kordoc 2.5.2 → 2.7.1

package/README.md

@@ -7,7 +7,7 @@
 
 > *대한민국에서 둘째가라면 서러울 문서지옥. 거기서 7년 버틴 공무원이 만들었습니다.*
 
-HWP, HWPX, PDF, XLSX, DOCX — 관공서에서 쏟아지는 모든 문서를 파싱하고, 비교하고, 분석하고, 생성합니다.
+HWP 3.x/5.x, HWPX, PDF, XLS, XLSX, DOCX — 관공서에서 쏟아지는 모든 문서를 파싱하고, 비교하고, 분석하고, 생성합니다.
 
 [English](./README-EN.md)
 
@@ -37,13 +37,28 @@ Windows 도 자동으로 `cmd /c npx` 래핑. 수동 JSON 편집 불필요. 재
 > npx -y kordoc@latest setup
 > ```
 
+> **Windows PowerShell 에서 `npx.ps1 파일을 로드할 수 없습니다 · PSSecurityException` 이 뜨면**: PowerShell 기본 보안 정책이 서명 없는 `.ps1` 을 차단하는 표준 동작입니다 (kordoc 무관). 아래 중 하나 쓰시면 됩니다.
+>
+> **방법 1 — 명령 프롬프트(cmd) 창에서 실행** (가장 안전)
+> 윈도우 키 → `cmd` 검색 → Enter → 검은 창에서 그대로:
+> ```
+> npx -y kordoc setup
+> ```
+>
+> **방법 2 — PowerShell 실행 정책 한 번만 완화**
+> 관리자 권한 PowerShell:
+> ```powershell
+> Set-ExecutionPolicy -Scope CurrentUser RemoteSigned
+> ```
+> 이후 PowerShell 재시작 → `npx -y kordoc setup` 그대로 됨.
+
 ---
 
 ## 💡 kordoc으로 무엇을 할 수 있나요?
 
 단순한 텍스트 추출을 넘어, **공문서 처리를 위한 모든 과정**을 자동화합니다.
 
-*   **📄 어떤 문서든 마크다운으로**: `HWP`, `HWPX`, `HWPML`, `PDF`, `XLSX`, `DOCX` 파일을 즉시 `Markdown`으로 변환합니다. AI(LLM)가 문서를 읽고 분석하기 가장 좋은 상태로 만들어줍니다.
+*   **📄 어떤 문서든 마크다운으로**: `HWP3` (구버전), `HWP`(5.x), `HWPX`, `HWPML`, `PDF`, `XLS`, `XLSX`, `DOCX` 파일을 즉시 `Markdown`으로 변환합니다. AI(LLM)가 문서를 읽고 분석하기 가장 좋은 상태로 만들어줍니다.
 *   **📊 복잡한 표(Table) 완벽 재현**: 선이 없는 PDF나 복잡하게 병합된 HWP 표도 구조를 분석하여 정확한 마크다운 테이블로 복원합니다.
 *   **🔍 신구대조표 자동 생성**: 두 문서의 차이점을 분석하여 무엇이 바뀌었는지 한눈에 보여줍니다. (HWP와 HWPX 간의 비교도 가능!)
 *   **📝 마크다운을 다시 HWPX로**: AI가 작성한 내용을 다시 보고서 양식(`HWPX`)으로 되돌려줍니다. 이제 복사-붙여넣기 노가다에서 해방되세요.
@@ -52,6 +67,10 @@ Windows 도 자동으로 `cmd /c npx` 래핑. 수동 JSON 편집 불필요. 재
 
 ---
 
+## v2.7.1 변경사항
+
+- **🕰️ HWP 3.0 (구버전) 파서 추가** — 1996~2002년 한컴이 쓰던 단일 binary 포맷 (`"HWP Document File V3.00"` 시그니처) 텍스트 추출. 기존 kordoc 이 거부하던 구버전 판결문/공문서 등이 검색 인덱싱 가능. 상용조합형(johab) → 유니코드 + 5,893개 한자/기호 lookup. 표 cell / 머리말 / 각주 의 nested paragraph 재귀 추출. [@edwardkim/rhwp](https://github.com/edwardkim/rhwp) 의 Rust 구현을 TypeScript 로 포팅.
+
 ## v2.5.0 변경사항
 
 - **🏛️ macOS 한컴오피스 호환 HWPX 생성** (#4) — `markdownToHwpx()` 가 만든 HWPX 가 macOS 한컴에서 "파일이 깨졌다"며 거부되던 문제 해결. 테이블 XML 을 최소 스켈레톤에서 완전 스펙 형태로 재작성 — `<hp:tbl>` 필수 속성 10종 + `<hp:sz>`/`<hp:pos>`/`<hp:outMargin>`/`<hp:inMargin>`, `<hp:tc>` 안에 `<hp:subList>` 래퍼 + `<hp:cellAddr>`/`<hp:cellSpan>`/`<hp:cellSz>`/`<hp:cellMargin>`, paragraph 래핑. `Preview/PrvText.txt` 추가 + `borderFill` id=1(SOLID 0.12mm) 추가.

package/dist/chunk-5CJGKKMZ.js

@@ -0,0 +1,266 @@
+#!/usr/bin/env node
+
+// src/detect.ts
+import JSZip from "jszip";
+
+// src/hwp5/cfb-lenient.ts
+var CFB_MAGIC = Buffer.from([208, 207, 17, 224, 161, 177, 26, 225]);
+var END_OF_CHAIN = 4294967294;
+var FREE_SECT = 4294967295;
+var MAX_CHAIN_LENGTH = 1e6;
+var MAX_DIR_ENTRIES = 1e5;
+var MAX_STREAM_SIZE = 100 * 1024 * 1024;
+function parseLenientCfb(data) {
+  if (data.length < 512) throw new Error("CFB \uD30C\uC77C\uC774 \uB108\uBB34 \uC9E7\uC2B5\uB2C8\uB2E4 (\uCD5C\uC18C 512\uBC14\uC774\uD2B8)");
+  if (!data.subarray(0, 8).equals(CFB_MAGIC)) throw new Error("CFB \uB9E4\uC9C1 \uBC14\uC774\uD2B8 \uBD88\uC77C\uCE58");
+  const sectorSizeShift = data.readUInt16LE(30);
+  if (sectorSizeShift < 7 || sectorSizeShift > 16) throw new Error("\uC720\uD6A8\uD558\uC9C0 \uC54A\uC740 \uC139\uD130 \uD06C\uAE30 \uC2DC\uD504\uD2B8: " + sectorSizeShift);
+  const sectorSize = 1 << sectorSizeShift;
+  const miniSectorSizeShift = data.readUInt16LE(32);
+  if (miniSectorSizeShift > 16) throw new Error("\uC720\uD6A8\uD558\uC9C0 \uC54A\uC740 \uBBF8\uB2C8 \uC139\uD130 \uD06C\uAE30 \uC2DC\uD504\uD2B8: " + miniSectorSizeShift);
+  const miniSectorSize = 1 << miniSectorSizeShift;
+  const fatSectorCount = data.readUInt32LE(44);
+  if (fatSectorCount > 1e4) throw new Error("FAT \uC139\uD130 \uC218\uAC00 \uB108\uBB34 \uB9CE\uC2B5\uB2C8\uB2E4: " + fatSectorCount);
+  const firstDirSector = data.readUInt32LE(48);
+  const miniStreamCutoff = data.readUInt32LE(56);
+  const firstMiniFatSector = data.readUInt32LE(60);
+  const miniFatSectorCount = data.readUInt32LE(64);
+  const firstDifatSector = data.readUInt32LE(68);
+  const difatSectorCount = data.readUInt32LE(72);
+  function sectorOffset(id) {
+    return 512 + id * sectorSize;
+  }
+  function readSectorData(id) {
+    const off = sectorOffset(id);
+    if (off + sectorSize > data.length) return Buffer.alloc(0);
+    return data.subarray(off, off + sectorSize);
+  }
+  const fatSectors = [];
+  for (let i = 0; i < 109 && fatSectors.length < fatSectorCount; i++) {
+    const sid = data.readUInt32LE(76 + i * 4);
+    if (sid === FREE_SECT || sid === END_OF_CHAIN) break;
+    fatSectors.push(sid);
+  }
+  let difatSector = firstDifatSector;
+  const visitedDifat = /* @__PURE__ */ new Set();
+  for (let d = 0; d < difatSectorCount && difatSector !== END_OF_CHAIN && difatSector !== FREE_SECT; d++) {
+    if (visitedDifat.has(difatSector)) break;
+    visitedDifat.add(difatSector);
+    const buf = readSectorData(difatSector);
+    const entriesPerSector = sectorSize / 4 - 1;
+    for (let i = 0; i < entriesPerSector && fatSectors.length < fatSectorCount; i++) {
+      const sid = buf.readUInt32LE(i * 4);
+      if (sid === FREE_SECT || sid === END_OF_CHAIN) continue;
+      fatSectors.push(sid);
+    }
+    difatSector = buf.readUInt32LE(entriesPerSector * 4);
+  }
+  const entriesPerFatSector = sectorSize / 4;
+  const fatTable = new Uint32Array(fatSectors.length * entriesPerFatSector);
+  for (let fi = 0; fi < fatSectors.length; fi++) {
+    const buf = readSectorData(fatSectors[fi]);
+    for (let i = 0; i < entriesPerFatSector; i++) {
+      fatTable[fi * entriesPerFatSector + i] = i * 4 + 3 < buf.length ? buf.readUInt32LE(i * 4) : FREE_SECT;
+    }
+  }
+  function readChain(startSector, maxBytes) {
+    if (startSector === END_OF_CHAIN || startSector === FREE_SECT) return Buffer.alloc(0);
+    if (maxBytes > MAX_STREAM_SIZE) throw new Error("\uC2A4\uD2B8\uB9BC\uC774 \uB108\uBB34 \uD07D\uB2C8\uB2E4");
+    const chunks = [];
+    let current = startSector;
+    let totalRead = 0;
+    const visited = /* @__PURE__ */ new Set();
+    while (current !== END_OF_CHAIN && current !== FREE_SECT && totalRead < maxBytes) {
+      if (visited.has(current)) break;
+      if (visited.size > MAX_CHAIN_LENGTH) break;
+      visited.add(current);
+      const buf = readSectorData(current);
+      const remaining = maxBytes - totalRead;
+      chunks.push(remaining < sectorSize ? buf.subarray(0, remaining) : buf);
+      totalRead += Math.min(buf.length, remaining);
+      current = current < fatTable.length ? fatTable[current] : END_OF_CHAIN;
+    }
+    return Buffer.concat(chunks);
+  }
+  let miniFatTable = null;
+  function getMiniFatTable() {
+    if (miniFatTable) return miniFatTable;
+    if (miniFatSectorCount === 0 || firstMiniFatSector === END_OF_CHAIN) {
+      miniFatTable = new Uint32Array(0);
+      return miniFatTable;
+    }
+    const miniFatData = readChain(firstMiniFatSector, miniFatSectorCount * sectorSize);
+    const entries = miniFatData.length / 4;
+    miniFatTable = new Uint32Array(entries);
+    for (let i = 0; i < entries; i++) {
+      miniFatTable[i] = miniFatData.readUInt32LE(i * 4);
+    }
+    return miniFatTable;
+  }
+  const dirData = readChain(firstDirSector, MAX_DIR_ENTRIES * 128);
+  const dirEntries = [];
+  for (let offset = 0; offset + 128 <= dirData.length && dirEntries.length < MAX_DIR_ENTRIES; offset += 128) {
+    const nameLen = dirData.readUInt16LE(offset + 64);
+    if (nameLen <= 0 || nameLen > 64) {
+      dirEntries.push({ name: "", type: 0, startSector: 0, size: 0 });
+      continue;
+    }
+    const nameBytes = nameLen - 2;
+    const name = nameBytes > 0 ? dirData.subarray(offset, offset + nameBytes).toString("utf16le") : "";
+    const type = dirData[offset + 66];
+    const startSector = dirData.readUInt32LE(offset + 116);
+    const size = dirData.readUInt32LE(offset + 120);
+    dirEntries.push({ name, type, startSector, size });
+  }
+  let miniStreamData = null;
+  function getMiniStream() {
+    if (miniStreamData) return miniStreamData;
+    const root = dirEntries[0];
+    if (!root || root.type !== 5) {
+      miniStreamData = Buffer.alloc(0);
+      return miniStreamData;
+    }
+    miniStreamData = readChain(root.startSector, root.size || MAX_STREAM_SIZE);
+    return miniStreamData;
+  }
+  function readMiniStream(startSector, size) {
+    const mft = getMiniFatTable();
+    const ms = getMiniStream();
+    if (mft.length === 0 || ms.length === 0) return Buffer.alloc(0);
+    const chunks = [];
+    let current = startSector;
+    let totalRead = 0;
+    const visited = /* @__PURE__ */ new Set();
+    while (current !== END_OF_CHAIN && current !== FREE_SECT && totalRead < size) {
+      if (visited.has(current)) break;
+      if (visited.size > MAX_CHAIN_LENGTH) break;
+      visited.add(current);
+      const off = current * miniSectorSize;
+      const remaining = size - totalRead;
+      const chunkSize = Math.min(miniSectorSize, remaining);
+      if (off + chunkSize <= ms.length) {
+        chunks.push(ms.subarray(off, off + chunkSize));
+      }
+      totalRead += chunkSize;
+      current = current < mft.length ? mft[current] : END_OF_CHAIN;
+    }
+    return Buffer.concat(chunks);
+  }
+  function readStreamData(entry) {
+    if (entry.size === 0) return Buffer.alloc(0);
+    if (entry.size < miniStreamCutoff) {
+      const miniResult = readMiniStream(entry.startSector, entry.size);
+      if (miniResult.length > 0) return miniResult;
+    }
+    return readChain(entry.startSector, entry.size);
+  }
+  function findEntryByPath(path) {
+    const parts = path.replace(/^\//, "").split("/");
+    if (parts.length === 1) {
+      return dirEntries.find((e) => e.name === parts[0] && e.type === 2) ?? null;
+    }
+    const storageName = parts[0];
+    const streamName = parts.slice(1).join("/");
+    for (const e of dirEntries) {
+      if (e.type === 2 && e.name === streamName) {
+        return e;
+      }
+    }
+    const lastPart = parts[parts.length - 1];
+    return dirEntries.find((e) => e.type === 2 && e.name === lastPart) ?? null;
+  }
+  return {
+    findStream(path) {
+      const normalized = path.replace(/^\//, "");
+      const entry = findEntryByPath(normalized);
+      if (!entry || entry.type !== 2) return null;
+      const stream = readStreamData(entry);
+      return stream.length > 0 ? stream : null;
+    },
+    entries() {
+      return dirEntries.filter((e) => e.type === 2);
+    }
+  };
+}
+
+// src/detect.ts
+function magicBytes(buffer) {
+  return new Uint8Array(buffer, 0, Math.min(4, buffer.byteLength));
+}
+function isZipFile(buffer) {
+  const b = magicBytes(buffer);
+  return b[0] === 80 && b[1] === 75 && b[2] === 3 && b[3] === 4;
+}
+function isHwpxFile(buffer) {
+  return isZipFile(buffer);
+}
+function isOldHwpFile(buffer) {
+  const b = magicBytes(buffer);
+  return b[0] === 208 && b[1] === 207 && b[2] === 17 && b[3] === 224;
+}
+var HWP3_PREFIX = new TextEncoder().encode("HWP Document File V3.00");
+function isHwp3File(buffer) {
+  if (buffer.byteLength < HWP3_PREFIX.length) return false;
+  const head = new Uint8Array(buffer, 0, HWP3_PREFIX.length);
+  for (let i = 0; i < HWP3_PREFIX.length; i++) {
+    if (head[i] !== HWP3_PREFIX[i]) return false;
+  }
+  return true;
+}
+function isPdfFile(buffer) {
+  const b = magicBytes(buffer);
+  return b[0] === 37 && b[1] === 80 && b[2] === 68 && b[3] === 70;
+}
+function isHwpmlFile(buffer) {
+  const bytes = new Uint8Array(buffer, 0, Math.min(512, buffer.byteLength));
+  const head = new TextDecoder("utf-8", { fatal: false }).decode(bytes).replace(/^\uFEFF/, "");
+  return head.trimStart().startsWith("<?xml") && head.includes("<HWPML");
+}
+function detectFormat(buffer) {
+  if (buffer.byteLength < 4) return "unknown";
+  if (isHwp3File(buffer)) return "hwp3";
+  if (isZipFile(buffer)) return "hwpx";
+  if (isOldHwpFile(buffer)) return "hwp";
+  if (isPdfFile(buffer)) return "pdf";
+  if (isHwpmlFile(buffer)) return "hwpml";
+  return "unknown";
+}
+function detectOle2Format(buffer) {
+  try {
+    const cfb = parseLenientCfb(Buffer.from(buffer));
+    const names = cfb.entries().map((e) => e.name);
+    if (names.includes("Workbook") || names.includes("Book")) return "xls";
+    if (names.includes("FileHeader")) return "hwp";
+    if (names.some((n) => n === "DocInfo" || n.startsWith("Section"))) return "hwp";
+    return "unknown";
+  } catch {
+    return "unknown";
+  }
+}
+async function detectZipFormat(buffer) {
+  try {
+    const zip = await JSZip.loadAsync(buffer);
+    if (zip.file("xl/workbook.xml")) return "xlsx";
+    if (zip.file("word/document.xml")) return "docx";
+    if (zip.file("Contents/content.hpf") || zip.file("mimetype")) return "hwpx";
+    const hasSection = Object.keys(zip.files).some((f) => f.startsWith("Contents/"));
+    if (hasSection) return "hwpx";
+    return "unknown";
+  } catch {
+    return "unknown";
+  }
+}
+
+export {
+  parseLenientCfb,
+  isZipFile,
+  isHwpxFile,
+  isOldHwpFile,
+  isHwp3File,
+  isPdfFile,
+  isHwpmlFile,
+  detectFormat,
+  detectOle2Format,
+  detectZipFormat
+};
+//# sourceMappingURL=chunk-5CJGKKMZ.js.map

