qBAAa,WAAW,YAAY,MAAM;AAAA,MAC5C;AAAA,IACF,SAAS,KAAK;AAEZ,YAAM,QAAQ,GAAG,MAAM,CAAC,EAAE,OAAO,IAAI,MAAM,MAAM,SAAS,CAAC,EAAE,OAAO;AACpE,gBAAU,KAAK;AAAA,QACb,SAAS,qDAAkB,KAAK,MAAM,eAAe,QAAQ,IAAI,UAAU,yCAAW;AAAA,QACtF,MAAM;AAAA,MACR,CAAC;AACD,mBAAa,MAAM;AACnB,mBAAa,WAAW,YAAY,MAAM;AAAA,IAC5C;AAAA,EACF;AAEA,SAAO;AACT;AAWA,eAAe,gBAAgB,MAAyC;AACtE,QAAM,EAAE,aAAa,IAAI,MAAM,OAAO,iBAAiB;AAEvD,QAAM,QAAQ;AACd,QAAM,WAAW,KAAK,YAAY,EAAE,MAAM,CAAC;AAC3C,QAAM,SAAS,aAAa,KAAK,MAAM,SAAS,KAAK,GAAG,KAAK,MAAM,SAAS,MAAM,CAAC;AACnF,QAAM,MAAM,OAAO,WAAW,IAAI;AAElC,QAAM,KAAK,OAAO,EAAE,eAAe,KAAgB,SAAS,CAAC,EAAE;AAC/D,SAAO,IAAI,WAAW,OAAO,SAAS,WAAW,CAAC;AACpD;","names":[]}