@clazic/kordoc 2.2.6 → 2.2.8

package/dist/mcp.js

@@ -10,13 +10,13 @@ import {
   markdownToHwpx,
   markdownToXlsx,
   parse
-} from "./chunk-KHUTUB7G.js";
+} from "./chunk-FC6BQOWD.js";
 import {
   KordocError,
   VERSION,
   sanitizeError,
   toArrayBuffer
-} from "./chunk-BU42ZFTN.js";
+} from "./chunk-6KLTURMA.js";
 import "./chunk-MOL7MDBG.js";
 import "./chunk-ZWE3DS7E.js";
 

package/dist/{provider-EPHXUWRL.js → provider-I3XGSVL6.js}

@@ -117,32 +117,77 @@ function parseMarkdownTable(lines) {
 }
 
 // src/ocr/provider.ts
-async function ocrPages(doc, provider, pageFilter, effectivePageCount, warnings) {
+async function runWithConcurrency(tasks, limit) {
+  const results = new Array(tasks.length);
+  let nextIndex = 0;
+  async function worker() {
+    while (nextIndex < tasks.length) {
+      const idx = nextIndex++;
+      results[idx] = await tasks[idx]();
+    }
+  }
+  await Promise.all(Array.from({ length: Math.min(limit, tasks.length) }, () => worker()));
+  return results;
+}
+function ocrResultToBlocks(result, pageNum) {
+  const pageBlocks = [];
+  if (typeof result === "string") {
+    if (result.trim()) {
+      pageBlocks.push({ type: "paragraph", text: result.trim(), pageNumber: pageNum });
+    }
+  } else if (result && typeof result === "object" && "markdown" in result) {
+    const structured = result;
+    if (structured.markdown.trim()) {
+      const converted = markdownToBlocks(structured.markdown, pageNum);
+      for (const b of converted) pageBlocks.push(b);
+    }
+  }
+  return pageBlocks;
+}
+async function ocrPages(doc, provider, pageFilter, effectivePageCount, warnings, concurrency = 1) {
   const blocks = [];
+  if (concurrency <= 1) {
+    for (let i = 1; i <= effectivePageCount; i++) {
+      if (pageFilter && !pageFilter.has(i)) continue;
+      const page = await doc.getPage(i);
+      try {
+        const imageData = await renderPageToPng(page);
+        const result = await provider(imageData, i, "image/png");
+        for (const b of ocrResultToBlocks(result, i)) blocks.push(b);
+      } catch (err) {
+        warnings?.push({
+          page: i,
+          message: `\uD398\uC774\uC9C0 ${i} OCR \uC2E4\uD328: ${err instanceof Error ? err.message : "\uC54C \uC218 \uC5C6\uB294 \uC624\uB958"}`,
+          code: "OCR_PAGE_FAILED"
+        });
+      }
+    }
+    return blocks;
+  }
+  const pageNumbers = [];
   for (let i = 1; i <= effectivePageCount; i++) {
     if (pageFilter && !pageFilter.has(i)) continue;
-    const page = await doc.getPage(i);
+    pageNumbers.push(i);
+  }
+  const tasks = pageNumbers.map((pageNum) => async () => {
     try {
+      const page = await doc.getPage(pageNum);
       const imageData = await renderPageToPng(page);
-      const result = await provider(imageData, i, "image/png");
-      if (typeof result === "string") {
-        if (result.trim()) {
-          blocks.push({ type: "paragraph", text: result.trim(), pageNumber: i });
-        }
-      } else if (result && typeof result === "object" && "markdown" in result) {
-        const structured = result;
-        if (structured.markdown.trim()) {
-          const pageBlocks = markdownToBlocks(structured.markdown, i);
-          for (const b of pageBlocks) blocks.push(b);
-        }
-      }
+      const result = await provider(imageData, pageNum, "image/png");
+      return { pageNum, pageBlocks: ocrResultToBlocks(result, pageNum) };
     } catch (err) {
       warnings?.push({
-        page: i,
-        message: `\uD398\uC774\uC9C0 ${i} OCR \uC2E4\uD328: ${err instanceof Error ? err.message : "\uC54C \uC218 \uC5C6\uB294 \uC624\uB958"}`,
+        page: pageNum,
+        message: `\uD398\uC774\uC9C0 ${pageNum} OCR \uC2E4\uD328: ${err instanceof Error ? err.message : "\uC54C \uC218 \uC5C6\uB294 \uC624\uB958"}`,
         code: "OCR_PAGE_FAILED"
       });
+      return null;
     }
+  });
+  const taskResults = await runWithConcurrency(tasks, concurrency);
+  for (const item of taskResults) {
+    if (!item) continue;
+    for (const b of item.pageBlocks) blocks.push(b);
   }
   return blocks;
 }
@@ -158,4 +203,4 @@ async function renderPageToPng(page) {
 export {
   ocrPages
 };
-//# sourceMappingURL=provider-EPHXUWRL.js.map
+//# sourceMappingURL=provider-I3XGSVL6.js.map

package/dist/provider-I3XGSVL6.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 } 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/**\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,\n  pageFilter: Set<number> | null,\n  effectivePageCount: number,\n  warnings?: ParseWarning[],\n  concurrency: number = 1    // 기본값 1 = 순차 처리 (하위 호환)\n): Promise<IRBlock[]> {\n  const blocks: IRBlock[] = []\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\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;AAcA,eAAsB,SACpB,KACA,UACA,YACA,oBACA,UACA,cAAsB,GACF;AACpB,QAAM,SAAoB,CAAC;AAG3B,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;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-77C5OWLO.js → resolve-UFUJEPCJ.js}

