@clazic/kordoc 2.3.1 → 2.3.3

package/dist/{chunk-ZOEUKD77.js → chunk-UDFKY7CH.js}

@@ -6,10 +6,19 @@ import {
   precheckZipSize,
   sanitizeHref,
   toArrayBuffer
-} from "./chunk-W5KUC23B.js";
+} from "./chunk-NU3KFVVZ.js";
 import {
   parsePageRange
 } from "./chunk-MOL7MDBG.js";
+import {
+  createTesseractProvider
+} from "./chunk-7FMKAV4P.js";
+import {
+  createCliOcrProvider
+} from "./chunk-JOGAFNIL.js";
+import {
+  markdownToBlocks
+} from "./chunk-4PP34NVQ.js";
 import {
   __commonJS,
   __require,
@@ -1918,24 +1927,29 @@ function isPdfFile(buffer) {
   const b = magicBytes(buffer);
   return b[0] === 37 && b[1] === 80 && b[2] === 68 && b[3] === 70;
 }
+function isPngFile(buffer) {
+  const b = magicBytes(buffer);
+  return b[0] === 137 && b[1] === 80 && b[2] === 78 && b[3] === 71;
+}
 function detectFormat(buffer) {
   if (buffer.byteLength < 4) return "unknown";
   if (isZipFile(buffer)) return "hwpx";
   if (isOldHwpFile(buffer)) return "hwp";
   if (isPdfFile(buffer)) return "pdf";
+  if (isPngFile(buffer)) return "image";
   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";
+    if (zip.file("xl/workbook.xml")) return { format: "xlsx", zip };
+    if (zip.file("word/document.xml")) return { format: "docx", zip };
+    if (zip.file("Contents/content.hpf") || zip.file("mimetype")) return { format: "hwpx", zip };
     const hasSection = Object.keys(zip.files).some((f) => f.startsWith("Contents/"));
-    if (hasSection) return "hwpx";
-    return "unknown";
+    if (hasSection) return { format: "hwpx", zip };
+    return { format: "unknown", zip: null };
   } catch {
-    return "unknown";
+    return { format: "unknown", zip: null };
   }
 }
 
@@ -2024,12 +2038,16 @@ function buildTableDirect(rows, numRows) {
   return trimAndReturn(grid, numRows, maxCols);
 }
 function trimAndReturn(grid, numRows, maxCols) {
-  let effectiveCols = maxCols;
-  while (effectiveCols > 0) {
-    const colEmpty = grid.every((row) => !row[effectiveCols - 1]?.text?.trim());
-    if (!colEmpty) break;
-    effectiveCols--;
+  let effectiveCols = 0;
+  for (const row of grid) {
+    for (let c = row.length - 1; c >= effectiveCols; c--) {
+      if (row[c]?.text?.trim()) {
+        effectiveCols = c + 1;
+        break;
+      }
+    }
   }
+  if (effectiveCols === 0) effectiveCols = maxCols;
   if (effectiveCols < maxCols && effectiveCols > 0) {
     const trimmed = grid.map((row) => row.slice(0, effectiveCols));
     return { rows: numRows, cols: effectiveCols, cells: trimmed, hasHeader: numRows > 1 };
@@ -2289,11 +2307,11 @@ function parseStyleElements(doc, map) {
 function stripDtd(xml) {
   return xml.replace(/<!DOCTYPE\s[^[>]*(\[[\s\S]*?\])?\s*>/gi, "");
 }
-async function parseHwpxDocument(buffer, options) {
+async function parseHwpxDocument(buffer, options, existingZip) {
   precheckZipSize(buffer, MAX_DECOMPRESS_SIZE, MAX_ZIP_ENTRIES);
   let zip;
   try {
-    zip = await JSZip2.loadAsync(buffer);
+    zip = existingZip ?? await JSZip2.loadAsync(buffer);
   } catch {
     return await extractFromBrokenZip(buffer);
   }
@@ -5328,8 +5346,15 @@ import { getDocument, GlobalWorkerOptions } from "pdfjs-dist/legacy/build/pdf.mj
 GlobalWorkerOptions.workerSrc = "";
 var MAX_PAGES = 5e3;
 var MAX_TOTAL_TEXT = 500 * 1024 * 1024;
-var PDF_LOAD_TIMEOUT_MS = 3e4;
