@clazic/kordoc 2.2.7 → 2.2.8

package/dist/provider-EPHXUWRL.js.map

@@ -1 +0,0 @@
-{"version":3,"sources":["../src/ocr/markdown-to-blocks.ts","../src/ocr/provider.ts"],"sourcesContent":["/**\n * Markdown → IRBlock[] 역파싱\n *\n * Vision LLM(gemini/claude/codex 등)이 반환한 Markdown 문자열을\n * kordoc의 IRBlock[] 중간 표현으로 변환.\n * 기존 blocksToMarkdown()의 역방향 처리.\n */\n\nimport type { IRBlock, IRTable, IRCell } from \"../types.js\"\n\n/**\n * Markdown 문자열을 IRBlock[] 배열로 변환.\n *\n * 지원 요소:\n * - 헤딩: # ~ ######\n * - 테이블: | col1 | col2 | (파이프 구분, |---|---| 구분선 포함)\n * - 순서/비순서 리스트: - / 1.\n * - 구분선: ---, ***, ___\n * - 일반 텍스트 (paragraph)\n */\nexport function markdownToBlocks(markdown: string, pageNumber: number): IRBlock[] {\n  const blocks: IRBlock[] = []\n  const lines = markdown.split(\"\\n\")\n  let i = 0\n\n  while (i < lines.length) {\n    const line = lines[i]\n\n    // 빈 줄 스킵\n    if (line.trim() === \"\") {\n      i++\n      continue\n    }\n\n    // 1. 헤딩: # ~ ######\n    const headingMatch = line.match(/^(#{1,6})\\s+(.+)$/)\n    if (headingMatch) {\n      blocks.push({\n        type: \"heading\",\n        level: headingMatch[1].length,\n        text: headingMatch[2].trim(),\n        pageNumber,\n      })\n      i++\n      continue\n    }\n\n    // 2. 구분선: ---, ***, ___\n    if (/^[-*_]{3,}\\s*$/.test(line.trim())) {\n      blocks.push({ type: \"separator\", pageNumber })\n      i++\n      continue\n    }\n\n    // 3. 테이블: | 로 시작하는 연속 행 수집\n    if (line.trim().startsWith(\"|\")) {\n      const tableLines: string[] = []\n      while (i < lines.length && lines[i].trim().startsWith(\"|\")) {\n        tableLines.push(lines[i])\n        i++\n      }\n      const table = parseMarkdownTable(tableLines)\n      if (table) {\n        blocks.push({ type: \"table\", table, pageNumber })\n      }\n      continue\n    }\n\n    // 4. 비순서 리스트: -, *, +\n    const ulMatch = line.match(/^(\\s*)[-*+]\\s+(.+)$/)\n    if (ulMatch) {\n      blocks.push({\n        type: \"list\",\n        listType: \"unordered\",\n        text: ulMatch[2].trim(),\n        pageNumber,\n      })\n      i++\n      continue\n    }\n\n    // 5. 순서 리스트: 1.\n    const olMatch = line.match(/^(\\s*)\\d+\\.\\s+(.+)$/)\n    if (olMatch) {\n      blocks.push({\n        type: \"list\",\n        listType: \"ordered\",\n        text: olMatch[2].trim(),\n        pageNumber,\n      })\n      i++\n      continue\n    }\n\n    // 6. 일반 텍스트 — 구조적 행이 나올 때까지 병합\n    const paraLines: string[] = []\n    while (i < lines.length && lines[i].trim() !== \"\" && !isStructuralLine(lines[i])) {\n      paraLines.push(lines[i].trim())\n      i++\n    }\n    if (paraLines.length > 0) {\n      blocks.push({\n        type: \"paragraph\",\n        text: paraLines.join(\"\\n\"),\n        pageNumber,\n      })\n    }\n  }\n\n  return blocks\n}\n\n/**\n * 구조적 행 판별 — paragraph 병합 중단 트리거.\n */\nfunction isStructuralLine(line: string): boolean {\n  if (/^#{1,6}\\s+/.test(line)) return true\n  if (line.trim().startsWith(\"|\")) return true\n  if (/^[-*_]{3,}\\s*$/.test(line.trim())) return true\n  if (/^\\s*[-*+]\\s+/.test(line)) return true\n  if (/^\\s*\\d+\\.\\s+/.test(line)) return true\n  return false\n}\n\n/**\n * Markdown 테이블 행 배열을 IRTable로 변환.\n *\n * 구분선 행(|---|---|)은 제거 후 데이터 행만 파싱.\n * hasHeader: 구분선이 있었으면 true.\n */\nfunction parseMarkdownTable(lines: string[]): IRTable | null {\n  const hasSeparator = lines.some(line => /^\\|[\\s:|-]+\\|$/.test(line.trim()))\n\n  const rows: IRCell[][] = []\n  let maxCols = 0\n\n  for (const line of lines) {\n    // 구분선 행 스킵: |---|---| 패턴\n    if (/^\\|\\s*:?-+:?\\s*(\\|\\s*:?-+:?\\s*)+\\|?