@@ -88,8 +88,10 @@ function callCli(mode, imagePath) {
   const args = buildCliArgs(mode, imagePath);
   const result = spawnSync(mode, args, {
     encoding: "utf-8",
-    timeout: 6e4,
-    maxBuffer: 10 * 1024 * 1024
+    timeout: 18e4,
+    maxBuffer: 10 * 1024 * 1024,
+    // claude: /tmp에서 실행하여 프로젝트 CLAUDE.md의 규칙 간섭 방지
+    ...mode === "claude" ? { cwd: tmpdir() } : {}
   });
   if (result.error) {
     throw new Error(`${mode} CLI \uC2E4\uD589 \uC2E4\uD328: ${result.error.message}`);
@@ -185,14 +187,17 @@ function stripCodeFence(text) {
 }
 
 // src/ocr/resolve.ts
-async function resolveOcrProvider(mode, warnings) {
+async function resolveOcrProvider(mode, warnings, concurrency) {
   if (mode === "off") {
     throw new Error("OCR\uC774 \uBE44\uD65C\uC131\uD654\uB418\uC5B4 \uC788\uC2B5\uB2C8\uB2E4 (--ocr off).");
   }
   if (mode !== "auto") {
     validateOcrMode(mode);
     if (mode === "tesseract") {
-      const { createTesseractProvider } = await import("./tesseract-provider-UNJOI25M.js");
+      const { createTesseractProvider, createTesseractPoolProvider } = await import("./tesseract-provider-WCVJWBUT.js");
+      if (concurrency && concurrency > 1) {
+        return createTesseractPoolProvider(concurrency);
+      }
       return createTesseractProvider();
     }
     return createCliOcrProvider(mode);
@@ -212,7 +217,10 @@ async function resolveOcrProvider(mode, warnings) {
     }
   }
   if (detected === "tesseract") {
-    const { createTesseractProvider } = await import("./tesseract-provider-UNJOI25M.js");
+    const { createTesseractProvider, createTesseractPoolProvider } = await import("./tesseract-provider-WCVJWBUT.js");
+    if (concurrency && concurrency > 1) {
+      return createTesseractPoolProvider(concurrency);
+    }
     return createTesseractProvider();
   }
   return createCliOcrProvider(detected);
@@ -220,4 +228,4 @@ async function resolveOcrProvider(mode, warnings) {
 export {
   resolveOcrProvider
 };
-//# sourceMappingURL=resolve-77C5OWLO.js.map
+//# sourceMappingURL=resolve-UFUJEPCJ.js.map