+function calcPdfTimeout(bufferSize) {
+  const base = 3e4;
+  const perMb = 500;
+  const mb = bufferSize / (1024 * 1024);
+  return Math.min(base + Math.ceil(mb * perMb), 3e5);
+}
 async function loadPdfWithTimeout(buffer) {
+  const timeoutMs = calcPdfTimeout(buffer.byteLength);
+  const timeoutSec = Math.round(timeoutMs / 1e3);
   const loadingTask = getDocument({
     data: new Uint8Array(buffer),
     useSystemFonts: true,
@@ -5343,8 +5368,8 @@ async function loadPdfWithTimeout(buffer) {
       new Promise((_, reject) => {
         timer = setTimeout(() => {
           loadingTask.destroy();
-          reject(new KordocError("PDF \uB85C\uB529 \uD0C0\uC784\uC544\uC6C3 (30\uCD08 \uCD08\uACFC)"));
-        }, PDF_LOAD_TIMEOUT_MS);
+          reject(new KordocError(`PDF \uB85C\uB529 \uD0C0\uC784\uC544\uC6C3 (${timeoutSec}\uCD08 \uCD08\uACFC)`));
+        }, timeoutMs);
       })
     ]);
   } finally {
@@ -5365,11 +5390,15 @@ async function parsePdfDocument(buffer, options) {
     const effectivePageCount = Math.min(pageCount, MAX_PAGES);
     const pageFilter = options?.pages ? parsePageRange(options.pages, effectivePageCount) : null;
     const totalTarget = pageFilter ? pageFilter.size : effectivePageCount;
-    const allFontSizes = [];
+    const fontSizeFreq = /* @__PURE__ */ new Map();
     const pageHeights = /* @__PURE__ */ new Map();
-    let parsedPages = 0;
+    const targetPageNums = [];
     for (let i = 1; i <= effectivePageCount; i++) {
       if (pageFilter && !pageFilter.has(i)) continue;
+      targetPageNums.push(i);
+    }
+    let parsedPages = 0;
+    const parseSinglePage = async (i) => {
       try {
         const page = await doc.getPage(i);
         const tc = await page.getTextContent();
@@ -5382,7 +5411,10 @@ async function parsePdfDocument(buffer, options) {
           warnings.push({ page: i, message: `${hiddenCount}\uAC1C \uC228\uACA8\uC9C4 \uD14D\uC2A4\uD2B8 \uC694\uC18C \uD544\uD130\uB9C1\uB428`, code: "HIDDEN_TEXT_FILTERED" });
         }
         for (const item of visible) {
-          if (item.fontSize > 0) allFontSizes.push(item.fontSize);
+          if (item.fontSize > 0) {
+            const rounded = Math.round(item.fontSize * 10) / 10;
+            fontSizeFreq.set(rounded, (fontSizeFreq.get(rounded) || 0) + 1);
+          }
         }
         const opList = await page.getOperatorList();
         const pageBlocks = extractPageBlocksWithLines(visible, i, opList, viewport.width, viewport.height);
@@ -5399,14 +5431,36 @@ async function parsePdfDocument(buffer, options) {
         if (pageErr instanceof KordocError) throw pageErr;
         warnings.push({ page: i, message: `\uD398\uC774\uC9C0 ${i} \uD30C\uC2F1 \uC2E4\uD328: ${pageErr instanceof Error ? pageErr.message : "\uC54C \uC218 \uC5C6\uB294 \uC624\uB958"}`, code: "PARTIAL_PARSE" });
       }
+    };
+    const SAMPLE_SIZE = Math.min(10, targetPageNums.length);
+    const sampledIndices = /* @__PURE__ */ new Set();
+    if (targetPageNums.length <= SAMPLE_SIZE) {
+      for (let i = 0; i < targetPageNums.length; i++) sampledIndices.add(i);
+    } else {
+      for (let i = 0; i < SAMPLE_SIZE; i++) {
+        const idx = Math.round(i * (targetPageNums.length - 1) / (SAMPLE_SIZE - 1));
+        sampledIndices.add(idx);
+      }
+    }
+    for (const si of sampledIndices) {
