kordoc 4.0.8 → 4.1.0

package/dist/{parser-TCTCSBYZ.js → parser-PEOYBXBO.js}

@@ -3,17 +3,19 @@ import {
   HEADING_RATIO_H1,
   HEADING_RATIO_H2,
   HEADING_RATIO_H3,
-  blocksToMarkdown,
   encodePng
-} from "./chunk-UEIYETCQ.js";
+} from "./chunk-UOO7LSDG.js";
 import {
   parsePageRange
 } from "./chunk-MOL7MDBG.js";
+import {
+  blocksToMarkdown
+} from "./chunk-SSTK6IKK.js";
 import {
   KordocError,
   safeMax,
   safeMin
-} from "./chunk-HMUEU7D7.js";
+} from "./chunk-O5VS7RIR.js";
 
 // src/pdf/line-extract.ts
 import { OPS } from "pdfjs-dist/legacy/build/pdf.mjs";
@@ -581,6 +583,18 @@ function buildVertices(horizontals, verticals) {
 }
 function mergeVertices(vertices) {
   if (vertices.length <= 1) return vertices;
+  let maxRadiusAll = 1;
+  for (const v of vertices) {
+    if (v.radius > maxRadiusAll) maxRadiusAll = v.radius;
+  }
+  const cell = Math.max(VERTEX_MERGE_FACTOR * maxRadiusAll, 1);
+  const buckets = /* @__PURE__ */ new Map();
+  for (let i = 0; i < vertices.length; i++) {
+    const key = Math.floor(vertices[i].x / cell) + "," + Math.floor(vertices[i].y / cell);
+    const arr = buckets.get(key);
+    if (arr) arr.push(i);
+    else buckets.set(key, [i]);
+  }
   const merged = [];
   const used = new Array(vertices.length).fill(false);
   for (let i = 0; i < vertices.length; i++) {
@@ -588,7 +602,20 @@ function mergeVertices(vertices) {
     let sumX = vertices[i].x, sumY = vertices[i].y;
     let maxRadius = vertices[i].radius;
     let count = 1;
-    for (let j = i + 1; j < vertices.length; j++) {
+    const cx = Math.floor(vertices[i].x / cell);
+    const cy = Math.floor(vertices[i].y / cell);
+    const candidates = [];
+    for (let dx = -1; dx <= 1; dx++) {
+      for (let dy = -1; dy <= 1; dy++) {
+        const arr = buckets.get(cx + dx + "," + (cy + dy));
+        if (!arr) continue;
+        for (const j of arr) {
+          if (j > i && !used[j]) candidates.push(j);
+        }
+      }
+    }
+    candidates.sort((a, b) => a - b);
+    for (const j of candidates) {
       if (used[j]) continue;
       const mergeTol = VERTEX_MERGE_FACTOR * Math.max(maxRadius, vertices[j].radius);
       if (Math.abs(vertices[i].x - vertices[j].x) <= mergeTol && Math.abs(vertices[i].y - vertices[j].y) <= mergeTol) {
@@ -815,6 +842,7 @@ function splitStackedGroup(group) {
   }
   return bands.filter((b) => b.length > 0);
 }
+var GROUP_BUCKET_CELL = 100;
 function groupConnectedLines(lines) {
   const parent = lines.map((_, i) => i);
   function find(x) {
@@ -828,10 +856,34 @@ function groupConnectedLines(lines) {
     const ra = find(a), rb = find(b);
     if (ra !== rb) parent[ra] = rb;
   }
+  const cellMap = /* @__PURE__ */ new Map();
   for (let i = 0; i < lines.length; i++) {
-    for (let j = i + 1; j < lines.length; j++) {
-      if (linesIntersect(lines[i], lines[j])) {
-        union(i, j);
+    const l = lines[i];
+    const cx1 = Math.floor((Math.min(l.x1, l.x2) - CONNECT_TOL) / GROUP_BUCKET_CELL);
+    const cx2 = Math.floor((Math.max(l.x1, l.x2) + CONNECT_TOL) / GROUP_BUCKET_CELL);
+    const cy1 = Math.floor((Math.min(l.y1, l.y2) - CONNECT_TOL) / GROUP_BUCKET_CELL);