package/dist/resolve-UFUJEPCJ.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/ocr/auto-detect.ts","../src/ocr/cli-provider.ts","../src/ocr/resolve.ts"],"sourcesContent":["/**\n * OCR CLI 자동 탐색\n *\n * 탐색 순서: gemini → codex → claude → ollama → tesseract.js\n * CLI는 which(unix) / where(win) 명령어로 PATH 존재 확인.\n * tesseract.js는 bundled 의존성이므로 항상 사용 가능 (최후 fallback).\n */\n\nimport { execSync } from \"child_process\"\nimport type { OcrMode } from \"../types.js\"\n\n/** CLI 탐색 우선순위 */\nconst CLI_PRIORITY = [\"gemini\", \"codex\", \"claude\", \"ollama\"] as const\n\n/**\n * 시스템에 설치된 OCR 도구를 우선순위대로 탐색.\n * tesseract.js는 bundled 의존성이므로 CLI를 찾지 못해도 항상 \"tesseract\" 반환.\n * @returns 사용 가능한 OcrMode (null 반환 없음)\n */\nexport function detectAvailableOcr(): OcrMode {\n  // 1. CLI 프로그램 탐색 (gemini → codex → claude → ollama)\n  for (const cli of CLI_PRIORITY) {\n    if (isCliInstalled(cli)) return cli\n  }\n\n  // 2. tesseract.js — bundled 의존성, 항상 사용 가능\n  return \"tesseract\"\n}\n\n/**\n * 특정 CLI가 시스템 PATH에 있는지 확인.\n * which(unix) 또는 where(win32) 사용.\n */\nfunction isCliInstalled(name: string): boolean {\n  try {\n    const cmd = process.platform === \"win32\" ? \"where\" : \"which\"\n    execSync(`${cmd} ${name}`, { stdio: \"ignore\", timeout: 3000 })\n    return true\n  } catch {\n    return false\n  }\n}\n\n/**\n * 수동 지정된 OcrMode 유효성 검증.\n * --ocr gemini 등 강제 지정 시 호출.\n * @throws 해당 CLI가 설치되지 않은 경우 Error (tesseract는 항상 통과)\n */\nexport function validateOcrMode(mode: OcrMode): void {\n  if (mode === \"auto\" || mode === \"off\" || mode === \"tesseract\") return\n\n  if (!isCliInstalled(mode)) {\n    throw new Error(`'${mode}' CLI가 설치되지 않았습니다.\\n${getInstallGuide(mode)}`)\n  }\n}\n\n/** CLI별 설치 안내 메시지 */\nfunction getInstallGuide(mode: string): string {\n  const guides: Record<string, string> = {\n    gemini: \"설치: https://ai.google.dev/gemini-api/docs/cli\",\n    claude: \"설치: npm install -g @anthropic-ai/claude-code 또는 https://claude.ai/code\",\n    codex:  \"설치: npm install -g @openai/codex 또는 https://github.com/openai/codex\",\n    ollama: \"설치: brew install ollama 또는 https://ollama.com/download\",\n  }\n  return guides[mode] || `'${mode}'을(를) 설치해주세요.`\n}\n\n/**\n * AI CLI가 없어 tesseract.js로 fallback할 때 표시할 안내 메시지.\n */\nexport function getTesseractFallbackMessage(): string {\n  return [\n    \"설치된 AI CLI가 없어 내장 tesseract.js로 OCR을 수행합니다.\",\n    \"더 나은 품질(테이블/헤딩 구조 보존)을 위해 AI CLI 설치를 권장합니다:\",\n    \"\",\n    \"  [권장] Gemini CLI:  https://ai.google.dev/gemini-api/docs/cli\",\n    \"         Codex CLI:   npm install -g @openai/codex\",\n    \"         Claude CLI:  npm install -g @anthropic-ai/claude-code\",\n    \"         Ollama:      brew install ollama (+ ollama pull gemma4:27b)\",\n  ].join(\"\\n\")\n}\n","/**\n * CLI 기반 OCR 프로바이더\n *\n * gemini / claude / codex / ollama CLI를 subprocess로 호출하여\n * PDF 페이지 이미지를 Markdown으로 변환.\n *\n * 이미지 전달 방식:\n * - gemini:  -p \"프롬프트 @이미지경로\"  (@ 파일 참조)\n * - claude:  -p \"프롬프트 @이미지경로\"  (@ 파일 참조, --print 모드)\n * - codex:   exec -i 이미지경로 \"프롬프트\"  (-i/--image 플래그)\n * - ollama:  REST API (localhost:11434) — CLI는 이미지 입력 미지원\n */\n\nimport { spawnSync } from \"child_process\"\nimport { writeFileSync, readFileSync, unlinkSync, mkdirSync } from \"fs\"\nimport { join } from \"path\"\nimport { tmpdir } from \"os\"\nimport type { OcrMode, StructuredOcrResult } from \"../types.js\"\n\n/** OCR 프롬프트 — 모든 CLI 공통 */\nconst OCR_PROMPT =\n  \"이 PDF 페이지 이미지에서 텍스트와 테이블을 추출하여 순수 Markdown으로 변환하세요.\\n\" +\n  \"규칙:\\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/** 임시 디렉토리 (프로세스당 1회 생성)\n *\n * gemini CLI는 /tmp/ 등 시스템 임시 디렉토리를 워크스페이스 외부로 간주하여\n * @파일참조 시 접근을 거부할 수 있음. cwd 하위 폴더를 사용하면 모든 CLI에서 접근 가능.\n */\nlet _tempDir: string | null = null\nfunction getTempDir(): string {\n  if (!_tempDir) {\n    _tempDir = join(process.cwd(), \".kordoc-tmp\")\n    mkdirSync(_tempDir, { recursive: true })\n  }\n  return _tempDir\n}\n\n/**\n * CLI OcrProvider 생성.