kordoc 4.10.0 → 4.11.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +11 -0
- package/dist/{-AN4MKOOO.js → -LFHXUYGZ.js} +8 -8
- package/dist/{chunk-HO7N6T36.js → chunk-3P6XONHY.js} +3 -3
- package/dist/{chunk-QYXQJPB7.js → chunk-476CF3WL.js} +3 -3
- package/dist/{chunk-YVPYCQOG.js → chunk-4F4PZQTW.js} +2 -2
- package/dist/{chunk-VR3ECURZ.js → chunk-7XDCXPWC.js} +52 -8
- package/dist/chunk-7XDCXPWC.js.map +1 -0
- package/dist/{chunk-SV2YYPBO.js → chunk-B6IA745V.js} +2 -2
- package/dist/{chunk-NMXDXHHW.js → chunk-CF2FQBVN.js} +2 -2
- package/dist/{chunk-2FNQN23O.cjs → chunk-FN6NBZ34.cjs} +3 -3
- package/dist/{chunk-2FNQN23O.cjs.map → chunk-FN6NBZ34.cjs.map} +1 -1
- package/dist/{chunk-B4MQXOU5.js → chunk-FNNOBF36.js} +52 -8
- package/dist/chunk-FNNOBF36.js.map +1 -0
- package/dist/{chunk-V7F5765M.cjs → chunk-FQTPFGVG.cjs} +54 -10
- package/dist/chunk-FQTPFGVG.cjs.map +1 -0
- package/dist/{chunk-PDXRLQD6.js → chunk-HXR7TSXU.js} +2 -2
- package/dist/{chunk-4SIKWOLQ.cjs → chunk-JO3W54E5.cjs} +2 -2
- package/dist/{chunk-4SIKWOLQ.cjs.map → chunk-JO3W54E5.cjs.map} +1 -1
- package/dist/{chunk-F3OPGGQ2.js → chunk-LM5A4DOY.js} +2 -2
- package/dist/{chunk-EWUFLCRP.js → chunk-LXMQUWGU.js} +3 -3
- package/dist/{chunk-KEF4UQYI.js → chunk-LYTFBD6P.js} +2 -2
- package/dist/{chunk-ZOLKTMBS.cjs → chunk-RSATURJN.cjs} +7 -7
- package/dist/{chunk-ZOLKTMBS.cjs.map → chunk-RSATURJN.cjs.map} +1 -1
- package/dist/{chunk-WLUDCIOK.js → chunk-SD3ITJLU.js} +2 -2
- package/dist/{chunk-34IBS3JN.js → chunk-V2TMM6OD.js} +2 -2
- package/dist/{chunk-4EJLRZ7A.js → chunk-W4DVZN5M.js} +187 -11
- package/dist/chunk-W4DVZN5M.js.map +1 -0
- package/dist/{chunk-AOTA4XTU.js → chunk-W6537QWB.js} +2 -2
- package/dist/{chunk-NE7GR65O.js → chunk-X25PKMMD.js} +3 -3
- package/dist/{chunk-OQNZCCJI.cjs → chunk-YBTTQQBH.cjs} +6 -6
- package/dist/{chunk-OQNZCCJI.cjs.map → chunk-YBTTQQBH.cjs.map} +1 -1
- package/dist/{chunk-JAY6BXZ6.js → chunk-Z37BVIRF.js} +2 -2
- package/dist/{chunk-7UUQXASL.js → chunk-Z6NFU6DA.js} +2 -2
- package/dist/chunks-Y32RX6IF.js +10 -0
- package/dist/cli.js +20 -20
- package/dist/{formula-LFDXPFQH.js → formula-2ORTZGRU.js} +4 -4
- package/dist/{formula-4QA6ZSFP.cjs → formula-KFSWOSCF.cjs} +9 -9
- package/dist/{formula-4QA6ZSFP.cjs.map → formula-KFSWOSCF.cjs.map} +1 -1
- package/dist/{formula-UYS7Z7IH.js → formula-YCFGCLUF.js} +3 -3
- package/dist/{image-ocr-NOU6O7ZY.js → image-ocr-2B7NVDRS.js} +7 -7
- package/dist/{image-ocr-3YP5I2H4.js → image-ocr-JCKC4QWU.js} +5 -5
- package/dist/{image-ocr-POPCJGL5.cjs → image-ocr-ZV4AP2I6.cjs} +11 -11
- package/dist/{image-ocr-POPCJGL5.cjs.map → image-ocr-ZV4AP2I6.cjs.map} +1 -1
- package/dist/index.cjs +364 -188
- package/dist/index.cjs.map +1 -1
- package/dist/index.js +183 -7
- package/dist/index.js.map +1 -1
- package/dist/mcp.js +15 -15
- package/dist/{model-bundle-AUKJF6GA.js → model-bundle-EX75CSEI.js} +5 -5
- package/dist/{models-QKSQSUWJ.js → models-KJCNM77P.js} +5 -5
- package/dist/{parser-6HKVNDZM.js → parser-2KFORGOT.js} +6 -6
- package/dist/{parser-VVGAZIAX.js → parser-TXBQATM7.js} +6 -6
- package/dist/{parser-VAIMONF2.cjs → parser-WY5PY5V4.cjs} +26 -26
- package/dist/{parser-VAIMONF2.cjs.map → parser-WY5PY5V4.cjs.map} +1 -1
- package/dist/pdf-ocr-GD6KYSH7.js +16 -0
- package/dist/pdf-ocr-JXYOCUGR.cjs +14 -0
- package/dist/{pdf-ocr-4CWHKKTX.cjs.map → pdf-ocr-JXYOCUGR.cjs.map} +1 -1
- package/dist/pdf-ocr-MNCYWT4Z.js +13 -0
- package/dist/{profile-io-PYCSBITE.js → profile-io-Z5NKKNGW.js} +3 -3
- package/dist/{rasterize-TUBHYALD.js → rasterize-66QIXE2S.js} +2 -2
- package/dist/render-ILBFLY3A.js +10 -0
- package/dist/seal-AKUPO4YT.js +10 -0
- package/dist/{setup-6EUKBDMC.js → setup-L2LYDZ6N.js} +3 -3
- package/dist/{watch-JVGXPTCI.js → watch-HJAEDIKZ.js} +9 -9
- package/package.json +1 -1
- package/dist/chunk-4EJLRZ7A.js.map +0 -1
- package/dist/chunk-B4MQXOU5.js.map +0 -1
- package/dist/chunk-V7F5765M.cjs.map +0 -1
- package/dist/chunk-VR3ECURZ.js.map +0 -1
- package/dist/chunks-JUBR2TW7.js +0 -10
- package/dist/pdf-ocr-4CWHKKTX.cjs +0 -14
- package/dist/pdf-ocr-D6HR23OB.js +0 -13
- package/dist/pdf-ocr-WIMNTEVH.js +0 -16
- package/dist/render-QW5Z4IMW.js +0 -10
- package/dist/seal-OTOP5CLA.js +0 -10
- /package/dist/{-AN4MKOOO.js.map → -LFHXUYGZ.js.map} +0 -0
- /package/dist/{chunk-HO7N6T36.js.map → chunk-3P6XONHY.js.map} +0 -0
- /package/dist/{chunk-QYXQJPB7.js.map → chunk-476CF3WL.js.map} +0 -0
- /package/dist/{chunk-YVPYCQOG.js.map → chunk-4F4PZQTW.js.map} +0 -0
- /package/dist/{chunk-SV2YYPBO.js.map → chunk-B6IA745V.js.map} +0 -0
- /package/dist/{chunk-NMXDXHHW.js.map → chunk-CF2FQBVN.js.map} +0 -0
- /package/dist/{chunk-PDXRLQD6.js.map → chunk-HXR7TSXU.js.map} +0 -0
- /package/dist/{chunk-F3OPGGQ2.js.map → chunk-LM5A4DOY.js.map} +0 -0
- /package/dist/{chunk-EWUFLCRP.js.map → chunk-LXMQUWGU.js.map} +0 -0
- /package/dist/{chunk-KEF4UQYI.js.map → chunk-LYTFBD6P.js.map} +0 -0
- /package/dist/{chunk-WLUDCIOK.js.map → chunk-SD3ITJLU.js.map} +0 -0
- /package/dist/{chunk-34IBS3JN.js.map → chunk-V2TMM6OD.js.map} +0 -0
- /package/dist/{chunk-AOTA4XTU.js.map → chunk-W6537QWB.js.map} +0 -0
- /package/dist/{chunk-NE7GR65O.js.map → chunk-X25PKMMD.js.map} +0 -0
- /package/dist/{chunk-JAY6BXZ6.js.map → chunk-Z37BVIRF.js.map} +0 -0
- /package/dist/{chunk-7UUQXASL.js.map → chunk-Z6NFU6DA.js.map} +0 -0
- /package/dist/{chunks-JUBR2TW7.js.map → chunks-Y32RX6IF.js.map} +0 -0
- /package/dist/{formula-LFDXPFQH.js.map → formula-2ORTZGRU.js.map} +0 -0
- /package/dist/{formula-UYS7Z7IH.js.map → formula-YCFGCLUF.js.map} +0 -0
- /package/dist/{image-ocr-NOU6O7ZY.js.map → image-ocr-2B7NVDRS.js.map} +0 -0
- /package/dist/{image-ocr-3YP5I2H4.js.map → image-ocr-JCKC4QWU.js.map} +0 -0
- /package/dist/{model-bundle-AUKJF6GA.js.map → model-bundle-EX75CSEI.js.map} +0 -0
- /package/dist/{models-QKSQSUWJ.js.map → models-KJCNM77P.js.map} +0 -0
- /package/dist/{parser-6HKVNDZM.js.map → parser-2KFORGOT.js.map} +0 -0
- /package/dist/{parser-VVGAZIAX.js.map → parser-TXBQATM7.js.map} +0 -0
- /package/dist/{pdf-ocr-D6HR23OB.js.map → pdf-ocr-GD6KYSH7.js.map} +0 -0
- /package/dist/{pdf-ocr-WIMNTEVH.js.map → pdf-ocr-MNCYWT4Z.js.map} +0 -0
- /package/dist/{profile-io-PYCSBITE.js.map → profile-io-Z5NKKNGW.js.map} +0 -0
- /package/dist/{rasterize-TUBHYALD.js.map → rasterize-66QIXE2S.js.map} +0 -0
- /package/dist/{render-QW5Z4IMW.js.map → render-ILBFLY3A.js.map} +0 -0
- /package/dist/{seal-OTOP5CLA.js.map → seal-AKUPO4YT.js.map} +0 -0
- /package/dist/{setup-6EUKBDMC.js.map → setup-L2LYDZ6N.js.map} +0 -0
- /package/dist/{watch-JVGXPTCI.js.map → watch-HJAEDIKZ.js.map} +0 -0
package/README.md
CHANGED
|
@@ -89,6 +89,17 @@ MCP 등록 대신 스킬(SKILL.md) 형태로 쓰려면:
|
|
|
89
89
|
|
|
90
90
|
---
|
|
91
91
|
|
|
92
|
+
## v4.11.0 변경사항
|
|
93
|
+
|
|
94
|
+
rhwp 최신 devel(v0.8.3~0.8.4)을 훑어 kordoc 의 HWP3 디코더에 반영했습니다. 네 결함 모두 **글자가 경고 없이 사라지는** 형태라, 변환은 성공했는데 본문 일부가 비어 있는 식으로만 드러납니다.
|
|
95
|
+
|
|
96
|
+
- **🈳 HWP3 기호·한자 유실 (완성형 좌표 규칙 누락)**: HWP3 은 KS X 1001 기호행(0xA1~0xAC)을 행 간격 96, 한자행(0xCA~0xFD)을 행 간격 94 로 편 코드에 문자를 담습니다. rhwp 에서 johab 디코더를 포팅할 때 하드코딩 표만 옮기고 **이 규칙 두 개를 빠뜨려**, 표에 없는 코드는 전부 삭제됐습니다 — `채권(債權)조서` 가 `채권()조서` 가 되고 `※ ○ 【】 ∴ ☞ ₩ ━` 같은 기호가 통째로 사라집니다. 규칙을 이식하고, 표준 매핑과 한컴 표기가 갈리는 두 자리(`∼→~`, `⊙→◉`)를 보정했습니다.
|
|
97
|
+
- **🔤 사적 코드 실측 표 확장**: 라틴 확장(`ü·ö·ä·ß` 가 지워져 "für"→"fr" 이 되던 것)과 rhwp #5860 의 실측 28코드를 이식하고, rhwp 표에도 없던 코드를 직접 실측해 더했습니다 — 괘선 조각 6종, 겹줄, 원문자·괄호문자 계열(`⓪①②`·`ⓐ~ⓕ`·`㉠~㉢`), 글머리표 `•` 등. 근거는 HWP3 원본과 같은 문서의 한컴 변환본을 문맥 정렬한 것으로, **코드 출현 횟수가 변환본의 해당 문자 개수와 정확히 맞는 것만** 실었습니다.
|
|
98
|
+
- **✒️ 초성 없는 아래아 음절 유실**: 초성이 '채움'인 아래아 음절(`ㆍ` 단독)을 무효 초성으로 보고 버렸습니다. 한컴 변환본과 같이 자모 하나로 남깁니다.
|
|
99
|
+
- **🧱 서식 run 이 PUA 정리를 건너뛰던 결함 (전 포맷)**: 굵게·밑줄이 걸린 문단은 span 경로를 타는데 여기서만 PUA 치환·제거를 하지 않아, 한컴 전용 글머리표·괘선 조각이 **원시 PUA(두부 문자) 그대로** 마크다운에 실렸습니다. HWP5·HWPX·HWP3 공통 경로입니다. 함께 rhwp 의 한컴 PUA 검증표(한컴 PDF 대조 확정)에서 빠져 있던 22종(`┌┬┐└┘│`·`┈═━`·`⓪①②④~⑨`·`✺◇▸□`·`⊞⊟`)을 동기화했습니다.
|
|
100
|
+
|
|
101
|
+
검증은 rhwp `samples/` 의 HWP3 원본 9종을 같은 문서의 한컴 변환본과 문자 멀티셋으로 대조했습니다. 변환본에만 있는 문자(=HWP3 경로에서 잃은 문자)가 `hwp3-sample11` 189→0, `hwp3-sample16` 1,139→618, `hwp3-sample10` 172→11 로 줄었습니다. 남은 618자는 **그리기 개체 안 글상자**(표지 제목·순서도 라벨)로, kordoc 은 아직 그리기 개체를 건너뜁니다.
|
|
102
|
+
|
|
92
103
|
## v4.10.0 변경사항
|
|
93
104
|
|
|
94
105
|
- **📡 실패 계약 전 포맷 확장 (#69)**: 실패 JSON(`success:false` + `code`)이 `--format json` 전용에서 **markdown·chunks 포함 전 포맷**으로 확장됐습니다. 헤드리스 파이프라인이 stderr 한국어 문구 파싱 없이 원인 코드로 분기할 수 있습니다. `code` 값 집합·종료 코드 규칙·안정성 보증을 README [실패 계약](#실패-계약--기계-판독-가능한-실패-json-v4100-69) 표로 문서화했습니다. (@sorbetsharkroundhand 제보)
|
|
@@ -56,7 +56,7 @@ import {
|
|
|
56
56
|
unknownFontWarnings,
|
|
57
57
|
usesGaejosikMunche,
|
|
58
58
|
validateHwpx
|
|
59
|
-
} from "./chunk-
|
|
59
|
+
} from "./chunk-W4DVZN5M.js";
|
|
60
60
|
import "./chunk-7FCNOKOR.js";
|
|
61
61
|
import "./chunk-N3DLQ2YH.js";
|
|
62
62
|
import "./chunk-MOL7MDBG.js";
|
|
@@ -66,7 +66,7 @@ import {
|
|
|
66
66
|
buildRangeSplices,
|
|
67
67
|
placeSealHwpx,
|
|
68
68
|
scanSectionXml
|
|
69
|
-
} from "./chunk-
|
|
69
|
+
} from "./chunk-LXMQUWGU.js";
|
|
70
70
|
import {
|
|
71
71
|
DEFAULT_REDACT_RULES,
|
|
72
72
|
redactMarkdown,
|
|
@@ -74,10 +74,10 @@ import {
|
|
|
74
74
|
} from "./chunk-DZIXKL7E.js";
|
|
75
75
|
import {
|
|
76
76
|
blocksToChunks
|
|
77
|
-
} from "./chunk-
|
|
77
|
+
} from "./chunk-LM5A4DOY.js";
|
|
78
78
|
import {
|
|
79
79
|
blocksToMarkdown
|
|
80
|
-
} from "./chunk-
|
|
80
|
+
} from "./chunk-FNNOBF36.js";
|
|
81
81
|
import {
|
|
82
82
|
SPACE_EM_FIXED,
|
|
83
83
|
SPACE_EM_FONT,
|
|
@@ -87,11 +87,11 @@ import {
|
|
|
87
87
|
renderHwpxToSvg,
|
|
88
88
|
simulateWrap,
|
|
89
89
|
simulateWrapKeepWord
|
|
90
|
-
} from "./chunk-
|
|
91
|
-
import "./chunk-
|
|
90
|
+
} from "./chunk-476CF3WL.js";
|
|
91
|
+
import "./chunk-Z37BVIRF.js";
|
|
92
92
|
import {
|
|
93
93
|
VERSION
|
|
94
|
-
} from "./chunk-
|
|
94
|
+
} from "./chunk-CF2FQBVN.js";
|
|
95
95
|
export {
|
|
96
96
|
BUILTIN_TEMPLATES,
|
|
97
97
|
DEFAULT_REDACT_RULES,
|
|
@@ -169,4 +169,4 @@ export {
|
|
|
169
169
|
usesGaejosikMunche,
|
|
170
170
|
validateHwpx
|
|
171
171
|
};
|
|
172
|
-
//# sourceMappingURL=-
|
|
172
|
+
//# sourceMappingURL=-LFHXUYGZ.js.map
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
#!/usr/bin/env node
|
|
2
2
|
import {
|
|
3
3
|
extractPageBlocksWithLines
|
|
4
|
-
} from "./chunk-
|
|
4
|
+
} from "./chunk-V2TMM6OD.js";
|
|
5
5
|
import {
|
|
6
6
|
OCR_DET_MODEL,
|
|
7
7
|
OCR_REC_DICT,
|
|
@@ -9,7 +9,7 @@ import {
|
|
|
9
9
|
ensureOcrModels,
|
|
10
10
|
getOcrModelsDir,
|
|
11
11
|
parseCharacterDict
|
|
12
|
-
} from "./chunk-
|
|
12
|
+
} from "./chunk-HXR7TSXU.js";
|
|
13
13
|
|
|
14
14
|
// src/ocr/ruling-lines.ts
|
|
15
15
|
var INK_LUMA_MAX = 190;
|
|
@@ -513,4 +513,4 @@ export {
|
|
|
513
513
|
runPdfOcr,
|
|
514
514
|
ocrItemsToBlocks
|
|
515
515
|
};
|
|
516
|
-
//# sourceMappingURL=chunk-
|
|
516
|
+
//# sourceMappingURL=chunk-3P6XONHY.js.map
|
|
@@ -7,11 +7,11 @@ import {
|
|
|
7
7
|
createXmlParser,
|
|
8
8
|
findChildByLocalName,
|
|
9
9
|
parseRenderStyles
|
|
10
|
-
} from "./chunk-
|
|
10
|
+
} from "./chunk-Z37BVIRF.js";
|
|
11
11
|
import {
|
|
12
12
|
KordocError,
|
|
13
13
|
precheckZipSize
|
|
14
|
-
} from "./chunk-
|
|
14
|
+
} from "./chunk-CF2FQBVN.js";
|
|
15
15
|
|
|
16
16
|
// src/render/svg-render.ts
|
|
17
17
|
import JSZip from "jszip";
|
|
@@ -1286,4 +1286,4 @@ export {
|
|
|
1286
1286
|
fitRatioForFewerLines,
|
|
1287
1287
|
renderHwpxToSvg
|
|
1288
1288
|
};
|
|
1289
|
-
//# sourceMappingURL=chunk-
|
|
1289
|
+
//# sourceMappingURL=chunk-476CF3WL.js.map
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
#!/usr/bin/env node
|
|
2
2
|
import {
|
|
3
3
|
KordocError
|
|
4
|
-
} from "./chunk-
|
|
4
|
+
} from "./chunk-CF2FQBVN.js";
|
|
5
5
|
|
|
6
6
|
// src/hwpx/profile-io.ts
|
|
7
7
|
import { z } from "zod";
|
|
@@ -89,4 +89,4 @@ export {
|
|
|
89
89
|
formatProfileSchema,
|
|
90
90
|
parseFormatProfileJson
|
|
91
91
|
};
|
|
92
|
-
//# sourceMappingURL=chunk-
|
|
92
|
+
//# sourceMappingURL=chunk-4F4PZQTW.js.map
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import {
|
|
2
2
|
sanitizeHref
|
|
3
|
-
} from "./chunk-
|
|
3
|
+
} from "./chunk-Z6NFU6DA.js";
|
|
4
4
|
|
|
5
5
|
// src/shared/pua.ts
|
|
6
6
|
var BMP_SYMBOL_MAP = {
|
|
@@ -139,7 +139,47 @@ var SUPPLEMENTARY_MAP = {
|
|
|
139
139
|
984049: "\uACFC",
|
|
140
140
|
984050: "\uCEF4",
|
|
141
141
|
984051: "\uD4E8",
|
|
142
|
-
984052: "\uD130"
|
|
142
|
+
984052: "\uD130",
|
|
143
|
+
// 아래는 rhwp VERIFIED_HANCOM_PUA_DISPLAY(한컴 PDF 대조 확정표)에서 kordoc 에만
|
|
144
|
+
// 빠져 있던 항목. 매핑이 없으면 sanitizeText 가 지워 글자가 사라진다 —
|
|
145
|
+
// hwp3-sample11 한 문서에서만 괘선 조각·원문자 57자가 그렇게 증발했다.