+    const cy2 = Math.floor((Math.max(l.y1, l.y2) + CONNECT_TOL) / GROUP_BUCKET_CELL);
+    for (let cx = cx1; cx <= cx2; cx++) {
+      for (let cy = cy1; cy <= cy2; cy++) {
+        const key = cx + "," + cy;
+        const arr = cellMap.get(key);
+        if (arr) arr.push(i);
+        else cellMap.set(key, [i]);
+      }
+    }
+  }
+  const tested = /* @__PURE__ */ new Set();
+  for (const arr of cellMap.values()) {
+    for (let a = 0; a < arr.length; a++) {
+      for (let b = a + 1; b < arr.length; b++) {
+        const i = Math.min(arr[a], arr[b]);
+        const j = Math.max(arr[a], arr[b]);
+        const key = i * lines.length + j;
+        if (tested.has(key)) continue;
+        tested.add(key);
+        if (linesIntersect(lines[i], lines[j])) {
+          union(i, j);
+        }
       }
     }
   }
@@ -1319,12 +1371,21 @@ async function extractPageImages(page, fnArray, argsArray, pageNumber, state, wa
       if (typeof objId !== "string" || pageSeenIds.has(objId)) continue;
       pageSeenIds.add(objId);
       dedupeId = objId;
-      imgData = await resolveImgData(page, objId);
-    } else if (op === OPS3.paintInlineImageXObject) {
-      imgData = argsArray[i]?.[0];
-    } else {
+    } else if (op !== OPS3.paintInlineImageXObject) {
       continue;
     }
+    if (state.imageIndex >= MAX_IMAGES_PER_DOC || state.totalBytes >= MAX_TOTAL_IMAGE_BYTES) {
+      if (!state.capWarned) {
+        state.capWarned = true;
+        warnings.push({ page: pageNumber, message: `\uC774\uBBF8\uC9C0 \uCD94\uCD9C \uC0C1\uD55C \uB3C4\uB2EC (${MAX_IMAGES_PER_DOC}\uAC1C/128MB) \u2014 \uC774\uD6C4 \uC774\uBBF8\uC9C0 \uC0DD\uB7B5`, code: "SKIPPED_IMAGE" });
+      }
+      continue;
+    }
+    if (dedupeId) {
+      imgData = await resolveImgData(page, dedupeId);
+    } else {
+      imgData = argsArray[i]?.[0];
+    }
     if (!imgData?.data || !imgData.width || !imgData.height) continue;
     const { width: w, height: h } = imgData;
     if (w < MIN_DIM || h < MIN_DIM) continue;
@@ -1334,13 +1395,6 @@ async function extractPageImages(page, fnArray, argsArray, pageNumber, state, wa
     }
     const hash = `${w}x${h}k${imgData.kind ?? "?"}h${fnv1a(imgData.data)}${dedupeId ? "" : "inline"}`;
     if (state.seen.has(hash)) continue;
-    if (state.imageIndex >= MAX_IMAGES_PER_DOC || state.totalBytes >= MAX_TOTAL_IMAGE_BYTES) {
-      if (!state.capWarned) {
-        state.capWarned = true;
-        warnings.push({ page: pageNumber, message: `\uC774\uBBF8\uC9C0 \uCD94\uCD9C \uC0C1\uD55C \uB3C4\uB2EC (${MAX_IMAGES_PER_DOC}\uAC1C/128MB) \u2014 \uC774\uD6C4 \uC774\uBBF8\uC9C0 \uC0DD\uB7B5`, code: "SKIPPED_IMAGE" });
-      }
-      continue;
-    }
     const rgba = toRgba(imgData);
     if (!rgba) continue;
     const png = encodePng(w, h, rgba);
@@ -2683,11 +2737,12 @@ function buildGridTable(lines, columns) {
   if (filledCells < totalCells * 0.35 || merged.length < 2 || merged.length <= 3 && numCols >= 7) {
     return merged.map((r) => r.filter((c) => c).join("	")).join("\n");
   }
+  const escCell = (c) => c.replace(/\t/g, " ").replace(/\|/g, "\\|");
   const md = [];
-  md.push("| " + merged[0].join(" | ") + " |");
+  md.push("| " + merged[0].map(escCell).join(" | ") + " |");