+      await parseSinglePage(targetPageNums[si]);
+    }
+    const sampleParsed = parsedPages || sampledIndices.size;
+    const isImageBased = totalChars / Math.max(sampleParsed, 1) < 10;
+    if (!isImageBased) {
+      for (let si = 0; si < targetPageNums.length; si++) {
+        if (!sampledIndices.has(si)) {
+          await parseSinglePage(targetPageNums[si]);
+        }
+      }
     }
     const parsedPageCount = parsedPages || (pageFilter ? pageFilter.size : effectivePageCount);
-    if (totalChars / Math.max(parsedPageCount, 1) < 10) {
+    if (isImageBased) {
       let ocrProvider = options?.ocr ?? null;
-      const ocrMode = options?.ocrMode;
-      if (!ocrProvider && ocrMode && ocrMode !== "off") {
+      const ocrMode = options?.ocrMode ?? "auto";
+      if (!ocrProvider && ocrMode !== "off") {
         try {
-          const { resolveOcrProvider } = await import("./resolve-4FSAQF2S.js");
+          const { resolveOcrProvider } = await import("./resolve-UOAOPQ4H.js");
           const concurrency = options?.ocrConcurrency ?? 1;
           const batchSize = options?.ocrBatchSize;
           ocrProvider = await resolveOcrProvider(ocrMode, warnings, concurrency, batchSize);
@@ -5422,7 +5476,7 @@ async function parsePdfDocument(buffer, options) {
       if (ocrProvider) {
         let ocrBlocks = [];
         try {
-          const { ocrPages } = await import("./provider-WYHC4NHI.js");
+          const { ocrPages } = await import("./provider-HE727F7Z.js");
           const concurrency = options?.ocrConcurrency ?? 1;
           ocrBlocks = await ocrPages(doc, ocrProvider, pageFilter, effectivePageCount, warnings, concurrency, options?.onProgress);
         } catch {
@@ -5456,7 +5510,7 @@ async function parsePdfDocument(buffer, options) {
         blocks.splice(removed[ri], 1);
       }
     }
-    const medianFontSize = computeMedianFontSize(allFontSizes);
+    const medianFontSize = computeMedianFromFreq(fontSizeFreq);
     if (medianFontSize > 0) {
       detectHeadings(blocks, medianFontSize);
     }
@@ -5520,11 +5574,18 @@ function filterHiddenText(items, pageWidth, pageHeight) {
   }
   return { visible, hiddenCount };
 }
-function computeMedianFontSize(sizes) {
-  if (sizes.length === 0) return 0;
-  const sorted = [...sizes].sort((a, b) => a - b);
-  const mid = Math.floor(sorted.length / 2);
-  return sorted.length % 2 === 0 ? (sorted[mid - 1] + sorted[mid]) / 2 : sorted[mid];
+function computeMedianFromFreq(freq) {
+  if (freq.size === 0) return 0;
+  const entries = [...freq.entries()].sort((a, b) => a[0] - b[0]);
+  let total = 0;
+  for (const [, count] of entries) total += count;
+  const mid = total / 2;
+  let cumulative = 0;
+  for (const [size, count] of entries) {
+    cumulative += count;
+    if (cumulative >= mid) return size;
+  }
+  return 0;
 }
 function detectHeadings(blocks, medianFontSize) {
   for (const block of blocks) {
@@ -6330,6 +6391,7 @@ var MAX_SHEETS = 100;
 var MAX_DECOMPRESS_SIZE3 = 500 * 1024 * 1024;
 var MAX_ROWS2 = 1e4;
 var MAX_COLS2 = 200;
+var MAX_TOTAL_CELLS = 2e6;
 function cleanNumericValue(raw) {
   if (!/^-?\d+\.\d+$/.test(raw)) return raw;
   const num = parseFloat(raw);
@@ -6513,9 +6575,9 @@ function sheetToBlocks(sheetName, grid, merges, maxRow, maxCol, sheetIndex) {
   }
   return blocks;
 }
-async function parseXlsxDocument(buffer, options) {
