kordoc 4.21.0 → 4.21.2

package/README.md

@@ -62,6 +62,8 @@ npx kordoc fill --template gian -j 값.json -o 기안문.hwpx
 
 `--jobs`는 파일 단위 병렬 변환이며 메모리 사용도 늘어납니다. [병렬 변환 안내](docs/parallel-batch.md)
 
+보고서·계획서·개조식·서울방침은 `#` 제목 바로 뒤 인용문(`> …하고자 함`)이 요약 상자가 됩니다(개조식은 표지가 있을 때(기본) 본문 첫 쪽 제목 상자 아래 — `cover:false` 면 첫 `#` 이 장 머리라 인용문은 ※ 참고). 보고 목적 한 문장(쉼표 허용) 3줄 이내로 쓰고, 넘거나 두 문장 이상이면 경고합니다. 업무보고는 `> ▪ …` 인용문이 자리와 관계없이 성과 요약 상자가 됩니다.
+
 ### JavaScript / TypeScript
 
 ```typescript
@@ -111,10 +113,12 @@ await writeFile("보고서.hwpx", Buffer.from(hwpx))
 
 ## 최근 업데이트
 
-최신 배포: **[4.21.0](https://github.com/chrisryugj/kordoc/releases/tag/v4.21.0)** · 2026-10-08
+최신 배포: **[4.21.2](https://github.com/chrisryugj/kordoc/releases/tag/v4.21.2)** · 2026-10-09
 
 | 버전 | 주요 변경 |
 | --- | --- |
+| 4.21.2 | 개조식 제목 뒤 요약이 1×1 요약 상자로(종전 ※ 참고)·요약 3줄·한 문장 검사, HWP3 중첩 한도·암호 HWPX 누적 압축 해제 상한 |
+| 4.21.1 | 이미지 OCR 에서 같은 줄 조각이 갈려 순서가 뒤집히던 것과 테두리 없는 영수증 표의 품목 행이 섞이던 것 (#141) |
 | 4.21.0 | 중앙부처 기관 서식 `--agency`(보도자료 2,670건·52개 기관 실측 단계 글꼴·크기·2단계 부호 ◦·❍·표 머리·기관 색), 보도자료 담당 표 사람별 행(`--press-people`), 폭표 없는 글꼴의 생성 폭 계산 보정 |
 | 4.20.0 | 표기법 검수 7룰(기간 대시·요일·금액 한글·공공언어·맞춤법·2타)과 문서 직접 검수·수치 대조·단계별 서식 편차, 견본 HWPX 단계별 서식 학습(`levels`·`--levels-from`), `compare`·`patch --json`, 패치 무결성 검사, 붙여넣기 흔적 정리 |
 | 4.19.2 | rhwp v0.8.7 파서 수정 반영 — HWP3 체크박스 □, HWP5 짝 없는 서로게이트 □, 손상 CFB 복구의 본문 경로 오인 수정, HWPX 압축 해제 상한을 엔트리 읽는 도중에 적용 |

package/dist/{-LJL6UNRO.js → -HOKFROKW.js}

@@ -54,17 +54,17 @@ import {
   unknownFontWarnings,
   usesGaejosikMunche,
   validateHwpx
-} from "./chunk-QQERGR6C.js";
+} from "./chunk-AIFZNEJ7.js";
 import {
   blocksToPages
-} from "./chunk-JEKZXUIJ.js";
+} from "./chunk-IZHZO5E5.js";
 import {
   applySplices,
   buildParagraphSplices,
   buildRangeSplices,
   placeSealHwpx,
   scanSectionXml
-} from "./chunk-D4EDAS3V.js";
+} from "./chunk-DD2MPABZ.js";
 import {
   DEFAULT_REDACT_RULES,
   redactMarkdown,
@@ -72,7 +72,7 @@ import {
 } from "./chunk-3WSNLH4B.js";
 import {
   blocksToChunks
-} from "./chunk-GYPEF6GQ.js";
+} from "./chunk-S327K362.js";
 import {
   SPACE_EM_FIXED,
   SPACE_EM_FONT,
@@ -97,20 +97,20 @@ import {
   renderSceneToHtml,
   simulateWrap,
   simulateWrapKeepWord
-} from "./chunk-VSSMJKOX.js";
+} from "./chunk-2R4EUTY2.js";
 import "./chunk-7FCNOKOR.js";
-import "./chunk-6SG56DDC.js";
+import "./chunk-YW42IW7C.js";
 import "./chunk-VQFA55IC.js";
-import "./chunk-G57LD7SG.js";
+import "./chunk-HMCHEPXZ.js";
 import {
   blocksToMarkdown,
   flattenLayoutTables,
   hasStructuredCellContent
-} from "./chunk-6EXVHTBH.js";
-import "./chunk-OVBIDELJ.js";
+} from "./chunk-VVWO34BY.js";
+import "./chunk-LW33OVQI.js";