   md.push("| " + merged[0].map(() => "---").join(" | ") + " |");
   for (let r = 1; r < merged.length; r++) {
-    md.push("| " + merged[r].join(" | ") + " |");
+    md.push("| " + merged[r].map(escCell).join(" | ") + " |");
   }
   return md.join("\n");
 }
@@ -2867,6 +2922,7 @@ function detectKoreanListBlocks(blocks) {
   }
   let familyStack = [];
   let lastTopLevelList = null;
+  let prevListIdx = -2;
   const toRemove = /* @__PURE__ */ new Set();
   for (let i = 0; i < blocks.length; i++) {
     const b = blocks[i];
@@ -2895,7 +2951,7 @@ function detectKoreanListBlocks(blocks) {
     if (depth === 0) {
       blocks[i] = listBlock;
       lastTopLevelList = listBlock;
-    } else if (lastTopLevelList) {
+    } else if (lastTopLevelList && i === prevListIdx + 1) {
       if (!lastTopLevelList.children) lastTopLevelList.children = [];
       lastTopLevelList.children.push(listBlock);
       toRemove.add(i);
@@ -2903,6 +2959,7 @@ function detectKoreanListBlocks(blocks) {
       blocks[i] = listBlock;
       lastTopLevelList = listBlock;
     }
+    prevListIdx = i;
   }
   if (toRemove.size > 0) {
     const sorted = [...toRemove].sort((a, b) => b - a);
@@ -3028,8 +3085,8 @@ function removeHeaderFooterBlocks(blocks, pageHeights, warnings) {
     const blockTop = ph - (b.bbox.y + b.bbox.height);
     const blockBottom = ph - b.bbox.y;
     const entry = { blockIdx: bi, page: b.pageNumber, text: b.text.trim() };
-    if (blockBottom <= ph * ZONE_RATIO) bottomEntries.push(entry);
-    else if (blockTop >= ph * (1 - ZONE_RATIO)) topEntries.push(entry);
+    if (blockBottom <= ph * ZONE_RATIO) topEntries.push(entry);
+    else if (blockTop >= ph * (1 - ZONE_RATIO)) bottomEntries.push(entry);
   }
   const removeSet = /* @__PURE__ */ new Set();
   for (const entries of [topEntries, bottomEntries]) {
@@ -3202,6 +3259,8 @@ function extractBlocksWithGrids(items, pageNum, grids, horizontals, verticals) {
       hasSpaceBefore: i.hasSpaceBefore
     }));
     const cellTextMap = mapTextToCells(textItems, cells);
+    const assignedItems = /* @__PURE__ */ new Set();
+    for (const arr of cellTextMap.values()) for (const it of arr) assignedItems.add(it);
     const numRows = grid.rowYs.length - 1;
     const numCols = grid.colXs.length - 1;
     const irGrid = Array.from(
@@ -3221,9 +3280,11 @@ function extractBlocksWithGrids(items, pageNum, grids, horizontals, verticals) {
     }
     let finalGrid = irGrid;
     let finalRows = numRows;
+    let rebuiltUsed = false;
     if (numRows <= 2 && numCols >= 3) {
       const rebuilt = normalizeUndersegmentedTable(irGrid, grid.colXs, textItems);
       if (rebuilt) {
+        rebuiltUsed = true;
         finalGrid = rebuilt.map((row) => row.map((rawText) => {
           const cleaned = rawText.replace(/^[\s]*[-–—]\s*\d+\s*[-–—][\s]*$/gm, "").trim();
           return {
@@ -3235,6 +3296,11 @@ function extractBlocksWithGrids(items, pageNum, grids, horizontals, verticals) {
         finalRows = finalGrid.length;
       }
     }
+    if (!rebuiltUsed) {
+      for (let ti = 0; ti < textItems.length; ti++) {
+        if (!assignedItems.has(textItems[ti])) usedItems.delete(tableItems[ti]);
