kordoc 4.21.17 → 4.21.18

package/README.md

@@ -12,7 +12,7 @@ HWP 3.x·5.x, HWPX, HWPML, PDF, XLS·XLSX, DOCX, PPTX, PNG·JPG·WebP를 Markdow
 [English](README-EN.md) · [상세 사용법](docs/usage.md) · [벤치마크](docs/benchmarks.md) · [변경 이력](CHANGELOG.md)
 
 - 📊 **PDF 공개 벤치 종합 0.963** — opendataloader-bench 200문서. 공개 12개 파서 비교 1위(2026-09-29 기록). [측정 조건](docs/benchmarks.md)
-- 🇰🇷 **한국 공문서 표 구조 100%** — 원본 HWPX 2,424문서의 보이는 표 **10,342개 전부 구조 일치**. [4.21.17 검증](docs/benchmarks.md)
+- 🇰🇷 **한국 공문서 표 구조 100%** — 원본 HWPX 2,424문서의 보이는 표 **10,342개 전부 구조 일치**. [4.21.18 검증](docs/benchmarks.md)
 
 쓸모 있었다면 GitHub ⭐ 하나 눌러주세요. 다른 사람이 이 도구를 찾는 데 도움이 됩니다.
 
@@ -100,7 +100,7 @@ await writeFile("보고서.hwpx", Buffer.from(hwpx))
 
 ## 검증 결과
 
-**4.21.17 · 2026-10-11 배포 검증**. 고정 코퍼스와 참조 기준으로 측정한 결과입니다.
+**4.21.18 · 2026-10-11 배포 검증**. 고정 코퍼스와 참조 기준으로 측정한 결과입니다.
 
 | 대상 | 규모 | 결과 |
 | --- | --- | --- |
@@ -113,14 +113,15 @@ await writeFile("보고서.hwpx", Buffer.from(hwpx))
 
 표 구조 점수와 셀 내용·화면 재현은 별도 지표입니다. 채점 범위·제외 기준·전 벤치 문서별 비교는 [벤치마크 상세](docs/benchmarks.md)에 있습니다.
 
-외부 PDF 벤치 **opendataloader-bench 200문서**는 4.21.17 배포 측정에서 기본값 종합 **0.963**, OCR 끔 **0.939**입니다. [옵션별 결과와 재현 방법](docs/benchmarks.md#pdf--markdown--opendataloader-bench)
+외부 PDF 벤치 **opendataloader-bench 200문서**는 4.21.18 배포 측정에서 기본값 종합 **0.963**, OCR 끔 **0.939**입니다. [옵션별 결과와 재현 방법](docs/benchmarks.md#pdf--markdown--opendataloader-bench)
 
 ## 최근 업데이트
 
-최신 배포: **[4.21.17](https://github.com/chrisryugj/kordoc/releases/tag/v4.21.17)** · 2026-10-11
+최신 배포: **[4.21.18](https://github.com/chrisryugj/kordoc/releases/tag/v4.21.18)** · 2026-10-11
 
 | 버전 | 주요 변경 |
 | --- | --- |
+| 4.21.18 | 쪽 넘김 문단 잇기도 줄 끝 공백 글리프를 어절 경계로 — "지원⏎활동이"·"계정⏎정보를" 을 띄워 씀 (PDF 어절 F1 98.986 → 98.987%, 5문서 상승) |
 | 4.21.17 | PDF 공백 없는 꽉 찬 줄 끝 + 영문 괄호 — "공적개발원조⏎(ODA)"·"하네스⏎(safety …)" 를 붙여 씀 (PDF 어절 F1 98.98 → 98.99%) |
 | 4.21.16 | PDF 표 쪽 넘김 갈린 행의 수량 이어짐 — 머리 칸 "30%" / 다음 쪽 "미만" 을 한 칸으로; 공문서 autoFit.safety(다른 조판기용 한 줄 판정 여유) (법령 별표 PDF 표 97.40 → 97.69%) |
 | 4.21.15 | PDF 줄 끝 공백 글리프로 어절 경계 꺾임 가르기 — 원문이 띄운 줄 꺾임("이루어⏎진다면")을 붙이던 것 (PDF 어절 F1 98.89 → 98.98%) |