|
|
146
|
+
983184: "\u273A",
|
|
147
|
+
// 물방울 asterisk 글머리
|
|
148
|
+
983688: "\u24EA",
|
|
149
|
+
// 별도 글리프 원숫자 — ③(F028B)는 근거 문서가 리터럴을 써 rhwp 표에도 없다
|
|
150
|
+
983689: "\u2460",
|
|
151
|
+
983690: "\u2461",
|
|
152
|
+
983692: "\u2463",
|
|
153
|
+
983693: "\u2464",
|
|
154
|
+
983694: "\u2465",
|
|
155
|
+
983695: "\u2466",
|
|
156
|
+
983696: "\u2467",
|
|
157
|
+
983697: "\u2468",
|
|
158
|
+
983788: "\u25C7",
|
|
159
|
+
// 작은 빈 마름모 글머리
|
|
160
|
+
983803: "\u25B8",
|
|
161
|
+
// 중첩 표 글머리
|
|
162
|
+
983975: "\u229F",
|
|
163
|
+
// 둥근 네모 안 −
|
|
164
|
+
983976: "\u229E",
|
|
165
|
+
// 둥근 네모 안 +
|
|
166
|
+
984026: "\u25A1",
|
|
167
|
+
// 둥근 모서리 빈 네모 글머리
|
|
168
|
+
984063: "\u25A1",
|
|
169
|
+
// 표 셀 제목 빈 네모 글머리
|
|
170
|
+
985094: "\u250C",
|
|
171
|
+
// 텍스트 다이어그램 괘선 조각
|
|
172
|
+
985095: "\u252C",
|
|
173
|
+
985096: "\u2510",
|
|
174
|
+
985100: "\u2514",
|
|
175
|
+
985102: "\u2518",
|
|
176
|
+
985104: "\u2502",
|
|
177
|
+
985116: "\u2508",
|
|
178
|
+
// 점선 괘선 조각
|
|
179
|
+
985138: "\u2550",
|
|
180
|
+
// 이중 가로 괘선 조각
|
|
181
|
+
985160: "\u2501"
|
|
182
|
+
// 굵은 가로 막대 글머리
|
|
143
183
|
};
|
|
144
184
|
var BOXED_NUMBER_START = 983729;
|
|
145
185
|
var BOXED_NUMBER_END = 983748;
|
|
@@ -283,8 +323,11 @@ function escapeGfm(text) {
|
|
|
283
323
|
return escaped.replace(new RegExp(NUL + "(\\d+)" + NUL, "g"), (_, n) => spans[Number(n)]);
|
|
284
324
|
}
|
|
285
325
|
var HWP_SHAPE_ALT_TEXT_RE = /^(?:모서리가 둥근 |둥근 )?(?:사각형|직사각형|정사각형|원|타원|삼각형|이등변 삼각형|직각 삼각형|선|직선|곡선|화살표|굵은 화살표|이중 화살표|오각형|육각형|팔각형|별|[4-8]점별|십자|십자형|구름|구름형|마름모|도넛|평행사변형|사다리꼴|부채꼴|호|반원|물결|번개|하트|빗금|블록 화살표|수식|표|그림|개체|그리기\s?개체|묶음\s?개체|글상자|수식\s?개체|OLE\s?개체)\s?입니다\.?$/gm;
|
|
326
|
+
function sanitizePua(text) {
|
|
327
|
+
return mapPuaText(text).replace(/[\u{F0000}-\u{FFFFD}]/gu, "");
|
|
328
|
+
}
|
|
286
329
|
function sanitizeText(text) {
|
|
287
|
-
let result =
|
|
330
|
+
let result = sanitizePua(text).replace(HWP_SHAPE_ALT_TEXT_RE, "").replace(/ +/g, " ").trim();
|
|
288
331
|
if (result.length <= 30 && result.includes(" ")) {
|
|
289
332
|
const tokens = result.split(" ");
|
|
290
333
|
const koreanSingleCharCount = tokens.filter((t) => t.length === 1 && /[\uAC00-\uD7AF\u3131-\u318E]/.test(t)).length;
|
|
@@ -373,18 +416,19 @@ function dedupeRunningHeaders(blocks) {
|
|
|
373
416
|
function spansToMarkdown(spans) {
|
|
374
417
|
let out = "";
|
|
375
418
|
for (const s of spans) {
|
|
376
|
-
|
|
419
|
+
const text = sanitizePua(s.text ?? "");
|
|
420
|
+
if (!text) continue;
|
|
377
421
|
let marker = s.code ? "`" : s.bold && s.italic ? "***" : s.bold ? "**" : s.italic ? "*" : "";
|
|
378
422
|
if (s.strike && !s.code) marker = `~~${marker}`;
|
|
379
423
|
const uWrap = !!s.underline && !s.code;
|
|
380
424
|
if (!marker && !uWrap) {
|
|
381
|
-
out += escapeGfm(
|
|
425
|
+
out += escapeGfm(text);
|
|
382
426
|
continue;
|
|
383
427
|
}
|
|
384
|
-
const m = /^(\s*)([\s\S]*?)(\s*)$/.exec(
|
|
428
|
+
const m = /^(\s*)([\s\S]*?)(\s*)$/.exec(text);
|
|
385
429
|
const core = m[2];
|
|
386
430
|
if (!core) {
|
|
387
|
-
out +=
|
|
431
|
+
out += text;
|
|
388
432
|
continue;
|
|
389
433
|
}
|
|
390
434
|
let open = marker;
|
|
@@ -778,4 +822,4 @@ export {
|
|
|
778
822
|
encodePng,
|
|
779
823
|
inlineImagesIntoMarkdown
|
|
780
824
|
};
|
|
781
|
-
//# sourceMappingURL=chunk-
|
|
825
|
+
//# sourceMappingURL=chunk-7XDCXPWC.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"sources":["../src/shared/pua.ts","../src/table/builder.ts","../src/image/transcode.ts"],"sourcesContent":["/**\n * 한컴 PUA(Private Use Area) 문자 → 유니코드 표준 문자 매핑.\n *\n * 한글(HWP)은 글머리표·문자표 기호를 Symbol/Wingdings 계열 PUA 코드포인트\n * (U+F020~U+F0FF)와 한컴 자체 Supplementary PUA-A 영역(U+F0000대)에 저장한다.\n * 매핑 없이 마크다운에 내보내면 빈 네모로 깨진다.\n *\n * 매핑 출처: rhwp (MIT, https://github.com/edwardkim/rhwp)\n * paragraph_layout.rs map_pua_bullet_char — 한컴 PDF 정답지 시각 검증 테이블.\n */\n\n/** BMP Symbol 영역 (U+F020~U+F0FF) 매핑 — 키는 (code - 0xF000) */\nconst BMP_SYMBOL_MAP: Record<number, string> = {\n // 도형/기호\n 0x6c: \"●\", // ●\n 0x6d: \"●\", // ● (그림자 원 근사)\n 0x6e: \"■\", // ■\n 0x6f: \"□\", // □\n 0x70: \"□\", // □ (굵은 흰 사각 근사)\n 0x71: \"□\", // □ (그림자 근사)\n 0x72: \"□\", // □ (그림자 근사)\n 0x73: \"⬧\", // ⬧\n 0x74: \"⧫\", // ⧫\n 0x75: \"◆\", // ◆\n 0x76: \"❖\", // ❖\n 0x77: \"⬥\", // ⬥\n // 체크/별/점\n 0x9e: \"·\", // ·\n 0x9f: \"•\", // •\n 0xa0: \"·\", // · (한컴 PDF 정답지 정합 — ▪ 아님)\n 0xa1: \"⚪\", // ⚪\n 0xa2: \"○\", // ○\n 0xa3: \"○\", // ○\n 0xa4: \"◉\", // ◉\n 0xa5: \"◎\", // ◎\n 0xa7: \"▪\", // ▪\n 0xa8: \"◻\", // ◻\n 0xaa: \"✦\", // ✦\n 0xab: \"★\", // ★\n 0xac: \"✶\", // ✶\n 0xad: \"✴\", // ✴\n 0xae: \"✹\", // ✹\n // 손 모양\n 0x45: \"☜\", // ☜\n 0x46: \"☞\", // ☞\n 0x47: \"☝\", // ☝\n 0x48: \"☟\", // ☟\n // 체크마크\n 0xfb: \"✗\", // ✗\n 0xfc: \"✔\", // ✔\n 0xfd: \"☒\", // ☒\n 0xfe: \"☑\", // ☑\n // 화살표\n 0xe8: \"➔\", // ➔ (heavy wide-headed — 한컴 PDF 정답지 정합)\n 0xef: \"⇦\", // ⇦\n 0xf0: \"⇨\", // ⇨\n 0xf1: \"⇧\", // ⇧\n 0xf2: \"⇩\", // ⇩\n // 기타\n 0x22: \"✂\", // ✂\n 0x36: \"⌛\", // ⌛\n 0x4a: \"☺\", // ☺\n 0x4e: \"☠\", // ☠\n 0x52: \"☼\", // ☼\n 0x54: \"❄\", // ❄\n 0x58: \"✠\", // ✠\n 0x59: \"✡\", // ✡\n}\n\n/** Supplementary PUA-A (U+F0000대) — 한컴 자체 영역 매핑.\n * 매핑 없는 코드는 호출부가 **삭제**하므로(sanitizeText) 누락은 곧 글자 증발이다.\n * 범위 추정으로 채우지 말 것 — PUA는 글꼴별 사적 영역이라 한컴 PDF 대조로 확인된 것만 넣는다. */\nconst SUPPLEMENTARY_MAP: Record<number, string> = {\n 0xf003b: \"↓\", // ↓\n 0xf02ef: \"·\", // ·\n 0xf0854: \"《\", // 《\n 0xf0855: \"》\", // 》\n 0xf00da: \"▸\", // ▸\n 0xf080f: \"━\", // ━\n 0xf0827: \"■\", // ■\n 0xf03c5: \"□\", // □ 글머리 — HWP3→HWP5 한컴 변환본 보존 코드, 한컴오피스 표시값 (rhwp #1105)\n // 아래 5종 — rhwp 44cabad9 verified_hancom_pua 표 (한컴 PDF 대조 확정)\n 0xf012b: \"(인)\", // 결재·서명란\n 0xf02fc: \"►\", // 2025 행정업무운영 편람 callout 글머리\n 0xf031c: \"■\", // 2025 행정업무운영 편람 목차 글머리\n 0xf03a0: \"↵\", // 하이퍼텍스트 안내문의 Enter 키 픽토그램\n // 머리말 회사명 6자 (HWP3 johab 0x37C0~0x37C5의 HWP5/HWPX 변환본 대응 코드)\n 0xf03ef: \"한\",\n 0xf03f0: \"글\",\n 0xf03f1: \"과\",\n 0xf03f2: \"컴\",\n 0xf03f3: \"퓨\",\n 0xf03f4: \"터\",\n // 아래는 rhwp VERIFIED_HANCOM_PUA_DISPLAY(한컴 PDF 대조 확정표)에서 kordoc 에만\n // 빠져 있던 항목. 매핑이 없으면 sanitizeText 가 지워 글자가 사라진다 —\n // hwp3-sample11 한 문서에서만 괘선 조각·원문자 57자가 그렇게 증발했다.\n 0xf0090: \"✺\", // 물방울 asterisk 글머리\n 0xf0288: \"⓪\", // 별도 글리프 원숫자 — ③(F028B)는 근거 문서가 리터럴을 써 rhwp 표에도 없다\n 0xf0289: \"①\",\n 0xf028a: \"②\",\n 0xf028c: \"④\",\n 0xf028d: \"⑤\",\n 0xf028e: \"⑥\",\n 0xf028f: \"⑦\",\n 0xf0290: \"⑧\",\n 0xf0291: \"⑨\",\n 0xf02ec: \"◇\", // 작은 빈 마름모 글머리\n 0xf02fb: \"▸\", // 중첩 표 글머리\n 0xf03a7: \"⊟\", // 둥근 네모 안 −\n 0xf03a8: \"⊞\", // 둥근 네모 안 +\n 0xf03da: \"□\", // 둥근 모서리 빈 네모 글머리\n 0xf03ff: \"□\", // 표 셀 제목 빈 네모 글머리\n 0xf0806: \"┌\", // 텍스트 다이어그램 괘선 조각\n 0xf0807: \"┬\",\n 0xf0808: \"┐\",\n 0xf080c: \"└\",\n 0xf080e: \"┘\",\n 0xf0810: \"│\",\n 0xf081c: \"┈\", // 점선 괘선 조각\n 0xf0832: \"═\", // 이중 가로 괘선 조각\n 0xf0848: \"━\", // 굵은 가로 막대 글머리\n}\n\n/** 사각 안 숫자 ①~⑳ (U+F02B1~F02C4) — 한컴 전용 글리프.\n * 렌더는 사각 글리프를 위해 원문 유지가 맞지만, 마크다운은 폰트 없는 소비자(RAG·grep)가\n * 읽으므로 둘러싸인 숫자로 옮긴다 (rhwp b74b5098 / #3385 — 실문서에서 5건 유출 확인). */\nconst BOXED_NUMBER_START = 0xf02b1\nconst BOXED_NUMBER_END = 0xf02c4\n\n/** 단일 코드포인트 매핑 — 매핑 없으면 원본 유지 */\nfunction mapPuaChar(code: number): string | undefined {\n if (code >= 0xf020 && code <= 0xf0ff) {\n return BMP_SYMBOL_MAP[code - 0xf000]\n }\n if (code >= BOXED_NUMBER_START && code <= BOXED_NUMBER_END) {\n return String.fromCodePoint(0x2460 + (code - BOXED_NUMBER_START))\n }\n if (code >= 0xf0000 && code <= 0xf09ff) {\n return SUPPLEMENTARY_MAP[code]\n }\n return undefined\n}\n\n/**\n * 문자열 내 한컴 PUA 문자를 표준 유니코드로 치환.\n * 매핑 없는 BMP PUA는 원본 유지(옛한글 한양PUA 가능성 — 제거하면 안 됨),\n * 매핑 없는 Supplementary PUA-A는 호출부의 기존 제거 로직에 위임한다.\n */\nexport function mapPuaText(text: string): string {\n let out = \"\"\n for (const ch of text) {\n const code = ch.codePointAt(0)!\n out += mapPuaChar(code) ?? ch\n }\n return out\n}\n","/** 2-pass colSpan/rowSpan 테이블 빌더 및 Markdown 변환 */\n\nimport type { CellContext, IRBlock, IRCell, IRSpan, IRTable } from \"../types.js\"\nimport { sanitizeHref } from \"../utils.js\"\nimport { mapPuaText } from \"../shared/pua.js\"\n\n/** 테이블 열 수 상한 — 한국 공공문서 기준 충분한 값 */\nexport const MAX_COLS = 200\n/** 테이블 행 수 상한 — 메모리 폭주 방지 */\nexport const MAX_ROWS = 10000\n\nexport interface BuildTableOptions {\n /** 실제 셀 앵커가 있는 빈 후행 열(서식 문서의 입력란)을 보존한다 (#47).\n * 앵커 없는 유령 열(span 인플레이션)은 이 옵션과 무관하게 트림.\n * 기본 false: 종전대로 텍스트 기준 전부 트림 (마크다운 가독성·벤치 계약). */\n keepAnchoredEmptyCols?: boolean\n /** 셀 텍스트를 trim하지 않고 선/후행 개행(빈 문단 줄)을 보존한다 (#57).\n * ParseOptions.keepEmptyParagraphs 연동 — 파서가 문단별로 이미 정돈한 텍스트를\n * `\\n` 결합해 넘기는 경로 전용. 기본 false: 종전대로 trim. */\n keepEmptyParagraphs?: boolean\n}\n\nexport function buildTable(rows: CellContext[][], options?: BuildTableOptions): IRTable {\n if (rows.length > MAX_ROWS) rows = rows.slice(0, MAX_ROWS)\n const numRows = rows.length\n\n // colAddr/rowAddr가 있으면 직접 배치 (HWPX cellAddr, HWP5 colAddr/rowAddr)\n const hasAddr = rows.some(row => row.some(c => c.colAddr !== undefined && c.rowAddr !