+async function parseXlsxDocument(buffer, options, existingZip) {
   precheckZipSize(buffer, MAX_DECOMPRESS_SIZE3);
-  const zip = await JSZip3.loadAsync(buffer);
+  const zip = existingZip ?? await JSZip3.loadAsync(buffer);
   const warnings = [];
   const workbookFile = zip.file("xl/workbook.xml");
   if (!workbookFile) {
@@ -6542,6 +6604,7 @@ async function parseXlsxDocument(buffer, options) {
   }
   const blocks = [];
   const processedSheets = Math.min(sheets.length, MAX_SHEETS);
+  let totalCells = 0;
   for (let i = 0; i < processedSheets; i++) {
     if (pageFilter && !pageFilter.has(i + 1)) continue;
     const sheet = sheets[i];
@@ -6568,6 +6631,11 @@ async function parseXlsxDocument(buffer, options) {
     try {
       const sheetXml = await sheetFile.async("text");
       const { grid, merges, maxRow, maxCol } = parseWorksheet(sheetXml, sharedStrings);
+      totalCells += maxRow * maxCol;
+      if (totalCells > MAX_TOTAL_CELLS) {
+        warnings.push({ message: `\uCD1D \uC140 \uC218 \uC81C\uD55C \uCD08\uACFC (${totalCells.toLocaleString()}\uC140), \uC774\uD6C4 \uC2DC\uD2B8 \uC0DD\uB7B5`, code: "PARTIAL_PARSE" });
+        break;
+      }
       const sheetBlocks = sheetToBlocks(sheet.name, grid, merges, maxRow, maxCol, i);
       blocks.push(...sheetBlocks);
     } catch (err) {
@@ -6651,10 +6719,35 @@ function getAttr(el, localName) {
 function parseXml2(text) {
   return new DOMParser3().parseFromString(text, "text/xml");
 }
+function buildElementIndex(root) {
+  const index = /* @__PURE__ */ new Map();
+  const walk = (node) => {
+    const children = node.childNodes;
+    for (let i = 0; i < children.length; i++) {
+      const child = children[i];
+      if (child.nodeType === 1) {
+        const el = child;
+        const name = el.localName ?? "";
+        if (name) {
+          let list = index.get(name);
+          if (!list) {
+            list = [];
+            index.set(name, list);
+          }
+          list.push(el);
+        }
+        walk(el);
+      }
+    }
+  };
+  walk(root);
+  return index;
+}
 function parseStyles(xml) {
   const doc = parseXml2(xml);
   const styles = /* @__PURE__ */ new Map();
-  const styleElements = findElements(doc, "style");
+  const idx = buildElementIndex(doc);
+  const styleElements = idx.get("style") ?? [];
   for (const el of styleElements) {
     const styleId = getAttr(el, "styleId");
     if (!styleId) continue;
@@ -6682,7 +6775,8 @@ function parseStyles(xml) {
 function parseNumbering(xml) {
   const doc = parseXml2(xml);
   const abstractNums = /* @__PURE__ */ new Map();
-  const abstractElements = findElements(doc, "abstractNum");
+  const idx = buildElementIndex(doc);
+  const abstractElements = idx.get("abstractNum") ?? [];
   for (const el of abstractElements) {
     const abstractNumId = getAttr(el, "abstractNumId");
     if (!abstractNumId) continue;
@@ -6697,7 +6791,7 @@ function parseNumbering(xml) {
     abstractNums.set(abstractNumId, levels);
   }
   const nums = /* @__PURE__ */ new Map();
-  const numElements = findElements(doc, "num");
+  const numElements = idx.get("num") ?? [];
   for (const el of numElements) {
     const numId = getAttr(el, "numId");
     if (!numId) continue;
@@ -6941,9 +7035,9 @@ async function extractImages(zip, rels, doc) {
   }