package/dist/{-GS6LEDZ6.js → -AQAGPODO.js}

@@ -54,17 +54,17 @@ import {
   unknownFontWarnings,
   usesGaejosikMunche,
   validateHwpx
-} from "./chunk-6F6L63RG.js";
+} from "./chunk-AIJCO3QD.js";
 import {
   blocksToPages
-} from "./chunk-L3VCBSFW.js";
+} from "./chunk-RV6FN6H5.js";
 import {
   applySplices,
   buildParagraphSplices,
   buildRangeSplices,
   placeSealHwpx,
   scanSectionXml
-} from "./chunk-7ASI3XKM.js";
+} from "./chunk-ES4VPZ3R.js";
 import {
   DEFAULT_REDACT_RULES,
   redactMarkdown,
@@ -72,7 +72,7 @@ import {
 } from "./chunk-3WSNLH4B.js";
 import {
   blocksToChunks
-} from "./chunk-FWQHYJZ2.js";
+} from "./chunk-LS535RMK.js";
 import {
   SPACE_EM_FIXED,
   SPACE_EM_FONT,
@@ -97,20 +97,20 @@ import {
   renderSceneToHtml,
   simulateWrap,
   simulateWrapKeepWord
-} from "./chunk-FMUBLLIZ.js";
+} from "./chunk-WKBQRD3C.js";
 import "./chunk-7FCNOKOR.js";
-import "./chunk-TZU7PXV2.js";
+import "./chunk-DVGX37KX.js";
 import "./chunk-VQFA55IC.js";
-import "./chunk-Q4GXF444.js";
+import "./chunk-NV3HZV5U.js";
 import {
   blocksToMarkdown,
   flattenLayoutTables,
   hasStructuredCellContent
-} from "./chunk-QLL7VG6L.js";
-import "./chunk-TUSREHO3.js";
+} from "./chunk-PIIW5A33.js";
+import "./chunk-M3SJKVRX.js";
 import {
   VERSION
-} from "./chunk-PEQ5K2JO.js";
+} from "./chunk-RRQTQNEZ.js";
 export {
   AGENCY_COMMON_NAME,
   BUILTIN_TEMPLATES,
@@ -204,4 +204,4 @@ export {
   usesGaejosikMunche,
   validateHwpx
 };
-//# sourceMappingURL=-GS6LEDZ6.js.map
+//# sourceMappingURL=-AQAGPODO.js.map

package/dist/{chunk-I43ZKTUM.cjs → chunk-6GSIGM7S.cjs}

@@ -4,15 +4,15 @@
 
 
 
-var _chunkVVFNVMFMcjs = require('./chunk-VVFNVMFM.cjs');
+var _chunkTKFGJLFScjs = require('./chunk-TKFGJLFS.cjs');
 
 
 
-var _chunkL3U6MKGVcjs = require('./chunk-L3U6MKGV.cjs');
+var _chunkNL5H3LN7cjs = require('./chunk-NL5H3LN7.cjs');
 
 
 
-var _chunkY4TO3GBFcjs = require('./chunk-Y4TO3GBF.cjs');
+var _chunkCYGTZS4Mcjs = require('./chunk-CYGTZS4M.cjs');
 
 // src/ocr/models.ts
 var _path = require('path');
@@ -40,10 +40,10 @@ var OCR_REC_DICT = {
 };
 var ALL_OCR_MODELS = [OCR_DET_MODEL, OCR_REC_MODEL, OCR_REC_DICT];
 function getOcrModelsDir() {
-  return _chunkL3U6MKGVcjs.getModelsDir.call(void 0, "ppocr");
+  return _chunkNL5H3LN7cjs.getModelsDir.call(void 0, "ppocr");
 }
 async function ensureOcrModels(onProgress) {
-  return _chunkL3U6MKGVcjs.ensureModelsIn.call(void 0, getOcrModelsDir(), ALL_OCR_MODELS, onProgress);
+  return _chunkNL5H3LN7cjs.ensureModelsIn.call(void 0, getOcrModelsDir(), ALL_OCR_MODELS, onProgress);
 }
 async function ocrModelsCached() {
   const dir = getOcrModelsDir();
@@ -2123,7 +2123,7 @@ function tagScripts(text, lines) {
     const [s, e, kind] = marks[m];
     out = out.slice(0, s) + `<${kind}>` + out.slice(s, e) + `</${kind}>` + out.slice(e);
   }
-  return _chunkVVFNVMFMcjs.tidyScriptTags.call(void 0, out);
+  return _chunkTKFGJLFScjs.tidyScriptTags.call(void 0, out);
 }
 