+      }
+    }
     const irTable = {
       rows: finalRows,
       cols: numCols,
@@ -3291,6 +3357,8 @@ function extractBlocksWithGrids(items, pageNum, grids, horizontals, verticals) {
       }
       remaining = remaining.filter((_, idx) => !usedClusterIndices.has(idx));
     }
+    const groupSizes = [];
+    let finalTextBlocks = [];
     if (remaining.length > 0) {
       const allY = remaining.map((i) => i.y);
       const pageH = safeMax(allY) - safeMin(allY);
@@ -3300,16 +3368,28 @@ function extractBlocksWithGrids(items, pageNum, grids, horizontals, verticals) {
         if (group.length === 0) continue;
         const groupBlocks = extractPageBlocksFallback(group, pageNum);
         for (const b of groupBlocks) textBlocks.push(b);
+        groupSizes.push(groupBlocks.length);
       }
-      const finalTextBlocks = detectListBlocks(textBlocks);
-      for (const b of finalTextBlocks) blocks.push(b);
+      finalTextBlocks = detectListBlocks(textBlocks);
     }
-    blocks.sort((a, b) => {
-      const ay = a.bbox ? a.bbox.y + a.bbox.height : 0;
-      const by = b.bbox ? b.bbox.y + b.bbox.height : 0;
-      return by - ay;
-    });
-    return mergeAdjacentTableBlocks(blocks);
+    const units = blocks.map((b) => [b]);
+    let off = 0;
+    for (const size of groupSizes) {
+      const unit = finalTextBlocks.slice(off, off + size);
+      off += size;
+      if (unit.length > 0) units.push(unit);
+    }
+    const unitTopY = (u) => {
+      let top = 0;
+      for (const b of u) {
+        if (b.bbox && b.bbox.y + b.bbox.height > top) top = b.bbox.y + b.bbox.height;
+      }
+      return top;
+    };
+    units.sort((a, b) => unitTopY(b) - unitTopY(a));
+    const ordered = [];
+    for (const u of units) for (const b of u) ordered.push(b);
+    return mergeAdjacentTableBlocks(ordered);
   }
   return mergeAdjacentTableBlocks(blocks);
 }
@@ -3798,7 +3878,11 @@ async function parsePdfDocument(buffer, options) {
         }
         let pageText = "";
         for (const b of pageBlocks) {
-          const t = b.text || "";
+          let t = b.text || "";
+          if (b.type === "table" && b.table) {
+            const cellText = b.table.cells.map((row) => row.map((c) => c.text).join(" ")).join("\n");
+            t = t ? t + "\n" + cellText : cellText;
+          }
           totalChars += t.replace(/\s/g, "").length;
           totalTextBytes += t.length * 2;
           pageText += pageText ? "\n" + t : t;
@@ -3817,7 +3901,7 @@ async function parsePdfDocument(buffer, options) {
     if (totalChars / Math.max(parsedPageCount, 1) < 10) {
       if (options?.ocr) {
         try {
-          const { ocrPages } = await import("./provider-AKROB7WQ.js");
+          const { ocrPages } = await import("./provider-UIBW2MIH.js");
           const ocrBlocks = await ocrPages(doc, options.ocr, pageFilter, effectivePageCount);
           if (ocrBlocks.length > 0) {
             const ocrMarkdown = ocrBlocks.map((b) => b.text || "").filter(Boolean).join("\n\n");
@@ -3938,4 +4022,4 @@ export {
   parsePdfDocument,
   removeHeaderFooterBlocks
 };
-//# sourceMappingURL=parser-TCTCSBYZ.js.map
+//# sourceMappingURL=parser-PEOYBXBO.js.map