   return { blocks, images };
 }
-async function parseDocxDocument(buffer, options) {
+async function parseDocxDocument(buffer, options, existingZip) {
   precheckZipSize(buffer, MAX_DECOMPRESS_SIZE4);
-  const zip = await JSZip4.loadAsync(buffer);
+  const zip = existingZip ?? await JSZip4.loadAsync(buffer);
   const warnings = [];
   const docFile = zip.file("word/document.xml");
   if (!docFile) {
@@ -9378,25 +9472,86 @@ async function parse2(input, options) {
   if (!buffer || buffer.byteLength === 0) {
     return { success: false, fileType: "unknown", error: "\uBE48 \uBC84\uD37C\uC774\uAC70\uB098 \uC720\uD6A8\uD558\uC9C0 \uC54A\uC740 \uC785\uB825\uC785\uB2C8\uB2E4.", code: "EMPTY_INPUT" };
   }
+  const MAX_FILE_SIZE = 500 * 1024 * 1024;
+  if (buffer.byteLength > MAX_FILE_SIZE) {
+    return { success: false, fileType: "unknown", error: `\uD30C\uC77C \uD06C\uAE30 \uCD08\uACFC: ${(buffer.byteLength / 1024 / 1024).toFixed(1)}MB (\uCD5C\uB300 500MB)`, code: "FILE_TOO_LARGE" };
+  }
   const format = detectFormat(buffer);
   switch (format) {
     case "hwpx": {
-      const zipFormat = await detectZipFormat(buffer);
-      if (zipFormat === "xlsx") return parseXlsx(buffer, options);
-      if (zipFormat === "docx") return parseDocx(buffer, options);
-      return parseHwpx(buffer, options);
+      const { format: zipFormat, zip } = await detectZipFormat(buffer);
+      if (zipFormat === "xlsx") return parseXlsx(buffer, options, zip ?? void 0);
+      if (zipFormat === "docx") return parseDocx(buffer, options, zip ?? void 0);
+      return parseHwpx(buffer, options, zip ?? void 0);
     }
     case "hwp":
       return parseHwp(buffer, options);
     case "pdf":
       return parsePdf(buffer, options);
+    case "image":
+      return parseImage(buffer, options);
     default:
       return { success: false, fileType: "unknown", error: "\uC9C0\uC6D0\uD558\uC9C0 \uC54A\uB294 \uD30C\uC77C \uD615\uC2DD\uC785\uB2C8\uB2E4.", code: "UNSUPPORTED_FORMAT" };
   }
 }
-async function parseHwpx(buffer, options) {
+async function parseImage(buffer, options) {
+  const ocrMode = options?.ocrMode || "auto";
+  if (ocrMode === "off") {
+    return { success: false, fileType: "image", error: "OCR\uC774 \uBE44\uD65C\uC131\uD654\uB418\uC5B4 \uC774\uBBF8\uC9C0 \uD30C\uC77C\uC744 \uCC98\uB9AC\uD560 \uC218 \uC5C6\uC2B5\uB2C8\uB2E4.", code: "IMAGE_BASED_PDF" };
+  }
+  let ocrProvider;
+  let actualOcrMode = "auto";
+  try {
+    if (ocrMode === "gemini" || ocrMode === "claude" || ocrMode === "codex" || ocrMode === "ollama") {
+      ocrProvider = createCliOcrProvider(ocrMode);
+      actualOcrMode = ocrMode;
+    } else if (ocrMode === "tesseract") {
+      ocrProvider = await createTesseractProvider();
+      actualOcrMode = ocrMode;
+    } else if (ocrMode === "auto") {
+      const modesToTry = ["gemini", "claude", "codex", "ollama"];
+      for (const mode of modesToTry) {
+        try {
+          ocrProvider = createCliOcrProvider(mode);
+          actualOcrMode = mode;
+          break;
+        } catch (e) {
+          console.warn(`[kordoc] OCR auto-detection: ${mode} CLI not available or failed. Trying next.`, e);
+        }
+      }
+      if (!ocrProvider) {
+        ocrProvider = await createTesseractProvider();
+        actualOcrMode = "tesseract";
+      }
+    }
+    if (!ocrProvider) {