== undefined))\n if (hasAddr) return buildTableDirect(rows, numRows, options)\n\n // Pass 1: maxCols 계산 — 2D 배열 사용 (동적 확장)\n let maxCols = 0\n const tempOccupied: boolean[][] = Array.from({ length: numRows }, () => [])\n\n for (let rowIdx = 0; rowIdx < numRows; rowIdx++) {\n let colIdx = 0\n for (const cell of rows[rowIdx]) {\n while (colIdx < MAX_COLS && tempOccupied[rowIdx][colIdx]) colIdx++\n if (colIdx >= MAX_COLS) break\n\n for (let r = rowIdx; r < Math.min(rowIdx + cell.rowSpan, numRows); r++) {\n for (let c = colIdx; c < Math.min(colIdx + cell.colSpan, MAX_COLS); c++) {\n tempOccupied[r][c] = true\n }\n }\n colIdx += cell.colSpan\n if (colIdx > maxCols) maxCols = colIdx\n }\n }\n\n if (maxCols === 0) return { rows: 0, cols: 0, cells: [], hasHeader: false }\n\n // Pass 2: 실제 배치\n const grid: IRCell[][] = Array.from({ length: numRows }, () =>\n Array.from({ length: maxCols }, () => ({ text: \"\", colSpan: 1, rowSpan: 1 }))\n )\n const occupied: boolean[][] = Array.from({ length: numRows }, () => Array(maxCols).fill(false))\n const anchorCols = new Set<number>()\n\n for (let rowIdx = 0; rowIdx < numRows; rowIdx++) {\n let colIdx = 0\n let cellIdx = 0\n\n while (colIdx < maxCols && cellIdx < rows[rowIdx].length) {\n while (colIdx < maxCols && occupied[rowIdx][colIdx]) colIdx++\n if (colIdx >= maxCols) break\n\n const cell = rows[rowIdx][cellIdx]\n anchorCols.add(colIdx)\n grid[rowIdx][colIdx] = {\n text: options?.keepEmptyParagraphs ? cell.text : cell.text.trim(),\n colSpan: cell.colSpan,\n rowSpan: cell.rowSpan,\n }\n\n for (let r = rowIdx; r < Math.min(rowIdx + cell.rowSpan, numRows); r++) {\n for (let c = colIdx; c < Math.min(colIdx + cell.colSpan, maxCols); c++) {\n occupied[r][c] = true\n }\n }\n\n colIdx += cell.colSpan\n cellIdx++\n }\n }\n\n return trimAndReturn(grid, numRows, maxCols, anchorCols, options)\n}\n\n/** colAddr/rowAddr 절대 좌표 기반 직접 배치 */\nfunction buildTableDirect(rows: CellContext[][], numRows: number, options?: BuildTableOptions): IRTable {\n // 전체 셀에서 maxCols 계산 (MAX_COLS 상한 적용)\n let maxCols = 0\n for (const row of rows) {\n for (const cell of row) {\n const end = (cell.colAddr ?? 0) + cell.colSpan\n if (end > maxCols) maxCols = end\n }\n }\n if (maxCols > MAX_COLS) maxCols = MAX_COLS\n if (maxCols === 0) return { rows: 0, cols: 0, cells: [], hasHeader: false }\n\n const grid: IRCell[][] = Array.from({ length: numRows }, () =>\n Array.from({ length: maxCols }, () => ({ text: \"\", colSpan: 1, rowSpan: 1 }))\n )\n\n const anchorCols = new Set<number>()\n for (const row of rows) {\n for (const cell of row) {\n const r = cell.rowAddr ?? 0\n const c = cell.colAddr ?? 0\n if (r >= numRows || c >= maxCols || r < 0 || c < 0) continue\n\n anchorCols.add(c)\n grid[r][c] = { text: options?.keepEmptyParagraphs ? cell.text : cell.text.trim(), colSpan: cell.colSpan, rowSpan: cell.rowSpan }\n\n // 병합 영역 마킹\n for (let dr = 0; dr < cell.rowSpan; dr++) {\n for (let dc = 0; dc < cell.colSpan; dc++) {\n if (dr === 0 && dc === 0) continue\n if (r + dr < numRows && c + dc < maxCols) {\n grid[r + dr][c + dc] = { text: \"\", colSpan: 1, rowSpan: 1 }\n }\n }\n }\n }\n }\n\n return trimAndReturn(grid, numRows, maxCols, anchorCols, options)\n}\n\n/** 빈 후행 열 제거 후 IRTable 반환 — 기본 텍스트 기준 전부, keepAnchoredEmptyCols면 앵커 없는 유령 열만 (#47) */\nfunction trimAndReturn(grid: IRCell[][], numRows: number, maxCols: number, anchorCols: Set<number>, options?: BuildTableOptions): IRTable {\n let effectiveCols = maxCols\n while (effectiveCols > 0) {\n const colEmpty = grid.every(row => !row[effectiveCols - 1]?.text?.trim())\n if (!colEmpty) break\n // 실제 셀 앵커가 있는 빈 열은 서식 문서의 입력란 — 옵션 시 트림하지 않는다 (#47)\n if (options?.keepAnchoredEmptyCols && anchorCols.has(effectiveCols - 1)) break\n effectiveCols--\n }\n if (effectiveCols < maxCols && effectiveCols > 0) {\n const trimmed = grid.map(row => row.slice(0, effectiveCols))\n return { rows: numRows, cols: effectiveCols, cells: trimmed, hasHeader: numRows > 1 }\n }\n return { rows: numRows, cols: maxCols, cells: grid, hasHeader: numRows > 1 }\n}\n\nexport function convertTableToText(rows: CellContext[][]): string {\n return rows\n .map(row =>\n row\n .map(c => c.text.trim().replace(/\\n/g, \" \").replace(/\\|/g, \"\\\\|\"))\n .filter(Boolean)\n .join(\" / \")\n )\n .filter(Boolean)\n .join(\"\\n\")\n}\n\n/** 마크다운 GFM 특수문자 이스케이프 — remark-gfm 오해석 방지 */\nfunction escapeGfm(text: string): string {\n // ~ → \\~ (GFM strikethrough 방지), * → \\* (emphasis/HR·마스킹 별표 \"******\" 방지),\n // _ → \\_ (emphasis 방지), ` → \\` (inline code 방지).\n // 단 $...$ / $$...$$ 수식 스팬은 KaTeX 문법이라 이스케이프하면 파스 에러가 나므로 보호한다\n // (스팬을 임시 필러로 가린 뒤 escape → 복원). NUL 필러는 마크다운 본문에 등장하지 않는다.\n //  이미지 참조 스팬과 링크 URL부 `](스킴...)`(sanitizeHref 허용\n // 스킴 한정 — 우연한 \"[라벨](식별자)\" 평문은 제외)도 동일 보호 — _ 이스케이프 시 문법 파괴.\n const NUL = String.fromCharCode(0) // 마크다운 본문에 없는 안전한 필러 (소스에 raw NUL 미기입)\n const spans: string[] = []\n const masked = text.replace(/!\\[[^\\]]*\\]\\([^)\\n]*\\)|\\]\\((?:https?:|mailto:|tel:|#)[^)\\n]*\\)|\\$\\$[^$]*\\$\\$|\\$[^$\\n]*\\$/gi, (m) => {\n spans.push(m)\n return NUL + (spans.length - 1) + NUL\n })\n const escaped = masked.replace(/([~*_`])/g, \"\\\\$1\")\n return escaped.replace(new RegExp(NUL + \"(\\\\d+)\" + NUL, \"g\"), (_, n) => spans[Number(n)])\n}\n\n/** HWP 자동생성 도형/개체 대체텍스트 정규식 — 한컴오피스가 삽입하는 모든 알려진 패턴.\n * 행 전체 일치(^…$m)로 한정 — 무앵커면 \"붙임 문서는 표 입니다.\" 같은 본문 중간을 오삭제한다 */\nconst HWP_SHAPE_ALT_TEXT_RE = /^(?:모서리가 둥근 |둥근 )?(?:사각형|직사각형|정사각형|원|타원|삼각형|이등변 삼각형|직각 삼각형|선|직선|곡선|화살표|굵은 화살표|이중 화살표|오각형|육각형|팔각형|별|[4-8]점별|십자|십자형|구름|구름형|마름모|도넛|평행사변형|사다리꼴|부채꼴|호|반원|물결|번개|하트|빗금|블록 화살표|수식|표|그림|개체|그리기\\s?개체|묶음\\s?개체|글상자|수식\\s?개체|OLE\\s?개체)\\s?입니다\\.?$/gm\n\n/** 한컴 PUA 축만 정리 — 표시값 치환 후 매핑 안 된 Supplementary PUA 제거.\n * span 처럼 문단 일부만 다루는 곳은 sanitizeText 의 trim·공백 접합을 쓰면 접합이\n * 깨지므로 이 축만 공유한다. */\nfunction sanitizePua(text: string): string {\n // 한컴 PUA → 표준 유니코드 매핑 (rhwp 검증 테이블) — 제거 regex보다 먼저 적용\n return mapPuaText(text)\n // Supplementary Private Use Area (U+F0000-U+FFFFD) — HWP 전용 기호 (매핑 안 된 잔여분)\n .replace(/[\\u{F0000}-\\u{FFFFD}]/gu, \"\")\n}\n\n/** HWP PUA 특수문자 및 도형 대체텍스트 제거 — 모든 포맷 공통 */\nfunction sanitizeText(text: string): string {\n let result = sanitizePua(text)\n // HWP 도형/개체 자동생성 대체텍스트 제거\n .replace(HWP_SHAPE_ALT_TEXT_RE, \"\")\n .replace(/ +/g, \" \")\n .trim()\n // 균등배분 스페이스 정리 (\"현 장 대 응 단 장\" → \"현장대응단장\")\n // 짧은 텍스트(30자 이하)에서 70%+ 토큰이 한글 1글자면 균등배분으로 판단\n if (result.length <= 30 && result.includes(\" \")) {\n const tokens = result.split(\" \")\n // 한글 1글자 토큰만 카운트 — ASCII 특수문자(< > & 등)는 균등배분이 아님\n const koreanSingleCharCount = tokens.filter(t => t.length === 1 && /[\\uAC00-\\uD7AF\\u3131-\\u318E]/.test(t)).length\n if (tokens.length >= 3 && koreanSingleCharCount / tokens.length >= 0.7) {\n result = tokens.join(\"\")\n }\n }\n return result\n}\n\n/**\n * 레이아웃 테이블 감지 및 해체 — IRBlock 레벨에서 수행\n * 적은 행(≤3) + 셀 내 줄바꿈 다량 → table 블록을 paragraph 블록들로 분해\n * heading 감지 전에 호출해야 해체된 텍스트에 heading 감지 적용 가능\n *\n * 호출 정책(의도): HWP5 파서만 호출한다. 구형 HWP5 문서는 제목/본문을 표로\n * 감싼 레이아웃 표가 흔하지만, HWPX는 그 관행이 드물고 무엇보다 patchHwpx/\n * fillHwpx 무손실 라운드트립이 \"파서 렌더 = 소스맵 표 서수\" 대응에 의존하므로\n * HWPX에서 표를 문단으로 해체하면 표 매핑이 깨진다. HWPX 적용은 코퍼스\n * 전/후 정량 비교 + 라운드트립 e2e 검증이 선행되어야 한다.\n */\nexport function flattenLayoutTables(blocks: IRBlock[]): IRBlock[] {\n const result: IRBlock[] = []\n\n for (const block of blocks) {\n if (block.type !== \"table\" || !block.table) {\n result.push(block)\n continue\n }\n\n const { rows: numRows, cols: numCols, cells } = block.table\n\n // 1x1 테이블은 기존 로직(tableToMarkdown)에서 처리\n if (numRows === 1 && numCols === 1) {\n result.push(block)\n continue\n }\n\n // 레이아웃 테이블 휴리스틱\n if (numRows <= 3) {\n let totalNewlines = 0\n let totalTextLen = 0\n for (let r = 0; r < numRows; r++) {\n for (let c = 0; c < numCols; c++) {\n const t = cells[r]?.[c]?.text || \"\"\n totalNewlines += (t.match(/\\n/g) || []).length\n totalTextLen += t.length\n }\n }\n\n // 레이아웃 테이블 판정: 많은 줄바꿈(>5), 또는 적은 행에 비해 총 텍스트 과다(>300)\n // 단, 열이 4개 이상이면 헤더-값 구조의 데이터 표일 가능성이 높아 해체하지 않는다\n // (실증: 2×10 모집프로그램 표가 문단으로 해체되어 헤더↔값 연결 파괴)\n if (numCols < 4 && (totalNewlines > 5 || (numRows <= 2 && totalTextLen > 300))) {\n // 레이아웃 테이블 → 각 셀을 paragraph 블록으로 분해\n for (let r = 0; r < numRows; r++) {\n for (let c = 0; c < numCols; c++) {\n const cell = cells[r]?.[c]\n if (!cell) continue\n // 셀에 구조화 블록(중첩표·이미지·다중문단)이 있으면 재귀 해체로 구조 보존.\n // 중첩 spec 표(numRows>3)는 해체되지 않고 실제 table 블록으로 살아남는다\n // (text의 \" / \" 평탄화 폴백으로만 남아 유실되던 버그 수정). 셀 blocks는\n // 이미 같은 pageNumber로 생성되므로 그대로 보존된다.\n if (cell.blocks?.length) {\n result.push(...flattenLayoutTables(cell.blocks))\n continue\n }\n const cellText = cell.text?.trim()\n if (!cellText) continue\n // 셀 내 줄바꿈을 별도 paragraph로 분리\n for (const line of cellText.split(\"\\n\")) {\n const trimmed = line.trim()\n if (!trimmed) continue\n result.push({ type: \"paragraph\", text: trimmed, pageNumber: block.pageNumber })\n }\n }\n }\n continue\n }\n }\n\n result.push(block)\n }\n\n return result\n}\n\n/** 러닝 헤더 후보 판정 최대 길이 — 긴 본문 문단을 헤더로 오인하지 않도록 */\nconst RUNNING_HEADER_MAX_LEN = 40\n/** 러닝 헤더로 판정할 최소 반복 횟수 — 페이지마다 재삽입되는 노이즈만 제거 */\nconst RUNNING_HEADER_MIN_FREQ = 3\n\n/** 러닝 헤더 후보 여부 — 짧은 번호매김 섹션 제목(\"2. 과제 구축 내용\")만 */\nfunction isRunningHeaderCandidate(block: IRBlock): boolean {\n if (block.type !== \"paragraph\" && block.type !== \"heading\") return false\n const text = block.text?.trim()\n if (!text || text.length > RUNNING_HEADER_MAX_LEN) return false\n return /^\\d+\\.\\s/.test(text)\n}\n\n/**\n * 페이지 레이아웃 표에서 유래한 반복 러닝 헤더 문단 중복 제거 — IRBlock 레벨.\n *\n * 배경: 구형 HWP5 문서(군 제안서 등)는 본문 전체를 \"페이지 = N×1 표\" 사슬로\n * 구성하고 각 표의 cell(0,0)에 동일한 섹션 헤더(\"2. 과제 구축 내용\")를 반복\n * 배치한다. flattenLayoutTables가 이 표들을 문단으로 해체하면 같은 헤더가\n * 페이지마다 1회씩 독립 문단으로 남아 본문 사이에 노이즈로 흩어진다. HWP\n * 머리말/꼬리말(CTRL_HEAD/CTRL_FOOT)은 applyHeaderFooterEffect에서 이미\n * dedupe되지만, 본문 레이아웃 표의 러닝 헤더는 그 대상이 아니다.\n *\n * 정책(보수적): \"짧은 번호매김 섹션 제목\"이 정확히 동일 텍스트로 3회 이상\n * 반복될 때만 최초 1회를 남기고 이후 중복을 제거한다. 비후보 블록과 3회 미만\n * 후보는 원형 그대로 통과시킨다(본문 삭제 위험 최소화). 위치 무관 — 평탄화된\n * 블록 스트림 전체를 대상으로 하며, 입력 블록을 변형하지 않고 새 배열을 반환한다.\n */\nexport function dedupeRunningHeaders(blocks: IRBlock[]): IRBlock[] {\n // Pass 1: 후보 텍스트 빈도 집계\n const freq = new Map<string, number>()\n for (const block of blocks) {\n if (!isRunningHeaderCandidate(block)) continue\n const text = block.text!.trim()\n freq.set(text, (freq.get(text) ?? 