package/dist/chunk-5CJGKKMZ.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/detect.ts","../src/hwp5/cfb-lenient.ts"],"sourcesContent":["/** 매직 바이트 기반 파일 포맷 감지 */\r\n\r\nimport JSZip from \"jszip\"\r\nimport type { FileType } from \"./types.js\"\r\nimport { parseLenientCfb } from \"./hwp5/cfb-lenient.js\"\r\n\r\n/** 매직 바이트 뷰 생성 (복사 없이 view) */\r\nfunction magicBytes(buffer: ArrayBuffer): Uint8Array {\r\n  return new Uint8Array(buffer, 0, Math.min(4, buffer.byteLength))\r\n}\r\n\r\n/** ZIP 파일 여부: PK\\x03\\x04 */\r\nexport function isZipFile(buffer: ArrayBuffer): boolean {\r\n  const b = magicBytes(buffer)\r\n  return b[0] === 0x50 && b[1] === 0x4b && b[2] === 0x03 && b[3] === 0x04\r\n}\r\n\r\n/** HWPX (ZIP 기반 한컴 문서): PK\\x03\\x04 — 하위 호환용 */\r\nexport function isHwpxFile(buffer: ArrayBuffer): boolean {\r\n  return isZipFile(buffer)\r\n}\r\n\r\n/** HWP 5.x (OLE2 바이너리 한컴 문서): \\xD0\\xCF\\x11\\xE0 */\r\nexport function isOldHwpFile(buffer: ArrayBuffer): boolean {\r\n  const b = magicBytes(buffer)\r\n  return b[0] === 0xd0 && b[1] === 0xcf && b[2] === 0x11 && b[3] === 0xe0\r\n}\r\n\r\n/**\r\n * HWP 3.x (한글 워드프로세서 3.0): \"HWP Document File V3.00 \\x1A\\x01\\x02\\x03\\x04\\x05\" 30 byte.\r\n * CFB(OLE2) 컨테이너 아닌 단일 binary stream — isOldHwpFile 과 magic 이 다르다.\r\n */\r\nconst HWP3_PREFIX = new TextEncoder().encode(\"HWP Document File V3.00\")\r\nexport function isHwp3File(buffer: ArrayBuffer): boolean {\r\n  if (buffer.byteLength < HWP3_PREFIX.length) return false\r\n  const head = new Uint8Array(buffer, 0, HWP3_PREFIX.length)\r\n  for (let i = 0; i < HWP3_PREFIX.length; i++) {\r\n    if (head[i] !== HWP3_PREFIX[i]) return false\r\n  }\r\n  return true\r\n}\r\n\r\n/** PDF 문서: %PDF */\r\nexport function isPdfFile(buffer: ArrayBuffer): boolean {\r\n  const b = magicBytes(buffer)\r\n  return b[0] === 0x25 && b[1] === 0x50 && b[2] === 0x44 && b[3] === 0x46\r\n}\r\n\r\n/** HWPML (XML 기반 한컴 문서): <?xml ... <HWPML */\r\nexport function isHwpmlFile(buffer: ArrayBuffer): boolean {\r\n  const bytes = new Uint8Array(buffer, 0, Math.min(512, buffer.byteLength))\r\n  const head = new TextDecoder(\"utf-8\", { fatal: false }).decode(bytes).replace(/^\\uFEFF/, \"\")\r\n  return head.trimStart().startsWith(\"<?xml\") && head.includes(\"<HWPML\")\r\n}\r\n\r\n/** 동기 포맷 감지 — ZIP은 모두 \"hwpx\"로 반환 (하위 호환) */\r\nexport function detectFormat(buffer: ArrayBuffer): FileType {\r\n  if (buffer.byteLength < 4) return \"unknown\"\r\n  if (isHwp3File(buffer)) return \"hwp3\"\r\n  if (isZipFile(buffer)) return \"hwpx\"\r\n  if (isOldHwpFile(buffer)) return \"hwp\"\r\n  if (isPdfFile(buffer)) return \"pdf\"\r\n  if (isHwpmlFile(buffer)) return \"hwpml\"\r\n  return \"unknown\"\r\n}\r\n\r\n/**\r\n * OLE2 컨테이너 내부 스트림 기반 포맷 세분화.\r\n * HWP 5.x, XLS 모두 OLE2이므로 스트림 이름으로 구분.\r\n *  - \"Workbook\" 또는 \"Book\" → 'xls'\r\n *  - 그 외 (FileHeader 등) → 'hwp'\r\n */\r\nexport function detectOle2Format(buffer: ArrayBuffer): \"hwp\" | \"xls\" | \"unknown\" {\r\n  try {\r\n    const cfb = parseLenientCfb(Buffer.from(buffer))\r\n    const names = cfb.entries().map(e => e.name)\r\n    if (names.includes(\"Workbook\") || names.includes(\"Book\")) return \"xls\"\r\n    if (names.includes(\"FileHeader\")) return \"hwp\"\r\n    // FileHeader 없어도 BodyText/DocInfo 있으면 hwp\r\n    if (names.some(n => n === \"DocInfo\" || n.startsWith(\"Section\"))) return \"hwp\"\r\n    return \"unknown\"\r\n  } catch {\r\n    return \"unknown\"\r\n  }\r\n}\r\n\r\n/**\r\n * ZIP 내부 구조 기반 포맷 세분화.\r\n * HWPX, XLSX, DOCX 모두 ZIP이므로 내부 파일로 구분.\r\n */\r\nexport async function detectZipFormat(buffer: ArrayBuffer): Promise<\"hwpx\" | \"xlsx\" | \"docx\" | \"unknown\"> {\r\n  try {\r\n    const zip = await JSZip.loadAsync(buffer)\r\n    // XLSX: xl/workbook.xml\r\n    if (zip.file(\"xl/workbook.xml\")) return \"xlsx\"\r\n    // DOCX: word/document.xml\r\n    if (zip.file(\"word/document.xml\")) return \"docx\"\r\n    // HWPX: Contents/ 또는 content.hpf 또는 mimetype\r\n    if (zip.file(\"Contents/content.hpf\") || zip.file(\"mimetype\")) return \"hwpx\"\r\n    // 기타 ZIP 내에 section 파일이 있으면 HWPX로 추정\r\n    const hasSection = Object.keys(zip.files).some(f => f.startsWith(\"Contents/\"))\r\n    if (hasSection) return \"hwpx\"\r\n    return \"unknown\"\r\n  } catch {\r\n    return \"unknown\"\r\n  }\r\n}\r\n","/**\n * Lenient CFB (Compound File Binary / OLE2) 파서.\n *\n * 표준 cfb 모듈이 FAT 검증 실패로 거부하는 손상된 HWP 파일을 열기 위한 폴백.\n * 직접 헤더/FAT/디렉토리를 파싱하여 스트림 데이터를 추출.\n *\n * 참조: rhwp (MIT) src/parser/cfb_reader.rs (LenientCfbReader)\n * 참조: MS-CFB spec (https://learn.microsoft.com/en-us/openspecs/windows_protocols/ms-cfb)\n */\n\nimport { decompressStream } from \"./record.js\"\n\n// ── 상수 ──\n\nconst CFB_MAGIC = Buffer.from([0xd0, 0xcf, 0x11, 0xe0, 0xa1, 0xb1, 0x1a, 0xe1])\nconst END_OF_CHAIN = 0xfffffffe\nconst FREE_SECT = 0xffffffff\n\n/** 순환 감지용 최대 체인 길이 */\nconst MAX_CHAIN_LENGTH = 1_000_000\n/** 최대 디렉토리 엔트리 수 */\nconst MAX_DIR_ENTRIES = 100_000\n/** 최대 스트림 크기 (100MB) */\nconst MAX_STREAM_SIZE = 100 * 1024 * 1024\n\n// ── 디렉토리 엔트리 ──\n\ninterface DirEntry {\n  name: string\n  type: number  // 0=unknown, 1=storage, 2=stream, 5=root\n  startSector: number\n  size: number\n}\n\n// ── CFB 컨테이너 ──\n\nexport interface LenientCfbContainer {\n  /** 이름 기반 스트림 탐색 */\n  findStream(path: string): Buffer | null\n  /** 디렉토리 엔트리 목록 */\n  entries(): DirEntry[]\n}\n\n// ── 구현 ──\n\nexport function parseLenientCfb(data: Buffer): LenientCfbContainer {\n  if (data.length < 512) throw new Error(\"CFB 파일이 너무 짧습니다 (최소 512바이트)\")\n  if (!data.subarray(0, 8).equals(CFB_MAGIC)) throw new Error(\"CFB 매직 바이트 불일치\")\n\n  // ── 헤더 파싱 ──\n\n  const sectorSizeShift = data.readUInt16LE(30)\n  if (sectorSizeShift < 7 || sectorSizeShift > 16) throw new Error(\"유효하지 않은 섹터 크기 시프트: \" + sectorSizeShift)\n  const sectorSize = 1 << sectorSizeShift  // 보통 512\n  const miniSectorSizeShift = data.readUInt16LE(32)\n  if (miniSectorSizeShift > 16) throw new Error(\"유효하지 않은 미니 섹터 크기 시프트: \" + miniSectorSizeShift)\n  const miniSectorSize = 1 << miniSectorSizeShift  // 보통 64\n\n  const fatSectorCount = data.readUInt32LE(44)\n  if (fatSectorCount > 10000) throw new Error(\"FAT 섹터 수가 너무 많습니다: \" + fatSectorCount)\n  const firstDirSector = data.readUInt32LE(48)\n  const miniStreamCutoff = data.readUInt32LE(56)  // 보통 4096\n  const firstMiniFatSector = data.readUInt32LE(60)\n  const miniFatSectorCount = data.readUInt32LE(64)\n  const firstDifatSector = data.readUInt32LE(68)\n  const difatSectorCount = data.readUInt32LE(72)\n\n  // ── 유틸 ──\n\n  function sectorOffset(id: number): number {\n    return 512 + id * sectorSize\n  }\n\n  function readSectorData(id: number): Buffer {\n    const off = sectorOffset(id)\n    if (off + sectorSize > data.length) return Buffer.alloc(0)\n    return data.subarray(off, off + sectorSize)\n  }\n\n  // ── DIFAT → FAT 섹터 목록 ──\n\n  const fatSectors: number[] = []\n\n  // 헤더 내 DIFAT (최대 109개)\n  for (let i = 0; i < 109 && fatSectors.length < fatSectorCount; i++) {\n    const sid = data.readUInt32LE(76 + i * 4)\n    if (sid === FREE_SECT || sid === END_OF_CHAIN) break\n    fatSectors.push(sid)\n  }\n\n  // 추가 DIFAT 섹터 체인\n  let difatSector = firstDifatSector\n  const visitedDifat = new Set<number>()\n  for (let d = 0; d < difatSectorCount && difatSector !== END_OF_CHAIN && difatSector !== FREE_SECT; d++) {\n    if (visitedDifat.has(difatSector)) break\n    visitedDifat.add(difatSector)\n\n    const buf = readSectorData(difatSector)\n    const entriesPerSector = (sectorSize / 4) - 1  // 마지막 4바이트는 다음 DIFAT 포인터\n    for (let i = 0; i < entriesPerSector && fatSectors.length < fatSectorCount; i++) {\n      const sid = buf.readUInt32LE(i * 4)\n      if (sid === FREE_SECT || sid === END_OF_CHAIN) continue\n      fatSectors.push(sid)\n    }\n    difatSector = buf.readUInt32LE(entriesPerSector * 4)\n  }\n\n  // ── FAT 테이블 구축 ──\n\n  const entriesPerFatSector = sectorSize / 4\n  const fatTable = new Uint32Array(fatSectors.length * entriesPerFatSector)\n\n  for (let fi = 0; fi < fatSectors.length; fi++) {\n    const buf = readSectorData(fatSectors[fi])\n    for (let i = 0; i < entriesPerFatSector; i++) {\n      fatTable[fi * entriesPerFatSector + i] = i * 4 + 3 < buf.length\n        ? buf.readUInt32LE(i * 4)\n        : FREE_SECT\n    }\n  }\n\n  // ── 체인 리더 (순환 방지) ──\n\n  function readChain(startSector: number, maxBytes: number): Buffer {\n    if (startSector === END_OF_CHAIN || startSector === FREE_SECT) return Buffer.alloc(0)\n    if (maxBytes > MAX_STREAM_SIZE) throw new Error(\"스트림이 너무 큽니다\")\n\n    const chunks: Buffer[] = []\n    let current = startSector\n    let totalRead = 0\n    const visited = new Set<number>()\n\n    while (current !== END_OF_CHAIN && current !== FREE_SECT && totalRead < maxBytes) {\n      if (visited.has(current)) break  // 순환 감지\n      if (visited.size > MAX_CHAIN_LENGTH) break\n      visited.add(current)\n\n      const buf = readSectorData(current)\n      const remaining = maxBytes - totalRead\n      chunks.push(remaining < sectorSize ? buf.subarray(0, remaining) : buf)\n      totalRead += Math.min(buf.length, remaining)\n\n      current = current < fatTable.length ? fatTable[current] : END_OF_CHAIN\n    }\n\n    return Buffer.concat(chunks)\n  }\n\n  // ── Mini-FAT 테이블 ──\n\n  let miniFatTable: Uint32Array | null = null\n\n  function getMiniFatTable(): Uint32Array {\n    if (miniFatTable) return miniFatTable\n\n    if (miniFatSectorCount === 0 || firstMiniFatSector === END_OF_CHAIN) {\n      miniFatTable = new Uint32Array(0)\n      return miniFatTable\n    }\n\n    const miniFatData = readChain(firstMiniFatSector, miniFatSectorCount * sectorSize)\n    const entries = miniFatData.length / 4\n    miniFatTable = new Uint32Array(entries)\n    for (let i = 0; i < entries; i++) {\n      miniFatTable[i] = miniFatData.readUInt32LE(i * 4)\n    }\n    return miniFatTable\n  }\n\n  // ── 디렉토리 엔트리 파싱 ──\n\n  const dirData = readChain(firstDirSector, MAX_DIR_ENTRIES * 128)\n  const dirEntries: DirEntry[] = []\n\n  for (let offset = 0; offset + 128 <= dirData.length && dirEntries.length < MAX_DIR_ENTRIES; offset += 128) {\n    const nameLen = dirData.readUInt16LE(offset + 64)  // 바이트 수 (null 포함)\n    if (nameLen <= 0 || nameLen > 64) {\n      dirEntries.push({ name: \"\", type: 0, startSector: 0, size: 0 })\n      continue\n    }\n\n    const nameBytes = nameLen - 2  // null terminator 제외\n    const name = nameBytes > 0\n      ? dirData.subarray(offset, offset + nameBytes).toString(\"utf16le\")\n      : \"\"\n\n    const type = dirData[offset + 66]\n    const startSector = dirData.readUInt32LE(offset + 116)\n    // CFBv3에서는 size가 u32 (offset 120), v4에서는 u64\n    const size = dirData.readUInt32LE(offset + 120)\n\n    dirEntries.push({ name, type, startSector, size })\n  }\n\n  // ── Root 엔트리에서 미니 스트림 추출 ──\n\n  let miniStreamData: Buffer | null = null\n\n  function getMiniStream(): Buffer {\n    if (miniStreamData) return miniStreamData\n    const root = dirEntries[0]\n    if (!root || root.type !== 5) {\n      miniStreamData = Buffer.alloc(0)\n      return miniStreamData\n    }\n    miniStreamData = readChain(root.startSector, root.size || MAX_STREAM_SIZE)\n    return miniStreamData\n  }\n\n  // ── 미니 스트림에서 읽기 ──\n\n  function readMiniStream(startSector: number, size: number): Buffer {\n    const mft = getMiniFatTable()\n    const ms = getMiniStream()\n    if (mft.length === 0 || ms.length === 0) return Buffer.alloc(0)\n\n    const chunks: Buffer[] = []\n    let current = startSector\n    let totalRead = 0\n    const visited = new Set<number>()\n\n    while (current !== END_OF_CHAIN && current !== FREE_SECT && totalRead < size) {\n      if (visited.has(current)) break\n      if (visited.size > MAX_CHAIN_LENGTH) break\n      visited.add(current)\n\n      const off = current * miniSectorSize\n      const remaining = size - totalRead\n      const chunkSize = Math.min(miniSectorSize, remaining)\n      if (off + chunkSize <= ms.length) {\n        chunks.push(ms.subarray(off, off + chunkSize))\n      }\n      totalRead += chunkSize\n\n      current = current < mft.length ? mft[current] : END_OF_CHAIN\n    }\n\n    return Buffer.concat(chunks)\n  }\n\n  // ── 스트림 읽기 (일반/미니 자동 분기) ──\n\n  function readStreamData(entry: DirEntry): Buffer {\n    if (entry.size === 0) return Buffer.alloc(0)\n    if (entry.size < miniStreamCutoff) {\n      const miniResult = readMiniStream(entry.startSector, entry.size)\n      // 미니스트림이 비어있으면 일반 체인으로 폴백 (lenient)\n      if (miniResult.length > 0) return miniResult\n    }\n    return readChain(entry.startSector, entry.size)\n  }\n\n  // ── 경로 기반 탐색 ──\n\n  // 전체 경로 맵 구축 (간이: 이름 기반 flat lookup)\n  // HWP 파일의 디렉토리 구조는 보통 1~2 depth이므로 이름 매칭으로 충분\n  function findEntryByPath(path: string): DirEntry | null {\n    // \"/FileHeader\" → \"FileHeader\"\n    // \"/BodyText/Section0\" → path component matching\n    const parts = path.replace(/^\\//, \"\").split(\"/\")\n\n    if (parts.length === 1) {\n      // 단일 이름 매칭\n      return dirEntries.find(e => e.name === parts[0] && e.type === 2) ?? null\n    }\n\n    // 2-depth: storage/stream\n    // HWP 구조: Root/BodyText/Section0, Root/DocInfo, Root/BinData/BIN0001 등\n    const storageName = parts[0]\n    const streamName = parts.slice(1).join(\"/\")\n\n    // 디렉토리 구조 대신 이름 패턴으로 찾기 (lenient)\n    for (const e of dirEntries) {\n      if (e.type === 2 && e.name === streamName) {\n        // 부모 확인은 생략 (lenient) — 중복 이름 시 첫 번째 반환\n        return e\n      }\n    }\n\n    // 정확한 이름이 아닌 경우 (ViewText/Section0 등)\n    const lastPart = parts[parts.length - 1]\n    return dirEntries.find(e => e.type === 2 && e.name === lastPart) ?? null\n  }\n\n  // ── 공개 API ──\n\n  return {\n    findStream(path: string): Buffer | null {\n      // \\005 prefix 처리 (SummaryInformation)\n      const normalized = path.replace(/^\\//, \"\")\n      const entry = findEntryByPath(normalized)\n      if (!entry || entry.type !== 2) return null\n      const stream = readStreamData(entry)\n      return stream.length > 0 ? stream : null\n    },\n\n    entries(): DirEntry[] {\n      return dirEntries.filter(e => e.type === 2)  // stream만\n    },\n  }\n}\n"],"mappings":";;;AAEA,OAAO,WAAW;;;ACYlB,IAAM,YAAY,OAAO,KAAK,CAAC,KAAM,KAAM,IAAM,KAAM,KAAM,KAAM,IAAM,GAAI,CAAC;AAC9E,IAAM,eAAe;AACrB,IAAM,YAAY;AAGlB,IAAM,mBAAmB;AAEzB,IAAM,kBAAkB;AAExB,IAAM,kBAAkB,MAAM,OAAO;AAsB9B,SAAS,gBAAgB,MAAmC;AACjE,MAAI,KAAK,SAAS,IAAK,OAAM,IAAI,MAAM,mGAA6B;AACpE,MAAI,CAAC,KAAK,SAAS,GAAG,CAAC,EAAE,OAAO,SAAS,EAAG,OAAM,IAAI,MAAM,wDAAgB;AAI5E,QAAM,kBAAkB,KAAK,aAAa,EAAE;AAC5C,MAAI,kBAAkB,KAAK,kBAAkB,GAAI,OAAM,IAAI,MAAM,yFAAwB,eAAe;AACxG,QAAM,aAAa,KAAK;AACxB,QAAM,sBAAsB,KAAK,aAAa,EAAE;AAChD,MAAI,sBAAsB,GAAI,OAAM,IAAI,MAAM,sGAA2B,mBAAmB;AAC5F,QAAM,iBAAiB,KAAK;AAE5B,QAAM,iBAAiB,KAAK,aAAa,EAAE;AAC3C,MAAI,iBAAiB,IAAO,OAAM,IAAI,MAAM,0EAAwB,cAAc;AAClF,QAAM,iBAAiB,KAAK,aAAa,EAAE;AAC3C,QAAM,mBAAmB,KAAK,aAAa,EAAE;AAC7C,QAAM,qBAAqB,KAAK,aAAa,EAAE;AAC/C,QAAM,qBAAqB,KAAK,aAAa,EAAE;AAC/C,QAAM,mBAAmB,KAAK,aAAa,EAAE;AAC7C,QAAM,mBAAmB,KAAK,aAAa,EAAE;AAI7C,WAAS,aAAa,IAAoB;AACxC,WAAO,MAAM,KAAK;AAAA,EACpB;AAEA,WAAS,eAAe,IAAoB;AAC1C,UAAM,MAAM,aAAa,EAAE;AAC3B,QAAI,MAAM,aAAa,KAAK,OAAQ,QAAO,OAAO,MAAM,CAAC;AACzD,WAAO,KAAK,SAAS,KAAK,MAAM,UAAU;AAAA,EAC5C;AAIA,QAAM,aAAuB,CAAC;AAG9B,WAAS,IAAI,GAAG,IAAI,OAAO,WAAW,SAAS,gBAAgB,KAAK;AAClE,UAAM,MAAM,KAAK,aAAa,KAAK,IAAI,CAAC;AACxC,QAAI,QAAQ,aAAa,QAAQ,aAAc;AAC/C,eAAW,KAAK,GAAG;AAAA,EACrB;AAGA,MAAI,cAAc;AAClB,QAAM,eAAe,oBAAI,IAAY;AACrC,WAAS,IAAI,GAAG,IAAI,oBAAoB,gBAAgB,gBAAgB,gBAAgB,WAAW,KAAK;AACtG,QAAI,aAAa,IAAI,WAAW,EAAG;AACnC,iBAAa,IAAI,WAAW;AAE5B,UAAM,MAAM,eAAe,WAAW;AACtC,UAAM,mBAAoB,aAAa,IAAK;AAC5C,aAAS,IAAI,GAAG,IAAI,oBAAoB,WAAW,SAAS,gBAAgB,KAAK;AAC/E,YAAM,MAAM,IAAI,aAAa,IAAI,CAAC;AAClC,UAAI,QAAQ,aAAa,QAAQ,aAAc;AAC/C,iBAAW,KAAK,GAAG;AAAA,IACrB;AACA,kBAAc,IAAI,aAAa,mBAAmB,CAAC;AAAA,EACrD;AAIA,QAAM,sBAAsB,aAAa;AACzC,QAAM,WAAW,IAAI,YAAY,WAAW,SAAS,mBAAmB;AAExE,WAAS,KAAK,GAAG,KAAK,WAAW,QAAQ,MAAM;AAC7C,UAAM,MAAM,eAAe,WAAW,EAAE,CAAC;AACzC,aAAS,IAAI,GAAG,IAAI,qBAAqB,KAAK;AAC5C,eAAS,KAAK,sBAAsB,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,SACrD,IAAI,aAAa,IAAI,CAAC,IACtB;AAAA,IACN;AAAA,EACF;AAIA,WAAS,UAAU,aAAqB,UAA0B;AAChE,QAAI,gBAAgB,gBAAgB,gBAAgB,UAAW,QAAO,OAAO,MAAM,CAAC;AACpF,QAAI,WAAW,gBAAiB,OAAM,IAAI,MAAM,0DAAa;AAE7D,UAAM,SAAmB,CAAC;AAC1B,QAAI,UAAU;AACd,QAAI,YAAY;AAChB,UAAM,UAAU,oBAAI,IAAY;AAEhC,WAAO,YAAY,gBAAgB,YAAY,aAAa,YAAY,UAAU;AAChF,UAAI,QAAQ,IAAI,OAAO,EAAG;AAC1B,UAAI,QAAQ,OAAO,iBAAkB;AACrC,cAAQ,IAAI,OAAO;AAEnB,YAAM,MAAM,eAAe,OAAO;AAClC,YAAM,YAAY,WAAW;AAC7B,aAAO,KAAK,YAAY,aAAa,IAAI,SAAS,GAAG,SAAS,IAAI,GAAG;AACrE,mBAAa,KAAK,IAAI,IAAI,QAAQ,SAAS;AAE3C,gBAAU,UAAU,SAAS,SAAS,SAAS,OAAO,IAAI;AAAA,IAC5D;AAEA,WAAO,OAAO,OAAO,MAAM;AAAA,EAC7B;AAIA,MAAI,eAAmC;AAEvC,WAAS,kBAA+B;AACtC,QAAI,aAAc,QAAO;AAEzB,QAAI,uBAAuB,KAAK,uBAAuB,cAAc;AACnE,qBAAe,IAAI,YAAY,CAAC;AAChC,aAAO;AAAA,IACT;AAEA,UAAM,cAAc,UAAU,oBAAoB,qBAAqB,UAAU;AACjF,UAAM,UAAU,YAAY,SAAS;AACrC,mBAAe,IAAI,YAAY,OAAO;AACtC,aAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,mBAAa,CAAC,IAAI,YAAY,aAAa,IAAI,CAAC;AAAA,IAClD;AACA,WAAO;AAAA,EACT;AAIA,QAAM,UAAU,UAAU,gBAAgB,kBAAkB,GAAG;AAC/D,QAAM,aAAyB,CAAC;AAEhC,WAAS,SAAS,GAAG,SAAS,OAAO,QAAQ,UAAU,WAAW,SAAS,iBAAiB,UAAU,KAAK;AACzG,UAAM,UAAU,QAAQ,aAAa,SAAS,EAAE;AAChD,QAAI,WAAW,KAAK,UAAU,IAAI;AAChC,iBAAW,KAAK,EAAE,MAAM,IAAI,MAAM,GAAG,aAAa,GAAG,MAAM,EAAE,CAAC;AAC9D;AAAA,IACF;AAEA,UAAM,YAAY,UAAU;AAC5B,UAAM,OAAO,YAAY,IACrB,QAAQ,SAAS,QAAQ,SAAS,SAAS,EAAE,SAAS,SAAS,IAC/D;AAEJ,UAAM,OAAO,QAAQ,SAAS,EAAE;AAChC,UAAM,cAAc,QAAQ,aAAa,SAAS,GAAG;AAErD,UAAM,OAAO,QAAQ,aAAa,SAAS,GAAG;AAE9C,eAAW,KAAK,EAAE,MAAM,MAAM,aAAa,KAAK,CAAC;AAAA,EACnD;AAIA,MAAI,iBAAgC;AAEpC,WAAS,gBAAwB;AAC/B,QAAI,eAAgB,QAAO;AAC3B,UAAM,OAAO,WAAW,CAAC;AACzB,QAAI,CAAC,QAAQ,KAAK,SAAS,GAAG;AAC5B,uBAAiB,OAAO,MAAM,CAAC;AAC/B,aAAO;AAAA,IACT;AACA,qBAAiB,UAAU,KAAK,aAAa,KAAK,QAAQ,eAAe;AACzE,WAAO;AAAA,EACT;AAIA,WAAS,eAAe,aAAqB,MAAsB;AACjE,UAAM,MAAM,gBAAgB;AAC5B,UAAM,KAAK,cAAc;AACzB,QAAI,IAAI,WAAW,KAAK,GAAG,WAAW,EAAG,QAAO,OAAO,MAAM,CAAC;AAE9D,UAAM,SAAmB,CAAC;AAC1B,QAAI,UAAU;AACd,QAAI,YAAY;AAChB,UAAM,UAAU,oBAAI,IAAY;AAEhC,WAAO,YAAY,gBAAgB,YAAY,aAAa,YAAY,MAAM;AAC5E,UAAI,QAAQ,IAAI,OAAO,EAAG;AAC1B,UAAI,QAAQ,OAAO,iBAAkB;AACrC,cAAQ,IAAI,OAAO;AAEnB,YAAM,MAAM,UAAU;AACtB,YAAM,YAAY,OAAO;AACzB,YAAM,YAAY,KAAK,IAAI,gBAAgB,SAAS;AACpD,UAAI,MAAM,aAAa,GAAG,QAAQ;AAChC,eAAO,KAAK,GAAG,SAAS,KAAK,MAAM,SAAS,CAAC;AAAA,MAC/C;AACA,mBAAa;AAEb,gBAAU,UAAU,IAAI,SAAS,IAAI,OAAO,IAAI;AAAA,IAClD;AAEA,WAAO,OAAO,OAAO,MAAM;AAAA,EAC7B;AAIA,WAAS,eAAe,OAAyB;AAC/C,QAAI,MAAM,SAAS,EAAG,QAAO,OAAO,MAAM,CAAC;AAC3C,QAAI,MAAM,OAAO,kBAAkB;AACjC,YAAM,aAAa,eAAe,MAAM,aAAa,MAAM,IAAI;AAE/D,UAAI,WAAW,SAAS,EAAG,QAAO;AAAA,IACpC;AACA,WAAO,UAAU,MAAM,aAAa,MAAM,IAAI;AAAA,EAChD;AAMA,WAAS,gBAAgB,MAA+B;AAGtD,UAAM,QAAQ,KAAK,QAAQ,OAAO,EAAE,EAAE,MAAM,GAAG;AAE/C,QAAI,MAAM,WAAW,GAAG;AAEtB,aAAO,WAAW,KAAK,OAAK,EAAE,SAAS,MAAM,CAAC,KAAK,EAAE,SAAS,CAAC,KAAK;AAAA,IACtE;AAIA,UAAM,cAAc,MAAM,CAAC;AAC3B,UAAM,aAAa,MAAM,MAAM,CAAC,EAAE,KAAK,GAAG;AAG1C,eAAW,KAAK,YAAY;AAC1B,UAAI,EAAE,SAAS,KAAK,EAAE,SAAS,YAAY;AAEzC,eAAO;AAAA,MACT;AAAA,IACF;AAGA,UAAM,WAAW,MAAM,MAAM,SAAS,CAAC;AACvC,WAAO,WAAW,KAAK,OAAK,EAAE,SAAS,KAAK,EAAE,SAAS,QAAQ,KAAK;AAAA,EACtE;AAIA,SAAO;AAAA,IACL,WAAW,MAA6B;AAEtC,YAAM,aAAa,KAAK,QAAQ,OAAO,EAAE;AACzC,YAAM,QAAQ,gBAAgB,UAAU;AACxC,UAAI,CAAC,SAAS,MAAM,SAAS,EAAG,QAAO;AACvC,YAAM,SAAS,eAAe,KAAK;AACnC,aAAO,OAAO,SAAS,IAAI,SAAS;AAAA,IACtC;AAAA,IAEA,UAAsB;AACpB,aAAO,WAAW,OAAO,OAAK,EAAE,SAAS,CAAC;AAAA,IAC5C;AAAA,EACF;AACF;;;ADrSA,SAAS,WAAW,QAAiC;AACnD,SAAO,IAAI,WAAW,QAAQ,GAAG,KAAK,IAAI,GAAG,OAAO,UAAU,CAAC;AACjE;AAGO,SAAS,UAAU,QAA8B;AACtD,QAAM,IAAI,WAAW,MAAM;AAC3B,SAAO,EAAE,CAAC,MAAM,MAAQ,EAAE,CAAC,MAAM,MAAQ,EAAE,CAAC,MAAM,KAAQ,EAAE,CAAC,MAAM;AACrE;AAGO,SAAS,WAAW,QAA8B;AACvD,SAAO,UAAU,MAAM;AACzB;AAGO,SAAS,aAAa,QAA8B;AACzD,QAAM,IAAI,WAAW,MAAM;AAC3B,SAAO,EAAE,CAAC,MAAM,OAAQ,EAAE,CAAC,MAAM,OAAQ,EAAE,CAAC,MAAM,MAAQ,EAAE,CAAC,MAAM;AACrE;AAMA,IAAM,cAAc,IAAI,YAAY,EAAE,OAAO,yBAAyB;AAC/D,SAAS,WAAW,QAA8B;AACvD,MAAI,OAAO,aAAa,YAAY,OAAQ,QAAO;AACnD,QAAM,OAAO,IAAI,WAAW,QAAQ,GAAG,YAAY,MAAM;AACzD,WAAS,IAAI,GAAG,IAAI,YAAY,QAAQ,KAAK;AAC3C,QAAI,KAAK,CAAC,MAAM,YAAY,CAAC,EAAG,QAAO;AAAA,EACzC;AACA,SAAO;AACT;AAGO,SAAS,UAAU,QAA8B;AACtD,QAAM,IAAI,WAAW,MAAM;AAC3B,SAAO,EAAE,CAAC,MAAM,MAAQ,EAAE,CAAC,MAAM,MAAQ,EAAE,CAAC,MAAM,MAAQ,EAAE,CAAC,MAAM;AACrE;AAGO,SAAS,YAAY,QAA8B;AACxD,QAAM,QAAQ,IAAI,WAAW,QAAQ,GAAG,KAAK,IAAI,KAAK,OAAO,UAAU,CAAC;AACxE,QAAM,OAAO,IAAI,YAAY,SAAS,EAAE,OAAO,MAAM,CAAC,EAAE,OAAO,KAAK,EAAE,QAAQ,WAAW,EAAE;AAC3F,SAAO,KAAK,UAAU,EAAE,WAAW,OAAO,KAAK,KAAK,SAAS,QAAQ;AACvE;AAGO,SAAS,aAAa,QAA+B;AAC1D,MAAI,OAAO,aAAa,EAAG,QAAO;AAClC,MAAI,WAAW,MAAM,EAAG,QAAO;AAC/B,MAAI,UAAU,MAAM,EAAG,QAAO;AAC9B,MAAI,aAAa,MAAM,EAAG,QAAO;AACjC,MAAI,UAAU,MAAM,EAAG,QAAO;AAC9B,MAAI,YAAY,MAAM,EAAG,QAAO;AAChC,SAAO;AACT;AAQO,SAAS,iBAAiB,QAAgD;AAC/E,MAAI;AACF,UAAM,MAAM,gBAAgB,OAAO,KAAK,MAAM,CAAC;AAC/C,UAAM,QAAQ,IAAI,QAAQ,EAAE,IAAI,OAAK,EAAE,IAAI;AAC3C,QAAI,MAAM,SAAS,UAAU,KAAK,MAAM,SAAS,MAAM,EAAG,QAAO;AACjE,QAAI,MAAM,SAAS,YAAY,EAAG,QAAO;AAEzC,QAAI,MAAM,KAAK,OAAK,MAAM,aAAa,EAAE,WAAW,SAAS,CAAC,EAAG,QAAO;AACxE,WAAO;AAAA,EACT,QAAQ;AACN,WAAO;AAAA,EACT;AACF;AAMA,eAAsB,gBAAgB,QAAoE;AACxG,MAAI;AACF,UAAM,MAAM,MAAM,MAAM,UAAU,MAAM;AAExC,QAAI,IAAI,KAAK,iBAAiB,EAAG,QAAO;AAExC,QAAI,IAAI,KAAK,mBAAmB,EAAG,QAAO;AAE1C,QAAI,IAAI,KAAK,sBAAsB,KAAK,IAAI,KAAK,UAAU,EAAG,QAAO;AAErE,UAAM,aAAa,OAAO,KAAK,IAAI,KAAK,EAAE,KAAK,OAAK,EAAE,WAAW,WAAW,CAAC;AAC7E,QAAI,WAAY,QAAO;AACvB,WAAO;AAAA,EACT,QAAQ;AACN,WAAO;AAAA,EACT;AACF;","names":[]}