 // src/pdf/line-wrap.ts
@@ -2339,7 +2339,8 @@ function joinPageBreakWraps(blocks, lex) {
     if (indented && SENTENCE_END.test(a.text.replace(MARKUP, "").trimEnd())) continue;
     if (Math.round(_nullishCoalesce(_optionalChain([b, 'access', _70 => _70.style, 'optionalAccess', _71 => _71.fontSize]), () => ( 0))) !== Math.round(fs) && titleLike(b.text)) continue;
     if (/\t/.test(a.text.slice(a.text.lastIndexOf("\n") + 1).replace(/^\S{1,8}\t/, "")) && titleLike(bt.trim().split("\n")[0])) continue;
-    const head = a.text, joiner = wrapJoiner(a.text, b.text, lex);
+    const joined = wrapJoiner(a.text, b.text, lex);
+    const head = a.text, joiner = joined === "" && last.spaceAfter ? " " : joined;
     a.text += joiner + b.text;
     const later = b.bbox && PAGE_BREAK_JOINS.get(b.bbox) || [];
     PAGE_BREAK_JOINS.set(a.bbox, [{ head, gap: joiner.length, tail: b }, ...later.map((c) => ({ ...c, head: head + joiner + c.head }))]);
@@ -5115,7 +5116,12 @@ function pushLineParagraphs(out, yLines, pageNum, lex) {
     const faces = /* @__PURE__ */ new Map();
     for (const it of items) faces.set(it.fontName, (_nullishCoalesce(faces.get(it.fontName), () => ( 0))) + it.text.length);
     FACE_CHARS.set(block, faces);
-    PARA_LAST_LINE.set(block.bbox, { right: geo[i].right, width: geo[i].right - geo[i].left, fontSize: geo[i].fontSize });
+    PARA_LAST_LINE.set(block.bbox, {
+      right: geo[i].right,
+      width: geo[i].right - geo[i].left,
+      fontSize: geo[i].fontSize,
+      spaceAfter: lines[i].items.reduce((a, b) => b.x + b.w > a.x + a.w ? b : a).spaceAfter
+    });
     PARA_FIRST_LEFT.set(block.bbox, geo[first2].left);
     out.push(block);
     i++;
@@ -5145,9 +5151,9 @@ function detectHeadings(blocks, medianFontSize) {
     if (/^\d+$/.test(text) && !isChapterNumber(block, blocks[bi + 1], medianFontSize)) continue;
     const ratio = block.style.fontSize / medianFontSize;
     let level = 0;
-    if (ratio >= _chunkVVFNVMFMcjs.HEADING_RATIO_H1) level = 1;
-    else if (ratio >= _chunkVVFNVMFMcjs.HEADING_RATIO_H2) level = 2;
-    else if (ratio >= _chunkVVFNVMFMcjs.HEADING_RATIO_H3) level = 3;
+    if (ratio >= _chunkTKFGJLFScjs.HEADING_RATIO_H1) level = 1;
+    else if (ratio >= _chunkTKFGJLFScjs.HEADING_RATIO_H2) level = 2;
+    else if (ratio >= _chunkTKFGJLFScjs.HEADING_RATIO_H3) level = 3;
     if (level > 0) {
       block.type = "heading";
       block.level = level;
@@ -5160,7 +5166,7 @@ function isChapterNumber(block, next, medianFontSize) {
   const nextSize = _nullishCoalesce(_optionalChain([next, 'optionalAccess', _143 => _143.style, 'optionalAccess', _144 => _144.fontSize]), () => ( 0));
   const text = _nullishCoalesce(_optionalChain([next, 'optionalAccess', _145 => _145.text, 'optionalAccess', _146 => _146.trim, 'call', _147 => _147()]), () => ( ""));
   const a = block.bbox, b = _optionalChain([next, 'optionalAccess', _148 => _148.bbox]);