0) + 1)\n }\n\n // Pass 2: 빈도 ≥ 임계값인 후보는 최초 1회만 유지, 이후 중복 제거\n const seen = new Set<string>()\n const result: IRBlock[] = []\n for (const block of blocks) {\n if (isRunningHeaderCandidate(block)) {\n const text = block.text!.trim()\n if ((freq.get(text) ?? 0) >= RUNNING_HEADER_MIN_FREQ) {\n if (seen.has(text)) continue // 이후 중복 — 제거\n seen.add(text)\n }\n }\n result.push(block)\n }\n\n return result\n}\n\n/**\n * 왕복 채널 run-span → 강조 마커 재방출. 마커는 텍스트에 밀착해야 유효하므로\n * span 가장자리 공백은 마커 밖으로 옮긴다. 코드 span 내부는 이스케이프 없이 원문.\n */\nfunction spansToMarkdown(spans: IRSpan[]): string {\n let out = \"\"\n for (const s of spans) {\n // 서식 run 도 문단 텍스트와 같은 PUA 계약을 탄다 — 종전엔 이 경로만 sanitize 를\n // 건너뛰어, 글머리표·괘선 조각이 span 을 타면 원시 PUA 가 마크다운에 그대로 실렸다\n // (hwp3-sample10-hwpx: U+F080F·U+F0827 116자)\n const text = sanitizePua(s.text ?? \"\")\n if (!text) continue\n let marker = s.code ? \"`\" : s.bold && s.italic ? \"***\" : s.bold ? \"**\" : s.italic ? \"*\" : \"\"\n if (s.strike && !s.code) marker = `~~${marker}` // 닫힘은 아래 close 에서 역순 조합 (~~**…**~~)\n const uWrap = !!s.underline && !s.code // 밑줄은 태그 쌍이라 역순 조합 불가 — 별도 최외곽 래핑\n if (!marker && !uWrap) {\n out += escapeGfm(text)\n continue\n }\n const m = /^(\\s*)([\\s\\S]*?)(\\s*)$/.exec(text)!\n const core = m[2]\n if (!core) {\n out += text\n continue\n }\n let open = marker\n let close = [...marker].reverse().join(\"\")\n if (uWrap) { open = \"<u>\" + open; close = close + \"</u>\" } // <u>~~**…**~~</u>\n out += m[1] + open + (s.code ? core : escapeGfm(core)) + close + m[3]\n }\n return out\n}\n\nexport function blocksToMarkdown(blocks: IRBlock[]): string {\n const lines: string[] = []\n\n for (let i = 0; i < blocks.length; i++) {\n const block = blocks[i]\n\n // 헤딩 블록 — escapeGfm 필수: 마스킹 별표(\"홍**\")가 볼드로 소비·삭제되는 것 방지\n if (block.type === \"heading\" && block.text) {\n const prefix = \"#\".repeat(Math.min(block.level || 2, 6))\n const headingText = sanitizeText(block.text)\n if (headingText) lines.push(\"\", `${prefix} ${escapeGfm(headingText)}`, \"\")\n continue\n }\n\n // 이미지 블록 —  참조\n if (block.type === \"image\" && block.text) {\n lines.push(\"\", ``, \"\")\n continue\n }\n\n // 구분선 블록\n if (block.type === \"separator\") {\n lines.push(\"\", \"---\", \"\")\n continue\n }\n\n // 리스트 블록\n if (block.type === \"list\" && block.text) {\n const listText = sanitizeText(block.text)\n if (!listText) continue\n // 텍스트가 이미 번호로 시작하면 그대로 출력 (원래 번호 보존)\n const alreadyNumbered = block.listType === \"ordered\" && /^\\d+\\.\\s/.test(listText)\n const prefix = alreadyNumbered ? \"\" : block.listType === \"ordered\" ? \"1. \" : \"- \"\n lines.push(`${prefix}${escapeGfm(listText)}`)\n if (block.children) {\n for (const child of block.children) {\n const childPrefix = child.listType === \"ordered\" ? \"1.\" : \"-\"\n lines.push(` ${childPrefix} ${escapeGfm(child.text || \"\")}`)\n }\n }\n continue\n }\n\n if (block.type === \"paragraph\" && block.text) {\n let text = sanitizeText(block.text)\n if (!text) continue\n\n // 별표 패턴 (기존 호환)\n if (/^\\[별표\\s*\\d+/.test(text)) {\n const nextBlock = blocks[i + 1]\n if (nextBlock?.type === \"paragraph\" && nextBlock.text && /관련\\)?$/.test(nextBlock.text)) {\n lines.push(\"\", `## ${escapeGfm(text)} ${escapeGfm(nextBlock.text)}`, \"\")\n i++\n } else {\n lines.push(\"\", `## ${escapeGfm(text)}`, \"\")\n }\n continue\n }\n\n if (/^\\([^)]*조[^)]*관련\\)$/.test(text)) {\n lines.push(`*${escapeGfm(text)}*`, \"\")\n continue\n }\n\n // gongmun 리스트 depth 선행 공백 (v4.0.5) — sanitizeText가 선두 공백을 지우므로\n // 텍스트가 아니라 방출 시점에 붙인다. 2칸/단계 = md-runs 리스트 그리드와 동일\n const listIndent = block.listDepth ? \" \".repeat(block.listDepth) : \"\"\n\n // 인라인 강조 run-span (왕복 채널 — 자사 생성 파일 + 외래 실속성 복원) → 마커 재방출.\n // 마커 자체는 이스케이프 대상이 아니므로 span 내부 텍스트만 escapeGfm 후 감싼다\n if (block.spans?.length) {\n let rendered = spansToMarkdown(block.spans)\n if (block.href) {\n const href = sanitizeHref(block.href)\n if (href) rendered = `[${rendered}](${href})`\n }\n if (block.footnoteText) rendered += ` (주: ${block.footnoteText})`\n lines.push(block.quote ? \"> \" + rendered : listIndent + rendered, \"\")\n continue\n }\n\n // 하이퍼링크가 있으면 텍스트에 링크 삽입 (javascript: 등 위험 스킴 제거)\n if (block.href) {\n const href = sanitizeHref(block.href)\n if (href) text = `[${text}](${href})`\n }\n\n // 각주가 있으면 괄호로 인라인 삽입\n if (block.footnoteText) {\n text += ` (주: ${block.footnoteText})`\n }\n\n lines.push(block.quote ? \"> \" + escapeGfm(text) : listIndent + escapeGfm(text), \"\")\n } else if (block.type === \"table\" && block.table) {\n // 테이블 앞에 빈 줄 보장 (마크다운 렌더링 필수)\n if (lines.length > 0 && lines[lines.length - 1] !== \"\") {\n lines.push(\"\")\n }\n // 표 캡션 — 표 위에 강조 문단으로 출력 (v3.0)\n if (block.table.caption) {\n const caption = sanitizeText(block.table.caption)\n if (caption) lines.push(`**${escapeGfm(caption)}**`, \"\")\n }\n const tableMd = tableToMarkdown(block.table)\n if (tableMd) {\n lines.push(tableMd)\n lines.push(\"\")\n }\n }\n }\n\n return lines.join(\"\\n\").trim()\n}\n\n/** 병합 셀 존재 여부 확인 */\nfunction hasMergedCells(table: IRTable): boolean {\n for (const row of table.cells) {\n for (const cell of row) {\n if (cell.colSpan > 1 || cell.rowSpan > 1) return true\n }\n }\n return false\n}\n\n/** 셀 내부에 중첩 표 블록 존재 여부 — v3.0 */\nfunction hasNestedTables(table: IRTable): boolean {\n for (const row of table.cells) {\n for (const cell of row) {\n if (cell.blocks?.some(b => b.type === \"table\" && b.table)) return true\n }\n }\n return false\n}\n\n/** 셀 내부 콘텐츠 → HTML — blocks(중첩표/다중문단) 있으면 구조 보존 재귀 렌더링 */\nfunction cellInnerHtml(cell: IRCell): string {\n if (cell.blocks?.length) {\n return cell.blocks\n .map(b => {\n if (b.type === \"table\" && b.table) {\n // 중첩표 캡션도 보존 — 표 위에 텍스트로\n const cap = b.table.caption ? sanitizeText(b.table.caption) : \"\"\n return (cap ? cap + \"<br>\" : \"\") + tableToHtml(b.table)\n }\n if (b.type === \"image\" && b.text) return `<img src=\"${b.text}\" alt=\"image\">`\n const t = sanitizeText(b.text ?? \"\")\n return t ? t.replace(/\\n/g, \"<br>\") : \"\"\n })\n .filter(Boolean)\n .join(\"<br>\")\n }\n return sanitizeText(cell.text).replace(/\\n/g, \"<br>\")\n}\n\nfunction containsInlineMath(text: string): boolean {\n // 교대 중복 금지: [^$\\n\\\\]가 백슬래시를 제외해 \\\\.와 겹치지 않는다 —\n // 겹치면 \"$\"+백슬래시 연속 입력에서 지수 백트래킹(ReDoS)\n return /(^|[^\\\\])\\$(?=\\S)(?:[^$\\n\\\\]|\\\\.)+?\\S\\$/.test(text)\n}\n\nfunction tableContainsInlineMath(table: IRTable): boolean {\n for (const row of table.cells) {\n for (const cell of row) {\n if (containsInlineMath(cell.text)) return true\n }\n }\n return false\n}\n\n/** 병합 테이블 → HTML <table> 출력 (rowspan/colspan 보존) */\nfunction tableToHtml(table: IRTable): string {\n const { cells, rows: numRows, cols: numCols } = table\n const skip = new Set<string>()\n const lines: string[] = [\"<table>\"]\n\n for (let r = 0; r < numRows; r++) {\n const tag = r === 0 ? \"th\" : \"td\"\n const rowHtml: string[] = []\n for (let c = 0; c < numCols; c++) {\n if (skip.has(`${r},${c}`)) continue\n const cell = cells[r]?.[c]\n if (!cell) continue\n\n // 병합 영역 skip 마킹\n for (let dr = 0; dr < cell.rowSpan; dr++) {\n for (let dc = 0; dc < cell.colSpan; dc++) {\n if (dr === 0 && dc === 0) continue\n if (r + dr < numRows && c + dc < numCols) skip.add(`${r + dr},${c + dc}`)\n }\n }\n\n const text = cellInnerHtml(cell)\n const attrs: string[] = []\n if (cell.colSpan > 1) attrs.push(`colspan=\"${cell.colSpan}\"`)\n if (cell.rowSpan > 1) attrs.push(`rowspan=\"${cell.rowSpan}\"`)\n const attrStr = attrs.length ? \" \" + attrs.join(\" \") : \"\"\n rowHtml.push(`<${tag}${attrStr}>${text}</${tag}>`)\n }\n if (rowHtml.length) lines.push(`<tr>${rowHtml.join(\"\")}</tr>`)\n }\n\n lines.push(\"</table>\")\n return lines.join(\"\\n\")\n}\n\nfunction tableToMarkdown(table: IRTable): string {\n if (table.rows === 0 || table.cols === 0) return \"\"\n\n const { cells, rows: numRows, cols: numCols } = table\n\n // 병합 셀·중첩표가 있으면 HTML 테이블로 출력하되, 수식이 있으면 GFM 표로 출력한다.\n // 많은 Markdown 렌더러가 raw HTML table 내부의 $...$를 수식으로 다시 처리하지 않는다.\n if ((hasMergedCells(table) || hasNestedTables(table)) && !tableContainsInlineMath(table)) {\n return tableToHtml(table)\n }\n\n // 1행 1열 → 구조화된 텍스트 (빈 셀이면 스킵)\n if (numRows === 1 && numCols === 1) {\n const content = sanitizeText(cells[0][0].text)\n if (!content) return \"\"\n return content\n .split(/\\n/)\n .map(line => {\n const trimmed = line.trim()\n if (!trimmed) return \"\"\n if (/^\\d+\\.\\s/.test(trimmed)) return `**${escapeGfm(trimmed)}**`\n if (/^[가-힣]\\.\\s/.test(trimmed)) return ` ${escapeGfm(trimmed)}`\n return escapeGfm(trimmed)\n })\n .filter(Boolean)\n .join(\"\\n\")\n }\n\n // 1열 다행 테이블 → 각 행을 별도 라인으로 출력 (목록성 데이터)\n if (numCols === 1 && numRows >= 2) {\n return cells\n .map(row => escapeGfm(sanitizeText(row[0].text)).replace(/\\n/g, \" \"))\n .filter(Boolean)\n .join(\"\\n\")\n }\n\n // 병합 셀: 행/열 병합된 셀은 빈 칸으로\n const display: string[][] = Array.from({ length: numRows }, () => Array(numCols).fill(\"\"))\n const skip = new Set<string>()\n\n for (let r = 0; r < numRows; r++) {\n for (let c = 0; c < numCols; c++) {\n if (skip.has(`${r},${c}`)) continue\n const cell = cells[r]?.