\\s*$/.test(line.trim())) continue\n\n    const parts = line.split(\"|\")\n    // 앞뒤 빈 요소 제거 (| 로 시작/종료하는 행)\n    const cells: IRCell[] = parts\n      .slice(1, parts[parts.length - 1].trim() === \"\" ? -1 : undefined)\n      .map(cell => ({\n        text: cell.trim(),\n        colSpan: 1,\n        rowSpan: 1,\n      }))\n\n    if (cells.length > 0) {\n      rows.push(cells)\n      maxCols = Math.max(maxCols, cells.length)\n    }\n  }\n\n  if (rows.length === 0) return null\n\n  // 열 수 통일 (부족한 셀은 빈 셀로 채움)\n  for (const row of rows) {\n    while (row.length < maxCols) {\n      row.push({ text: \"\", colSpan: 1, rowSpan: 1 })\n    }\n  }\n\n  return {\n    rows: rows.length,\n    cols: maxCols,\n    cells: rows,\n    hasHeader: hasSeparator && rows.length > 1,\n  }\n}\n","/**\n * OCR 프로바이더 브릿지 — PDF 페이지를 이미지로 렌더링하여 OCR 호출\n *\n * kordoc은 OCR 라이브러리를 번들하지 않음.\n * 사용자가 OcrProvider 함수를 제공하면 이미지 기반 PDF도 텍스트 추출 가능.\n *\n * @example\n * ```ts\n * import { parse } from \"kordoc\"\n *\n * const result = await parse(buffer, {\n *   ocr: async (pageImage, pageNumber, mimeType) => {\n *     // Tesseract, Claude Vision, Google Vision 등 사용\n *     return await myOcrService.recognize(pageImage)\n *   }\n * })\n * ```\n */\n\nimport type { OcrProvider, IRBlock, ParseWarning, StructuredOcrResult } from \"../types.js\"\nimport { markdownToBlocks } from \"./markdown-to-blocks.js\"\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 *\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): Promise<IRBlock[]> {\n  const blocks: IRBlock[] = []\n\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\n      if (typeof result === \"string\") {\n        // 기존 동작: 순수 텍스트 → paragraph 블록\n        if (result.trim()) {\n          blocks.push({ type: \"paragraph\", text: result.trim(), pageNumber: i })\n        }\n      } else if (result && typeof result === \"object\" && \"markdown\" in result) {\n        // 신규: 구조화된 결과 → Markdown → IRBlock[]\n        const structured = result as StructuredOcrResult\n        if (structured.markdown.trim()) {\n          const pageBlocks = markdownToBlocks(structured.markdown, i)\n          for (const b of pageBlocks) blocks.push(b)\n        }\n      }\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\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;;;AC1IA,eAAsB,SACpB,KACA,UACA,YACA,oBACA,UACoB;AACpB,QAAM,SAAoB,CAAC;AAE3B,WAAS,IAAI,GAAG,KAAK,oBAAoB,KAAK;AAC5C,QAAI,cAAc,CAAC,WAAW,IAAI,CAAC,EAAG;AACtC,UAAM,OAAO,MAAM,IAAI,QAAQ,CAAC;AAChC,QAAI;AACF,YAAM,YAAY,MAAM,gBAAgB,IAAI;AAC5C,YAAM,SAAS,MAAM,SAAS,WAAW,GAAG,WAAW;AAEvD,UAAI,OAAO,WAAW,UAAU;AAE9B,YAAI,OAAO,KAAK,GAAG;AACjB,iBAAO,KAAK,EAAE,MAAM,aAAa,MAAM,OAAO,KAAK,GAAG,YAAY,EAAE,CAAC;AAAA,QACvE;AAAA,MACF,WAAW,UAAU,OAAO,WAAW,YAAY,cAAc,QAAQ;AAEvE,cAAM,aAAa;AACnB,YAAI,WAAW,SAAS,KAAK,GAAG;AAC9B,gBAAM,aAAa,iBAAiB,WAAW,UAAU,CAAC;AAC1D,qBAAW,KAAK,WAAY,QAAO,KAAK,CAAC;AAAA,QAC3C;AAAA,MACF;AAAA,IACF,SAAS,KAAK;AAEZ,gBAAU,KAAK;AAAA,QACb,MAAM;AAAA,QACN,SAAS,sBAAO,CAAC,sBAAY,eAAe,QAAQ,IAAI,UAAU,yCAAW;AAAA,QAC7E,MAAM;AAAA,MACR,CAAC;AAAA,IACH;AAAA,EACF;AAEA,SAAO;AACT;AAWA,eAAe,gBAAgB,MAAyC;AACtE,QAAM,EAAE,aAAa,IAAI,MAAM,OAAO,iBAAiB;AAEvD,QAAM,QAAQ;AACd,QAAM,WAAW,KAAK,YAAY,EAAE,MAAM,CAAC;AAC3C,QAAM,SAAS,aAAa,KAAK,MAAM,SAAS,KAAK,GAAG,KAAK,MAAM,SAAS,MAAM,CAAC;AACnF,QAAM,MAAM,OAAO,WAAW,IAAI;AAElC,QAAM,KAAK,OAAO,EAAE,eAAe,KAAgB,SAAS,CAAC,EAAE;AAC/D,SAAO,IAAI,WAAW,OAAO,SAAS,WAAW,CAAC;AACpD;","names":[]}

package/dist/tesseract-provider-UNJOI25M.js

@@ -1,24 +0,0 @@
-#!/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;
-}
-export {
-  createTesseractProvider
-};
-//# sourceMappingURL=tesseract-provider-UNJOI25M.js.map

package/dist/tesseract-provider-UNJOI25M.js.map

@@ -1 +0,0 @@
-{"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"],"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;","names":[]}