package/dist/{chunk-24NKFRB4.js → chunk-GNN6MHH4.js}

@@ -1,5 +1,5 @@
 // src/utils.ts
-var VERSION = true ? "2.5.2" : "0.0.0-dev";
+var VERSION = true ? "2.7.1" : "0.0.0-dev";
 function toArrayBuffer(buf) {
   if (buf.byteOffset === 0 && buf.byteLength === buf.buffer.byteLength) {
     return buf.buffer;
@@ -325,6 +325,17 @@ function hasMergedCells(table) {
   }
   return false;
 }
+function containsInlineMath(text) {
+  return /(^|[^\\])\$(?=\S)(?:\\.|[^$\n])+?\S\$/.test(text);
+}
+function tableContainsInlineMath(table) {
+  for (const row of table.cells) {
+    for (const cell of row) {
+      if (containsInlineMath(cell.text)) return true;
+    }
+  }
+  return false;
+}
 function tableToHtml(table) {
   const { cells, rows: numRows, cols: numCols } = table;
   const skip = /* @__PURE__ */ new Set();
@@ -357,7 +368,7 @@ function tableToHtml(table) {
 function tableToMarkdown(table) {
   if (table.rows === 0 || table.cols === 0) return "";
   const { cells, rows: numRows, cols: numCols } = table;
-  if (hasMergedCells(table)) return tableToHtml(table);
+  if (hasMergedCells(table) && !tableContainsInlineMath(table)) return tableToHtml(table);
   if (numRows === 1 && numCols === 1) {
     const content = sanitizeText(cells[0][0].text);
     if (!content) return "";
@@ -447,4 +458,4 @@ export {
   HEADING_RATIO_H2,
   HEADING_RATIO_H3
 };
-//# sourceMappingURL=chunk-24NKFRB4.js.map
+//# sourceMappingURL=chunk-GNN6MHH4.js.map