[c]\n if (!cell) continue\n // 왕복 채널 셀 spans (v4.0.4) — 강조 마커 재방출 (문단별, 개행은 <br> 규약)\n display[r][c] = (cell.blocks?.some(b => b.spans)\n ? cell.blocks\n .map(b => b.type === \"image\" && b.text\n ? ``\n : b.spans ? spansToMarkdown(b.spans) : escapeGfm(sanitizeText(b.text ?? \"\")))\n .filter(Boolean)\n .join(\"<br>\")\n : escapeGfm(sanitizeText(cell.text)).replace(/\\n/g, \"<br>\")\n ).replace(/\\|/g, \"\\\\|\")\n\n // colSpan/rowSpan: 병합된 열은 빈 칸으로 유지 (텍스트 중복 방지)\n for (let dr = 0; dr < cell.rowSpan; dr++) {\n for (let dc = 0; dc < cell.colSpan; dc++) {\n if (dr === 0 && dc === 0) continue\n if (r + dr < numRows && c + dc < numCols) {\n skip.add(`${r + dr},${c + dc}`)\n }\n }\n }\n // colSpan > 1이면 display 열 인덱스를 건너뜀\n c += cell.colSpan - 1\n }\n }\n\n // rowSpan 잔류 처리:\n // 1) 완전 빈 행 제거\n // 2) \"첫 열만 값, 나머지 빈\" 행 → 다음 데이터 행의 첫 열에 값을 전파\n // 단, colSpan으로 인한 빈 열(skip 셀)은 이 대상이 아님\n const uniqueRows: string[][] = []\n let pendingLabelRow: string[] | null = null\n for (let r = 0; r < display.length; r++) {\n const row = display[r]\n if (row.every(cell => cell === \"\")) {\n // rowSpan 잔류 행(병합 커버 셀 포함 — 병합+수식 GFM 경로)만 제거.\n // 앵커 셀이 전부 빈 텍스트인 진짜 빈 행은 문서 구조 — 보존해야 왕복이 성립한다.\n if (row.some((_, c) => skip.has(`${r},${c}`))) continue\n if (pendingLabelRow) { uniqueRows.push(pendingLabelRow); pendingLabelRow = null }\n uniqueRows.push(row)\n continue\n }\n\n // 첫 열만 값이 있고 나머지 모두 빈 행 → 다음 데이터 행의 첫 열에 전파\n // 단, colSpan으로 인한 빈 열(skip 셀)은 \"진짜 빈\"이 아니므로 제외\n const nonEmptyCols = row.filter(cell => cell !== \"\")\n const hasSkipInRow = row.some((_, c) => skip.has(`${r},${c}`))\n if (!hasSkipInRow && nonEmptyCols.length === 1 && row[0] !== \"\" && row.slice(1).every(c => c === \"\")) {\n if (pendingLabelRow) uniqueRows.push(pendingLabelRow) // 연속 보류 — 앞 행 소실 방지\n pendingLabelRow = row\n continue\n }\n\n // 보류한 첫 열 값을 현재 행의 빈 첫 열에 전파, 전파 불가면 보류 행 그대로 출력\n if (pendingLabelRow) {\n if (row[0] === \"\") row[0] = pendingLabelRow[0]\n else uniqueRows.push(pendingLabelRow)\n pendingLabelRow = null\n }\n uniqueRows.push(row)\n }\n if (pendingLabelRow) uniqueRows.push(pendingLabelRow) // 표 끝 보류 행 소실 방지\n\n if (uniqueRows.length === 0) return \"\"\n\n const md: string[] = []\n md.push(\"| \" + uniqueRows[0].join(\" | \") + \" |\")\n md.push(\"| \" + uniqueRows[0].map(() => \"---\").join(\" | \") + \" |\")\n for (let i = 1; i < uniqueRows.length; i++) {\n md.push(\"| \" + uniqueRows[i].join(\" | \") + \" |\")\n }\n return md.join(\"\\n\")\n}\n","/**\n * 순수 JS BMP → PNG 트랜스코더 + 마크다운 이미지 인라이너.\n *\n * HWP5 BinData 임베드 이미지는 비압축 BMP(수 MB)인 경우가 많다. AI 에이전트가\n * 자체 완결형 마크다운을 받도록, BMP 를 PNG(무손실 압축)로 변환한 뒤 base64\n * data URI 로 마크다운에 인라인한다.\n *\n * 의존성 최소화 원칙 — Node 내장(node:zlib, Buffer)만 사용한다.\n */\n\nimport { deflateSync } from \"node:zlib\"\n\n// ─── CRC32 (PNG 청크용) ──────────────────────────────\n\nconst CRC_TABLE = (() => {\n const table = new Uint32Array(256)\n for (let n = 0; n < 256; n++) {\n let c = n\n for (let k = 0; k < 8; k++) {\n c = (c & 1) ? (0xedb88320 ^ (c >>> 1)) : (c >>> 1)\n }\n table[n] = c >>> 0\n }\n return table\n})()\n\n/** 표준 PNG CRC32 (다항식 0xEDB88320) */\nfunction crc32(bytes: Uint8Array): number {\n let c = 0xffffffff\n for (let i = 0; i < bytes.length; i++) {\n c = CRC_TABLE[(c ^ bytes[i]) & 0xff] ^ (c >>> 8)\n }\n return (c ^ 0xffffffff) >>> 0\n}\n\n// ─── BMP 디코드 ──────────────────────────────────────\n\nconst BI_RGB = 0\n/** 폭·높이 상한 — 비정상 헤더에 의한 과대 할당 방지 */\nconst MAX_DIM = 0x7fff\n/**\n * 총 픽셀(W×H) 상한 — MAX_DIM 은 각 변만 제한하므로 곱의 폭주(과대 할당 + 긴 deflateSync\n * 동기 블로킹)를 별도로 차단한다. raw 저장 BMP 는 상위 inflate 100MB 캡을 우회하므로 필요.\n * 36MP = A4 전면 600dpi 스캔(≈35MP)까지 허용 — 그 이상은 rgba+raw 로 300MB 를 넘겨\n * MCP 서버 이벤트루프를 수 초 세우므로 원본 바이트 폴백으로 돌린다.\n */\nconst MAX_PIXELS = 36_000_000\n\n/**\n * 비압축 BMP(24/32bpp, BI_RGB) → PNG(8bit RGBA).\n *\n * 지원: BITMAPFILEHEADER + BITMAPINFOHEADER(biSize>=40), biBitCount 24/32,\n * biCompression BI_RGB, bottom-up(양수 biHeight)/top-down(음수), 24bpp 4바이트 행 패딩.\n * 미지원 변형(팔레트/8bpp·RLE·BITFIELDS·OS/2 헤더 등)은 null 을 반환하여\n * 호출부가 원본 바이트로 폴백하게 한다.\n */\nexport function bmpToPng(bmp: Uint8Array): Uint8Array | null {\n // BITMAPFILEHEADER(14) + BITMAPINFOHEADER(40)\n if (bmp.length < 54) return null\n if (bmp[0] !== 0x42 || bmp[1] !== 0x4d) return null // \"BM\"\n\n const dv = new DataView(bmp.buffer, bmp.byteOffset, bmp.byteLength)\n const dataOffset = dv.getUint32(10, true)\n const headerSize = dv.getUint32(14, true)\n if (headerSize < 40) return null // OS/2 BITMAPCOREHEADER 등 미지원\n\n const width = dv.getInt32(18, true)\n const rawHeight = dv.getInt32(22, true)\n const bitCount = dv.getUint16(28, true)\n const compression = dv.getUint32(30, true)\n\n if (compression !== BI_RGB) return null // RLE / BITFIELDS 미지원\n if (bitCount !== 24 && bitCount !== 32) return null // 팔레트 / 8bpp 등 미지원\n if (width <= 0 || rawHeight === 0) return null\n if (width > MAX_DIM || Math.abs(rawHeight) > MAX_DIM) return null\n\n const topDown = rawHeight < 0\n const height = Math.abs(rawHeight)\n // 총 픽셀 상한 검사 — rgba/raw 할당·deflate 전에 조기 차단 (각 변은 MAX_DIM 이하여도 곱은 폭주 가능)\n if (width * height > MAX_PIXELS) return null\n const bytesPerPixel = bitCount >> 3\n // 행 스트라이드 — 4바이트 경계 패딩 (24bpp 에서 유효)\n const rowStride = (width * bytesPerPixel + 3) & ~3\n if (dataOffset + rowStride * height > bmp.length) return null // 잘린 파일\n\n const rgba = new Uint8Array(width * height * 4)\n let anyAlpha = 0\n\n for (let y = 0; y < height; y++) {\n // bottom-up 은 첫 행이 이미지 맨 아래 → 뒤집어 읽는다\n const srcRow = topDown ? y : height - 1 - y\n let src = dataOffset + srcRow * rowStride\n let dst = y * width * 4\n for (let x = 0; x < width; x++) {\n // BMP 픽셀은 BGR(A) 순\n rgba[dst] = bmp[src + 2] // R\n rgba[dst + 1] = bmp[src + 1] // G\n rgba[dst + 2] = bmp[src] // B\n const a = bitCount === 32 ? bmp[src + 3] : 255\n rgba[dst + 3] = a\n anyAlpha |= a\n src += bytesPerPixel\n dst += 4\n }\n }\n\n // 32bpp BI_RGB 의 4번째 바이트는 \"미정의\" — 전부 0 이면 전면 투명으로 오해되므로\n // 불투명(255)으로 승격한다. 실제 알파가 있으면(anyAlpha!=0) 그대로 보존.\n if (bitCount === 32 && anyAlpha === 0) {\n for (let i = 3; i < rgba.length; i += 4) rgba[i] = 255\n }\n\n return encodePng(width, height, rgba)\n}\n\n// ─── PNG 인코드 (8bit RGBA, color type 6) ────────────\n\nconst PNG_SIGNATURE = Uint8Array.of(0x89, 0x50, 0x4e, 0x47, 0x0d, 0x0a, 0x1a, 0x0a)\n\n/** [4B 길이][4B 타입][데이터][4B CRC] 청크 조립. CRC 는 (타입+데이터) 기준. */\nfunction chunk(type: string, data: Uint8Array): Uint8Array {\n const body = new Uint8Array(4 + data.length)\n body[0] = type.charCodeAt(0)\n body[1] = type.charCodeAt(1)\n body[2] = type.charCodeAt(2)\n body[3] = type.charCodeAt(3)\n body.set(data, 4)\n\n const out = new Uint8Array(8 + data.length + 4)\n const dv = new DataView(out.buffer)\n dv.setUint32(0, data.length, false) // 길이(데이터만, big-endian)\n out.set(body, 4)\n dv.setUint32(8 + data.length, crc32(body), false)\n return out\n}\n\n/** 8bit RGBA 픽셀 버퍼 → PNG (PDF 이미지 XObject 추출에서도 사용) */\nexport function encodePng(width: number, height: number, rgba: Uint8Array): Uint8Array {\n // IHDR: width, height, bit depth 8, color type 6(RGBA), compression 0, filter 0, interlace 0\n const ihdr = new Uint8Array(13)\n const iv = new DataView(ihdr.buffer)\n iv.setUint32(0, width, false)\n iv.setUint32(4, height, false)\n ihdr[8] = 8\n ihdr[9] = 6\n\n // IDAT: 각 스캔라인 앞에 필터 바이트 0(None) 을 붙인 raw 데이터를 zlib deflate\n const stride = width * 4\n const raw = new Uint8Array((stride + 1) * height)\n for (let y = 0; y < height; y++) {\n const rowStart = y * (stride + 1)\n raw[rowStart] = 0 // filter: None\n raw.set(rgba.subarray(y * stride, y * stride + stride), rowStart + 1)\n }\n const idat = deflateSync(raw)\n\n const ihdrChunk = chunk(\"IHDR\", ihdr)\n const idatChunk = chunk(\"IDAT\", idat)\n const iendChunk = chunk(\"IEND\", new Uint8Array(0))\n\n const out = new Uint8Array(PNG_SIGNATURE.length + ihdrChunk.length + idatChunk.length + iendChunk.length)\n let o = 0\n out.set(PNG_SIGNATURE, o); o += PNG_SIGNATURE.length\n out.set(ihdrChunk, o); o += ihdrChunk.length\n out.set(idatChunk, o); o += idatChunk.length\n out.set(iendChunk, o)\n return out\n}\n\n// ─── 마크다운 이미지 인라이너 ────────────────────────\n\ninterface InlineImage {\n filename: string\n data: Uint8Array\n mimeType: string\n}\n\n/**\n * 마크다운의 이미지 참조를 base64 data URI 로 치환해 자체 완결형 마크다운을 만든다.\n * 두 방출 형태를 모두 커버한다:\n * - `` / `` — 일반 문단 (CLI 접두사 포함)\n * - `<img src=\"FILENAME\">` — 병합/중첩 표 셀 (table/builder.ts HTML 표 경로).\n * 이걸 빼면 CLI `--inline-images` 가 이미지 저장을 생략하면서 표 셀 이미지만\n * dangling 참조로 유실된다.\n *\n * @param opts.compress 기본 true. `image/bmp` 는 PNG 로 트랜스코딩 후 인라인하며,\n * 트랜스코딩 실패 시 원본 바이트+원본 MIME 으로 폴백한다.\n */\nexport function inlineImagesIntoMarkdown(\n markdown: string,\n images: ReadonlyArray<InlineImage>,\n opts?: { compress?: boolean },\n): string {\n const compress = opts?.compress !== false\n // 파일명 → data URI 사전 구축 후 단일 패스 치환 — 이미지 k개마다 전문 재스캔하던\n // O(k²) 제거. 맵에 없는 파일명 참조는 원문 그대로 둔다 (기존 계약 동일).\n const uris = new Map<string, string>()\n for (const img of images) {\n let bytes = img.data\n let mime = img.mimeType\n if (compress && mime === \"image/bmp\") {\n const png = bmpToPng(img.data)\n if (png) {\n bytes = png\n mime = \"image/png\"\n }\n }\n uris.set(img.filename, `data:${mime};base64,${Buffer.from(bytes).toString(\"base64\")}`)\n }\n if (uris.size === 0) return markdown\n // FILENAME 과 images/FILENAME 접두사를 모두 정확한 파일명 기준으로 치환\n let out = markdown.replace(/!\\[image\\]\\((?:images\\/)?([^()\\s]+)\\)/g, (m, name: string) => {\n const uri = uris.get(name)\n return uri !== undefined ? `` : m\n })\n out = out.replace(/(<img\\b[^>]*\\bsrc=\")(?:images\\/)?([^\"]*)(\")/g, (m, pre: string, name: string, post: string) => {\n const uri = uris.get(name)\n return uri !