-  return /^\d{1,2}$/.test(_nullishCoalesce(_optionalChain([block, 'access', _149 => _149.text, 'optionalAccess', _150 => _150.trim, 'call', _151 => _151()]), () => ( ""))) && size >= medianFontSize * 2.5 && !!next && (next.type === "paragraph" || next.type === "heading") && next.pageNumber === block.pageNumber && nextSize >= medianFontSize * _chunkVVFNVMFMcjs.HEADING_RATIO_H1 && text.length > 0 && text.length <= 80 && !/^[\d\s.]+$/.test(text) && !!a && !!b && a.y > b.y && a.y - (b.y + b.height) <= size * 2;
+  return /^\d{1,2}$/.test(_nullishCoalesce(_optionalChain([block, 'access', _149 => _149.text, 'optionalAccess', _150 => _150.trim, 'call', _151 => _151()]), () => ( ""))) && size >= medianFontSize * 2.5 && !!next && (next.type === "paragraph" || next.type === "heading") && next.pageNumber === block.pageNumber && nextSize >= medianFontSize * _chunkTKFGJLFScjs.HEADING_RATIO_H1 && text.length > 0 && text.length <= 80 && !/^[\d\s.]+$/.test(text) && !!a && !!b && a.y > b.y && a.y - (b.y + b.height) <= size * 2;
 }
 function mergeStackedHeadingLines(blocks, medianFontSize) {
   for (let i = 0; i < blocks.length - 1; ) {
@@ -6533,16 +6539,16 @@ function isProseSpread(items) {
   for (let i = 1; i < sorted.length; i++) {
     gaps.push(sorted[i].x - (sorted[i - 1].x + sorted[i - 1].w));
   }
-  const maxGap = _chunkY4TO3GBFcjs.safeMax.call(void 0, gaps);
+  const maxGap = _chunkCYGTZS4Mcjs.safeMax.call(void 0, gaps);
   const avgLen = items.reduce((s, i) => s + i.text.length, 0) / items.length;
   if (items.length >= 4 && maxGap < 40 && avgLen < 5) return true;
-  const minGap = _chunkY4TO3GBFcjs.safeMin.call(void 0, gaps);
+  const minGap = _chunkCYGTZS4Mcjs.safeMin.call(void 0, gaps);
   return minGap > 0 && maxGap - minGap <= Math.max(2, maxGap * 0.15) && items.filter((i) => /^\P{L}*\p{Ll}/u.test(i.text)).length * 2 >= items.length;
 }
 function detectColumns(yLines) {
   const allItems = yLines.flat();
   if (allItems.length === 0) return null;
-  const pageWidth = _chunkY4TO3GBFcjs.safeMax.call(void 0, allItems.map((i) => i.x + i.w)) - _chunkY4TO3GBFcjs.safeMin.call(void 0, allItems.map((i) => i.x));
+  const pageWidth = _chunkCYGTZS4Mcjs.safeMax.call(void 0, allItems.map((i) => i.x + i.w)) - _chunkCYGTZS4Mcjs.safeMin.call(void 0, allItems.map((i) => i.x));
   if (pageWidth < 100) return null;
   let bigoLineIdx = -1;
   for (let i = 0; i < yLines.length; i++) {
@@ -7086,12 +7092,12 @@ function recordClipCellEdges(grids, horizontals, verticals, nonRules) {
 }
 function takeClipCellEdges(from, to) {
   const e = CLIP_CELL_EDGES.get(from);
-  if (e) _chunkVVFNVMFMcjs.CELL_EDGES.set(to, e);
+  if (e) _chunkTKFGJLFScjs.CELL_EDGES.set(to, e);
 }
 function joinCellEdges(a, b) {
-  const u = _chunkVVFNVMFMcjs.CELL_EDGES.get(a), d = _chunkVVFNVMFMcjs.CELL_EDGES.get(b);
+  const u = _chunkTKFGJLFScjs.CELL_EDGES.get(a), d = _chunkTKFGJLFScjs.CELL_EDGES.get(b);
   if (!u || !d) return;
-  _chunkVVFNVMFMcjs.CELL_EDGES.set(a, { t: u.t, b: d.b, l: u.l || d.l, r: u.r || d.r });
+  _chunkTKFGJLFScjs.CELL_EDGES.set(a, { t: u.t, b: d.b, l: u.l || d.l, r: u.r || d.r });
 }
 