+      return { success: false, fileType: "image", error: "\uC0AC\uC6A9 \uAC00\uB2A5\uD55C OCR \uD504\uB85C\uBC14\uC774\uB354\uB97C \uCC3E\uC744 \uC218 \uC5C6\uC2B5\uB2C8\uB2E4.", code: "PARSE_ERROR" };
+    }
+    const imageUint8Array = new Uint8Array(buffer);
+    const ocrResult = await ocrProvider(imageUint8Array, 1, "image/png");
+    if (ocrProvider.terminate) {
+      await ocrProvider.terminate();
+    }
+    const markdown = typeof ocrResult === "string" ? ocrResult : ocrResult.markdown;
+    const blocks = markdownToBlocks(markdown, 1);
+    return {
+      success: true,
+      fileType: "image",
+      markdown,
+      blocks,
+      isImageBased: true,
+      warnings: [{ message: `OCR \uCC98\uB9AC\uB428 (${actualOcrMode})`, code: "OCR_FALLBACK" }]
+    };
+  } catch (err) {
+    if (ocrProvider && ocrProvider.terminate) {
+      await ocrProvider.terminate();
+    }
+    return { success: false, fileType: "image", error: err instanceof Error ? err.message : "\uC774\uBBF8\uC9C0 OCR \uC2E4\uD328", code: classifyError(err) };
+  }
+}
+async function parseHwpx(buffer, options, zip) {
   try {
-    const { markdown, blocks, metadata, outline, warnings, images } = await parseHwpxDocument(buffer, options);
+    const { markdown, blocks, metadata, outline, warnings, images } = await parseHwpxDocument(buffer, options, zip);
     return { success: true, fileType: "hwpx", markdown, blocks, metadata, outline, warnings, images: images?.length ? images : void 0 };
   } catch (err) {
     return { success: false, fileType: "hwpx", error: err instanceof Error ? err.message : "HWPX \uD30C\uC2F1 \uC2E4\uD328", code: classifyError(err) };
@@ -9419,17 +9574,17 @@ async function parsePdf(buffer, options) {
     return { success: false, fileType: "pdf", error: err instanceof Error ? err.message : "PDF \uD30C\uC2F1 \uC2E4\uD328", code: classifyError(err), isImageBased };
   }
 }
-async function parseXlsx(buffer, options) {
+async function parseXlsx(buffer, options, zip) {
   try {
-    const { markdown, blocks, metadata, warnings } = await parseXlsxDocument(buffer, options);
+    const { markdown, blocks, metadata, warnings } = await parseXlsxDocument(buffer, options, zip);
     return { success: true, fileType: "xlsx", markdown, blocks, metadata, warnings };
   } catch (err) {
     return { success: false, fileType: "xlsx", error: err instanceof Error ? err.message : "XLSX \uD30C\uC2F1 \uC2E4\uD328", code: classifyError(err) };
   }
 }
-async function parseDocx(buffer, options) {
+async function parseDocx(buffer, options, zip) {
   try {
-    const { markdown, blocks, metadata, outline, warnings, images } = await parseDocxDocument(buffer, options);
+    const { markdown, blocks, metadata, outline, warnings, images } = await parseDocxDocument(buffer, options, zip);
     return { success: true, fileType: "docx", markdown, blocks, metadata, outline, warnings, images: images?.length ? images : void 0 };
   } catch (err) {
     return { success: false, fileType: "docx", error: err instanceof Error ? err.message : "DOCX \uD30C\uC2F1 \uC2E4\uD328", code: classifyError(err) };
@@ -9624,4 +9779,4 @@ export {
 cfb/cfb.js:
   (*! crc32.js (C) 2014-present SheetJS -- http://sheetjs.com *)
 */
-//# sourceMappingURL=chunk-ZOEUKD77.js.map
+//# sourceMappingURL=chunk-UDFKY7CH.js.map