== undefined ? `${pre}${uri}${post}` : m\n })\n return out\n}\n"],"mappings":";;;;;AAYA,IAAM,iBAAyC;AAAA;AAAA,EAE7C,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA;AAAA,EAEN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA;AAAA,EAEN,IAAM;AAAA;AAAA,EACN,IAAM;AAAA;AAAA,EACN,IAAM;AAAA;AAAA,EACN,IAAM;AAAA;AAAA;AAAA,EAEN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA;AAAA,EAEN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA,EACN,KAAM;AAAA;AAAA;AAAA,EAEN,IAAM;AAAA;AAAA,EACN,IAAM;AAAA;AAAA,EACN,IAAM;AAAA;AAAA,EACN,IAAM;AAAA;AAAA,EACN,IAAM;AAAA;AAAA,EACN,IAAM;AAAA;AAAA,EACN,IAAM;AAAA;AAAA,EACN,IAAM;AAAA;AACR;AAKA,IAAM,oBAA4C;AAAA,EACh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qBAAqB;AAC3B,IAAM,mBAAmB;AAGzB,SAAS,WAAW,MAAkC;AACpD,MAAI,QAAQ,SAAU,QAAQ,OAAQ;AACpC,WAAO,eAAe,OAAO,KAAM;AAAA,EACrC;AACA,MAAI,QAAQ,sBAAsB,QAAQ,kBAAkB;AAC1D,WAAO,OAAO,cAAc,QAAU,OAAO,mBAAmB;AAAA,EAClE;AACA,MAAI,QAAQ,UAAW,QAAQ,QAAS;AACtC,WAAO,kBAAkB,IAAI;AAAA,EAC/B;AACA,SAAO;AACT;AAOO,SAAS,WAAW,MAAsB;AAC/C,MAAI,MAAM;AACV,aAAW,MAAM,MAAM;AACrB,UAAM,OAAO,GAAG,YAAY,CAAC;AAC7B,WAAO,WAAW,IAAI,KAAK;AAAA,EAC7B;AACA,SAAO;AACT;;;ACpJO,IAAM,WAAW;AAEjB,IAAM,WAAW;AAajB,SAAS,WAAW,MAAuB,SAAsC;AACtF,MAAI,KAAK,SAAS,SAAU,QAAO,KAAK,MAAM,GAAG,QAAQ;AACzD,QAAM,UAAU,KAAK;AAGrB,QAAM,UAAU,KAAK,KAAK,SAAO,IAAI,KAAK,OAAK,EAAE,YAAY,UAAa,EAAE,YAAY,MAAS,CAAC;AAClG,MAAI,QAAS,QAAO,iBAAiB,MAAM,SAAS,OAAO;AAG3D,MAAI,UAAU;AACd,QAAM,eAA4B,MAAM,KAAK,EAAE,QAAQ,QAAQ,GAAG,MAAM,CAAC,CAAC;AAE1E,WAAS,SAAS,GAAG,SAAS,SAAS,UAAU;AAC/C,QAAI,SAAS;AACb,eAAW,QAAQ,KAAK,MAAM,GAAG;AAC/B,aAAO,SAAS,YAAY,aAAa,MAAM,EAAE,MAAM,EAAG;AAC1D,UAAI,UAAU,SAAU;AAExB,eAAS,IAAI,QAAQ,IAAI,KAAK,IAAI,SAAS,KAAK,SAAS,OAAO,GAAG,KAAK;AACtE,iBAAS,IAAI,QAAQ,IAAI,KAAK,IAAI,SAAS,KAAK,SAAS,QAAQ,GAAG,KAAK;AACvE,uBAAa,CAAC,EAAE,CAAC,IAAI;AAAA,QACvB;AAAA,MACF;AACA,gBAAU,KAAK;AACf,UAAI,SAAS,QAAS,WAAU;AAAA,IAClC;AAAA,EACF;AAEA,MAAI,YAAY,EAAG,QAAO,EAAE,MAAM,GAAG,MAAM,GAAG,OAAO,CAAC,GAAG,WAAW,MAAM;AAG1E,QAAM,OAAmB,MAAM;AAAA,IAAK,EAAE,QAAQ,QAAQ;AAAA,IAAG,MACvD,MAAM,KAAK,EAAE,QAAQ,QAAQ,GAAG,OAAO,EAAE,MAAM,IAAI,SAAS,GAAG,SAAS,EAAE,EAAE;AAAA,EAC9E;AACA,QAAM,WAAwB,MAAM,KAAK,EAAE,QAAQ,QAAQ,GAAG,MAAM,MAAM,OAAO,EAAE,KAAK,KAAK,CAAC;AAC9F,QAAM,aAAa,oBAAI,IAAY;AAEnC,WAAS,SAAS,GAAG,SAAS,SAAS,UAAU;AAC/C,QAAI,SAAS;AACb,QAAI,UAAU;AAEd,WAAO,SAAS,WAAW,UAAU,KAAK,MAAM,EAAE,QAAQ;AACxD,aAAO,SAAS,WAAW,SAAS,MAAM,EAAE,MAAM,EAAG;AACrD,UAAI,UAAU,QAAS;AAEvB,YAAM,OAAO,KAAK,MAAM,EAAE,OAAO;AACjC,iBAAW,IAAI,MAAM;AACrB,WAAK,MAAM,EAAE,MAAM,IAAI;AAAA,QACrB,MAAM,SAAS,sBAAsB,KAAK,OAAO,KAAK,KAAK,KAAK;AAAA,QAChE,SAAS,KAAK;AAAA,QACd,SAAS,KAAK;AAAA,MAChB;AAEA,eAAS,IAAI,QAAQ,IAAI,KAAK,IAAI,SAAS,KAAK,SAAS,OAAO,GAAG,KAAK;AACtE,iBAAS,IAAI,QAAQ,IAAI,KAAK,IAAI,SAAS,KAAK,SAAS,OAAO,GAAG,KAAK;AACtE,mBAAS,CAAC,EAAE,CAAC,IAAI;AAAA,QACnB;AAAA,MACF;AAEA,gBAAU,KAAK;AACf;AAAA,IACF;AAAA,EACF;AAEA,SAAO,cAAc,MAAM,SAAS,SAAS,YAAY,OAAO;AAClE;AAGA,SAAS,iBAAiB,MAAuB,SAAiB,SAAsC;AAEtG,MAAI,UAAU;AACd,aAAW,OAAO,MAAM;AACtB,eAAW,QAAQ,KAAK;AACtB,YAAM,OAAO,KAAK,WAAW,KAAK,KAAK;AACvC,UAAI,MAAM,QAAS,WAAU;AAAA,IAC/B;AAAA,EACF;AACA,MAAI,UAAU,SAAU,WAAU;AAClC,MAAI,YAAY,EAAG,QAAO,EAAE,MAAM,GAAG,MAAM,GAAG,OAAO,CAAC,GAAG,WAAW,MAAM;AAE1E,QAAM,OAAmB,MAAM;AAAA,IAAK,EAAE,QAAQ,QAAQ;AAAA,IAAG,MACvD,MAAM,KAAK,EAAE,QAAQ,QAAQ,GAAG,OAAO,EAAE,MAAM,IAAI,SAAS,GAAG,SAAS,EAAE,EAAE;AAAA,EAC9E;AAEA,QAAM,aAAa,oBAAI,IAAY;AACnC,aAAW,OAAO,MAAM;AACtB,eAAW,QAAQ,KAAK;AACtB,YAAM,IAAI,KAAK,WAAW;AAC1B,YAAM,IAAI,KAAK,WAAW;AAC1B,UAAI,KAAK,WAAW,KAAK,WAAW,IAAI,KAAK,IAAI,EAAG;AAEpD,iBAAW,IAAI,CAAC;AAChB,WAAK,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,SAAS,sBAAsB,KAAK,OAAO,KAAK,KAAK,KAAK,GAAG,SAAS,KAAK,SAAS,SAAS,KAAK,QAAQ;AAG/H,eAAS,KAAK,GAAG,KAAK,KAAK,SAAS,MAAM;AACxC,iBAAS,KAAK,GAAG,KAAK,KAAK,SAAS,MAAM;AACxC,cAAI,OAAO,KAAK,OAAO,EAAG;AAC1B,cAAI,IAAI,KAAK,WAAW,IAAI,KAAK,SAAS;AACxC,iBAAK,IAAI,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,MAAM,IAAI,SAAS,GAAG,SAAS,EAAE;AAAA,UAC5D;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAEA,SAAO,cAAc,MAAM,SAAS,SAAS,YAAY,OAAO;AAClE;AAGA,SAAS,cAAc,MAAkB,SAAiB,SAAiB,YAAyB,SAAsC;AACxI,MAAI,gBAAgB;AACpB,SAAO,gBAAgB,GAAG;AACxB,UAAM,WAAW,KAAK,MAAM,SAAO,CAAC,IAAI,gBAAgB,CAAC,GAAG,MAAM,KAAK,CAAC;AACxE,QAAI,CAAC,SAAU;AAEf,QAAI,SAAS,yBAAyB,WAAW,IAAI,gBAAgB,CAAC,EAAG;AACzE;AAAA,EACF;AACA,MAAI,gBAAgB,WAAW,gBAAgB,GAAG;AAChD,UAAM,UAAU,KAAK,IAAI,SAAO,IAAI,MAAM,GAAG,aAAa,CAAC;AAC3D,WAAO,EAAE,MAAM,SAAS,MAAM,eAAe,OAAO,SAAS,WAAW,UAAU,EAAE;AAAA,EACtF;AACA,SAAO,EAAE,MAAM,SAAS,MAAM,SAAS,OAAO,MAAM,WAAW,UAAU,EAAE;AAC7E;AAEO,SAAS,mBAAmB,MAA+B;AAChE,SAAO,KACJ;AAAA,IAAI,SACH,IACG,IAAI,OAAK,EAAE,KAAK,KAAK,EAAE,QAAQ,OAAO,GAAG,EAAE,QAAQ,OAAO,KAAK,CAAC,EAChE,OAAO,OAAO,EACd,KAAK,KAAK;AAAA,EACf,EACC,OAAO,OAAO,EACd,KAAK,IAAI;AACd;AAGA,SAAS,UAAU,MAAsB;AAOvC,QAAM,MAAM,OAAO,aAAa,CAAC;AACjC,QAAM,QAAkB,CAAC;AACzB,QAAM,SAAS,KAAK,QAAQ,8FAA8F,CAAC,MAAM;AAC/H,UAAM,KAAK,CAAC;AACZ,WAAO,OAAO,MAAM,SAAS,KAAK;AAAA,EACpC,CAAC;AACD,QAAM,UAAU,OAAO,QAAQ,aAAa,MAAM;AAClD,SAAO,QAAQ,QAAQ,IAAI,OAAO,MAAM,WAAW,KAAK,GAAG,GAAG,CAAC,GAAG,MAAM,MAAM,OAAO,CAAC,CAAC,CAAC;AAC1F;AAIA,IAAM,wBAAwB;AAK9B,SAAS,YAAY,MAAsB;AAEzC,SAAO,WAAW,IAAI,EAEnB,QAAQ,2BAA2B,EAAE;AAC1C;AAGA,SAAS,aAAa,MAAsB;AAC1C,MAAI,SAAS,YAAY,IAAI,EAE1B,QAAQ,uBAAuB,EAAE,EACjC,QAAQ,QAAQ,GAAG,EACnB,KAAK;AAGR,MAAI,OAAO,UAAU,MAAM,OAAO,SAAS,GAAG,GAAG;AAC/C,UAAM,SAAS,OAAO,MAAM,GAAG;AAE/B,UAAM,wBAAwB,OAAO,OAAO,OAAK,EAAE,WAAW,KAAK,+BAA+B,KAAK,CAAC,CAAC,EAAE;AAC3G,QAAI,OAAO,UAAU,KAAK,wBAAwB,OAAO,UAAU,KAAK;AACtE,eAAS,OAAO,KAAK,EAAE;AAAA,IACzB;AAAA,EACF;AACA,SAAO;AACT;AAaO,SAAS,oBAAoB,QAA8B;AAChE,QAAM,SAAoB,CAAC;AAE3B,aAAW,SAAS,QAAQ;AAC1B,QAAI,MAAM,SAAS,WAAW,CAAC,MAAM,OAAO;AAC1C,aAAO,KAAK,KAAK;AACjB;AAAA,IACF;AAEA,UAAM,EAAE,MAAM,SAAS,MAAM,SAAS,MAAM,IAAI,MAAM;AAGtD,QAAI,YAAY,KAAK,YAAY,GAAG;AAClC,aAAO,KAAK,KAAK;AACjB;AAAA,IACF;AAGA,QAAI,WAAW,GAAG;AAChB,UAAI,gBAAgB;AACpB,UAAI,eAAe;AACnB,eAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,iBAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,gBAAM,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,QAAQ;AACjC,4BAAkB,EAAE,MAAM,KAAK,KAAK,CAAC,GAAG;AACxC,0BAAgB,EAAE;AAAA,QACpB;AAAA,MACF;AAKA,UAAI,UAAU,MAAM,gBAAgB,KAAM,WAAW,KAAK,eAAe,MAAO;AAE9E,iBAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,mBAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,kBAAM,OAAO,MAAM,CAAC,IAAI,CAAC;AACzB,gBAAI,CAAC,KAAM;AAKX,gBAAI,KAAK,QAAQ,QAAQ;AACvB,qBAAO,KAAK,GAAG,oBAAoB,KAAK,MAAM,CAAC;AAC/C;AAAA,YACF;AACA,kBAAM,WAAW,KAAK,MAAM,KAAK;AACjC,gBAAI,CAAC,SAAU;AAEf,uBAAW,QAAQ,SAAS,MAAM,IAAI,GAAG;AACvC,oBAAM,UAAU,KAAK,KAAK;AAC1B,kBAAI,CAAC,QAAS;AACd,qBAAO,KAAK,EAAE,MAAM,aAAa,MAAM,SAAS,YAAY,MAAM,WAAW,CAAC;AAAA,YAChF;AAAA,UACF;AAAA,QACF;AACA;AAAA,MACF;AAAA,IACF;AAEA,WAAO,KAAK,KAAK;AAAA,EACnB;AAEA,SAAO;AACT;AAGA,IAAM,yBAAyB;AAE/B,IAAM,0BAA0B;AAGhC,SAAS,yBAAyB,OAAyB;AACzD,MAAI,MAAM,SAAS,eAAe,MAAM,SAAS,UAAW,QAAO;AACnE,QAAM,OAAO,MAAM,MAAM,KAAK;AAC9B,MAAI,CAAC,QAAQ,KAAK,SAAS,uBAAwB,QAAO;AAC1D,SAAO,WAAW,KAAK,IAAI;AAC7B;AAiBO,SAAS,qBAAqB,QAA8B;AAEjE,QAAM,OAAO,oBAAI,IAAoB;AACrC,aAAW,SAAS,QAAQ;AAC1B,QAAI,CAAC,yBAAyB,KAAK,EAAG;AACtC,UAAM,OAAO,MAAM,KAAM,KAAK;AAC9B,SAAK,IAAI,OAAO,KAAK,IAAI,IAAI,KAAK,KAAK,CAAC;AAAA,EAC1C;AAGA,QAAM,OAAO,oBAAI,IAAY;AAC7B,QAAM,SAAoB,CAAC;AAC3B,aAAW,SAAS,QAAQ;AAC1B,QAAI,yBAAyB,KAAK,GAAG;AACnC,YAAM,OAAO,MAAM,KAAM,KAAK;AAC9B,WAAK,KAAK,IAAI,IAAI,KAAK,MAAM,yBAAyB;AACpD,YAAI,KAAK,IAAI,IAAI,EAAG;AACpB,aAAK,IAAI,IAAI;AAAA,MACf;AAAA,IACF;AACA,WAAO,KAAK,KAAK;AAAA,EACnB;AAEA,SAAO;AACT;AAMA,SAAS,gBAAgB,OAAyB;AAChD,MAAI,MAAM;AACV,aAAW,KAAK,OAAO;AAIrB,UAAM,OAAO,YAAY,EAAE,QAAQ,EAAE;AACrC,QAAI,CAAC,KAAM;AACX,QAAI,SAAS,EAAE,OAAO,MAAM,EAAE,QAAQ,EAAE,SAAS,QAAQ,EAAE,OAAO,OAAO,EAAE,SAAS,MAAM;AAC1F,QAAI,EAAE,UAAU,CAAC,EAAE,KAAM,UAAS,KAAK,MAAM;AAC7C,UAAM,QAAQ,CAAC,CAAC,EAAE,aAAa,CAAC,EAAE;AAClC,QAAI,CAAC,UAAU,CAAC,OAAO;AACrB,aAAO,UAAU,IAAI;AACrB;AAAA,IACF;AACA,UAAM,IAAI,yBAAyB,KAAK,IAAI;AAC5C,UAAM,OAAO,EAAE,CAAC;AAChB,QAAI,CAAC,MAAM;AACT,aAAO;AACP;AAAA,IACF;AACA,QAAI,OAAO;AACX,QAAI,QAAQ,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK,EAAE;AACzC,QAAI,OAAO;AAAE,aAAO,QAAQ;AAAM,cAAQ,QAAQ;AAAA,IAAO;AACzD,WAAO,EAAE,CAAC,IAAI,QAAQ,EAAE,OAAO,OAAO,UAAU,IAAI,KAAK,QAAQ,EAAE,CAAC;AAAA,EACtE;AACA,SAAO;AACT;AAEO,SAAS,iBAAiB,QAA2B;AAC1D,QAAM,QAAkB,CAAC;AAEzB,WAAS,IAAI,GAAG,IAAI,OAAO,QAAQ,KAAK;AACtC,UAAM,QAAQ,OAAO,CAAC;AAGtB,QAAI,MAAM,SAAS,aAAa,MAAM,MAAM;AAC1C,YAAM,SAAS,IAAI,OAAO,KAAK,IAAI,MAAM,SAAS,GAAG,CAAC,CAAC;AACvD,YAAM,cAAc,aAAa,MAAM,IAAI;AAC3C,UAAI,YAAa,OAAM,KAAK,IAAI,GAAG,MAAM,IAAI,UAAU,WAAW,CAAC,IAAI,EAAE;AACzE;AAAA,IACF;AAGA,QAAI,MAAM,SAAS,WAAW,MAAM,MAAM;AACxC,YAAM,KAAK,IAAI,YAAY,MAAM,IAAI,KAAK,EAAE;AAC5C;AAAA,IACF;AAGA,QAAI,MAAM,SAAS,aAAa;AAC9B,YAAM,KAAK,IAAI,OAAO,EAAE;AACxB;AAAA,IACF;AAGA,QAAI,MAAM,SAAS,UAAU,MAAM,MAAM;AACvC,YAAM,WAAW,aAAa,MAAM,IAAI;AACxC,UAAI,CAAC,SAAU;AAEf,YAAM,kBAAkB,MAAM,aAAa,aAAa,WAAW,KAAK,QAAQ;AAChF,YAAM,SAAS,kBAAkB,KAAK,MAAM,aAAa,YAAY,QAAQ;AAC7E,YAAM,KAAK,GAAG,MAAM,GAAG,UAAU,QAAQ,CAAC,EAAE;AAC5C,UAAI,MAAM,UAAU;AAClB,mBAAW,SAAS,MAAM,UAAU;AAClC,gBAAM,cAAc,MAAM,aAAa,YAAY,OAAO;AAC1D,gBAAM,KAAK,KAAK,WAAW,IAAI,UAAU,MAAM,QAAQ,EAAE,CAAC,EAAE;AAAA,QAC9D;AAAA,MACF;AACA;AAAA,IACF;AAEA,QAAI,MAAM,SAAS,eAAe,MAAM,MAAM;AAC5C,UAAI,OAAO,aAAa,MAAM,IAAI;AAClC,UAAI,CAAC,KAAM;AAGX,UAAI,cAAc,KAAK,IAAI,GAAG;AAC5B,cAAM,YAAY,OAAO,IAAI,CAAC;AAC9B,YAAI,WAAW,SAAS,eAAe,UAAU,QAAQ,SAAS,KAAK,UAAU,IAAI,GAAG;AACtF,gBAAM,KAAK,IAAI,MAAM,UAAU,IAAI,CAAC,IAAI,UAAU,UAAU,IAAI,CAAC,IAAI,EAAE;AACvE;AAAA,QACF,OAAO;AACL,gBAAM,KAAK,IAAI,MAAM,UAAU,IAAI,CAAC,IAAI,EAAE;AAAA,QAC5C;AACA;AAAA,MACF;AAEA,UAAI,sBAAsB,KAAK,IAAI,GAAG;AACpC,cAAM,KAAK,IAAI,UAAU,IAAI,CAAC,KAAK,EAAE;AACrC;AAAA,MACF;AAIA,YAAM,aAAa,MAAM,YAAY,KAAK,OAAO,MAAM,SAAS,IAAI;AAIpE,UAAI,MAAM,OAAO,QAAQ;AACvB,YAAI,WAAW,gBAAgB,MAAM,KAAK;AAC1C,YAAI,MAAM,MAAM;AACd,gBAAM,OAAO,aAAa,MAAM,IAAI;AACpC,cAAI,KAAM,YAAW,IAAI,QAAQ,KAAK,IAAI;AAAA,QAC5C;AACA,YAAI,MAAM,aAAc,aAAY,aAAQ,MAAM,YAAY;AAC9D,cAAM,KAAK,MAAM,QAAQ,OAAO,WAAW,aAAa,UAAU,EAAE;AACpE;AAAA,MACF;AAGA,UAAI,MAAM,MAAM;AACd,cAAM,OAAO,aAAa,MAAM,IAAI;AACpC,YAAI,KAAM,QAAO,IAAI,IAAI,KAAK,IAAI;AAAA,MACpC;AAGA,UAAI,MAAM,cAAc;AACtB,gBAAQ,aAAQ,MAAM,YAAY;AAAA,MACpC;AAEA,YAAM,KAAK,MAAM,QAAQ,OAAO,UAAU,IAAI,IAAI,aAAa,UAAU,IAAI,GAAG,EAAE;AAAA,IACpF,WAAW,MAAM,SAAS,WAAW,MAAM,OAAO;AAEhD,UAAI,MAAM,SAAS,KAAK,MAAM,MAAM,SAAS,CAAC,MAAM,IAAI;AACtD,cAAM,KAAK,EAAE;AAAA,MACf;AAEA,UAAI,MAAM,MAAM,SAAS;AACvB,cAAM,UAAU,aAAa,MAAM,MAAM,OAAO;AAChD,YAAI,QAAS,OAAM,KAAK,KAAK,UAAU,OAAO,CAAC,MAAM,EAAE;AAAA,MACzD;AACA,YAAM,UAAU,gBAAgB,MAAM,KAAK;AAC3C,UAAI,SAAS;AACX,cAAM,KAAK,OAAO;AAClB,cAAM,KAAK,EAAE;AAAA,MACf;AAAA,IACF;AAAA,EACF;AAEA,SAAO,MAAM,KAAK,IAAI,EAAE,KAAK;AAC/B;AAGA,SAAS,eAAe,OAAyB;AAC/C,aAAW,OAAO,MAAM,OAAO;AAC7B,eAAW,QAAQ,KAAK;AACtB,UAAI,KAAK,UAAU,KAAK,KAAK,UAAU,EAAG,QAAO;AAAA,IACnD;AAAA,EACF;AACA,SAAO;AACT;AAGA,SAAS,gBAAgB,OAAyB;AAChD,aAAW,OAAO,MAAM,OAAO;AAC7B,eAAW,QAAQ,KAAK;AACtB,UAAI,KAAK,QAAQ,KAAK,OAAK,EAAE,SAAS,WAAW,EAAE,KAAK,EAAG,QAAO;AAAA,IACpE;AAAA,EACF;AACA,SAAO;AACT;AAGA,SAAS,cAAc,MAAsB;AAC3C,MAAI,KAAK,QAAQ,QAAQ;AACvB,WAAO,KAAK,OACT,IAAI,OAAK;AACR,UAAI,EAAE,SAAS,WAAW,EAAE,OAAO;AAEjC,cAAM,MAAM,EAAE,MAAM,UAAU,aAAa,EAAE,MAAM,OAAO,IAAI;AAC9D,gBAAQ,MAAM,MAAM,SAAS,MAAM,YAAY,EAAE,KAAK;AAAA,MACxD;AACA,UAAI,EAAE,SAAS,WAAW,EAAE,KAAM,QAAO,aAAa,EAAE,IAAI;AAC5D,YAAM,IAAI,aAAa,EAAE,QAAQ,EAAE;AACnC,aAAO,IAAI,EAAE,QAAQ,OAAO,MAAM,IAAI;AAAA,IACxC,CAAC,EACA,OAAO,OAAO,EACd,KAAK,MAAM;AAAA,EAChB;AACA,SAAO,aAAa,KAAK,IAAI,EAAE,QAAQ,OAAO,MAAM;AACtD;AAEA,SAAS,mBAAmB,MAAuB;AAGjD,SAAO,0CAA0C,KAAK,IAAI;AAC5D;AAEA,SAAS,wBAAwB,OAAyB;AACxD,aAAW,OAAO,MAAM,OAAO;AAC7B,eAAW,QAAQ,KAAK;AACtB,UAAI,mBAAmB,KAAK,IAAI,EAAG,QAAO;AAAA,IAC5C;AAAA,EACF;AACA,SAAO;AACT;AAGA,SAAS,YAAY,OAAwB;AAC3C,QAAM,EAAE,OAAO,MAAM,SAAS,MAAM,QAAQ,IAAI;AAChD,QAAM,OAAO,oBAAI,IAAY;AAC7B,QAAM,QAAkB,CAAC,SAAS;AAElC,WAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,UAAM,MAAM,MAAM,IAAI,OAAO;AAC7B,UAAM,UAAoB,CAAC;AAC3B,aAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,UAAI,KAAK,IAAI,GAAG,CAAC,IAAI,CAAC,EAAE,EAAG;AAC3B,YAAM,OAAO,MAAM,CAAC,IAAI,CAAC;AACzB,UAAI,CAAC,KAAM;AAGX,eAAS,KAAK,GAAG,KAAK,KAAK,SAAS,MAAM;AACxC,iBAAS,KAAK,GAAG,KAAK,KAAK,SAAS,MAAM;AACxC,cAAI,OAAO,KAAK,OAAO,EAAG;AAC1B,cAAI,IAAI,KAAK,WAAW,IAAI,KAAK,QAAS,MAAK,IAAI,GAAG,IAAI,EAAE,IAAI,IAAI,EAAE,EAAE;AAAA,QAC1E;AAAA,MACF;AAEA,YAAM,OAAO,cAAc,IAAI;AAC/B,YAAM,QAAkB,CAAC;AACzB,UAAI,KAAK,UAAU,EAAG,OAAM,KAAK,YAAY,KAAK,OAAO,GAAG;AAC5D,UAAI,KAAK,UAAU,EAAG,OAAM,KAAK,YAAY,KAAK,OAAO,GAAG;AAC5D,YAAM,UAAU,MAAM,SAAS,MAAM,MAAM,KAAK,GAAG,IAAI;AACvD,cAAQ,KAAK,IAAI,GAAG,GAAG,OAAO,IAAI,IAAI,KAAK,GAAG,GAAG;AAAA,IACnD;AACA,QAAI,QAAQ,OAAQ,OAAM,KAAK,OAAO,QAAQ,KAAK,EAAE,CAAC,OAAO;AAAA,EAC/D;AAEA,QAAM,KAAK,UAAU;AACrB,SAAO,MAAM,KAAK,IAAI;AACxB;AAEA,SAAS,gBAAgB,OAAwB;AAC/C,MAAI,MAAM,SAAS,KAAK,MAAM,SAAS,EAAG,QAAO;AAEjD,QAAM,EAAE,OAAO,MAAM,SAAS,MAAM,QAAQ,IAAI;AAIhD,OAAK,eAAe,KAAK,KAAK,gBAAgB,KAAK,MAAM,CAAC,wBAAwB,KAAK,GAAG;AACxF,WAAO,YAAY,KAAK;AAAA,EAC1B;AAGA,MAAI,YAAY,KAAK,YAAY,GAAG;AAClC,UAAM,UAAU,aAAa,MAAM,CAAC,EAAE,CAAC,EAAE,IAAI;AAC7C,QAAI,CAAC,QAAS,QAAO;AACrB,WAAO,QACJ,MAAM,IAAI,EACV,IAAI,UAAQ;AACX,YAAM,UAAU,KAAK,KAAK;AAC1B,UAAI,CAAC,QAAS,QAAO;AACrB,UAAI,WAAW,KAAK,OAAO,EAAG,QAAO,KAAK,UAAU,OAAO,CAAC;AAC5D,UAAI,aAAa,KAAK,OAAO,EAAG,QAAO,KAAK,UAAU,OAAO,CAAC;AAC9D,aAAO,UAAU,OAAO;AAAA,IAC1B,CAAC,EACA,OAAO,OAAO,EACd,KAAK,IAAI;AAAA,EACd;AAGA,MAAI,YAAY,KAAK,WAAW,GAAG;AACjC,WAAO,MACJ,IAAI,SAAO,UAAU,aAAa,IAAI,CAAC,EAAE,IAAI,CAAC,EAAE,QAAQ,OAAO,GAAG,CAAC,EACnE,OAAO,OAAO,EACd,KAAK,IAAI;AAAA,EACd;AAGA,QAAM,UAAsB,MAAM,KAAK,EAAE,QAAQ,QAAQ,GAAG,MAAM,MAAM,OAAO,EAAE,KAAK,EAAE,CAAC;AACzF,QAAM,OAAO,oBAAI,IAAY;AAE7B,WAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,aAAS,IAAI,GAAG,IAAI,SAAS,KAAK;AAChC,UAAI,KAAK,IAAI,GAAG,CAAC,IAAI,CAAC,EAAE,EAAG;AAC3B,YAAM,OAAO,MAAM,CAAC,IAAI,CAAC;AACzB,UAAI,CAAC,KAAM;AAEX,cAAQ,CAAC,EAAE,CAAC,KAAK,KAAK,QAAQ,KAAK,OAAK,EAAE,KAAK,IAC3C,KAAK,OACJ,IAAI,OAAK,EAAE,SAAS,WAAW,EAAE,OAC9B,YAAY,EAAE,IAAI,MAClB,EAAE,QAAQ,gBAAgB,EAAE,KAAK,IAAI,UAAU,aAAa,EAAE,QAAQ,EAAE,CAAC,CAAC,EAC7E,OAAO,OAAO,EACd,KAAK,MAAM,IACZ,UAAU,aAAa,KAAK,IAAI,CAAC,EAAE,QAAQ,OAAO,MAAM,GAC1D,QAAQ,OAAO,KAAK;AAGtB,eAAS,KAAK,GAAG,KAAK,KAAK,SAAS,MAAM;AACxC,iBAAS,KAAK,GAAG,KAAK,KAAK,SAAS,MAAM;AACxC,cAAI,OAAO,KAAK,OAAO,EAAG;AAC1B,cAAI,IAAI,KAAK,WAAW,IAAI,KAAK,SAAS;AACxC,iBAAK,IAAI,GAAG,IAAI,EAAE,IAAI,IAAI,EAAE,EAAE;AAAA,UAChC;AAAA,QACF;AAAA,MACF;AAEA,WAAK,KAAK,UAAU;AAAA,IACtB;AAAA,EACF;AAMA,QAAM,aAAyB,CAAC;AAChC,MAAI,kBAAmC;AACvC,WAAS,IAAI,GAAG,IAAI,QAAQ,QAAQ,KAAK;AACvC,UAAM,MAAM,QAAQ,CAAC;AACrB,QAAI,IAAI,MAAM,UAAQ,SAAS,EAAE,GAAG;AAGlC,UAAI,IAAI,KAAK,CAAC,GAAG,MAAM,KAAK,IAAI,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,EAAG;AAC/C,UAAI,iBAAiB;AAAE,mBAAW,KAAK,eAAe;AAAG,0BAAkB;AAAA,MAAK;AAChF,iBAAW,KAAK,GAAG;AACnB;AAAA,IACF;AAIA,UAAM,eAAe,IAAI,OAAO,UAAQ,SAAS,EAAE;AACnD,UAAM,eAAe,IAAI,KAAK,CAAC,GAAG,MAAM,KAAK,IAAI,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC;AAC7D,QAAI,CAAC,gBAAgB,aAAa,WAAW,KAAK,IAAI,CAAC,MAAM,MAAM,IAAI,MAAM,CAAC,EAAE,MAAM,OAAK,MAAM,EAAE,GAAG;AACpG,UAAI,gBAAiB,YAAW,KAAK,eAAe;AACpD,wBAAkB;AAClB;AAAA,IACF;AAGA,QAAI,iBAAiB;AACnB,UAAI,IAAI,CAAC,MAAM,GAAI,KAAI,CAAC,IAAI,gBAAgB,CAAC;AAAA,UACxC,YAAW,KAAK,eAAe;AACpC,wBAAkB;AAAA,IACpB;AACA,eAAW,KAAK,GAAG;AAAA,EACrB;AACA,MAAI,gBAAiB,YAAW,KAAK,eAAe;AAEpD,MAAI,WAAW,WAAW,EAAG,QAAO;AAEpC,QAAM,KAAe,CAAC;AACtB,KAAG,KAAK,OAAO,WAAW,CAAC,EAAE,KAAK,KAAK,IAAI,IAAI;AAC/C,KAAG,KAAK,OAAO,WAAW,CAAC,EAAE,IAAI,MAAM,KAAK,EAAE,KAAK,KAAK,IAAI,IAAI;AAChE,WAAS,IAAI,GAAG,IAAI,WAAW,QAAQ,KAAK;AAC1C,OAAG,KAAK,OAAO,WAAW,CAAC,EAAE,KAAK,KAAK,IAAI,IAAI;AAAA,EACjD;AACA,SAAO,GAAG,KAAK,IAAI;AACrB;;;AC/qBA,SAAS,mBAAmB;AAI5B,IAAM,aAAa,MAAM;AACvB,QAAM,QAAQ,IAAI,YAAY,GAAG;AACjC,WAAS,IAAI,GAAG,IAAI,KAAK,KAAK;AAC5B,QAAI,IAAI;AACR,aAAS,IAAI,GAAG,IAAI,GAAG,KAAK;AAC1B,UAAK,IAAI,IAAM,aAAc,MAAM,IAAO,MAAM;AAAA,IAClD;AACA,UAAM,CAAC,IAAI,MAAM;AAAA,EACnB;AACA,SAAO;AACT,GAAG;AAGH,SAAS,MAAM,OAA2B;AACxC,MAAI,IAAI;AACR,WAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACrC,QAAI,WAAW,IAAI,MAAM,CAAC,KAAK,GAAI,IAAK,MAAM;AAAA,EAChD;AACA,UAAQ,IAAI,gBAAgB;AAC9B;AAIA,IAAM,SAAS;AAEf,IAAM,UAAU;AAOhB,IAAM,aAAa;AAUZ,SAAS,SAAS,KAAoC;AAE3D,MAAI,IAAI,SAAS,GAAI,QAAO;AAC5B,MAAI,IAAI,CAAC,MAAM,MAAQ,IAAI,CAAC,MAAM,GAAM,QAAO;AAE/C,QAAM,KAAK,IAAI,SAAS,IAAI,QAAQ,IAAI,YAAY,IAAI,UAAU;AAClE,QAAM,aAAa,GAAG,UAAU,IAAI,IAAI;AACxC,QAAM,aAAa,GAAG,UAAU,IAAI,IAAI;AACxC,MAAI,aAAa,GAAI,QAAO;AAE5B,QAAM,QAAQ,GAAG,SAAS,IAAI,IAAI;AAClC,QAAM,YAAY,GAAG,SAAS,IAAI,IAAI;AACtC,QAAM,WAAW,GAAG,UAAU,IAAI,IAAI;AACtC,QAAM,cAAc,GAAG,UAAU,IAAI,IAAI;AAEzC,MAAI,gBAAgB,OAAQ,QAAO;AACnC,MAAI,aAAa,MAAM,aAAa,GAAI,QAAO;AAC/C,MAAI,SAAS,KAAK,cAAc,EAAG,QAAO;AAC1C,MAAI,QAAQ,WAAW,KAAK,IAAI,SAAS,IAAI,QAAS,QAAO;AAE7D,QAAM,UAAU,YAAY;AAC5B,QAAM,SAAS,KAAK,IAAI,SAAS;AAEjC,MAAI,QAAQ,SAAS,WAAY,QAAO;AACxC,QAAM,gBAAgB,YAAY;AAElC,QAAM,YAAa,QAAQ,gBAAgB,IAAK,CAAC;AACjD,MAAI,aAAa,YAAY,SAAS,IAAI,OAAQ,QAAO;AAEzD,QAAM,OAAO,IAAI,WAAW,QAAQ,SAAS,CAAC;AAC9C,MAAI,WAAW;AAEf,WAAS,IAAI,GAAG,IAAI,QAAQ,KAAK;AAE/B,UAAM,SAAS,UAAU,IAAI,SAAS,IAAI;AAC1C,QAAI,MAAM,aAAa,SAAS;AAChC,QAAI,MAAM,IAAI,QAAQ;AACtB,aAAS,IAAI,GAAG,IAAI,OAAO,KAAK;AAE9B,WAAK,GAAG,IAAI,IAAI,MAAM,CAAC;AACvB,WAAK,MAAM,CAAC,IAAI,IAAI,MAAM,CAAC;AAC3B,WAAK,MAAM,CAAC,IAAI,IAAI,GAAG;AACvB,YAAM,IAAI,aAAa,KAAK,IAAI,MAAM,CAAC,IAAI;AAC3C,WAAK,MAAM,CAAC,IAAI;AAChB,kBAAY;AACZ,aAAO;AACP,aAAO;AAAA,IACT;AAAA,EACF;AAIA,MAAI,aAAa,MAAM,aAAa,GAAG;AACrC,aAAS,IAAI,GAAG,IAAI,KAAK,QAAQ,KAAK,EAAG,MAAK,CAAC,IAAI;AAAA,EACrD;AAEA,SAAO,UAAU,OAAO,QAAQ,IAAI;AACtC;AAIA,IAAM,gBAAgB,WAAW,GAAG,KAAM,IAAM,IAAM,IAAM,IAAM,IAAM,IAAM,EAAI;AAGlF,SAAS,MAAM,MAAc,MAA8B;AACzD,QAAM,OAAO,IAAI,WAAW,IAAI,KAAK,MAAM;AAC3C,OAAK,CAAC,IAAI,KAAK,WAAW,CAAC;AAC3B,OAAK,CAAC,IAAI,KAAK,WAAW,CAAC;AAC3B,OAAK,CAAC,IAAI,KAAK,WAAW,CAAC;AAC3B,OAAK,CAAC,IAAI,KAAK,WAAW,CAAC;AAC3B,OAAK,IAAI,MAAM,CAAC;AAEhB,QAAM,MAAM,IAAI,WAAW,IAAI,KAAK,SAAS,CAAC;AAC9C,QAAM,KAAK,IAAI,SAAS,IAAI,MAAM;AAClC,KAAG,UAAU,GAAG,KAAK,QAAQ,KAAK;AAClC,MAAI,IAAI,MAAM,CAAC;AACf,KAAG,UAAU,IAAI,KAAK,QAAQ,MAAM,IAAI,GAAG,KAAK;AAChD,SAAO;AACT;AAGO,SAAS,UAAU,OAAe,QAAgB,MAA8B;AAErF,QAAM,OAAO,IAAI,WAAW,EAAE;AAC9B,QAAM,KAAK,IAAI,SAAS,KAAK,MAAM;AACnC,KAAG,UAAU,GAAG,OAAO,KAAK;AAC5B,KAAG,UAAU,GAAG,QAAQ,KAAK;AAC7B,OAAK,CAAC,IAAI;AACV,OAAK,CAAC,IAAI;AAGV,QAAM,SAAS,QAAQ;AACvB,QAAM,MAAM,IAAI,YAAY,SAAS,KAAK,MAAM;AAChD,WAAS,IAAI,GAAG,IAAI,QAAQ,KAAK;AAC/B,UAAM,WAAW,KAAK,SAAS;AAC/B,QAAI,QAAQ,IAAI;AAChB,QAAI,IAAI,KAAK,SAAS,IAAI,QAAQ,IAAI,SAAS,MAAM,GAAG,WAAW,CAAC;AAAA,EACtE;AACA,QAAM,OAAO,YAAY,GAAG;AAE5B,QAAM,YAAY,MAAM,QAAQ,IAAI;AACpC,QAAM,YAAY,MAAM,QAAQ,IAAI;AACpC,QAAM,YAAY,MAAM,QAAQ,IAAI,WAAW,CAAC,CAAC;AAEjD,QAAM,MAAM,IAAI,WAAW,cAAc,SAAS,UAAU,SAAS,UAAU,SAAS,UAAU,MAAM;AACxG,MAAI,IAAI;AACR,MAAI,IAAI,eAAe,CAAC;AAAG,OAAK,cAAc;AAC9C,MAAI,IAAI,WAAW,CAAC;AAAG,OAAK,UAAU;AACtC,MAAI,IAAI,WAAW,CAAC;AAAG,OAAK,UAAU;AACtC,MAAI,IAAI,WAAW,CAAC;AACpB,SAAO;AACT;AAqBO,SAAS,yBACd,UACA,QACA,MACQ;AACR,QAAM,WAAW,MAAM,aAAa;AAGpC,QAAM,OAAO,oBAAI,IAAoB;AACrC,aAAW,OAAO,QAAQ;AACxB,QAAI,QAAQ,IAAI;AAChB,QAAI,OAAO,IAAI;AACf,QAAI,YAAY,SAAS,aAAa;AACpC,YAAM,MAAM,SAAS,IAAI,IAAI;AAC7B,UAAI,KAAK;AACP,gBAAQ;AACR,eAAO;AAAA,MACT;AAAA,IACF;AACA,SAAK,IAAI,IAAI,UAAU,QAAQ,IAAI,WAAW,OAAO,KAAK,KAAK,EAAE,SAAS,QAAQ,CAAC,EAAE;AAAA,EACvF;AACA,MAAI,KAAK,SAAS,EAAG,QAAO;AAE5B,MAAI,MAAM,SAAS,QAAQ,0CAA0C,CAAC,GAAG,SAAiB;AACxF,UAAM,MAAM,KAAK,IAAI,IAAI;AACzB,WAAO,QAAQ,SAAY,YAAY,GAAG,MAAM;AAAA,EAClD,CAAC;AACD,QAAM,IAAI,QAAQ,gDAAgD,CAAC,GAAG,KAAa,MAAc,SAAiB;AAChH,UAAM,MAAM,KAAK,IAAI,IAAI;AACzB,WAAO,QAAQ,SAAY,GAAG,GAAG,GAAG,GAAG,GAAG,IAAI,KAAK;AAAA,EACrD,CAAC;AACD,SAAO;AACT;","names":[]}
|
|
@@ -6,7 +6,7 @@ import {
|
|
|
6
6
|
import {
|
|
7
7
|
safeMax,
|
|
8
8
|
safeMin
|
|
9
|
-
} from "./chunk-
|
|
9
|
+
} from "./chunk-Z6NFU6DA.js";
|
|
10
10
|
|
|
11
11
|
// src/pdf/image-regions.ts
|
|
12
12
|
import { OPS } from "pdfjs-dist/legacy/build/pdf.mjs";
|
|
@@ -3686,4 +3686,4 @@ export {
|
|
|
3686
3686
|
extractPageBlocksWithLines,
|
|
3687
3687
|
mergeCrossPageTables
|
|
3688
3688
|
};
|
|
3689
|
-
//# sourceMappingURL=chunk-
|
|
3689
|
+
//# sourceMappingURL=chunk-B6IA745V.js.map
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
#!/usr/bin/env node
|
|
2
2
|
|
|
3
3
|
// src/utils.ts
|
|
4
|
-
var VERSION = true ? "4.