package/dist/chunk-GNN6MHH4.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/utils.ts","../src/table/builder.ts","../src/types.ts"],"sourcesContent":["/** kordoc 공용 유틸리티 */\n\n/** 빌드 타임에 tsup define으로 주입되는 버전 */\ndeclare const __KORDOC_VERSION__: string\nexport const VERSION: string = typeof __KORDOC_VERSION__ !== \"undefined\" ? __KORDOC_VERSION__ : \"0.0.0-dev\"\n\n/**\n * Node.js Buffer → ArrayBuffer 변환\n * pool Buffer의 공유 ArrayBuffer 문제를 안전하게 처리.\n * offset=0이고 전체 ArrayBuffer를 차지하면 복사 없이 직접 반환.\n */\nexport function toArrayBuffer(buf: Buffer): ArrayBuffer {\n  if (buf.byteOffset === 0 && buf.byteLength === buf.buffer.byteLength) {\n    return buf.buffer as ArrayBuffer\n  }\n  return buf.buffer.slice(buf.byteOffset, buf.byteOffset + buf.byteLength) as ArrayBuffer\n}\n\n/**\n * kordoc 내부 에러 클래스 — 사용자에게 노출해도 안전한 메시지만 포함.\n * MCP 에러 정제에서 instanceof로 판별하여 allowlist 패턴 매칭 없이 안전하게 통과.\n */\nexport class KordocError extends Error {\n  constructor(message: string) {\n    super(message)\n    this.name = \"KordocError\"\n  }\n}\n\n/**\n * 에러 메시지 정제 — KordocError는 그대로, 나머지는 일반 메시지로 대체.\n * 파일시스템 경로, 스택 트레이스 등 내부 정보 노출 방지.\n */\nexport function sanitizeError(err: unknown): string {\n  if (err instanceof KordocError) return err.message\n  return \"문서 처리 중 오류가 발생했습니다\"\n}\n\n/**\n * ZIP 엔트리 경로의 경로 순회 여부 판별.\n * 백슬래시 정규화, .., 절대경로, Windows 드라이브 문자 모두 차단.\n */\nexport function isPathTraversal(name: string): boolean {\n  if (name.includes(\"\\x00\")) return true\n  const normalized = name.replace(/\\\\/g, \"/\")\n  const segments = normalized.split(\"/\")\n  return segments.some(s => s === \"..\") || normalized.startsWith(\"/\") || /^[A-Za-z]:/.test(normalized)\n}\n\n// ─── ZIP 안전 로딩 (ZIP bomb 방지) ────────────────────\n\n/**\n * ZIP bomb 사전 검사 — Central Directory에서 비압축 합계와 엔트리 수 확인.\n * HWPX/XLSX/DOCX 등 모든 ZIP 기반 포맷에서 공통 사용.\n */\nexport function precheckZipSize(\n  buffer: ArrayBuffer,\n  maxUncompressedSize = 100 * 1024 * 1024,\n  maxEntries = 500,\n): { totalUncompressed: number; entryCount: number } {\n  try {\n    const data = new DataView(buffer)\n    const len = buffer.byteLength\n    // EOCD 시그니처 역방향 스캔\n    let eocdOffset = -1\n    for (let i = len - 22; i >= Math.max(0, len - 65557); i--) {\n      if (data.getUint32(i, true) === 0x06054b50) { eocdOffset = i; break }\n    }\n    if (eocdOffset < 0) return { totalUncompressed: 0, entryCount: 0 }\n\n    const entryCount = data.getUint16(eocdOffset + 10, true)\n    if (entryCount > maxEntries) {\n      throw new KordocError(`ZIP 엔트리 수 초과: ${entryCount} (최대 ${maxEntries})`)\n    }\n\n    const cdSize = data.getUint32(eocdOffset + 12, true)\n    const cdOffset = data.getUint32(eocdOffset + 16, true)\n    if (cdOffset + cdSize > len) return { totalUncompressed: 0, entryCount }\n\n    let totalUncompressed = 0\n    let pos = cdOffset\n    for (let i = 0; i < entryCount && pos + 46 <= cdOffset + cdSize; i++) {\n      if (data.getUint32(pos, true) !== 0x02014b50) break\n      totalUncompressed += data.getUint32(pos + 24, true)\n      const nameLen = data.getUint16(pos + 28, true)\n      const extraLen = data.getUint16(pos + 30, true)\n      const commentLen = data.getUint16(pos + 32, true)\n      pos += 46 + nameLen + extraLen + commentLen\n    }\n\n    if (totalUncompressed > maxUncompressedSize) {\n      throw new KordocError(`ZIP 비압축 크기 초과: ${(totalUncompressed / 1024 / 1024).toFixed(1)}MB (최대 ${maxUncompressedSize / 1024 / 1024}MB)`)\n    }\n\n    return { totalUncompressed, entryCount }\n  } catch (err) {\n    if (err instanceof KordocError) throw err\n    return { totalUncompressed: 0, entryCount: 0 }\n  }\n}\n\n/** XXE/Billion Laughs 방지 — DOCTYPE 제거 (내부 DTD 서브셋 포함) */\nexport function stripDtd(xml: string): string {\n  return xml.replace(/<!DOCTYPE\\s[^[>]*(\\[[\\s\\S]*?\\])?\\s*>/gi, \"\")\n}\n\n/** 하이퍼링크 URL 살균 — javascript: 등 XSS 위험 스킴 차단 */\nconst SAFE_HREF_RE = /^(?:https?:|mailto:|tel:|#)/i\nexport function sanitizeHref(href: string): string | null {\n  const trimmed = href.trim()\n  if (!trimmed || !SAFE_HREF_RE.test(trimmed)) return null\n  return trimmed\n}\n\n// ─── 안전한 min/max (스택 오버플로 방지) ─────────────\n\n/** Math.min(...arr) 대체 — 대형 배열에서 스택 오버플로 방지 */\nexport function safeMin(arr: number[]): number {\n  let min = Infinity\n  for (let i = 0; i < arr.length; i++) if (arr[i] < min) min = arr[i]\n  return min\n}\n\n/** Math.max(...arr) 대체 — 대형 배열에서 스택 오버플로 방지 */\nexport function safeMax(arr: number[]): number {\n  let max = -Infinity\n  for (let i = 0; i < arr.length; i++) if (arr[i] > max) max = arr[i]\n  return max\n}\n\n// ─── 에러 분류 ──────────────────────────────────────\n\nimport type { ErrorCode } from \"./types.js\"\n\n/** 에러를 구조화된 ErrorCode로 분류 — KordocError 메시지 패턴 매칭 */\nexport function classifyError(err: unknown): ErrorCode {\n  if (!(err instanceof Error)) return \"PARSE_ERROR\"\n  const msg = err.message\n  if (msg.includes(\"암호화\")) return \"ENCRYPTED\"\n  if (msg.includes(\"DRM\")) return \"DRM_PROTECTED\"\n  if (msg.includes(\"ZIP bomb\") || msg.includes(\"ZIP 비압축 크기 초과\") || msg.includes(\"ZIP 엔트리 수 초과\")) return \"ZIP_BOMB\"\n  if (msg.includes(\"bomb\") || msg.includes(\"크기 초과\") || msg.includes(\"압축 해제\")) return \"DECOMPRESSION_BOMB\"\n  if (msg.includes(\"이미지 기반\")) return \"IMAGE_BASED_PDF\"\n  if (msg.includes(\"섹션\") && (msg.includes(\"찾을 수 없\") || msg.includes(\"없음\"))) return \"NO_SECTIONS\"\n  if (msg.includes(\"시그니처\") || msg.includes(\"복구할 수 없\")) return \"CORRUPTED\"\n  return \"PARSE_ERROR\"\n}\n","/** 2-pass colSpan/rowSpan 테이블 빌더 및 Markdown 변환 */\r\n\r\nimport type { CellContext, IRBlock, IRCell, IRTable } from \"../types.js\"\r\nimport { sanitizeHref } from \"../utils.js\"\r\n\r\n/** 테이블 열 수 상한 — 한국 공공문서 기준 충분한 값 */\r\nexport const MAX_COLS = 200\r\n/** 테이블 행 수 상한 — 메모리 폭주 방지 */\r\nexport const MAX_ROWS = 10000\r\n\r\nexport function buildTable(rows: CellContext[][]): IRTable {\r\n  if (rows.length > MAX_ROWS) rows = rows.slice(0, MAX_ROWS)\r\n  const numRows = rows.length\r\n\r\n  // colAddr/rowAddr가 있으면 직접 배치 (HWPX cellAddr, HWP5 colAddr/rowAddr)\r\n  const hasAddr = rows.some(row => row.some(c => c.colAddr !== undefined && c.rowAddr !== undefined))\r\n  if (hasAddr) return buildTableDirect(rows, numRows)\r\n\r\n  // Pass 1: maxCols 계산 — 2D 배열 사용 (동적 확장)\r\n  let maxCols = 0\r\n  const tempOccupied: boolean[][] = Array.from({ length: numRows }, () => [])\r\n\r\n  for (let rowIdx = 0; rowIdx < numRows; rowIdx++) {\r\n    let colIdx = 0\r\n    for (const cell of rows[rowIdx]) {\r\n      while (colIdx < MAX_COLS && tempOccupied[rowIdx][colIdx]) colIdx++\r\n      if (colIdx >= MAX_COLS) break\r\n\r\n      for (let r = rowIdx; r < Math.min(rowIdx + cell.rowSpan, numRows); r++) {\r\n        for (let c = colIdx; c < Math.min(colIdx + cell.colSpan, MAX_COLS); c++) {\r\n          tempOccupied[r][c] = true\r\n        }\r\n      }\r\n      colIdx += cell.colSpan\r\n      if (colIdx > maxCols) maxCols = colIdx\r\n    }\r\n  }\r\n\r\n  if (maxCols === 0) return { rows: 0, cols: 0, cells: [], hasHeader: false }\r\n\r\n  // Pass 2: 실제 배치\r\n  const grid: IRCell[][] = Array.from({ length: numRows }, () =>\r\n    Array.from({ length: maxCols }, () => ({ text: \"\", colSpan: 1, rowSpan: 1 }))\r\n  )\r\n  const occupied: boolean[][] = Array.from({ length: numRows }, () => Array(maxCols).fill(false))\r\n\r\n  for (let rowIdx = 0; rowIdx < numRows; rowIdx++) {\r\n    let colIdx = 0\r\n    let cellIdx = 0\r\n\r\n    while (colIdx < maxCols && cellIdx < rows[rowIdx].length) {\r\n      while (colIdx < maxCols && occupied[rowIdx][colIdx]) colIdx++\r\n      if (colIdx >= maxCols) break\r\n\r\n      const cell = rows[rowIdx][cellIdx]\r\n      grid[rowIdx][colIdx] = {\r\n        text: cell.text.trim(),\r\n        colSpan: cell.colSpan,\r\n        rowSpan: cell.rowSpan,\r\n      }\r\n\r\n      for (let r = rowIdx; r < Math.min(rowIdx + cell.rowSpan, numRows); r++) {\r\n        for (let c = colIdx; c < Math.min(colIdx + cell.colSpan, maxCols); c++) {\r\n          occupied[r][c] = true\r\n        }\r\n      }\r\n\r\n      colIdx += cell.colSpan\r\n      cellIdx++\r\n    }\r\n  }\r\n\r\n  return trimAndReturn(grid, numRows, maxCols)\r\n}\r\n\r\n/** colAddr/rowAddr 절대 좌표 기반 직접 배치 */\r\nfunction buildTableDirect(rows: CellContext[][], numRows: number): IRTable {\r\n  // 전체 셀에서 maxCols 계산 (MAX_COLS 상한 적용)\r\n  let maxCols = 0\r\n  for (const row of rows) {\r\n    for (const cell of row) {\r\n      const end = (cell.colAddr ?? 0) + cell.colSpan\r\n      if (end > maxCols) maxCols = end\r\n    }\r\n  }\r\n  if (maxCols > MAX_COLS) maxCols = MAX_COLS\r\n  if (maxCols === 0) return { rows: 0, cols: 0, cells: [], hasHeader: false }\r\n\r\n  const grid: IRCell[][] = Array.from({ length: numRows }, () =>\r\n    Array.from({ length: maxCols }, () => ({ text: \"\", colSpan: 1, rowSpan: 1 }))\r\n  )\r\n\r\n  for (const row of rows) {\r\n    for (const cell of row) {\r\n      const r = cell.rowAddr ?? 0\r\n      const c = cell.colAddr ?? 0\r\n      if (r >= numRows || c >= maxCols || r < 0 || c < 0) continue\r\n\r\n      grid[r][c] = { text: cell.text.trim(), colSpan: cell.colSpan, rowSpan: cell.rowSpan }\r\n\r\n      // 병합 영역 마킹\r\n      for (let dr = 0; dr < cell.rowSpan; dr++) {\r\n        for (let dc = 0; dc < cell.colSpan; dc++) {\r\n          if (dr === 0 && dc === 0) continue\r\n          if (r + dr < numRows && c + dc < maxCols) {\r\n            grid[r + dr][c + dc] = { text: \"\", colSpan: 1, rowSpan: 1 }\r\n          }\r\n        }\r\n      }\r\n    }\r\n  }\r\n\r\n  return trimAndReturn(grid, numRows, maxCols)\r\n}\r\n\r\n/** 빈 후행 열 제거 후 IRTable 반환 */\r\nfunction trimAndReturn(grid: IRCell[][], numRows: number, maxCols: number): IRTable {\r\n  let effectiveCols = maxCols\r\n  while (effectiveCols > 0) {\r\n    const colEmpty = grid.every(row => !row[effectiveCols - 1]?.text?.trim())\r\n    if (!colEmpty) break\r\n    effectiveCols--\r\n  }\r\n  if (effectiveCols < maxCols && effectiveCols > 0) {\r\n    const trimmed = grid.map(row => row.slice(0, effectiveCols))\r\n    return { rows: numRows, cols: effectiveCols, cells: trimmed, hasHeader: numRows > 1 }\r\n  }\r\n  return { rows: numRows, cols: maxCols, cells: grid, hasHeader: numRows > 1 }\r\n}\r\n\r\nexport function convertTableToText(rows: CellContext[][]): string {\r\n  return rows\r\n    .map(row =>\r\n      row\r\n        .map(c => c.text.trim().replace(/\\n/g, \" \").replace(/\\|/g, \"\\\\|\"))\r\n        .filter(Boolean)\r\n        .join(\" / \")\r\n    )\r\n    .filter(Boolean)\r\n    .join(\"\\n\")\r\n}\r\n\r\n/** 마크다운 GFM 특수문자 이스케이프 — remark-gfm 오해석 방지 */\r\nfunction escapeGfm(text: string): string {\r\n  // ~ → \\~ (GFM strikethrough 방지)\r\n  return text.replace(/~/g, \"\\\\~\")\r\n}\r\n\r\n/** HWP 자동생성 도형/개체 대체텍스트 정규식 — 한컴오피스가 삽입하는 모든 알려진 패턴 */\r\nconst HWP_SHAPE_ALT_TEXT_RE = /(?:모서리가 둥근 |둥근 )?(?:사각형|직사각형|정사각형|원|타원|삼각형|이등변 삼각형|직각 삼각형|선|직선|곡선|화살표|굵은 화살표|이중 화살표|오각형|육각형|팔각형|별|[4-8]점별|십자|십자형|구름|구름형|마름모|도넛|평행사변형|사다리꼴|부채꼴|호|반원|물결|번개|하트|빗금|블록 화살표|수식|표|그림|개체|그리기\\s?