\n *\n * @param mode - 사용할 CLI (gemini, claude, codex, ollama)\n * @returns OcrProvider 함수 (StructuredOcrResult 반환)\n */\nexport function createCliOcrProvider(\n  mode: Exclude<OcrMode, \"auto\" | \"off\" | \"tesseract\">\n): (pageImage: Uint8Array, pageNumber: number, mimeType: \"image/png\") => Promise<StructuredOcrResult> {\n  return async (pageImage: Uint8Array, pageNumber: number): Promise<StructuredOcrResult> => {\n    const tempPath = join(getTempDir(), `page-${pageNumber}.png`)\n\n    try {\n      writeFileSync(tempPath, pageImage)\n\n      let output: string\n      if (mode === \"ollama\") {\n        output = await callOllamaApi(tempPath)\n      } else {\n        output = callCli(mode, tempPath)\n      }\n\n      return { markdown: stripCodeFence(output.trim()) }\n    } finally {\n      try { unlinkSync(tempPath) } catch { /* 임시 파일 정리 실패 무시 */ }\n    }\n  }\n}\n\n/**\n * CLI 실행 — gemini / claude / codex\n *\n * @throws CLI 실행 실패 또는 타임아웃(180초) 시 Error\n */\nfunction callCli(mode: string, imagePath: string): string {\n  // codex는 --output-last-message로 대화 헤더 없는 깔끔한 출력 사용\n  if (mode === \"codex\") {\n    return callCodexCli(imagePath)\n  }\n\n  const args = buildCliArgs(mode, imagePath)\n\n  const result = spawnSync(mode, args, {\n    encoding: \"utf-8\",\n    timeout: 180_000,\n    maxBuffer: 10 * 1024 * 1024,\n    // claude: /tmp에서 실행하여 프로젝트 CLAUDE.md의 규칙 간섭 방지\n    ...(mode === \"claude\" ? { cwd: tmpdir() } : {}),\n  })\n\n  if (result.error) {\n    throw new Error(`${mode} CLI 실행 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} OCR 실패: ${errMsg}`)\n  }\n\n  return result.stdout || \"\"\n}\n\n/**\n * codex exec 실행 — --output-last-message로 대화 헤더 없는 깔끔한 출력.\n * 인자 순서: `codex exec <prompt> --image <file> --output-last-message <outfile>`\n */\nfunction callCodexCli(imagePath: string): string {\n  // 출력 파일은 /tmp/ 사용 — codex sandbox는 cwd 내 쓰기를 막을 수 있음\n  const outPath = join(tmpdir(), `kordoc-codex-out-${Date.now()}.txt`)\n  try {\n    const args = [\"exec\", OCR_PROMPT, \"--image\", imagePath, \"--output-last-message\", outPath]\n    const model = process.env.KORDOC_CODEX_MODEL\n    if (model) args.push(\"--model\", model)\n\n    const result = spawnSync(\"codex\", args, {\n      encoding: \"utf-8\",\n      timeout: 180_000,\n      maxBuffer: 10 * 1024 * 1024,\n      input: \"\",  // stdin EOF 즉시 전달 (대화형 입력 차단)\n    })\n\n    if (result.error) {\n      throw new Error(`codex CLI 실행 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 OCR 실패: ${errMsg}`)\n    }\n\n    // --output-last-message 파일에서 읽기 (없으면 stdout 폴백)\n    try {\n      return readFileSync(outPath, \"utf-8\")\n    } catch {\n      return result.stdout || \"\"\n    }\n  } finally {\n    try { unlinkSync(outPath) } catch { /* 무시 */ }\n  }\n}\n\n/**\n * CLI별 인자 배열 생성.\n *\n * gemini:  [\"--prompt\", \"프롬프트 @이미지경로\", \"--yolo\"]\n *   - -y/--yolo: 자동 승인 (OCR은 도구 사용 없으므로 실질적 영향 없음)\n *   - @ 파일 참조로 이미지를 컨텍스트에 포함\n *\n * claude:  [\"--print\", \"프롬프트 @이미지경로\"]\n *   - --print(-p): 비대화형 출력 모드\n *   - @ 파일 참조로 이미지를 컨텍스트에 포함\n *\n * codex: callCodexCli()에서 별도 처리\n *   - `codex exec <prompt> --image <file> --output-last-message <outfile>`\n *   - 프롬프트가 --image보다 앞에 위치해야 함 (인자 순서 중요)\n *\n * ⚠️ CLI 버전에 따라 문법이 다를 수 있음. 업데이트 시 --help 재확인 필요.\n */\nfunction buildCliArgs(mode: string, imagePath: string): string[] {\n  const promptWithImage = `${OCR_PROMPT}\\n\\n이미지: @${imagePath}`\n\n  switch (mode) {\n    case \"gemini\": {\n      const args = [\"--prompt\", promptWithImage, \"--yolo\"]\n      const model = process.env.KORDOC_GEMINI_MODEL\n      if (model) args.push(\"--model\", model)\n      return args\n    }\n\n    case \"claude\": {\n      const args = [\"--print\", promptWithImage]\n      const model = process.env.KORDOC_CLAUDE_MODEL\n      if (model) args.push(\"--model\", model)\n      return args\n    }\n\n    default:\n      throw new Error(`지원하지 않는 CLI: ${mode}`)\n  }\n}\n\n/**\n * Ollama REST API 호출 — CLI는 이미지 입력을 지원하지 않으므로 API 직접 사용.