 // src/pdf/table-unit-row.ts
@@ -7107,7 +7113,7 @@ function rebuildUnitLine(items, grid) {
 function prependUnitRow(cells, cols, line, usedItems) {
   const text = cleanCellText(line.map((item) => item.text).join(""));
   cells.unshift(Array.from({ length: cols }, (_, c) => ({ text: c === 0 ? text : "", colSpan: c === 0 ? cols : 1, rowSpan: 1 })));
-  for (const cell of cells[0]) _chunkVVFNVMFMcjs.CELL_EDGES.set(cell, NO_EDGES);
+  for (const cell of cells[0]) _chunkTKFGJLFScjs.CELL_EDGES.set(cell, NO_EDGES);
   for (const item of line) usedItems.add(item);
 }
 function attachUnitRow(items, grid, cells, cols, usedItems) {
@@ -7637,7 +7643,7 @@ function recordFrameTitle(table, source, pageNum) {
   const [label, spacer, title] = cells;
   if (!/^(?:붙임|별첨|첨부|부록)\s*\d{1,3}$/.test(label.text.trim()) || spacer.text.trim() || !title.text.trim()) return;
   if (cells.some((c) => {
-    const e = _chunkVVFNVMFMcjs.CELL_EDGES.get(c);
+    const e = _chunkTKFGJLFScjs.CELL_EDGES.get(c);
     return !e || e.t || e.b || e.l || e.r;
   })) return;
   const xs = TABLE_COLXS.get(table), style = dominantStyle(source), fs = _nullishCoalesce(_optionalChain([style, 'optionalAccess', _344 => _344.fontSize]), () => ( 0));
@@ -8296,7 +8302,7 @@ function extractBlocksWithGrids(items, pageNum, pageWidth, pageHeight, grids, ho
         groupSizes.push(1);
       }
       const allY = remaining.map((i) => i.y);
-      const pageH = _chunkY4TO3GBFcjs.safeMax.call(void 0, allY) - _chunkY4TO3GBFcjs.safeMin.call(void 0, allY);
+      const pageH = _chunkCYGTZS4Mcjs.safeMax.call(void 0, allY) - _chunkCYGTZS4Mcjs.safeMin.call(void 0, allY);
       const gapThreshold = Math.max(15, pageH * 0.03);
       const sides = frameColumnTextBands(remaining, spanning, gx);
       const textBlocks = [];
@@ -8312,7 +8318,7 @@ function extractBlocksWithGrids(items, pageNum, pageWidth, pageHeight, grids, ho
       finalTextBlocks.push(...detectListBlocks(textBlocks));
     } else {
       const allY = remaining.map((i) => i.y);
-      const pageH = _chunkY4TO3GBFcjs.safeMax.call(void 0, allY) - _chunkY4TO3GBFcjs.safeMin.call(void 0, allY);
+      const pageH = _chunkCYGTZS4Mcjs.safeMax.call(void 0, allY) - _chunkCYGTZS4Mcjs.safeMin.call(void 0, allY);
       const gapThreshold = Math.max(15, pageH * 0.03);
       const groups = xyCutOrder(remaining, gapThreshold);
       const textBlocks = [];
@@ -8640,7 +8646,7 @@ function extractPageBlocksFallback(items, pageNum, fullPage = false, detectTable
       blocks.push(...columnTextToBlocks(tableText, pageNum, bbox, dominantStyle(items)));
     } else {
       const allY = items.map((i) => i.y);
-      const pageHeight = _chunkY4TO3GBFcjs.safeMax.call(void 0, allY) - _chunkY4TO3GBFcjs.safeMin.call(void 0, allY);
+      const pageHeight = _chunkCYGTZS4Mcjs.safeMax.call(void 0, allY) - _chunkCYGTZS4Mcjs.safeMin.call(void 0, allY);
       const gapThreshold = Math.max(15, pageHeight * 0.03);
       const orderedGroups = proseCutX !== null ? splitTwoColumnProse(items, proseCutX) : xyCutOrder(items, gapThreshold, 0, void 0, minGap);
       for (const group of orderedGroups) {
@@ -8851,4 +8857,4 @@ function rowParagraphs(block) {
 