 import {
   VERSION
-} from "./chunk-FNHELO37.js";
+} from "./chunk-Q4XAUAGE.js";
 export {
   AGENCY_COMMON_NAME,
   BUILTIN_TEMPLATES,
@@ -204,4 +204,4 @@ export {
   usesGaejosikMunche,
   validateHwpx
 };
-//# sourceMappingURL=-LJL6UNRO.js.map
+//# sourceMappingURL=-HOKFROKW.js.map

package/dist/{chunk-VSSMJKOX.js → chunk-2R4EUTY2.js}

@@ -9,7 +9,7 @@ import {
   HEADING_RATIO_H3,
   tidyScriptTags,
   unframeLayoutTables
-} from "./chunk-6SG56DDC.js";
+} from "./chunk-YW42IW7C.js";
 import {
   parsePageRange
 } from "./chunk-VQFA55IC.js";
@@ -22,7 +22,7 @@ import {
   findChildByLocalName,
   hwpFaceToCssStack,
   parseRenderStyles
-} from "./chunk-G57LD7SG.js";
+} from "./chunk-HMCHEPXZ.js";
 import {
   MAX_COLS,
   MAX_ROWS,
@@ -33,18 +33,18 @@ import {
   flattenLayoutTables,
   mapPuaText,
   markNonLayoutTable
-} from "./chunk-6EXVHTBH.js";
+} from "./chunk-VVWO34BY.js";
 import {
   loadSharp,
   rasterizePageSvg
-} from "./chunk-OVBIDELJ.js";
+} from "./chunk-LW33OVQI.js";
 import {
   KordocError,
   escapeHtml,
   precheckZipSize,
   sanitizeHref,
   toArrayBuffer
-} from "./chunk-FNHELO37.js";
+} from "./chunk-Q4XAUAGE.js";
 
 // src/render/svg-render.ts
 import JSZip from "jszip";
@@ -7803,4 +7803,4 @@ export {
   cropRect,
   extractRenderedRegions
 };
-//# sourceMappingURL=chunk-VSSMJKOX.js.map
+//# sourceMappingURL=chunk-2R4EUTY2.js.map

package/dist/{chunk-WB6RROCW.js → chunk-3G3LTCET.js}

@@ -1,7 +1,7 @@
 #!/usr/bin/env node
 import {
   KordocError
-} from "./chunk-FNHELO37.js";
+} from "./chunk-Q4XAUAGE.js";
 
 // src/hwpx/profile-io.ts
 import { z } from "zod";
@@ -89,4 +89,4 @@ export {
   formatProfileSchema,
   parseFormatProfileJson
 };
-//# sourceMappingURL=chunk-WB6RROCW.js.map
+//# sourceMappingURL=chunk-3G3LTCET.js.map

package/dist/{chunk-3ITKDX7E.cjs → chunk-7FVTFUXA.cjs}

@@ -9,10 +9,10 @@ var _chunkTK3FDAEEcjs = require('./chunk-TK3FDAEE.cjs');
 
 
 
-var _chunk7QL3G7UFcjs = require('./chunk-7QL3G7UF.cjs');
+var _chunkPZU22IY6cjs = require('./chunk-PZU22IY6.cjs');
 
 
-var _chunkUYKDMNLDcjs = require('./chunk-UYKDMNLD.cjs');
+var _chunkNRMHYBTCcjs = require('./chunk-NRMHYBTC.cjs');
 
 // src/ocr/ruling-lines.ts
 var INK_LUMA_MAX = 205;
@@ -1414,7 +1414,7 @@ var OcrEngine = class _OcrEngine {
     ]);
     const sharpAny = sharpModRaw;
     const sharpMod = typeof sharpAny === "function" ? sharpAny : _nullishCoalesce(sharpAny.default, () => ( sharpAny));
-    const dir = _chunk7QL3G7UFcjs.getOcrModelsDir.call(void 0, );
+    const dir = _chunkPZU22IY6cjs.getOcrModelsDir.call(void 0, );
     const sessionOpts = {
       graphOptimizationLevel: "all",