\n *\n * 기본 모델: KORDOC_OLLAMA_MODEL 환경변수 또는 \"gemma4:27b\"\n * 기본 호스트: KORDOC_OLLAMA_HOST 환경변수 또는 \"http://localhost:11434\"\n *\n * @throws Ollama 서버 미실행 또는 응답 오류 시 Error\n */\nasync function callOllamaApi(imagePath: string): Promise<string> {\n  const { readFileSync } = await import(\"fs\")\n  const imageBase64 = readFileSync(imagePath).toString(\"base64\")\n\n  const model = process.env.KORDOC_OLLAMA_MODEL || \"qwen3-vl:8b\"\n  const host = process.env.KORDOC_OLLAMA_HOST || \"http://localhost:11434\"\n  const timeoutMs = Number(process.env.KORDOC_OLLAMA_TIMEOUT) || 120_000\n\n  const response = await fetch(`${host}/api/chat`, {\n    method: \"POST\",\n    headers: { \"Content-Type\": \"application/json\" },\n    body: JSON.stringify({\n      model,\n      messages: [{\n        role: \"user\",\n        content: OCR_PROMPT,\n        images: [imageBase64],\n      }],\n      stream: false,\n    }),\n    signal: AbortSignal.timeout(timeoutMs),\n  })\n\n  if (!response.ok) {\n    throw new Error(`Ollama API 오류: ${response.status} ${response.statusText}`)\n  }\n\n  const data = await response.json() as { message?: { content?: string } }\n  return data.message?.content || \"\"\n}\n\n/**\n * LLM 출력에서 코드 펜스 제거.\n * LLM이 가끔 결과를 ```markdown ... ``` 으로 감싸는 경우 처리.\n */\nfunction stripCodeFence(text: string): string {\n  const match = text.match(/^```(?:markdown|md)?\\s*\\n([\\s\\S]*?)\\n```\\s*$/m)\n  return match ? match[1].trim() : text\n}\n","/**\n * OCR 프로바이더 팩토리\n *\n * ocrMode에 따라 적절한 OcrProvider를 생성하여 반환.\n * - \"auto\": 설치된 CLI 자동 탐색 (gemini → claude → codex → ollama → tesseract)\n *           tesseract.js는 bundled 의존성이므로 항상 사용 가능 (null 반환 없음)\n * - 특정 CLI: 해당 CLI 사용 (미설치 시 에러)\n * - \"tesseract\": 내장 tesseract.js 직접 사용\n * - \"off\": 에러 throw\n */\n\nimport type { OcrMode, OcrProvider, ParseWarning } from \"../types.js\"\nimport { detectAvailableOcr, validateOcrMode, getTesseractFallbackMessage } from \"./auto-detect.js\"\nimport { createCliOcrProvider } from \"./cli-provider.js\"\n\n/**\n * ocrMode에 따라 OcrProvider를 생성.\n *\n * @param mode - OCR 모드\n * @param warnings - 경고 수집 배열 (fallback 발생 시 경고 추가)\n * @param concurrency - 병렬 처리 수 (tesseract 전용, 기본: 1=순차)\n * @returns OcrProvider 함수\n * @throws mode=\"off\"이거나 지정 CLI 미설치 시 Error\n */\nexport async function resolveOcrProvider(\n  mode: OcrMode,\n  warnings?: ParseWarning[],\n  concurrency?: number\n): Promise<OcrProvider> {\n  if (mode === \"off\") {\n    throw new Error(\"OCR이 비활성화되어 있습니다 (--ocr off).\")\n  }\n\n  // ── 수동 지정 모드 ──────────────────────────────────\n  if (mode !== \"auto\") {\n    validateOcrMode(mode)  // tesseract는 항상 통과\n\n    if (mode === \"tesseract\") {\n      const { createTesseractProvider, createTesseractPoolProvider } = await import(\"./tesseract-provider.js\")\n      // concurrency > 1이면 워커 풀 사용, 그 외 단일 워커 사용\n      if (concurrency && concurrency > 1) {\n        return createTesseractPoolProvider(concurrency)\n      }\n      return createTesseractProvider()\n    }\n\n    // CLI 프로바이더는 rate limit 보호를 위해 concurrency 무시 (항상 순차)\n    return createCliOcrProvider(mode)\n  }\n\n  // ── 자동 탐색 모드 ───────────────────────────────────\n  // detectAvailableOcr()는 항상 값을 반환 (tesseract fallback으로 null 없음)\n  const detected = detectAvailableOcr()\n\n  // gemini가 아닌 경우 fallback 경고\n  if (detected !== \"gemini\") {\n    if (detected === \"tesseract\") {\n      // 내장 tesseract로 fallback — 구조 복원 제한 안내\n      warnings?.push({\n        message: getTesseractFallbackMessage(),\n        code: \"OCR_CLI_FALLBACK\",\n      })\n    } else {\n      warnings?.push({\n        message: `OCR: '${detected}' 사용 중 (gemini CLI가 없어 fallback). 더 나은 품질을 위해 gemini CLI 설치를 권장합니다.`,\n        code: \"OCR_CLI_FALLBACK\",\n      })\n    }\n  }\n\n  if (detected === \"tesseract\") {\n    const { createTesseractProvider, createTesseractPoolProvider } = await import(\"./tesseract-provider.js\")\n    // concurrency > 1이면 워커 풀 사용, 그 외 단일 워커 사용\n    if (concurrency && concurrency > 1) {\n      return createTesseractPoolProvider(concurrency)\n    }\n    return createTesseractProvider()\n  }\n\n  // CLI 프로바이더는 rate limit 보호를 위해 concurrency 무시 (항상 순차)\n  return