개체|묶음\\s?개체|글상자|수식\\s?개체|OLE\\s?개체)\\s?입니다\\.?/g\r\n\r\n/** HWP PUA 특수문자 및 도형 대체텍스트 제거 — 모든 포맷 공통 */\r\nfunction sanitizeText(text: string): string {\r\n  let result = text\r\n    // Supplementary Private Use Area (U+F0000-U+FFFFD) — HWP 전용 기호\r\n    .replace(/[\\u{F0000}-\\u{FFFFD}]/gu, \"\")\r\n    // HWP 도형/개체 자동생성 대체텍스트 제거\r\n    .replace(HWP_SHAPE_ALT_TEXT_RE, \"\")\r\n    .replace(/  +/g, \" \")\r\n    .trim()\r\n  // 균등배분 스페이스 정리 (\"현 장 대 응 단 장\" → \"현장대응단장\")\r\n  // 짧은 텍스트(30자 이하)에서 70%+ 토큰이 한글 1글자면 균등배분으로 판단\r\n  if (result.length <= 30 && result.includes(\" \")) {\r\n    const tokens = result.split(\" \")\r\n    // 한글 1글자 토큰만 카운트 — ASCII 특수문자(< > & 등)는 균등배분이 아님\r\n    const koreanSingleCharCount = tokens.filter(t => t.length === 1 && /[\\uAC00-\\uD7AF\\u3131-\\u318E]/.test(t)).length\r\n    if (tokens.length >= 3 && koreanSingleCharCount / tokens.length >= 0.7) {\r\n      result = tokens.join(\"\")\r\n    }\r\n  }\r\n  return result\r\n}\r\n\r\n/**\r\n * 레이아웃 테이블 감지 및 해체 — IRBlock 레벨에서 수행\r\n * 적은 행(≤3) + 셀 내 줄바꿈 다량 → table 블록을 paragraph 블록들로 분해\r\n * heading 감지 전에 호출해야 해체된 텍스트에 heading 감지 적용 가능\r\n */\r\nexport function flattenLayoutTables(blocks: IRBlock[]): IRBlock[] {\r\n  const result: IRBlock[] = []\r\n\r\n  for (const block of blocks) {\r\n    if (block.type !== \"table\" || !block.table) {\r\n      result.push(block)\r\n      continue\r\n    }\r\n\r\n    const { rows: numRows, cols: numCols, cells } = block.table\r\n\r\n    // 1x1 테이블은 기존 로직(tableToMarkdown)에서 처리\r\n    if (numRows === 1 && numCols === 1) {\r\n      result.push(block)\r\n      continue\r\n    }\r\n\r\n    // 레이아웃 테이블 휴리스틱\r\n    if (numRows <= 3) {\r\n      let totalNewlines = 0\r\n      let totalTextLen = 0\r\n      for (let r = 0; r < numRows; r++) {\r\n        for (let c = 0; c < numCols; c++) {\r\n          const t = cells[r]?.[c]?.text || \"\"\r\n          totalNewlines += (t.match(/\\n/g) || []).length\r\n          totalTextLen += t.length\r\n        }\r\n      }\r\n\r\n      // 레이아웃 테이블 판정: 많은 줄바꿈(>5), 또는 적은 행에 비해 총 텍스트 과다(>300)\r\n      if (totalNewlines > 5 || (numRows <= 2 && totalTextLen > 300)) {\r\n        // 레이아웃 테이블 → 각 셀을 paragraph 블록으로 분해\r\n        for (let r = 0; r < numRows; r++) {\r\n          for (let c = 0; c < numCols; c++) {\r\n            const cellText = cells[r]?.[c]?.text?.trim()\r\n            if (!cellText) continue\r\n            // 셀 내 줄바꿈을 별도 paragraph로 분리\r\n            for (const line of cellText.split(\"\\n\")) {\r\n              const trimmed = line.trim()\r\n              if (!trimmed) continue\r\n              result.push({ type: \"paragraph\", text: trimmed, pageNumber: block.pageNumber })\r\n            }\r\n          }\r\n        }\r\n        continue\r\n      }\r\n    }\r\n\r\n    result.push(block)\r\n  }\r\n\r\n  return result\r\n}\r\n\r\nexport function blocksToMarkdown(blocks: IRBlock[]): string {\r\n  const lines: string[] = []\r\n\r\n  for (let i = 0; i < blocks.length; i++) {\r\n    const block = blocks[i]\r\n\r\n    // 헤딩 블록\r\n    if (block.type === \"heading\" && block.text) {\r\n      const prefix = \"#\".repeat(Math.min(block.level || 2, 6))\r\n      const headingText = sanitizeText(block.text)\r\n      if (headingText) lines.push(\"\", `${prefix} ${headingText}`, \"\")\r\n      continue\r\n    }\r\n\r\n    // 이미지 블록 — ![alt](filename) 참조\r\n    if (block.type === \"image\" && block.text) {\r\n      lines.push(\"\", `![image](${block.text})`, \"\")\r\n      continue\r\n    }\r\n\r\n    // 구분선 블록\r\n    if (block.type === \"separator\") {\r\n      lines.push(\"\", \"---\", \"\")\r\n      continue\r\n    }\r\n\r\n    // 리스트 블록\r\n    if (block.type === \"list\" && block.text) {\r\n      const listText = sanitizeText(block.text)\r\n      if (!listText) continue\r\n      // 텍스트가 이미 번호로 시작하면 그대로 출력 (원래 번호 보존)\r\n      const alreadyNumbered = block.listType === \"ordered\" && /^\\d+\\.\\s/.test(listText)\r\n      const prefix = alreadyNumbered ? \"\" : block.listType === \"ordered\" ? \"1. \" : \"- \"\r\n      lines.push(`${prefix}${listText}`)\r\n      if (block.children) {\r\n        for (const child of block.children) {\r\n          const childPrefix = child.listType === \"ordered\" ? \"1.\" : \"-\"\r\n          lines.push(`  ${childPrefix} ${child.text || \"\"}`)\r\n        }\r\n      }\r\n      continue\r\n    }\r\n\r\n    if (block.type === \"paragraph\" && block.text) {\r\n      let text = sanitizeText(block.text)\r\n      if (!text) continue\r\n\r\n      // 별표 패턴 (기존 호환)\r\n      if (/^\\[별표\\s*\\d+/.test(text)) {\r\n        const nextBlock = blocks[i + 1]\r\n        if (nextBlock?.type === \"paragraph\" && nextBlock.text && /관련\\)?$/.test(nextBlock.text)) {\r\n          lines.push(\"\", `## ${text} ${nextBlock.text}`, \"\")\r\n          i++\r\n        } else {\r\n          lines.push(\"\", `## ${text}`, \"\")\r\n        }\r\n        continue\r\n      }\r\n\r\n      if (/^\\([^)]*조[^)]*관련\\)$/.test(text)) {\r\n        lines.push(`*${text}*`, \"\")\r\n        continue\r\n      }\r\n\r\n      // 하이퍼링크가 있으면 텍스트에 링크 삽입 (javascript: 등 위험 스킴 제거)\r\n      if (block.href) {\r\n        const href = sanitizeHref(block.href)\r\n        if (href) text = `[${text}](${href})`\r\n      }\r\n\r\n      // 각주가 있으면 괄호로 인라인 삽입\r\n      if (block.footnoteText) {\r\n        text += ` (주: ${block.footnoteText})`\r\n      }\r\n\r\n      lines.push(escapeGfm(text), \"\")\r\n    } else if (block.type === \"table\" && block.table) {\r\n      // 테이블 앞에 빈 줄 보장 (마크다운 렌더링 필수)\r\n      if (lines.length > 0 && lines[lines.length - 1] !== \"\") {\r\n        lines.push(\"\")\r\n      }\r\n      const tableMd = tableToMarkdown(block.table)\r\n      if (tableMd) {\r\n        lines.push(tableMd)\r\n        lines.push(\"\")\r\n      }\r\n    }\r\n  }\r\n\r\n  return lines.join(\"\\n\").trim()\r\n}\r\n\r\n/** 병합 셀 존재 여부 확인 */\r\nfunction hasMergedCells(table: IRTable): boolean {\r\n  for (const row of table.cells) {\r\n    for (const cell of row) {\r\n      if (cell.colSpan > 1 || cell.rowSpan > 1) return true\r\n    }\r\n  }\r\n  return false\r\n}\r\n\r\nfunction containsInlineMath(text: string): boolean {\r\n  return /(^|[^\\\\])\\$(?=\\S)(?:\\\\.|[^$\\n])+?\\S\\$/.test(text)\r\n}\r\n\r\nfunction tableContainsInlineMath(table: IRTable): boolean {\r\n  for (const row of table.cells) {\r\n    for (const cell of row) {\r\n      if (containsInlineMath(cell.text)) return true\r\n    }\r\n  }\r\n  return false\r\n}\r\n\r\n/** 병합 테이블 → HTML <table> 출력 (rowspan/colspan 보존) */\r\nfunction tableToHtml(table: IRTable): string {\r\n  const { cells, rows: numRows, cols: numCols } = table\r\n  const skip = new Set<string>()\r\n  const lines: string[] = [\"<table>\"]\r\n\r\n  for (let r = 0; r < numRows; r++) {\r\n    const tag = r === 0 ? \"th\" : \"td\"\r\n    const rowHtml: string[] = []\r\n    for (let c = 0; c < numCols; c++) {\r\n      if (skip.has(`${r},${c}`)) continue\r\n      const cell = cells[r]?.[c]\r\n      if (!cell) continue\r\n\r\n      // 병합 영역 skip 마킹\r\n      for (let dr = 0; dr < cell.rowSpan; dr++) {\r\n        for (let dc = 0; dc < cell.colSpan; dc++) {\r\n          if (dr === 0 && dc === 0) continue\r\n          if (r + dr < numRows && c + dc < numCols) skip.add(`${r + dr},${c + dc}`)\r\n        }\r\n      }\r\n\r\n      const text = sanitizeText(cell.text).replace(/\\n/g, \"<br>\")\r\n      const attrs: string[] = []\r\n      if (cell.colSpan > 1) attrs.push(`colspan=\"${cell.colSpan}\"`)\r\n      if (cell.rowSpan > 1) attrs.push(`rowspan=\"${cell.rowSpan}\"`)\r\n      const attrStr = attrs.length ? \" \" + attrs.join(\" \") : \"\"\r\n      rowHtml.push(`<${tag}${attrStr}>${text}</${tag}>`)\r\n    }\r\n    if (rowHtml.length) lines.push(`<tr>${rowHtml.join(\"\")}</tr>`)\r\n  }\r\n\r\n  lines.push(\"</table>\")\r\n  return lines.join(\"\\n\")\r\n}\r\n\r\nfunction tableToMarkdown(table: IRTable): string {\r\n  if (table.rows === 0 || table.cols === 0) return \"\"\r\n\r\n  const { cells, rows: numRows, cols: numCols } = table\r\n\r\n  // 병합 셀이 있으면 HTML 테이블로 출력하되, 수식이 있으면 GFM 표로 출력한다.\r\n  // 많은 Markdown 렌더러가 raw HTML table 내부의 $...$를 수식으로 다시 처리하지 않는다.\r\n  if (hasMergedCells(table) && !tableContainsInlineMath(table)) return tableToHtml(table)\r\n\r\n  // 1행 1열 → 구조화된 텍스트 (빈 셀이면 스킵)\r\n  if (numRows === 1 && numCols === 1) {\r\n    const content = sanitizeText(cells[0][0].text)\r\n    if (!content) return \"\"\r\n    return content\r\n      .split(/\\n/)\r\n      .map(line => {\r\n        const trimmed = line.trim()\r\n        if (!trimmed) return \"\"\r\n        if (/^\\d+\\.\\s/.test(trimmed)) return `**${escapeGfm(trimmed)}**`\r\n        if (/^[가-힣]\\.\\s/.test(trimmed)) return `  ${escapeGfm(trimmed)}`\r\n        return escapeGfm(trimmed)\r\n      })\r\n      .filter(Boolean)\r\n      .join(\"\\n\")\r\n  }\r\n\r\n  // 1열 다행 테이블 → 각 행을 별도 라인으로 출력 (목록성 데이터)\r\n  if (numCols === 1 && numRows >= 2) {\r\n    return cells\r\n      .map(row => escapeGfm(sanitizeText(row[0].text)).replace(/\\n/g, \" \"))\r\n      .filter(Boolean)\r\n      .join(\"\\n\")\r\n  }\r\n\r\n  // 병합 셀: 행/열 병합된 셀은 빈 칸으로\r\n  const display: string[][] = Array.from({ length: numRows }, () => Array(numCols).fill(\"\"))\r\n  const skip = new Set<string>()\r\n\r\n  for (let r = 0; r < numRows; r++) {\r\n    for (let c = 0; c < numCols; c++) {\r\n      if (skip.has(`${r},${c}`)) continue\r\n      const cell = cells[r]?.[c]\r\n      if (!cell) continue\r\n      display[r][c] = escapeGfm(sanitizeText(cell.text)).replace(/\\|/g, \"\\\\|\").replace(/\\n/g, \"<br>\")\r\n\r\n      // colSpan/rowSpan: 병합된 열은 빈 칸으로 유지 (텍스트 중복 방지)\r\n      for (let dr = 0; dr < cell.rowSpan; dr++) {\r\n        for (let dc = 0; dc < cell.colSpan; dc++) {\r\n          if (dr === 0 && dc === 0) continue\r\n          if (r + dr < numRows && c + dc < numCols) {\r\n            skip.add(`${r + dr},${c + dc}`)\r\n          }\r\n        }\r\n      }\r\n      // colSpan > 1이면 display 열 인덱스를 건너뜀\r\n      c += cell.colSpan - 1\r\n    }\r\n  }\r\n\r\n  // rowSpan 잔류 처리:\r\n  // 1) 완전 빈 행 제거\r\n  // 2) \"첫 열만 값, 나머지 빈\" 행 → 다음 데이터 행의 첫 열에 값을 전파\r\n  //    단, colSpan으로 인한 빈 열(skip 셀)은 이 대상이 아님\r\n  const uniqueRows: string[][] = []\r\n  let pendingFirstCol = \"\"\r\n  for (let r = 0; r < display.length; r++) {\r\n    const row = display[r]\r\n    const isEmptyPlaceholder = row.every(cell => cell === \"\")\r\n    if (isEmptyPlaceholder) continue\r\n\r\n    // 첫 열만 값이 있고 나머지 모두 빈 행 → 다음 데이터 행의 첫 열에 전파\r\n    // 단, colSpan으로 인한 빈 열(skip 셀)은 \"진짜 빈\"이 아니므로 제외\r\n    const nonEmptyCols = row.filter(cell => cell !== \"\")\r\n    const hasSkipInRow = row.some((_, c) => skip.has(`${r},${c}`))\r\n    if (!hasSkipInRow && nonEmptyCols.length === 1 && row[0] !