 
 exports.OCR_DET_MODEL = OCR_DET_MODEL; exports.OCR_REC_MODEL = OCR_REC_MODEL; exports.OCR_REC_DICT = OCR_REC_DICT; exports.getOcrModelsDir = getOcrModelsDir; exports.ensureOcrModels = ensureOcrModels; exports.ocrModelsCached = ocrModelsCached; exports.parseCharacterDict = parseCharacterDict; exports.extractLines = extractLines; exports.extractImageRegions = extractImageRegions; exports.filterHiddenText = filterHiddenText; exports.collapseEvenSpacing = collapseEvenSpacing; exports.computeBBox = computeBBox; exports.dominantStyle = dominantStyle; exports.normalizeItems = normalizeItems; exports.groupByY = groupByY; exports.mergeLineSimple = mergeLineSimple; exports.WrapLexicon = WrapLexicon; exports.startsNewItem = startsNewItem; exports.wrapJoiner = wrapJoiner; exports.joinPageBreakWraps = joinPageBreakWraps; exports.splitPageBreakWraps = splitPageBreakWraps; exports.latinSoftWrap = latinSoftWrap; exports.TOC_BLOCKS = TOC_BLOCKS; exports.computeMedianFontSizeFromFreq = computeMedianFontSizeFromFreq; exports.detectHeadings = detectHeadings; exports.mergeStackedHeadingLines = mergeStackedHeadingLines; exports.detectTypographyHeadings = detectTypographyHeadings; exports.detectDocumentStyleHeadings = detectDocumentStyleHeadings; exports.detectSiblingStyleHeadings = detectSiblingStyleHeadings; exports.detectRepeatedPageLabels = detectRepeatedPageLabels; exports.detectPageLeadHeadings = detectPageLeadHeadings; exports.refineDocumentStyleHeadings = refineDocumentStyleHeadings; exports.detectMarkerHeadings = detectMarkerHeadings; exports.detectTableCaptions = detectTableCaptions; exports.detectKoreanListBlocks = detectKoreanListBlocks; exports.removeHeaderFooterBlocks = removeHeaderFooterBlocks; exports.CLIP_TABLES = CLIP_TABLES; exports.FRAME_TITLE_BLOCKS = FRAME_TITLE_BLOCKS; exports.TABLE_COLXS = TABLE_COLXS; exports.TABLE_ROWYS = TABLE_ROWYS; exports.TABLE_TAIL = TABLE_TAIL; exports.PART_COLXS = PART_COLXS; exports.EMPTY_PARTS = EMPTY_PARTS; exports.IMAGE_CELLS = IMAGE_CELLS; exports.FILLER_CELLS = FILLER_CELLS; exports.CONT_PARTS = CONT_PARTS; exports.CELL_LINES = CELL_LINES; exports.ROW_RULES = ROW_RULES; exports.trimTrailingEmptyTableCols = trimTrailingEmptyTableCols; exports.NO_EDGES = NO_EDGES; exports.joinCellEdges = joinCellEdges; exports.restoreImageBullets = restoreImageBullets; exports.detectRightArrowRegions = detectRightArrowRegions; exports.removeSideTabs = removeSideTabs; exports.extractPageBlocksWithLines = extractPageBlocksWithLines; exports.UNRULED_REGION_TABLES = UNRULED_REGION_TABLES; exports.mergeOcrImageRegions = mergeOcrImageRegions;
-//# sourceMappingURL=chunk-I43ZKTUM.cjs.map
+//# sourceMappingURL=chunk-6GSIGM7S.cjs.map