createCliOcrProvider(detected)\n}\n"],"mappings":";;;;AAQA,SAAS,gBAAgB;AAIzB,IAAM,eAAe,CAAC,UAAU,SAAS,UAAU,QAAQ;AAOpD,SAAS,qBAA8B;AAE5C,aAAW,OAAO,cAAc;AAC9B,QAAI,eAAe,GAAG,EAAG,QAAO;AAAA,EAClC;AAGA,SAAO;AACT;AAMA,SAAS,eAAe,MAAuB;AAC7C,MAAI;AACF,UAAM,MAAM,QAAQ,aAAa,UAAU,UAAU;AACrD,aAAS,GAAG,GAAG,IAAI,IAAI,IAAI,EAAE,OAAO,UAAU,SAAS,IAAK,CAAC;AAC7D,WAAO;AAAA,EACT,QAAQ;AACN,WAAO;AAAA,EACT;AACF;AAOO,SAAS,gBAAgB,MAAqB;AACnD,MAAI,SAAS,UAAU,SAAS,SAAS,SAAS,YAAa;AAE/D,MAAI,CAAC,eAAe,IAAI,GAAG;AACzB,UAAM,IAAI,MAAM,IAAI,IAAI;AAAA,EAAuB,gBAAgB,IAAI,CAAC,EAAE;AAAA,EACxE;AACF;AAGA,SAAS,gBAAgB,MAAsB;AAC7C,QAAM,SAAiC;AAAA,IACrC,QAAQ;AAAA,IACR,QAAQ;AAAA,IACR,OAAQ;AAAA,IACR,QAAQ;AAAA,EACV;AACA,SAAO,OAAO,IAAI,KAAK,IAAI,IAAI;AACjC;AAKO,SAAS,8BAAsC;AACpD,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,EACF,EAAE,KAAK,IAAI;AACb;;;ACnEA,SAAS,iBAAiB;AAC1B,SAAS,eAAe,cAAc,YAAY,iBAAiB;AACnE,SAAS,YAAY;AACrB,SAAS,cAAc;AAIvB,IAAM,aACJ;AAeF,IAAI,WAA0B;AAC9B,SAAS,aAAqB;AAC5B,MAAI,CAAC,UAAU;AACb,eAAW,KAAK,QAAQ,IAAI,GAAG,aAAa;AAC5C,cAAU,UAAU,EAAE,WAAW,KAAK,CAAC;AAAA,EACzC;AACA,SAAO;AACT;AAQO,SAAS,qBACd,MACoG;AACpG,SAAO,OAAO,WAAuB,eAAqD;AACxF,UAAM,WAAW,KAAK,WAAW,GAAG,QAAQ,UAAU,MAAM;AAE5D,QAAI;AACF,oBAAc,UAAU,SAAS;AAEjC,UAAI;AACJ,UAAI,SAAS,UAAU;AACrB,iBAAS,MAAM,cAAc,QAAQ;AAAA,MACvC,OAAO;AACL,iBAAS,QAAQ,MAAM,QAAQ;AAAA,MACjC;AAEA,aAAO,EAAE,UAAU,eAAe,OAAO,KAAK,CAAC,EAAE;AAAA,IACnD,UAAE;AACA,UAAI;AAAE,mBAAW,QAAQ;AAAA,MAAE,QAAQ;AAAA,MAAuB;AAAA,IAC5D;AAAA,EACF;AACF;AAOA,SAAS,QAAQ,MAAc,WAA2B;AAExD,MAAI,SAAS,SAAS;AACpB,WAAO,aAAa,SAAS;AAAA,EAC/B;AAEA,QAAM,OAAO,aAAa,MAAM,SAAS;AAEzC,QAAM,SAAS,UAAU,MAAM,MAAM;AAAA,IACnC,UAAU;AAAA,IACV,SAAS;AAAA,IACT,WAAW,KAAK,OAAO;AAAA;AAAA,IAEvB,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,sBAAY,MAAM,EAAE;AAAA,EAC7C;AAEA,SAAO,OAAO,UAAU;AAC1B;AAMA,SAAS,aAAa,WAA2B;AAE/C,QAAM,UAAU,KAAK,OAAO,GAAG,oBAAoB,KAAK,IAAI,CAAC,MAAM;AACnE,MAAI;AACF,UAAM,OAAO,CAAC,QAAQ,YAAY,WAAW,WAAW,yBAAyB,OAAO;AACxF,UAAM,QAAQ,QAAQ,IAAI;AAC1B,QAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AAErC,UAAM,SAAS,UAAU,SAAS,MAAM;AAAA,MACtC,UAAU;AAAA,MACV,SAAS;AAAA,MACT,WAAW,KAAK,OAAO;AAAA,MACvB,OAAO;AAAA;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,2BAAiB,MAAM,EAAE;AAAA,IAC3C;AAGA,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,IAAW;AAAA,EAC/C;AACF;AAmBA,SAAS,aAAa,MAAc,WAA6B;AAC/D,QAAM,kBAAkB,GAAG,UAAU;AAAA;AAAA,uBAAa,SAAS;AAE3D,UAAQ,MAAM;AAAA,IACZ,KAAK,UAAU;AACb,YAAM,OAAO,CAAC,YAAY,iBAAiB,QAAQ;AACnD,YAAM,QAAQ,QAAQ,IAAI;AAC1B,UAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AACrC,aAAO;AAAA,IACT;AAAA,IAEA,KAAK,UAAU;AACb,YAAM,OAAO,CAAC,WAAW,eAAe;AACxC,YAAM,QAAQ,QAAQ,IAAI;AAC1B,UAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AACrC,aAAO;AAAA,IACT;AAAA,IAEA;AACE,YAAM,IAAI,MAAM,8CAAgB,IAAI,EAAE;AAAA,EAC1C;AACF;AAUA,eAAe,cAAc,WAAoC;AAC/D,QAAM,EAAE,cAAAA,cAAa,IAAI,MAAM,OAAO,IAAI;AAC1C,QAAM,cAAcA,cAAa,SAAS,EAAE,SAAS,QAAQ;AAE7D,QAAM,QAAQ,QAAQ,IAAI,uBAAuB;AACjD,QAAM,OAAO,QAAQ,IAAI,sBAAsB;AAC/C,QAAM,YAAY,OAAO,QAAQ,IAAI,qBAAqB,KAAK;AAE/D,QAAM,WAAW,MAAM,MAAM,GAAG,IAAI,aAAa;AAAA,IAC/C,QAAQ;AAAA,IACR,SAAS,EAAE,gBAAgB,mBAAmB;AAAA,IAC9C,MAAM,KAAK,UAAU;AAAA,MACnB;AAAA,MACA,UAAU,CAAC;AAAA,QACT,MAAM;AAAA,QACN,SAAS;AAAA,QACT,QAAQ,CAAC,WAAW;AAAA,MACtB,CAAC;AAAA,MACD,QAAQ;AAAA,IACV,CAAC;AAAA,IACD,QAAQ,YAAY,QAAQ,SAAS;AAAA,EACvC,CAAC;AAED,MAAI,CAAC,SAAS,IAAI;AAChB,UAAM,IAAI,MAAM,4BAAkB,SAAS,MAAM,IAAI,SAAS,UAAU,EAAE;AAAA,EAC5E;AAEA,QAAM,OAAO,MAAM,SAAS,KAAK;AACjC,SAAO,KAAK,SAAS,WAAW;AAClC;AAMA,SAAS,eAAe,MAAsB;AAC5C,QAAM,QAAQ,KAAK,MAAM,+CAA+C;AACxE,SAAO,QAAQ,MAAM,CAAC,EAAE,KAAK,IAAI;AACnC;;;AC9MA,eAAsB,mBACpB,MACA,UACA,aACsB;AACtB,MAAI,SAAS,OAAO;AAClB,UAAM,IAAI,MAAM,sFAA+B;AAAA,EACjD;AAGA,MAAI,SAAS,QAAQ;AACnB,oBAAgB,IAAI;AAEpB,QAAI,SAAS,aAAa;AACxB,YAAM,EAAE,yBAAyB,4BAA4B,IAAI,MAAM,OAAO,kCAAyB;AAEvG,UAAI,eAAe,cAAc,GAAG;AAClC,eAAO,4BAA4B,WAAW;AAAA,MAChD;AACA,aAAO,wBAAwB;AAAA,IACjC;AAGA,WAAO,qBAAqB,IAAI;AAAA,EAClC;AAIA,QAAM,WAAW,mBAAmB;AAGpC,MAAI,aAAa,UAAU;AACzB,QAAI,aAAa,aAAa;AAE5B,gBAAU,KAAK;AAAA,QACb,SAAS,4BAA4B;AAAA,QACrC,MAAM;AAAA,MACR,CAAC;AAAA,IACH,OAAO;AACL,gBAAU,KAAK;AAAA,QACb,SAAS,SAAS,QAAQ;AAAA,QAC1B,MAAM;AAAA,MACR,CAAC;AAAA,IACH;AAAA,EACF;AAEA,MAAI,aAAa,aAAa;AAC5B,UAAM,EAAE,yBAAyB,4BAA4B,IAAI,MAAM,OAAO,kCAAyB;AAEvG,QAAI,eAAe,cAAc,GAAG;AAClC,aAAO,4BAA4B,WAAW;AAAA,IAChD;AACA,WAAO,wBAAwB;AAAA,EACjC;AAGA,SAAO,qBAAqB,QAAQ;AACtC;","names":["readFileSync"]}