== \"\" && row.slice(1).every(c => c === \"\")) {\r\n      pendingFirstCol = row[0]\r\n      continue\r\n    }\r\n\r\n    // 저장된 첫 열 값을 현재 행의 빈 첫 열에 전파\r\n    if (pendingFirstCol && row[0] === \"\") {\r\n      row[0] = pendingFirstCol\r\n      pendingFirstCol = \"\"\r\n    } else {\r\n      pendingFirstCol = \"\"\r\n    }\r\n    uniqueRows.push(row)\r\n  }\r\n\r\n  if (uniqueRows.length === 0) return \"\"\r\n\r\n  const md: string[] = []\r\n  md.push(\"| \" + uniqueRows[0].join(\" | \") + \" |\")\r\n  md.push(\"| \" + uniqueRows[0].map(() => \"---\").join(\" | \") + \" |\")\r\n  for (let i = 1; i < uniqueRows.length; i++) {\r\n    md.push(\"| \" + uniqueRows[i].join(\" | \") + \" |\")\r\n  }\r\n  return md.join(\"\\n\")\r\n}\r\n","/** kordoc 공통 타입 정의 */\r\n\r\n// ─── 중간 표현 (Intermediate Representation) ─────────\r\n\r\nexport interface CellContext {\r\n  text: string\r\n  colSpan: number\r\n  rowSpan: number\r\n  /** HWP5 셀 열 주소 (0-based) — 병합 테이블 배치용 */\r\n  colAddr?: number\r\n  /** HWP5 셀 행 주소 (0-based) — 병합 테이블 배치용 */\r\n  rowAddr?: number\r\n}\r\n\r\n/** 블록 타입 — v2.0에서 heading, list, image, separator 추가 */\r\nexport type IRBlockType = \"paragraph\" | \"table\" | \"heading\" | \"list\" | \"image\" | \"separator\"\r\n\r\nexport interface IRBlock {\r\n  type: IRBlockType\r\n  text?: string\r\n  table?: IRTable\r\n  /** 헤딩 레벨 (1-6), type=\"heading\"일 때 사용 */\r\n  level?: number\r\n  /** 원본 페이지 번호 (1-based) */\r\n  pageNumber?: number\r\n  /** 바운딩 박스 — PDF에서만 제공 */\r\n  bbox?: BoundingBox\r\n  /** 텍스트 스타일 정보 (선택) */\r\n  style?: InlineStyle\r\n  /** 리스트 타입, type=\"list\"일 때 사용 */\r\n  listType?: \"ordered\" | \"unordered\"\r\n  /** 중첩 리스트 아이템 */\r\n  children?: IRBlock[]\r\n  /** 하이퍼링크 URL */\r\n  href?: string\r\n  /** 각주/미주 텍스트 (인라인 삽입용) */\r\n  footnoteText?: string\r\n  /** 이미지 데이터 (type=\"image\"일 때) */\r\n  imageData?: ImageData\r\n}\r\n\r\n/** 추출된 이미지 바이너리 데이터 */\r\nexport interface ImageData {\r\n  /** 이미지 바이너리 */\r\n  data: Uint8Array\r\n  /** MIME 타입 (image/png, image/jpeg, image/gif, image/bmp, image/wmf, image/emf) */\r\n  mimeType: string\r\n  /** 원본 파일명 (있는 경우) */\r\n  filename?: string\r\n}\r\n\r\n/** 바운딩 박스 — PDF 포인트 단위 (72pt = 1인치) */\r\nexport interface BoundingBox {\r\n  page: number\r\n  x: number\r\n  y: number\r\n  width: number\r\n  height: number\r\n}\r\n\r\n/** 인라인 텍스트 스타일 */\r\nexport interface InlineStyle {\r\n  bold?: boolean\r\n  italic?: boolean\r\n  fontSize?: number\r\n  fontName?: string\r\n}\r\n\r\nexport interface IRTable {\r\n  rows: number\r\n  cols: number\r\n  cells: IRCell[][]\r\n  /** 첫 행을 헤더로 렌더링할지 여부 (현재: rows > 1이면 true — 의미적 감지가 아닌 레이아웃 힌트) */\r\n  hasHeader: boolean\r\n}\r\n\r\nexport interface IRCell {\r\n  text: string\r\n  colSpan: number\r\n  rowSpan: number\r\n}\r\n\r\n// ─── 메타데이터 ─────────────────────────────────────\r\n\r\n/** 문서 메타데이터 — 각 포맷에서 추출 가능한 필드만 채워짐 */\r\nexport interface DocumentMetadata {\r\n  /** 문서 제목 */\r\n  title?: string\r\n  /** 작성자 */\r\n  author?: string\r\n  /** 작성 프로그램 (예: \"한글 2020\", \"Adobe Acrobat\") */\r\n  creator?: string\r\n  /** 생성일시 (ISO 8601) */\r\n  createdAt?: string\r\n  /** 수정일시 (ISO 8601) */\r\n  modifiedAt?: string\r\n  /** 페이지/섹션 수 */\r\n  pageCount?: number\r\n  /** 문서 포맷 버전 (예: HWP \"5.1.0.1\") */\r\n  version?: string\r\n  /** 설명 */\r\n  description?: string\r\n  /** 키워드 */\r\n  keywords?: string[]\r\n}\r\n\r\n// ─── 파싱 옵션 ──────────────────────────────────────\r\n\r\n/** 파싱 옵션 — parse() 함수에 전달 */\r\nexport interface ParseOptions {\r\n  /**\r\n   * 파싱할 페이지/섹션 범위 (1-based).\r\n   * - 배열: [1, 2, 3]\r\n   * - 문자열: \"1-3\", \"1,3,5-7\"\r\n   *\r\n   * PDF: 정확한 페이지 단위. HWP/HWPX: 섹션 단위 근사치.\r\n   */\r\n  pages?: number[] | string\r\n  /** 이미지 기반 PDF용 OCR 프로바이더 (선택) */\r\n  ocr?: OcrProvider\r\n  /** 진행률 콜백 — current: 현재 페이지/섹션, total: 전체 수 */\r\n  onProgress?: (current: number, total: number) => void\r\n  /** PDF 머리글/바닥글 자동 제거 */\r\n  removeHeaderFooter?: boolean\r\n  /** 원본 파일 경로 (DRM COM fallback에 필요, 내부 전용) */\r\n  filePath?: string\r\n  /**\r\n   * PDF 수식 OCR 활성화 (기본 false).\r\n   *\r\n   * 활성화 시 각 PDF 페이지를 이미지로 렌더링 → YOLOv8 기반 수식 영역 검출 →\r\n   * TrOCR 기반 LaTeX 인식. 감지된 수식은 `$...$` (inline) / `$$...$$` (display) 로\r\n   * 블록 텍스트에 삽입된다.\r\n   *\r\n   * 필수 optional 의존성: `onnxruntime-node`, `@huggingface/transformers`,\r\n   * `@hyzyla/pdfium`, `sharp`. 미설치 시 parse 에 실패하지 않고 **경고만** 남기고\r\n   * 수식 인식은 skip 한다 (일반 텍스트 추출은 정상 동작).\r\n   *\r\n   * 모델(~155MB) 은 첫 사용 시 HuggingFace 에서 자동 다운로드 되어\r\n   * `~/.cache/kordoc/models/pix2text/` 에 SHA-256 검증과 함께 저장된다.\r\n   */\r\n  formulaOcr?: boolean\r\n}\r\n\r\n// ─── 파싱 경고 ──────────────────────────────────────\r\n\r\n/** 파싱 중 스킵/실패한 요소 보고 */\r\nexport interface ParseWarning {\r\n  /** 관련 페이지 번호 (알 수 있는 경우) */\r\n  page?: number\r\n  /** 경고 메시지 */\r\n  message: string\r\n  /** 구조화된 경고 코드 */\r\n  code: WarningCode\r\n}\r\n\r\nexport type WarningCode =\r\n  | \"SKIPPED_IMAGE\"\r\n  | \"SKIPPED_OLE\"\r\n  | \"TRUNCATED_TABLE\"\r\n  | \"OCR_FALLBACK\"\r\n  | \"UNSUPPORTED_ELEMENT\"\r\n  | \"BROKEN_ZIP_RECOVERY\"\r\n  | \"HIDDEN_TEXT_FILTERED\"\r\n  | \"MALFORMED_XML\"\r\n  | \"PARTIAL_PARSE\"\r\n  | \"LENIENT_CFB_RECOVERY\"\r\n\r\n/** 문서 구조 (헤딩 트리) */\r\nexport interface OutlineItem {\r\n  level: number\r\n  text: string\r\n  pageNumber?: number\r\n}\r\n\r\n// ─── 에러 코드 ──────────────────────────────────────\r\n\r\n/** 구조화된 에러 코드 — 프로그래밍적 에러 핸들링용 */\r\nexport type ErrorCode =\r\n  | \"EMPTY_INPUT\"\r\n  | \"UNSUPPORTED_FORMAT\"\r\n  | \"ENCRYPTED\"\r\n  | \"DRM_PROTECTED\"\r\n  | \"CORRUPTED\"\r\n  | \"DECOMPRESSION_BOMB\"\r\n  | \"ZIP_BOMB\"\r\n  | \"IMAGE_BASED_PDF\"\r\n  | \"NO_SECTIONS\"\r\n  | \"PARSE_ERROR\"\r\n  | \"MISSING_DEPENDENCY\"\r\n\r\n// ─── 파싱 결과 (discriminated union) ────────────────\r\n\r\nexport type FileType = \"hwpx\" | \"hwp\" | \"hwp3\" | \"hwpml\" | \"pdf\" | \"xlsx\" | \"xls\" | \"docx\" | \"unknown\"\r\n\r\ninterface ParseResultBase {\r\n  fileType: FileType\r\n  /** 페이지/섹션 수 — PDF: 실제 페이지 수, HWP/HWPX: 섹션 수, XLSX: 시트 수 */\r\n  pageCount?: number\r\n  /** 이미지 기반 PDF 여부 (텍스트 추출 불가) */\r\n  isImageBased?: boolean\r\n}\r\n\r\nexport interface ParseSuccess extends ParseResultBase {\r\n  success: true\r\n  /** 추출된 마크다운 텍스트 */\r\n  markdown: string\r\n  /** 중간 표현 블록 (구조화된 데이터 접근용) */\r\n  blocks: IRBlock[]\r\n  /** 문서 메타데이터 */\r\n  metadata?: DocumentMetadata\r\n  /** 문서 구조 (헤딩 트리) — v2.0 */\r\n  outline?: OutlineItem[]\r\n  /** 파싱 중 발생한 경고 — v2.0 */\r\n  warnings?: ParseWarning[]\r\n  /** 추출된 이미지 목록 — 마크다운에서 파일명으로 참조됨 */\r\n  images?: ExtractedImage[]\r\n}\r\n\r\n/** 추출된 이미지 — ParseSuccess.images에 포함 */\r\nexport interface ExtractedImage {\r\n  /** 마크다운에서 참조되는 파일명 (예: image_001.png) */\r\n  filename: string\r\n  /** 이미지 바이너리 */\r\n  data: Uint8Array\r\n  /** MIME 타입 */\r\n  mimeType: string\r\n}\r\n\r\nexport interface ParseFailure extends ParseResultBase {\r\n  success: false\r\n  /** 오류 메시지 */\r\n  error: string\r\n  /** 구조화된 에러 코드 */\r\n  code?: ErrorCode\r\n}\r\n\r\nexport type ParseResult = ParseSuccess | ParseFailure\r\n\r\n// ─── 문서 비교 (Diff) ───────────────────────────────\r\n\r\nexport type DiffChangeType = \"added\" | \"removed\" | \"modified\" | \"unchanged\"\r\n\r\nexport interface BlockDiff {\r\n  type: DiffChangeType\r\n  /** 원본 블록 (added이면 undefined) */\r\n  before?: IRBlock\r\n  /** 변경 후 블록 (removed이면 undefined) */\r\n  after?: IRBlock\r\n  /** modified 테이블의 셀 단위 diff */\r\n  cellDiffs?: CellDiff[][]\r\n  /** 유사도 (0-1) */\r\n  similarity?: number\r\n}\r\n\r\nexport interface CellDiff {\r\n  type: DiffChangeType\r\n  before?: string\r\n  after?: string\r\n}\r\n\r\nexport interface DiffResult {\r\n  stats: { added: number; removed: number; modified: number; unchanged: number }\r\n  diffs: BlockDiff[]\r\n}\r\n\r\n// ─── 양식 인식 ──────────────────────────────────────\r\n\r\nexport interface FormField {\r\n  label: string\r\n  value: string\r\n  /** 0-based 소스 행 */\r\n  row: number\r\n  /** 0-based 소스 열 */\r\n  col: number\r\n}\r\n\r\nexport interface FormResult {\r\n  fields: FormField[]\r\n  /** 양식 확신도 (0-1) */\r\n  confidence: number\r\n}\r\n\r\n// ─── OCR 프로바이더 ─────────────────────────────────\r\n\r\n/** 사용자 제공 OCR 함수 — 페이지 이미지를 받아 텍스트 반환 */\r\nexport type OcrProvider = (\r\n  pageImage: Uint8Array,\r\n  pageNumber: number,\r\n  mimeType: \"image/png\"\r\n) => Promise<string>\r\n\r\n// ─── Watch 모드 ─────────────────────────────────────\r\n\r\nexport interface WatchOptions {\r\n  dir: string\r\n  outDir?: string\r\n  webhook?: string\r\n  format?: \"markdown\" | \"json\"\r\n  pages?: string\r\n  silent?: boolean\r\n}\r\n\r\n// ─── 헤딩 감지 공통 임계값 ──────────────────────────\r\n\r\n/** 폰트 크기 비율 → heading level (전 파서 공통) */\r\nexport const HEADING_RATIO_H1 = 1.5\r\nexport const HEADING_RATIO_H2 = 1.3\r\nexport const HEADING_RATIO_H3 = 1.15\r\n\r\n// ─── 내부 파서 반환 타입 ─────────────────────────────\r\n\r\n/** 내부 파서가 index.ts에 반환하는 공통 타입 (HWP5/HWPX/PDF/XLSX/DOCX) */\r\nexport interface InternalParseResult {\r\n  markdown: string\r\n  blocks: IRBlock[]\r\n  metadata?: DocumentMetadata\r\n  outline?: OutlineItem[]\r\n  warnings?: ParseWarning[]\r\n  images?: ExtractedImage[]\r\n  /** PDF 전용: 이미지 기반 PDF 여부 */\r\n  isImageBased?: boolean\r\n}\r\n"],"mappings":";AAIO,IAAM,UAAkB,OAA4C,UAAqB;AAOzF,SAAS,cAAc,KAA0B;AACtD,MAAI,IAAI,eAAe,KAAK,IAAI,eAAe,IAAI,OAAO,YAAY;AACpE,WAAO,IAAI;AAAA,EACb;AACA,SAAO,IAAI,OAAO,MAAM,IAAI,YAAY,IAAI,aAAa,IAAI,UAAU;AACzE;AAMO,IAAM,cAAN,cAA0B,MAAM;AAAA,EACrC,YAAY,SAAiB;AAC3B,UAAM,OAAO;AACb,SAAK,OAAO;AAAA,EACd;AACF;AAeO,SAAS,gBAAgB,MAAuB;AACrD,MAAI,KAAK,SAAS,IAAM,EAAG,QAAO;AAClC,QAAM,aAAa,KAAK,QAAQ,OAAO,GAAG;AAC1C,QAAM,WAAW,WAAW,MAAM,GAAG;AACrC,SAAO,SAAS,KAAK,OAAK,MAAM,IAAI,KAAK,WAAW,WAAW,GAAG,KAAK,aAAa,KAAK,UAAU;AACrG;AAQO,SAAS,gBACd,QACA,sBAAsB,MAAM,OAAO,MACnC,aAAa,KACsC;AACnD,MAAI;AACF,UAAM,OAAO,IAAI,SAAS,MAAM;AAChC,UAAM,MAAM,OAAO;AAEnB,QAAI,aAAa;AACjB,aAAS,IAAI,MAAM,IAAI,KAAK,KAAK,IAAI,GAAG,MAAM,KAAK,GAAG,KAAK;AACzD,UAAI,KAAK,UAAU,GAAG,IAAI,MAAM,WAAY;AAAE,qBAAa;AAAG;AAAA,MAAM;AAAA,IACtE;AACA,QAAI,aAAa,EAAG,QAAO,EAAE,mBAAmB,GAAG,YAAY,EAAE;AAEjE,UAAM,aAAa,KAAK,UAAU,aAAa,IAAI,IAAI;AACvD,QAAI,aAAa,YAAY;AAC3B,YAAM,IAAI,YAAY,+CAAiB,UAAU,kBAAQ,UAAU,GAAG;AAAA,IACxE;AAEA,UAAM,SAAS,KAAK,UAAU,aAAa,IAAI,IAAI;AACnD,UAAM,WAAW,KAAK,UAAU,aAAa,IAAI,IAAI;AACrD,QAAI,WAAW,SAAS,IAAK,QAAO,EAAE,mBAAmB,GAAG,WAAW;AAEvE,QAAI,oBAAoB;AACxB,QAAI,MAAM;AACV,aAAS,IAAI,GAAG,IAAI,cAAc,MAAM,MAAM,WAAW,QAAQ,KAAK;AACpE,UAAI,KAAK,UAAU,KAAK,IAAI,MAAM,SAAY;AAC9C,2BAAqB,KAAK,UAAU,MAAM,IAAI,IAAI;AAClD,YAAM,UAAU,KAAK,UAAU,MAAM,IAAI,IAAI;AAC7C,YAAM,WAAW,KAAK,UAAU,MAAM,IAAI,IAAI;AAC9C,YAAM,aAAa,KAAK,UAAU,MAAM,IAAI,IAAI;AAChD,aAAO,KAAK,UAAU,WAAW;AAAA,IACnC;AAEA,QAAI,oBAAoB,qBAAqB;AAC3C,YAAM,IAAI,YAAY,sDAAmB,oBAAoB,OAAO,MAAM,QAAQ,CAAC,CAAC,oBAAU,sBAAsB,OAAO,IAAI,KAAK;AAAA,IACtI;AAEA,WAAO,EAAE,mBAAmB,WAAW;AAAA,EACzC,SAAS,KAAK;AACZ,QAAI,eAAe,YAAa,OAAM;AACtC,WAAO,EAAE,mBAAmB,GAAG,YAAY,EAAE;AAAA,EAC/C;AACF;AAGO,SAAS,SAAS,KAAqB;AAC5C,SAAO,IAAI,QAAQ,0CAA0C,EAAE;AACjE;AAGA,IAAM,eAAe;AACd,SAAS,aAAa,MAA6B;AACxD,QAAM,UAAU,KAAK,KAAK;AAC1B,MAAI,CAAC,WAAW,CAAC,aAAa,KAAK,OAAO,EAAG,QAAO;AACpD,SAAO;AACT;AAKO,SAAS,QAAQ,KAAuB;AAC7C,MAAI,MAAM;AACV,WAAS,IAAI,GAAG,IAAI,IAAI,QAAQ,IAAK,KAAI,IAAI,CAAC,IAAI,IAAK,OAAM,IAAI,CAAC;AAClE,SAAO;AACT;AAGO,SAAS,QAAQ,KAAuB;AAC7C,MAAI,MAAM;AACV,WAAS,IAAI,GAAG,IAAI,IAAI,QAAQ,IAAK,KAAI,IAAI,CAAC,IAAI,IAAK,OAAM,IAAI,CAAC;AAClE,SAAO;AACT;AAOO,SAAS,cAAc,KAAyB;AACrD,MAAI,EAAE,eAAe,OAAQ,QAAO;AACpC,QAAM,MAAM,IAAI;AAChB,MAAI,IAAI,SAAS,oBAAK,EAAG,QAAO;AAChC,MAAI,IAAI,SAAS,KAAK,EAAG,QAAO;AAChC,MAAI,IAAI,SAAS,UAAU,KAAK,IAAI,SAAS,kDAAe,KAAK,IAAI,SAAS,4CAAc,EAAG,QAAO;AACtG,MAAI,IAAI,SAAS,MAAM,KAAK,IAAI,SAAS,2BAAO,KAAK,IAAI,SAAS,2BAAO,EAAG,QAAO;AACnF,MAAI,IAAI,SAAS,iCAAQ,EAAG,QAAO;AACnC,MAAI,IAAI,SAAS,cAAI,MAAM,IAAI,SAAS,4BAAQ,KAAK,IAAI,SAAS,cAAI,GAAI,QAAO;AACjF,MAAI,IAAI,SAAS,0BAAM,KAAK,IAAI,SAAS,kCAAS,EAAG,QAAO;AAC5D,SAAO;AACT;;;AC5IO,IAAM,WAAW;AAEjB,IAAM,WAAW;AAEjB,SAAS,WAAW,MAAgC;AACzD,MAAI,KAAK,SAAS,SAAU,QAAO,KAAK,MAAM,GAAG,QAAQ;AACzD,QAAM,UAAU,KAAK;AAGrB,QAAM,UAAU,KAAK,KAAK,SAAO,IAAI,KAAK,OAAK,EAAE,YAAY,UAAa,EAAE,YAAY,MAAS,CAAC;AAClG,MAAI,QAAS,QAAO,iBAAiB,MAAM,OAAO;AAGlD,MAAI,UAAU;AACd,QAAM,eAA4B,MAAM,KAAK,EAAE,QAAQ,QAAQ,GAAG,MAAM,CAAC,CAAC;AAE1E,WAAS,SAAS,GAAG,SAAS,SAAS,UAAU;AAC/C,QAAI,SAAS;AACb,eAAW,QAAQ,KAAK,MAAM,GAAG;AAC/B,aAAO,SAAS,YAAY,aAAa,MAAM,EAAE,MAAM,EAAG;AAC1D,UAAI,UAAU,SAAU;AAExB,eAAS,IAAI,QAAQ,IAAI,KAAK,IAAI,SAAS,KAAK,SAAS,OAAO,GAAG,KAAK;AACtE,iBAAS,IAAI,QAAQ,IAAI,KAAK,IAAI,SAAS,KAAK,SAAS,QAAQ,GAAG,KAAK;AACvE,uBAAa,CAAC,EAAE,CAAC,IAAI;AAAA,QACvB;AAAA,MACF;AACA,gBAAU,KAAK;AACf,UAAI,SAAS,QAAS,WAAU;AAAA,IAClC;AAAA,EACF;AAEA,MAAI,YAAY,EAAG,QAAO,EAAE,MAAM,GAAG,MAAM,GAAG,OAAO,CAAC,GAAG,WAAW,MAAM;AAG1E,QAAM,OAAmB,MAAM;AAAA,IAAK,EAAE,QAAQ,QAAQ;AAAA,IAAG,MACvD,MAAM,KAAK,EAAE,QAAQ,QAAQ,GAAG,OAAO,EAAE,MAAM,IAAI,SAAS,GAAG,SAAS,EAAE,EAAE;AAAA,EAC9E;AACA,QAAM,WAAwB,MAAM,KAAK,EAAE,QAAQ,QAAQ,GAAG,MAAM,MAAM,OAAO,EAAE,KAAK,KAAK,CAAC;AAE9F,WAAS,SAAS,GAAG,SAAS,SAAS,UAAU;AAC/C,QAAI,SAAS;AACb,QAAI,UAAU;AAEd,WAAO,SAAS,WAAW,UAAU,KAAK,MAAM,EAAE,QAAQ;AACxD,aAAO,SAAS,WAAW,SAAS,MAAM,EAAE,MAAM,EAAG;AACrD,UAAI,UAAU,QAAS;AAEvB,YAAM,OAAO,KAAK,MAAM,EAAE,OAAO;AACjC,WAAK,MAAM,EAAE,MAAM,IAAI;AAAA,QACrB,MAAM,KAAK,KAAK,KAAK;AAAA,QACrB,SAAS,KAAK;AAAA,QACd,SAAS,KAAK;AAAA,MAChB;AAEA,eAAS,IAAI,QAAQ,IAAI,KAAK,IAAI,SAAS,KAAK,SAAS,OAAO,GAAG,KAAK;AACtE,iBAAS,IAAI,QAAQ,IAAI,KAAK,IAAI,SAAS,KAAK,SAAS,OAAO,GAAG,KAAK;AACtE,mBAAS,CAAC,EAAE,CAAC,IAAI;AAAA,QACnB;AAAA,MACF;AAEA,gBAAU,KAAK;AACf;AAAA,IACF;AAAA,EACF;AAEA,SAAO,cAAc,MAAM,SAAS,OAAO;AAC7C;AAGA,SAAS,iBAAiB,MAAuB,SAA0B;AAEzE,MAAI,UAAU;AACd,aAAW,OAAO,MAAM;AACtB,eAAW,QAAQ,KAAK;AACtB,YAAM,OAAO,KAAK,WAAW,KAAK,KAAK;AACvC,UAAI,MAAM,QAAS,WAAU;AAAA,IAC/B;AAAA,EACF;AACA,MAAI,UAAU,SAAU,WAAU;AAClC,MAAI,YAAY,EAAG,QAAO,EAAE,MAAM,GAAG,MAAM,GAAG,OAAO,CAAC,GAAG,WAAW,MAAM;AAE1E,QAAM,OAAmB,MAAM;AAAA,IAAK,EAAE,QAAQ,QAAQ;AAAA,IAAG,MACvD,MAAM,KAAK,EAAE,QAAQ,QAAQ,GAAG,OAAO,EAAE,MAAM,IAAI,SAAS,GAAG,SAAS,EAAE,EAAE;AAAA,EAC9E;AAEA,aAAW,OAAO,MAAM;AACtB,eAAW,QAAQ,KAAK;AACtB,YAAM,IAAI,KAAK,WAAW;AAC1B,YAAM,IAAI,KAAK,WAAW;AAC1B,UAAI,KAAK,WAAW,KAAK,WAAW,IAAI,KAAK,IAAI,EAAG;AAEpD,WAAK,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,KAAK,KAAK,GAAG,SAAS,KAAK,SAAS,SAAS,KAAK,QAAQ;AAGpF,eAAS,KAAK,GAAG,KAAK,KAAK,SAAS,MAAM;AACxC,iBAAS,KAAK,GAAG,KAAK,KAAK,SAAS,MAAM;AACxC,cAAI,OAAO,KAAK,OAAO,EAAG;AAC1B,cAAI,IAAI,KAAK,WAAW,IAAI,KAAK,SAAS;AACxC,iBAAK,IAAI,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,MAAM,IAAI,SAAS,GAAG,SAAS,EAAE;AAAA,UAC5D;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,SAAO,cAAc,MAAM,SAAS,OAAO;AAC7C;AAGA,SAAS,cAAc,MAAkB,SAAiB,SAA0B;AAClF,MAAI,gBAAgB;AACpB,SAAO,gBAAgB,GAAG;AACxB,UAAM,WAAW,KAAK,MAAM,SAAO,CAAC,IAAI,gBAAgB,CAAC,GAAG,MAAM,KAAK,CAAC;AACxE,QAAI,CAAC,SAAU;AACf;AAAA,EACF;AACA,MAAI,gBAAgB,WAAW,gBAAgB,GAAG;AAChD,UAAM,UAAU,KAAK,IAAI,SAAO,IAAI,MAAM,GAAG,aAAa,CAAC;AAC3D,WAAO,EAAE,MAAM,SAAS,MAAM,eAAe,OAAO,SAAS,WAAW,UAAU,EAAE;AAAA,EACtF;AACA,SAAO,EAAE,MAAM,SAAS,MAAM,SAAS,OAAO,MAAM,WAAW,UAAU,EAAE;AAC7E;AAEO,SAAS,mBAAmB,MAA+B;AAChE,SAAO,KACJ;AAAA,IAAI,SACH,IACG,IAAI,OAAK,EAAE,KAAK,KAAK,EAAE,QAAQ,OAAO,GAAG,EAAE,QAAQ,OAAO,KAAK,CAAC,EAChE,OAAO,OAAO,EACd,KAAK,KAAK;AAAA,EACf,EACC,OAAO,OAAO,EACd,KAAK,IAAI;AACd;AAGA,SAAS,UAAU,MAAsB;AAEvC,SAAO,KAAK,QAAQ,MAAM,KAAK;AACjC;AAGA,IAAM,wBAAwB;AAG9B,SAAS,aAAa,MAAsB;AAC1C,MAAI,SAAS,KAEV,QAAQ,2BAA2B,EAAE,EAErC,QAAQ,uBAAuB,EAAE,EACjC,QAAQ,QAAQ,GAAG,EACnB,KAAK;AAGR,MAAI,OAAO,UAAU,MAAM,OAAO,SAAS,GAAG,GAAG;AAC/C,UAAM,SAAS,OAAO,MAAM,GAAG;AAE/B,UAAM,wBAAwB,OAAO,OAAO,OAAK,EAAE,WAAW,KAAK,+BAA+B,KAAK,CAAC,CAAC,EAAE;AAC3G,QAAI,OAAO,UAAU,KAAK,wBAAwB,OAAO,UAAU,KAAK;AACtE,eAAS,OAAO,KAAK,EAAE;AAAA,IACzB;AAAA,EACF;AACA,SAAO;AACT;AAOO,SAAS,oBAAoB,QAA8B;AAChE,QAAM,SAAoB,CAAC;AAE3B,aAAW,SAAS,QAAQ;AAC1B,QAAI,MAAM,SAAS,WAAW,CAAC,MAAM,OAAO;AAC1C,aAAO,KAAK,KAAK;AACjB;AAAA,IACF;AAEA,UAAM,EAAE,MAAM,SAAS,MAAM,SAAS,MAAM,IAAI,MAAM;AAGtD,QAAI,YAAY,KAAK,YAAY,GAAG;AAClC,aAAO,KAAK,KAAK;AACjB;AAAA,IACF;AAGA,QAAI,WAAW,GAAG;AAChB,UAAI,gBAAgB;AACpB,UAAI,eAAe;AACnB,eAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,iBAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,gBAAM,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,QAAQ;AACjC,4BAAkB,EAAE,MAAM,KAAK,KAAK,CAAC,GAAG;AACxC,0BAAgB,EAAE;AAAA,QACpB;AAAA,MACF;AAGA,UAAI,gBAAgB,KAAM,WAAW,KAAK,eAAe,KAAM;AAE7D,iBAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,mBAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,kBAAM,WAAW,MAAM,CAAC,IAAI,CAAC,GAAG,MAAM,KAAK;AAC3C,gBAAI,CAAC,SAAU;AAEf,uBAAW,QAAQ,SAAS,MAAM,IAAI,GAAG;AACvC,oBAAM,UAAU,KAAK,KAAK;AAC1B,kBAAI,CAAC,QAAS;AACd,qBAAO,KAAK,EAAE,MAAM,aAAa,MAAM,SAAS,YAAY,MAAM,WAAW,CAAC;AAAA,YAChF;AAAA,UACF;AAAA,QACF;AACA;AAAA,MACF;AAAA,IACF;AAEA,WAAO,KAAK,KAAK;AAAA,EACnB;AAEA,SAAO;AACT;AAEO,SAAS,iBAAiB,QAA2B;AAC1D,QAAM,QAAkB,CAAC;AAEzB,WAAS,IAAI,GAAG,IAAI,OAAO,QAAQ,KAAK;AACtC,UAAM,QAAQ,OAAO,CAAC;AAGtB,QAAI,MAAM,SAAS,aAAa,MAAM,MAAM;AAC1C,YAAM,SAAS,IAAI,OAAO,KAAK,IAAI,MAAM,SAAS,GAAG,CAAC,CAAC;AACvD,YAAM,cAAc,aAAa,MAAM,IAAI;AAC3C,UAAI,YAAa,OAAM,KAAK,IAAI,GAAG,MAAM,IAAI,WAAW,IAAI,EAAE;AAC9D;AAAA,IACF;AAGA,QAAI,MAAM,SAAS,WAAW,MAAM,MAAM;AACxC,YAAM,KAAK,IAAI,YAAY,MAAM,IAAI,KAAK,EAAE;AAC5C;AAAA,IACF;AAGA,QAAI,MAAM,SAAS,aAAa;AAC9B,YAAM,KAAK,IAAI,OAAO,EAAE;AACxB;AAAA,IACF;AAGA,QAAI,MAAM,SAAS,UAAU,MAAM,MAAM;AACvC,YAAM,WAAW,aAAa,MAAM,IAAI;AACxC,UAAI,CAAC,SAAU;AAEf,YAAM,kBAAkB,MAAM,aAAa,aAAa,WAAW,KAAK,QAAQ;AAChF,YAAM,SAAS,kBAAkB,KAAK,MAAM,aAAa,YAAY,QAAQ;AAC7E,YAAM,KAAK,GAAG,MAAM,GAAG,QAAQ,EAAE;AACjC,UAAI,MAAM,UAAU;AAClB,mBAAW,SAAS,MAAM,UAAU;AAClC,gBAAM,cAAc,MAAM,aAAa,YAAY,OAAO;AAC1D,gBAAM,KAAK,KAAK,WAAW,IAAI,MAAM,QAAQ,EAAE,EAAE;AAAA,QACnD;AAAA,MACF;AACA;AAAA,IACF;AAEA,QAAI,MAAM,SAAS,eAAe,MAAM,MAAM;AAC5C,UAAI,OAAO,aAAa,MAAM,IAAI;AAClC,UAAI,CAAC,KAAM;AAGX,UAAI,cAAc,KAAK,IAAI,GAAG;AAC5B,cAAM,YAAY,OAAO,IAAI,CAAC;AAC9B,YAAI,WAAW,SAAS,eAAe,UAAU,QAAQ,SAAS,KAAK,UAAU,IAAI,GAAG;AACtF,gBAAM,KAAK,IAAI,MAAM,IAAI,IAAI,UAAU,IAAI,IAAI,EAAE;AACjD;AAAA,QACF,OAAO;AACL,gBAAM,KAAK,IAAI,MAAM,IAAI,IAAI,EAAE;AAAA,QACjC;AACA;AAAA,MACF;AAEA,UAAI,sBAAsB,KAAK,IAAI,GAAG;AACpC,cAAM,KAAK,IAAI,IAAI,KAAK,EAAE;AAC1B;AAAA,MACF;AAGA,UAAI,MAAM,MAAM;AACd,cAAM,OAAO,aAAa,MAAM,IAAI;AACpC,YAAI,KAAM,QAAO,IAAI,IAAI,KAAK,IAAI;AAAA,MACpC;AAGA,UAAI,MAAM,cAAc;AACtB,gBAAQ,aAAQ,MAAM,YAAY;AAAA,MACpC;AAEA,YAAM,KAAK,UAAU,IAAI,GAAG,EAAE;AAAA,IAChC,WAAW,MAAM,SAAS,WAAW,MAAM,OAAO;AAEhD,UAAI,MAAM,SAAS,KAAK,MAAM,MAAM,SAAS,CAAC,MAAM,IAAI;AACtD,cAAM,KAAK,EAAE;AAAA,MACf;AACA,YAAM,UAAU,gBAAgB,MAAM,KAAK;AAC3C,UAAI,SAAS;AACX,cAAM,KAAK,OAAO;AAClB,cAAM,KAAK,EAAE;AAAA,MACf;AAAA,IACF;AAAA,EACF;AAEA,SAAO,MAAM,KAAK,IAAI,EAAE,KAAK;AAC/B;AAGA,SAAS,eAAe,OAAyB;AAC/C,aAAW,OAAO,MAAM,OAAO;AAC7B,eAAW,QAAQ,KAAK;AACtB,UAAI,KAAK,UAAU,KAAK,KAAK,UAAU,EAAG,QAAO;AAAA,IACnD;AAAA,EACF;AACA,SAAO;AACT;AAEA,SAAS,mBAAmB,MAAuB;AACjD,SAAO,wCAAwC,KAAK,IAAI;AAC1D;AAEA,SAAS,wBAAwB,OAAyB;AACxD,aAAW,OAAO,MAAM,OAAO;AAC7B,eAAW,QAAQ,KAAK;AACtB,UAAI,mBAAmB,KAAK,IAAI,EAAG,QAAO;AAAA,IAC5C;AAAA,EACF;AACA,SAAO;AACT;AAGA,SAAS,YAAY,OAAwB;AAC3C,QAAM,EAAE,OAAO,MAAM,SAAS,MAAM,QAAQ,IAAI;AAChD,QAAM,OAAO,oBAAI,IAAY;AAC7B,QAAM,QAAkB,CAAC,SAAS;AAElC,WAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,UAAM,MAAM,MAAM,IAAI,OAAO;AAC7B,UAAM,UAAoB,CAAC;AAC3B,aAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,UAAI,KAAK,IAAI,GAAG,CAAC,IAAI,CAAC,EAAE,EAAG;AAC3B,YAAM,OAAO,MAAM,CAAC,IAAI,CAAC;AACzB,UAAI,CAAC,KAAM;AAGX,eAAS,KAAK,GAAG,KAAK,KAAK,SAAS,MAAM;AACxC,iBAAS,KAAK,GAAG,KAAK,KAAK,SAAS,MAAM;AACxC,cAAI,OAAO,KAAK,OAAO,EAAG;AAC1B,cAAI,IAAI,KAAK,WAAW,IAAI,KAAK,QAAS,MAAK,IAAI,GAAG,IAAI,EAAE,IAAI,IAAI,EAAE,EAAE;AAAA,QAC1E;AAAA,MACF;AAEA,YAAM,OAAO,aAAa,KAAK,IAAI,EAAE,QAAQ,OAAO,MAAM;AAC1D,YAAM,QAAkB,CAAC;AACzB,UAAI,KAAK,UAAU,EAAG,OAAM,KAAK,YAAY,KAAK,OAAO,GAAG;AAC5D,UAAI,KAAK,UAAU,EAAG,OAAM,KAAK,YAAY,KAAK,OAAO,GAAG;AAC5D,YAAM,UAAU,MAAM,SAAS,MAAM,MAAM,KAAK,GAAG,IAAI;AACvD,cAAQ,KAAK,IAAI,GAAG,GAAG,OAAO,IAAI,IAAI,KAAK,GAAG,GAAG;AAAA,IACnD;AACA,QAAI,QAAQ,OAAQ,OAAM,KAAK,OAAO,QAAQ,KAAK,EAAE,CAAC,OAAO;AAAA,EAC/D;AAEA,QAAM,KAAK,UAAU;AACrB,SAAO,MAAM,KAAK,IAAI;AACxB;AAEA,SAAS,gBAAgB,OAAwB;AAC/C,MAAI,MAAM,SAAS,KAAK,MAAM,SAAS,EAAG,QAAO;AAEjD,QAAM,EAAE,OAAO,MAAM,SAAS,MAAM,QAAQ,IAAI;AAIhD,MAAI,eAAe,KAAK,KAAK,CAAC,wBAAwB,KAAK,EAAG,QAAO,YAAY,KAAK;AAGtF,MAAI,YAAY,KAAK,YAAY,GAAG;AAClC,UAAM,UAAU,aAAa,MAAM,CAAC,EAAE,CAAC,EAAE,IAAI;AAC7C,QAAI,CAAC,QAAS,QAAO;AACrB,WAAO,QACJ,MAAM,IAAI,EACV,IAAI,UAAQ;AACX,YAAM,UAAU,KAAK,KAAK;AAC1B,UAAI,CAAC,QAAS,QAAO;AACrB,UAAI,WAAW,KAAK,OAAO,EAAG,QAAO,KAAK,UAAU,OAAO,CAAC;AAC5D,UAAI,aAAa,KAAK,OAAO,EAAG,QAAO,KAAK,UAAU,OAAO,CAAC;AAC9D,aAAO,UAAU,OAAO;AAAA,IAC1B,CAAC,EACA,OAAO,OAAO,EACd,KAAK,IAAI;AAAA,EACd;AAGA,MAAI,YAAY,KAAK,WAAW,GAAG;AACjC,WAAO,MACJ,IAAI,SAAO,UAAU,aAAa,IAAI,CAAC,EAAE,IAAI,CAAC,EAAE,QAAQ,OAAO,GAAG,CAAC,EACnE,OAAO,OAAO,EACd,KAAK,IAAI;AAAA,EACd;AAGA,QAAM,UAAsB,MAAM,KAAK,EAAE,QAAQ,QAAQ,GAAG,MAAM,MAAM,OAAO,EAAE,KAAK,EAAE,CAAC;AACzF,QAAM,OAAO,oBAAI,IAAY;AAE7B,WAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,aAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,UAAI,KAAK,IAAI,GAAG,CAAC,IAAI,CAAC,EAAE,EAAG;AAC3B,YAAM,OAAO,MAAM,CAAC,IAAI,CAAC;AACzB,UAAI,CAAC,KAAM;AACX,cAAQ,CAAC,EAAE,CAAC,IAAI,UAAU,aAAa,KAAK,IAAI,CAAC,EAAE,QAAQ,OAAO,KAAK,EAAE,QAAQ,OAAO,MAAM;AAG9F,eAAS,KAAK,GAAG,KAAK,KAAK,SAAS,MAAM;AACxC,iBAAS,KAAK,GAAG,KAAK,KAAK,SAAS,MAAM;AACxC,cAAI,OAAO,KAAK,OAAO,EAAG;AAC1B,cAAI,IAAI,KAAK,WAAW,IAAI,KAAK,SAAS;AACxC,iBAAK,IAAI,GAAG,IAAI,EAAE,IAAI,IAAI,EAAE,EAAE;AAAA,UAChC;AAAA,QACF;AAAA,MACF;AAEA,WAAK,KAAK,UAAU;AAAA,IACtB;AAAA,EACF;AAMA,QAAM,aAAyB,CAAC;AAChC,MAAI,kBAAkB;AACtB,WAAS,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAAK;AACvC,UAAM,MAAM,QAAQ,CAAC;AACrB,UAAM,qBAAqB,IAAI,MAAM,UAAQ,SAAS,EAAE;AACxD,QAAI,mBAAoB;AAIxB,UAAM,eAAe,IAAI,OAAO,UAAQ,SAAS,EAAE;AACnD,UAAM,eAAe,IAAI,KAAK,CAAC,GAAG,MAAM,KAAK,IAAI,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;AAC7D,QAAI,CAAC,gBAAgB,aAAa,WAAW,KAAK,IAAI,CAAC,MAAM,MAAM,IAAI,MAAM,CAAC,EAAE,MAAM,OAAK,MAAM,EAAE,GAAG;AACpG,wBAAkB,IAAI,CAAC;AACvB;AAAA,IACF;AAGA,QAAI,mBAAmB,IAAI,CAAC,MAAM,IAAI;AACpC,UAAI,CAAC,IAAI;AACT,wBAAkB;AAAA,IACpB,OAAO;AACL,wBAAkB;AAAA,IACpB;AACA,eAAW,KAAK,GAAG;AAAA,EACrB;AAEA,MAAI,WAAW,WAAW,EAAG,QAAO;AAEpC,QAAM,KAAe,CAAC;AACtB,KAAG,KAAK,OAAO,WAAW,CAAC,EAAE,KAAK,KAAK,IAAI,IAAI;AAC/C,KAAG,KAAK,OAAO,WAAW,CAAC,EAAE,IAAI,MAAM,KAAK,EAAE,KAAK,KAAK,IAAI,IAAI;AAChE,WAAS,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK;AAC1C,OAAG,KAAK,OAAO,WAAW,CAAC,EAAE,KAAK,KAAK,IAAI,IAAI;AAAA,EACjD;AACA,SAAO,GAAG,KAAK,IAAI;AACrB;;;AChLO,IAAM,mBAAmB;AACzB,IAAM,mBAAmB;AACzB,IAAM,mBAAmB;","names":[]}