       executionProviders: ["cpu"],
@@ -1422,11 +1422,11 @@ var OcrEngine = class _OcrEngine {
       // paddle2onnx 변환 잔여물 W 로그 폭주 억제
     };
     const [det, rec, dictYml] = await Promise.all([
-      ortMod.InferenceSession.create(_path.join.call(void 0, dir, _chunk7QL3G7UFcjs.OCR_DET_MODEL.filename), sessionOpts),
-      ortMod.InferenceSession.create(_path.join.call(void 0, dir, _chunk7QL3G7UFcjs.OCR_REC_MODEL.filename), sessionOpts),
-      _promises.readFile.call(void 0, _path.join.call(void 0, dir, _chunk7QL3G7UFcjs.OCR_REC_DICT.filename), "utf-8")
+      ortMod.InferenceSession.create(_path.join.call(void 0, dir, _chunkPZU22IY6cjs.OCR_DET_MODEL.filename), sessionOpts),
+      ortMod.InferenceSession.create(_path.join.call(void 0, dir, _chunkPZU22IY6cjs.OCR_REC_MODEL.filename), sessionOpts),
+      _promises.readFile.call(void 0, _path.join.call(void 0, dir, _chunkPZU22IY6cjs.OCR_REC_DICT.filename), "utf-8")
     ]);
-    const dict = _chunk7QL3G7UFcjs.parseCharacterDict.call(void 0, dictYml);
+    const dict = _chunkPZU22IY6cjs.parseCharacterDict.call(void 0, dictYml);
     if (dict.length === 0) throw new Error("OCR \uC0AC\uC804 \uD30C\uC2F1 \uC2E4\uD328 \u2014 \uBAA8\uB378 \uCE90\uC2DC\uB97C \uC0AD\uC81C \uD6C4 \uC7AC\uB2E4\uC6B4\uB85C\uB4DC\uD558\uC138\uC694");
     return new _OcrEngine({ det, rec, dict, ort: ortMod, sharp: sharpMod });
   }
@@ -1754,7 +1754,7 @@ async function tryImport(name, loader) {
     return await loader();
   } catch (e) {
     throw new Error(
-      `\uB0B4\uC7A5 OCR \uC744 \uC0AC\uC6A9\uD558\uB824\uBA74 optional dependency '${name}' \uC774 \uD544\uC694\uD569\uB2C8\uB2E4. \`npm install ${name}\` \uD6C4 \uB2E4\uC2DC \uC2E4\uD589\uD558\uC138\uC694.${_chunkUYKDMNLDcjs.OPTIONAL_DEP_INSTALL_HINT} \uC6D0\uC778: ${e.message}`
+      `\uB0B4\uC7A5 OCR \uC744 \uC0AC\uC6A9\uD558\uB824\uBA74 optional dependency '${name}' \uC774 \uD544\uC694\uD569\uB2C8\uB2E4. \`npm install ${name}\` \uD6C4 \uB2E4\uC2DC \uC2E4\uD589\uD558\uC138\uC694.${_chunkNRMHYBTCcjs.OPTIONAL_DEP_INSTALL_HINT} \uC6D0\uC778: ${e.message}`
     );
   }
 }
@@ -1868,7 +1868,7 @@ async function runPdfOcr(buffer, targets, mode, warnings, onProgress, detectTabl
     () => Promise.resolve().then(() => _interopRequireWildcard(require("@hyzyla/pdfium")))
   );
   if (mode === "builtin") {
-    await _chunk7QL3G7UFcjs.ensureOcrModels.call(void 0, (p) => {
+    await _chunkPZU22IY6cjs.ensureOcrModels.call(void 0, (p) => {
       if (p.phase === "download" && p.downloaded === 0) {
         process.stderr.write(`[kordoc-ocr] ${p.spec.name} \uB2E4\uC6B4\uB85C\uB4DC \uC911 (~${p.spec.sizeMb}MB)...