package/dist/tesseract-provider-WCVJWBUT.js

@@ -0,0 +1,56 @@
+#!/usr/bin/env node
+import "./chunk-ZWE3DS7E.js";
+
+// src/ocr/tesseract-provider.ts
+import { createWorker } from "tesseract.js";
+async function createTesseractProvider() {
+  const worker = await createWorker("kor+eng");
+  let terminated = false;
+  const provider = async (pageImage, _pageNumber, _mimeType) => {
+    const { data } = await worker.recognize(pageImage);
+    return data.text;
+  };
+  provider.terminate = async () => {
+    if (!terminated) {
+      await worker.terminate();
+      terminated = true;
+    }
+  };
+  return provider;
+}
+async function createTesseractPoolProvider(concurrency) {
+  const workers = await Promise.all(
+    Array.from({ length: concurrency }, () => createWorker("kor+eng"))
+  );
+  const idle = [...workers];
+  const waitQueue = [];
+  function acquire() {
+    if (idle.length > 0) return Promise.resolve(idle.pop());
+    return new Promise((resolve) => waitQueue.push(resolve));
+  }
+  function release(w) {
+    if (waitQueue.length > 0) {
+      waitQueue.shift()(w);
+    } else {
+      idle.push(w);
+    }
+  }
+  const provider = async (pageImage, _pageNumber, _mimeType) => {
+    const w = await acquire();
+    try {
+      const { data } = await w.recognize(pageImage);
+      return data.text;
+    } finally {
+      release(w);
+    }
+  };
+  provider.terminate = async () => {
+    await Promise.all(workers.map((w) => w.terminate()));
+  };
+  return provider;
+}
+export {
+  createTesseractPoolProvider,
+  createTesseractProvider
+};
+//# sourceMappingURL=tesseract-provider-WCVJWBUT.js.map