package/dist/{chunk-NKKLA43G.js → chunk-LA66FVBN.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 
 // src/utils.ts
-var VERSION = true ? "2.5.2" : "0.0.0-dev";
+var VERSION = true ? "2.7.1" : "0.0.0-dev";
 function toArrayBuffer(buf) {
   if (buf.byteOffset === 0 && buf.byteLength === buf.buffer.byteLength) {
     return buf.buffer;
@@ -331,6 +331,17 @@ function hasMergedCells(table) {
   }
   return false;
 }
+function containsInlineMath(text) {
+  return /(^|[^\\])\$(?=\S)(?:\\.|[^$\n])+?\S\$/.test(text);
+}
+function tableContainsInlineMath(table) {
+  for (const row of table.cells) {
+    for (const cell of row) {
+      if (containsInlineMath(cell.text)) return true;
+    }
+  }
+  return false;
+}
 function tableToHtml(table) {
   const { cells, rows: numRows, cols: numCols } = table;
   const skip = /* @__PURE__ */ new Set();
@@ -363,7 +374,7 @@ function tableToHtml(table) {
 function tableToMarkdown(table) {
   if (table.rows === 0 || table.cols === 0) return "";
   const { cells, rows: numRows, cols: numCols } = table;
-  if (hasMergedCells(table)) return tableToHtml(table);
+  if (hasMergedCells(table) && !tableContainsInlineMath(table)) return tableToHtml(table);
   if (numRows === 1 && numCols === 1) {
     const content = sanitizeText(cells[0][0].text);
     if (!content) return "";
@@ -454,4 +465,4 @@ export {
   HEADING_RATIO_H2,
   HEADING_RATIO_H3
 };
-//# sourceMappingURL=chunk-NKKLA43G.js.map
+//# sourceMappingURL=chunk-LA66FVBN.js.map