 `);
@@ -2000,30 +2000,65 @@ async function closerReads(rgba, rw, rh, pdfH, hi, regions, engine) {
   }
   return out;
 }
+function groupOcrLines(items) {
+  const order = items.map((_, i) => i).sort((a, b) => items[a].y + items[a].h / 2 - (items[b].y + items[b].h / 2));
+  const lines = [];
+  for (const i of order) {
+    const it = items[i], c = it.y + it.h / 2;
+    const line = lines.find((l) => {
+      const lc = l.reduce((s, j) => s + items[j].y + items[j].h / 2, 0) / l.length;
+      const minH = Math.min(it.h, ...l.map((j) => items[j].h));
+      return Math.abs(c - lc) <= minH / 2 && l.every((j) => {
+        const ov = Math.min(it.x + it.w, items[j].x + items[j].w) - Math.max(it.x, items[j].x);
+        return ov < Math.min(it.w, items[j].w) / 2;
+      });
+    });
+    if (line) line.push(i);
+    else lines.push([i]);
+  }
+  return lines;
+}
 function ocrItemsToBlocks(items, pageNumber, pdfW, pdfH, scale, extraLines, detectTables = true, opList = { fnArray: [], argsArray: [] }) {
-  const norm = items.map((it) => {
+  const bottoms = items.map((it) => it.y + it.h);
+  const xs = items.map((it) => ({ x: it.x, r: it.x + it.w, space: false }));
+  for (const line of groupOcrLines(items)) {
+    if (line.length < 2) continue;
+    const sorted = line.map((j) => items[j].y + items[j].h).sort((a, b) => a - b);
+    for (const j of line) bottoms[j] = sorted[Math.floor(sorted.length / 2)];
+    const byX = [...line].sort((a, b) => items[a].x - items[b].x);
+    for (let k = 1; k < byX.length; k++) {
+      const L = xs[byX[k - 1]], R = xs[byX[k]];
+      if (R.x >= L.r) continue;
+      const mid = (R.x + L.r) / 2, half = Math.min(items[byX[k - 1]].h, items[byX[k]].h) * 0.15;
+      L.r = Math.max(L.x + 1, mid - half);
+      R.x = Math.min(R.r - 1, mid + half);
+      R.space = true;
+    }
+  }
+  const norm = items.map((it, i) => {
     const h = it.h / scale;
     return {
       text: it.text,
-      x: Math.round(it.x / scale),
+      x: Math.round(xs[i].x / scale),
       // NormItem.y 는 pdfjs transform[5] = 베이스라인 (bottom-up) — 잉크 하단으로 근사
-      y: Math.round(pdfH - (it.y + it.h) / scale),
-      w: Math.round(it.w / scale),
+      y: Math.round(pdfH - bottoms[i] / scale),
+      w: Math.round((xs[i].r - xs[i].x) / scale),
       h: Math.round(h),
       // 박스는 잉크 외곽(engine tightBoxes) — 한글 줄 잉크 높이 ≈ 0.9em
       fontSize: Math.max(1, Math.round(h / 0.9)),
       fontName: "ocr",
-      isHidden: false
+      isHidden: false,
+      ...xs[i].space ? { hasSpaceBefore: true } : {}
     };
   });
-  return _chunk7QL3G7UFcjs.extractPageBlocksWithLines.call(void 0, norm, pageNumber, opList, pdfW, pdfH, extraLines, detectTables);
+  return _chunkPZU22IY6cjs.extractPageBlocksWithLines.call(void 0, norm, pageNumber, opList, pdfW, pdfH, extraLines, detectTables);
 }
 async function tryImport2(name, loader) {
   try {
     return await loader();
   } catch (e) {
     throw new Error(
-      `OCR \uC744 \uC0AC\uC6A9\uD558\uB824\uBA74 optional dependency '${name}' \uC774 \uD544\uC694\uD569\uB2C8\uB2E4. \`npm install ${name}\` \uD6C4 \uB2E4\uC2DC \uC2E4\uD589\uD558\uC138\uC694.${_chunkUYKDMNLDcjs.OPTIONAL_DEP_INSTALL_HINT} \uC6D0\uC778: ${e.message}`
+      `OCR \uC744 \uC0AC\uC6A9\uD558\uB824\uBA74 optional dependency '${name}' \uC774 \uD544\uC694\uD569\uB2C8\uB2E4. \`npm install ${name}\` \uD6C4 \uB2E4\uC2DC \uC2E4\uD589\uD558\uC138\uC694.${_chunkNRMHYBTCcjs.OPTIONAL_DEP_INSTALL_HINT} \uC6D0\uC778: ${e.message}`
     );
   }
 }
@@ -2061,5 +2096,6 @@ function bgraToRgba(bgra) {
 
 
 
-exports.getOcrEngine = getOcrEngine; exports.detectRulingLines = detectRulingLines; exports.rulingToPdfLines = rulingToPdfLines; exports.deskewPage = deskewPage; exports.MAX_OCR_PIXELS = MAX_OCR_PIXELS; exports.runPdfOcr = runPdfOcr; exports.ocrItemsToBlocks = ocrItemsToBlocks;
-//# sourceMappingURL=chunk-3ITKDX7E.cjs.map
+
+exports.getOcrEngine = getOcrEngine; exports.detectRulingLines = detectRulingLines; exports.rulingToPdfLines = rulingToPdfLines; exports.deskewPage = deskewPage; exports.MAX_OCR_PIXELS = MAX_OCR_PIXELS; exports.runPdfOcr = runPdfOcr; exports.groupOcrLines = groupOcrLines; exports.ocrItemsToBlocks = ocrItemsToBlocks;
+//# sourceMappingURL=chunk-7FVTFUXA.cjs.map