package/dist/tesseract-provider-WCVJWBUT.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/ocr/tesseract-provider.ts"],"sourcesContent":["/**\n * Tesseract.js 기반 OCR 프로바이더\n *\n * tesseract.js는 bundled 의존성으로 별도 설치 불필요.\n * Vision LLM CLI가 없을 때의 최후 fallback으로 자동 사용.\n *\n * 특성:\n * - 순수 텍스트만 반환 (테이블/헤딩 구조 복원 불가)\n * - 한글 인식률 약 85-90% (깨끗한 이미지 기준)\n * - 완전 오프라인 동작 (API 키 불필요)\n */\n\nimport { createWorker } from \"tesseract.js\"\nimport type { OcrProvider } from \"../types.js\"\n\n/**\n * Tesseract.js OcrProvider 생성.\n *\n * 워커를 1회 생성하여 재사용 (매 페이지마다 초기화 방지).\n * 문서 처리 완료 후 terminate()로 워커 정리.\n *\n * @returns OcrProvider 함수 (+ terminate 메서드)\n */\nexport async function createTesseractProvider(): Promise<OcrProvider & { terminate: () => Promise<void> }> {\n  // kor+eng: 한글 + 영문 동시 인식 (한국 공문서 특성)\n  const worker = await createWorker(\"kor+eng\")\n  let terminated = false\n\n  const provider = async (\n    pageImage: Uint8Array,\n    _pageNumber: number,\n    _mimeType: \"image/png\"\n  ): Promise<string> => {\n    const { data } = await worker.recognize(pageImage)\n    return data.text\n  }\n\n  ;(provider as OcrProvider & { terminate: () => Promise<void> }).terminate = async () => {\n    if (!terminated) {\n      await worker.terminate()\n      terminated = true\n    }\n  }\n\n  return provider as OcrProvider & { terminate: () => Promise<void> }\n}\n\n/**\n * Tesseract.js 워커 풀 기반 병렬 OcrProvider 생성.\n *\n * N개 워커를 동시 생성하여 여러 페이지를 병렬로 OCR 처리.\n * acquire/release 패턴으로 동시 실행 안전성 보장.\n * 대량 이미지 기반 PDF(예: 300페이지) 처리 속도 향상에 효과적.\n *\n * @param concurrency - 동시 워커 수 (권장: CPU 코어 수)\n * @returns OcrProvider 함수 (+ terminate 메서드)\n */\nexport async function createTesseractPoolProvider(\n  concurrency: number\n): Promise<OcrProvider & { terminate: () => Promise<void> }> {\n  // N개 워커를 동시 초기화 (Promise.all로 병렬 생성)\n  const workers = await Promise.all(\n    Array.from({ length: concurrency }, () => createWorker(\"kor+eng\"))\n  )\n\n  // 유휴 워커 큐 (초기에는 모든 워커가 유휴 상태)\n  const idle: typeof workers = [...workers]\n  // 워커를 기다리는 콜백 큐\n  const waitQueue: ((w: (typeof workers)[0]) => void)[] = []\n\n  /** 유휴 워커를 획득 (없으면 대기) */\n  function acquire(): Promise<(typeof workers)[0]> {\n    if (idle.length > 0) return Promise.resolve(idle.pop()!)\n    return new Promise(resolve => waitQueue.push(resolve))\n  }\n\n  /** 사용 완료된 워커를 반환 (대기 중인 호출자에게 우선 전달) */\n  function release(w: (typeof workers)[0]): void {\n    if (waitQueue.length > 0) {\n      // 대기 중인 요청자에게 즉시 전달\n      waitQueue.shift()!(w)\n    } else {\n      idle.push(w)\n    }\n  }\n\n  // 병렬 호출 안전 프로바이더 — 각 호출이 acquire/release로 워커를 독점 사용\n  const provider = async (\n    pageImage: Uint8Array,\n    _pageNumber: number,\n    _mimeType: \"image/png\"\n  ): Promise<string> => {\n    const w = await acquire()\n    try {\n      const { data } = await w.recognize(pageImage)\n      return data.text\n    } finally {\n      // 성공/실패 무관하게 워커 반환\n      release(w)\n    }\n  }\n\n  // 모든 워커 정리 메서드\n  ;(provider as OcrProvider & { terminate: () => Promise<void> }).terminate = async () => {\n    await Promise.all(workers.map(w => w.terminate()))\n  }\n\n  return provider as OcrProvider & { terminate: () => Promise<void> }\n}\n"],"mappings":";;;;AAYA,SAAS,oBAAoB;AAW7B,eAAsB,0BAAqF;AAEzG,QAAM,SAAS,MAAM,aAAa,SAAS;AAC3C,MAAI,aAAa;AAEjB,QAAM,WAAW,OACf,WACA,aACA,cACoB;AACpB,UAAM,EAAE,KAAK,IAAI,MAAM,OAAO,UAAU,SAAS;AACjD,WAAO,KAAK;AAAA,EACd;AAEC,EAAC,SAA8D,YAAY,YAAY;AACtF,QAAI,CAAC,YAAY;AACf,YAAM,OAAO,UAAU;AACvB,mBAAa;AAAA,IACf;AAAA,EACF;AAEA,SAAO;AACT;AAYA,eAAsB,4BACpB,aAC2D;AAE3D,QAAM,UAAU,MAAM,QAAQ;AAAA,IAC5B,MAAM,KAAK,EAAE,QAAQ,YAAY,GAAG,MAAM,aAAa,SAAS,CAAC;AAAA,EACnE;AAGA,QAAM,OAAuB,CAAC,GAAG,OAAO;AAExC,QAAM,YAAkD,CAAC;AAGzD,WAAS,UAAwC;AAC/C,QAAI,KAAK,SAAS,EAAG,QAAO,QAAQ,QAAQ,KAAK,IAAI,CAAE;AACvD,WAAO,IAAI,QAAQ,aAAW,UAAU,KAAK,OAAO,CAAC;AAAA,EACvD;AAGA,WAAS,QAAQ,GAA8B;AAC7C,QAAI,UAAU,SAAS,GAAG;AAExB,gBAAU,MAAM,EAAG,CAAC;AAAA,IACtB,OAAO;AACL,WAAK,KAAK,CAAC;AAAA,IACb;AAAA,EACF;AAGA,QAAM,WAAW,OACf,WACA,aACA,cACoB;AACpB,UAAM,IAAI,MAAM,QAAQ;AACxB,QAAI;AACF,YAAM,EAAE,KAAK,IAAI,MAAM,EAAE,UAAU,SAAS;AAC5C,aAAO,KAAK;AAAA,IACd,UAAE;AAEA,cAAQ,CAAC;AAAA,IACX;AAAA,EACF;AAGC,EAAC,SAA8D,YAAY,YAAY;AACtF,UAAM,QAAQ,IAAI,QAAQ,IAAI,OAAK,EAAE,UAAU,CAAC,CAAC;AAAA,EACnD;AAEA,SAAO;AACT;","names":[]}

package/dist/{utils-PGXOUDRW.js → utils-JRBHPKTC.js}

@@ -8,7 +8,7 @@ import {
   sanitizeError,
   sanitizeHref,
   toArrayBuffer
-} from "./chunk-BU42ZFTN.js";
+} from "./chunk-6KLTURMA.js";
 import "./chunk-ZWE3DS7E.js";
 export {
   KordocError,
@@ -20,4 +20,4 @@ export {
   sanitizeHref,
   toArrayBuffer
 };
-//# sourceMappingURL=utils-PGXOUDRW.js.map
+//# sourceMappingURL=utils-JRBHPKTC.js.map

package/dist/{watch-VDBYOMEJ.js → watch-JANDW746.js}

@@ -2,10 +2,10 @@
 import {
   detectFormat,
   parse
-} from "./chunk-KHUTUB7G.js";
+} from "./chunk-FC6BQOWD.js";
 import {
   toArrayBuffer
-} from "./chunk-BU42ZFTN.js";
+} from "./chunk-6KLTURMA.js";
 import "./chunk-MOL7MDBG.js";
 import "./chunk-ZWE3DS7E.js";
 
@@ -126,4 +126,4 @@ async function sendWebhook(url, payload) {
 export {
   watchDirectory
 };
-//# sourceMappingURL=watch-VDBYOMEJ.js.map
+//# sourceMappingURL=watch-JANDW746.js.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@clazic/kordoc",
-  "version": "2.2.6",
+  "version": "2.2.8",
   "description": "Parse Korean documents (HWP, HWPX, PDF, XLSX, DOCX) to Markdown",
   "type": "module",
   "exports": {