|
|
4
|
+
var VERSION = true ? "4.11.0" : "0.0.0-dev";
|
|
5
5
|
function toArrayBuffer(buf) {
|
|
6
6
|
if (buf.byteOffset === 0 && buf.byteLength === buf.buffer.byteLength) {
|
|
7
7
|
return buf.buffer;
|
|
@@ -123,4 +123,4 @@ export {
|
|
|
123
123
|
safeMax,
|
|
124
124
|
classifyError
|
|
125
125
|
};
|
|
126
|
-
//# sourceMappingURL=chunk-
|
|
126
|
+
//# sourceMappingURL=chunk-CF2FQBVN.js.map
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
"use strict";Object.defineProperty(exports, "__esModule", {value: true}); function _optionalChain(ops) { let lastAccessLHS = undefined; let value = ops[0]; let i = 1; while (i < ops.length) { const op = ops[i]; const fn = ops[i + 1]; i += 2; if ((op === 'optionalAccess' || op === 'optionalCall') && value == null) { return undefined; } if (op === 'access' || op === 'optionalAccess') { lastAccessLHS = value; value = fn(value); } else if (op === 'call' || op === 'optionalCall') { value = fn((...args) => value.call(lastAccessLHS, ...args)); lastAccessLHS = undefined; } } return value; }
|
|
2
2
|
|
|
3
|
-
var
|
|
3
|
+
var _chunkJO3W54E5cjs = require('./chunk-JO3W54E5.cjs');
|
|
4
4
|
|
|
5
5
|
// src/pdf/formula/models.ts
|
|
6
6
|
var _crypto = require('crypto');
|
|
@@ -23,7 +23,7 @@ function isOfflineMode() {
|
|
|
23
23
|
}
|
|
24
24
|
function assertNetworkAllowed(what, hint) {
|
|
25
25
|
if (!isOfflineMode()) return;
|
|
26
|
-
throw new (0,
|
|
26
|
+
throw new (0, _chunkJO3W54E5cjs.KordocError)(
|
|
27
27
|
`\uD3D0\uC1C4\uB9DD \uBAA8\uB4DC(KORDOC_OFFLINE)\uC5D0\uC11C \uCC28\uB2E8\uB428: ${what}` + (hint ? ` \u2014 ${hint}` : "")
|
|
28
28
|
);
|
|
29
29
|
}
|
|
@@ -266,4 +266,4 @@ async function sha256OfFile(p) {
|
|
|
266
266
|
|
|
267
267
|
|
|
268
268
|
exports.MFD_MODEL = MFD_MODEL; exports.MFR_ENCODER_MODEL = MFR_ENCODER_MODEL; exports.MFR_DECODER_MODEL = MFR_DECODER_MODEL; exports.MFR_TOKENIZER = MFR_TOKENIZER; exports.ALL_FORMULA_MODELS = ALL_FORMULA_MODELS; exports.getModelsDir = getModelsDir; exports.getFormulaModelsDir = getFormulaModelsDir; exports.getFormulaModelStatus = getFormulaModelStatus; exports.ensureModelsIn = ensureModelsIn; exports.ensureFormulaModels = ensureFormulaModels; exports.ensureSingleModel = ensureSingleModel;
|
|
269
|
-
//# sourceMappingURL=chunk-
|
|
269
|
+
//# sourceMappingURL=chunk-FN6NBZ34.cjs.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"sources":["/Users/mong-e/workspace/kordoc/dist/chunk-2FNQN23O.cjs","../src/pdf/formula/models.ts","../src/shared/offline.ts"],"names":[],"mappings":"AAAA;AACE;AACF,wDAA6B;AAC7B;AACA;ACMA,gCAA2B;AAC3B,wBAAiC;AACjC,uCAA4C;AAC5C;AACA,wBAAwB;AACxB,4BAA8B;AAC9B,4CAAyB;AACzB,gCAAyB;ADJzB;AACA;AEKA,IAAM,OAAA,kBAAS,IAAI,GAAA,CAAI,CAAC,GAAA,EAAK,MAAA,EAAQ,KAAA,EAAO,IAAI,CAAC,CAAA;AAEjD,SAAS,OAAA,CAAQ,IAAA,EAAuB;AACtC,EAAA,MAAM,EAAA,EAAI,OAAA,CAAQ,GAAA,CAAI,IAAI,CAAA;AAC1B,EAAA,OAAO,CAAC,CAAC,EAAA,GAAK,MAAA,CAAO,GAAA,CAAI,CAAA,CAAE,IAAA,CAAK,CAAA,CAAE,WAAA,CAAY,CAAC,CAAA;AACjD;AAGO,SAAS,aAAA,CAAA,EAAyB;AACvC,EAAA,OAAO,OAAA,CAAQ,gBAAgB,CAAA;AACjC;AAOO,SAAS,oBAAA,CAAqB,IAAA,EAAc,IAAA,EAAqB;AACtE,EAAA,GAAA,CAAI,CAAC,aAAA,CAAc,CAAA,EAAG,MAAA;AACtB,EAAA,MAAM,IAAI,kCAAA;AAAA,IACR,CAAA,gFAAA,EAAiC,IAAI,CAAA,EAAA;AACvC,EAAA;AACF;AFZ4F;AACA;ACDxD;AAC5B,EAAA;AACI,EAAA;AACL,EAAA;AACG,EAAA;AACA,EAAA;AACV;AAE4C;AACpC,EAAA;AACI,EAAA;AACL,EAAA;AACG,EAAA;AACA,EAAA;AACV;AAE4C;AACpC,EAAA;AACI,EAAA;AACL,EAAA;AACG,EAAA;AACA,EAAA;AACV;AAEwC;AAChC,EAAA;AACI,EAAA;AACL,EAAA;AACG,EAAA;AACA,EAAA;AACV;AAE4D;AAC1D,EAAA;AACA,EAAA;AACA,EAAA;AACA,EAAA;AACF;AAOqD;AACtB,EAAA;AACI,EAAA;AACH,IAAA;AAC9B,EAAA;AAC2D,EAAA;AAC7D;AAE8C;AACd,EAAA;AAChC;AAa6G;AAC5E,EAAA;AACL,EAAA;AACiB,IAAA;AAC5B,IAAA;AACT,IAAA;AAC4B,MAAA;AACE,MAAA;AAC1B,IAAA;AACG,MAAA;AACX,IAAA;AACa,IAAA;AACoD,MAAA;AAC/D,MAAA;AACF,IAAA;AACI,IAAA;AACyC,MAAA;AACf,MAAA;AACmC,QAAA;AACxD,MAAA;AACO,QAAA;AACV,UAAA;AACA,UAAA;AACQ,UAAA;AACE,UAAA;AAC4D,UAAA;AACvE,QAAA;AACH,MAAA;AACU,IAAA;AACE,MAAA;AACV,QAAA;AACA,QAAA;AACQ,QAAA;AACE,QAAA;AACgD,QAAA;AAC3D,MAAA;AACH,IAAA;AACF,EAAA;AACO,EAAA;AACT;AAGsE;AACH,EAAA;AACnE;AAoBgI;AAC1F,EAAA;AAEV,EAAA;AACiB,IAAA;AAEU,IAAA;AACpC,sBAAA;AACX,QAAA;AACY,QAAA;AACL,QAAA;AACA,QAAA;AACE,QAAA;AACV,MAAA;AACD,MAAA;AACF,IAAA;AAGI,IAAA;AACoB,MAAA;AAChB,IAAA;AAER,IAAA;AAEgD,IAAA;AAClD,EAAA;AACF;AAGuF;AACV,EAAA;AAC7E;AAGsG;AACpE,EAAA;AACI,EAAA;AACK,EAAA;AACU,EAAA;AACe,oBAAA;AAChE,IAAA;AACF,EAAA;AACI,EAAA;AACoB,IAAA;AAChB,EAAA;AAER,EAAA;AACgD,EAAA;AAClD;AAE4F;AACtF,EAAA;AAC4B,IAAA;AACU,IAAA;AAClC,EAAA;AACC,IAAA;AACT,EAAA;AACI,EAAA;AACyC,IAAA;AACzB,IAAA;AACZ,EAAA;AACC,IAAA;AACT,EAAA;AACF;AAMiB;AAEf,EAAA;AACc,IAAA;AACZ,IAAA;AACF,EAAA;AAG6B,EAAA;AACsB,EAAA;AAEhB,EAAA;AACxB,IAAA;AAAA;AAEO,MAAA;AAChB,IAAA;AACD,EAAA;AAC2B,EAAA;AAChB,IAAA;AAC0C,MAAA;AACpD,IAAA;AACF,EAAA;AAEmD,EAAA;AACQ,EAAA;AAC1C,EAAA;AAEoB,EAAA;AACjC,EAAA;AACyF,IAAA;AACzC,IAAA;AAC5B,MAAA;AACP,sBAAA;AACX,QAAA;AACA,QAAA;AACA,QAAA;AACO,QAAA;AACR,MAAA;AACF,IAAA;AACwB,IAAA;AACf,EAAA;AACN,IAAA;AACmB,MAAA;AACf,IAAA;AAER,IAAA;AAC+D,IAAA;AACjE,EAAA;AAEa,kBAAA;AACX,IAAA;AACA,IAAA;AACA,IAAA;AACO,IAAA;AACR,EAAA;AAGG,EAAA;AACA,EAAA;AACkC,IAAA;AAC1B,EAAA;AACN,IAAA;AACmB,MAAA;AACf,IAAA;AAER,IAAA;AACiE,IAAA;AACnE,EAAA;AAE4B,EAAA;AACtB,IAAA;AACmB,MAAA;AACf,IAAA;AAER,IAAA;AACU,IAAA;AAC4D,MAAA;AACtE,IAAA;AACF,EAAA;AAEgC,EAAA;AACnB,kBAAA;AACX,IAAA;AACA,IAAA;AACA,IAAA;AACO,IAAA;AACR,EAAA;AACH;AAEwD;AACzB,EAAA;AACI,EAAA;AACY,EAAA;AACZ,IAAA;AACf,MAAA;AAEhB,IAAA;AACD,EAAA;AACoB,EAAA;AACvB;AD1E4F;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA","file":"/Users/mong-e/workspace/kordoc/dist/chunk-2FNQN23O.cjs","sourcesContent":[null,"/**\n * 수식 OCR 모델 자동 다운로드 + SHA-256 검증\n *\n * 캐시 위치: `~/.cache/kordoc/models/pix2text/`\n * 총 ~155MB (MFD 44MB + MFR encoder 87MB + MFR decoder 30MB + tokenizer 40KB)\n *\n * HF URL + SHA 는 Docufinder archive/rust-formula-ocr-wip 에서 검증 완료됨.\n * 모델 버전 갱신 시 URL + SHA 를 함께 갱신해야 한다.\n */\n\nimport { createHash } from \"crypto\"\nimport { createReadStream } from \"fs\"\nimport { mkdir, stat, unlink, rename } from \"fs/promises\"\nimport { createWriteStream } from \"fs\"\nimport { homedir } from \"os\"\nimport { join, dirname } from \"path\"\nimport { pipeline } from \"stream/promises\"\nimport { Readable } from \"stream\"\nimport { assertNetworkAllowed } from \"../../shared/offline.js\"\n\nexport interface ModelSpec {\n name: string\n filename: string\n url: string\n sha256: string\n sizeMb: number\n}\n\n/** 모델 스펙 (SHA-256 은 실제 다운로드로 검증됨 — 변경 금지) */\nexport const MFD_MODEL: ModelSpec = {\n name: \"Pix2Text MFD\",\n filename: \"mfd.onnx\",\n url: \"https://huggingface.co/breezedeus/pix2text-mfd/resolve/main/mfd-v20240618.onnx\",\n sha256: \"51a8854743b17ae654729af8db82a630c1ccfa06debf4856c8b28055f87d02c1\",\n sizeMb: 42,\n}\n\nexport const MFR_ENCODER_MODEL: ModelSpec = {\n name: \"Pix2Text MFR encoder\",\n filename: \"encoder_model.onnx\",\n url: \"https://huggingface.co/breezedeus/pix2text-mfr/resolve/main/encoder_model.onnx\",\n sha256: \"bd8d5c322792e9ec45793af5569e9748f82a3d728a9e00213dbfc56c1486f37d\",\n sizeMb: 87,\n}\n\nexport const MFR_DECODER_MODEL: ModelSpec = {\n name: \"Pix2Text MFR decoder\",\n filename: \"decoder_model.onnx\",\n url: \"https://huggingface.co/breezedeus/pix2text-mfr/resolve/main/decoder_model.onnx\",\n sha256: \"fd0f92d7a012f3dae41e1ac79421aea0ea888b5a66cb3f9a004e424f82f3daed\",\n sizeMb: 30,\n}\n\nexport const MFR_TOKENIZER: ModelSpec = {\n name: \"Pix2Text MFR tokenizer\",\n filename: \"tokenizer.json\",\n url: \"https://huggingface.co/breezedeus/pix2text-mfr/resolve/main/tokenizer.json\",\n sha256: \"3e2ab757277d22639bec28c9d7972e352d3d1dba223051fa674002dc5ab64df3\",\n sizeMb: 1,\n}\n\nexport const ALL_FORMULA_MODELS: ReadonlyArray<ModelSpec> = [\n MFD_MODEL,\n MFR_ENCODER_MODEL,\n MFR_DECODER_MODEL,\n MFR_TOKENIZER,\n]\n\n/**\n * 모델 캐시 하위 디렉토리 경로. 환경 변수 `KORDOC_MODEL_CACHE` 가 설정되면\n * 해당 경로를 우선 사용, 없으면 `~/.cache/kordoc/models/<subdir>/`.\n * (수식 OCR 외 텍스트 OCR 등 다른 모델군도 공용 — subdir 로 격리)\n */\nexport function getModelsDir(subdir: string): string {\n const override = process.env.KORDOC_MODEL_CACHE\n if (override && override.trim()) {\n return join(override, subdir)\n }\n return join(homedir(), \".cache\", \"kordoc\", \"models\", subdir)\n}\n\nexport function getFormulaModelsDir(): string {\n return getModelsDir(\"pix2text\")\n}\n\nexport interface ModelStatus {\n spec: ModelSpec\n localPath: string\n exists: boolean\n /** SHA 검증까지 끝났는가 (exists && valid) */\n verified: boolean\n /** 검증 실패 시 이유 */\n invalidReason?: string\n}\n\n/** 지정 모델군의 존재/유효성 상태 반환 (다운로드 없이 확인만) */\nexport async function getModelStatusIn(dir: string, specs: ReadonlyArray<ModelSpec>): Promise<ModelStatus[]> {\n const result: ModelStatus[] = []\n for (const spec of specs) {\n const localPath = join(dir, spec.filename)\n let exists = false\n try {\n const s = await stat(localPath)\n exists = s.isFile() && s.size > 0\n } catch {\n exists = false\n }\n if (!exists) {\n result.push({ spec, localPath, exists: false, verified: false })\n continue\n }\n try {\n const actual = await sha256OfFile(localPath)\n if (actual === spec.sha256) {\n result.push({ spec, localPath, exists: true, verified: true })\n } else {\n result.push({\n spec,\n localPath,\n exists: true,\n verified: false,\n invalidReason: `SHA256 mismatch: expected ${spec.sha256}, got ${actual}`,\n })\n }\n } catch (e) {\n result.push({\n spec,\n localPath,\n exists: true,\n verified: false,\n invalidReason: `SHA compute failed: ${(e as Error).message}`,\n })\n }\n }\n return result\n}\n\n/** 모든 수식 모델의 존재/유효성 상태 반환 (다운로드 없이 확인만) */\nexport async function getFormulaModelStatus(): Promise<ModelStatus[]> {\n return getModelStatusIn(getFormulaModelsDir(), ALL_FORMULA_MODELS)\n}\n\nexport interface DownloadProgress {\n spec: ModelSpec\n /** 현재까지 다운로드한 바이트 */\n downloaded: number\n /** 전체 바이트 (알 수 있으면) */\n total: number | null\n /** \"download\" | \"verify\" | \"done\" | \"skip\" */\n phase: \"download\" | \"verify\" | \"done\" | \"skip\" | \"error\"\n message?: string\n}\n\nexport type ProgressHandler = (p: DownloadProgress) => void\n\n/**\n * 지정 모델군을 다운로드/검증한다.\n * - 이미 있고 SHA 일치 → skip\n * - 없음 → 다운로드 후 SHA 검증. 실패 시 파일 삭제.\n */\nexport async function ensureModelsIn(dir: string, specs: ReadonlyArray<ModelSpec>, onProgress?: ProgressHandler): Promise<void> {\n await mkdir(dir, { recursive: true })\n\n for (const spec of specs) {\n const localPath = join(dir, spec.filename)\n\n if (await isExistingValid(localPath, spec.sha256)) {\n onProgress?.({\n spec,\n downloaded: 0,\n total: null,\n phase: \"skip\",\n message: \"이미 존재 + SHA 일치\",\n })\n continue\n }\n\n // 기존 파일 있지만 SHA 불일치 → 삭제\n try {\n await unlink(localPath)\n } catch {\n // 없을 수 있음\n }\n\n await downloadToFile(spec, localPath, onProgress)\n }\n}\n\n/** 필요한 모든 수식 모델을 다운로드/검증한다. */\nexport async function ensureFormulaModels(onProgress?: ProgressHandler): Promise<void> {\n return ensureModelsIn(getFormulaModelsDir(), ALL_FORMULA_MODELS, onProgress)\n}\n\n/** 단일 모델만 확인/다운로드 (진단 UI 에서 개별 상태 새로고침용) */\nexport async function ensureSingleModel(spec: ModelSpec, onProgress?: ProgressHandler): Promise<void> {\n const dir = getFormulaModelsDir()\n await mkdir(dir, { recursive: true })\n const localPath = join(dir, spec.filename)\n if (await isExistingValid(localPath, spec.sha256)) {\n onProgress?.({ spec, downloaded: 0, total: null, phase: \"skip\" })\n return\n }\n try {\n await unlink(localPath)\n } catch {\n // 재다운로드 전 손상/잔존 파일 정리 best-effort — 애초에 없으면 실패해도 무방\n }\n await downloadToFile(spec, localPath, onProgress)\n}\n\nasync function isExistingValid(localPath: string, sha256Expected: string): Promise<boolean> {\n try {\n const s = await stat(localPath)\n if (!s.isFile() || s.size === 0) return false\n } catch {\n return false\n }\n try {\n const actual = await sha256OfFile(localPath)\n return actual === sha256Expected\n } catch {\n return false\n }\n}\n\nasync function downloadToFile(\n spec: ModelSpec,\n localPath: string,\n onProgress?: ProgressHandler,\n): Promise<void> {\n // 폐쇄망 모드에서는 요청 자체를 보내지 않는다 (사이드로드 안내로 대체)\n assertNetworkAllowed(\n `${spec.name} 모델 다운로드`,\n \"온라인 PC에서 `kordoc models --export <디렉토리>` 로 내보낸 뒤 이 PC에서 `kordoc models --import <디렉토리>` 하거나, KORDOC_MODEL_CACHE 로 모델 캐시 경로를 지정하세요\",\n )\n\n // 먼저 .part 로 받고 검증 후 rename — 중단된 다운로드가 \"정상 파일\"로 오인되는 걸 방지\n const partPath = `${localPath}.part`\n await mkdir(dirname(localPath), { recursive: true })\n\n const resp = await fetch(spec.url, {\n headers: {\n // HF CDN 은 UA 없으면 가끔 403 을 뱉는다\n \"User-Agent\": \"kordoc-formula-ocr/1.0 (+https://github.com/chrisryugj/kordoc)\",\n },\n })\n if (!resp.ok || !resp.body) {\n throw new Error(\n `${spec.name} 다운로드 실패: HTTP ${resp.status} ${resp.statusText} (${spec.url})`,\n )\n }\n\n const lenHeader = resp.headers.get(\"content-length\")\n const total = lenHeader ? Number.parseInt(lenHeader, 10) : null\n let downloaded = 0\n\n const ws = createWriteStream(partPath)\n try {\n const reader = Readable.fromWeb(resp.body as unknown as import(\"stream/web\").ReadableStream)\n reader.on(\"data\", (chunk: Buffer | Uint8Array) => {\n downloaded += chunk.length\n onProgress?.({\n spec,\n downloaded,\n total,\n phase: \"download\",\n })\n })\n await pipeline(reader, ws)\n } catch (e) {\n try {\n await unlink(partPath)\n } catch {\n // .part 임시파일 정리 best-effort — 본 에러(스트리밍 실패)를 가리지 않도록 무시\n }\n throw new Error(`${spec.name} 스트리밍 실패: ${(e as Error).message}`)\n }\n\n onProgress?.({\n spec,\n downloaded,\n total,\n phase: \"verify\",\n })\n\n // SHA 검증\n let actual: string\n try {\n actual = await sha256OfFile(partPath)\n } catch (e) {\n try {\n await unlink(partPath)\n } catch {\n // .part 임시파일 정리 best-effort — 본 에러(SHA 계산 실패)를 가리지 않도록 무시\n }\n throw new Error(`${spec.name} SHA 계산 실패: ${(e as Error).message}`)\n }\n\n if (actual !== spec.sha256) {\n try {\n await unlink(partPath)\n } catch {\n // 오염된 .part 정리 best-effort — 본 에러(SHA mismatch)를 가리지 않도록 무시\n }\n throw new Error(\n `${spec.name} SHA256 mismatch: expected ${spec.sha256}, got ${actual} — 모델 URL 이 오염되었거나 전송 중 손상되었습니다.`,\n )\n }\n\n await rename(partPath, localPath)\n onProgress?.({\n spec,\n downloaded,\n total,\n phase: \"done\",\n })\n}\n\nasync function sha256OfFile(p: string): Promise<string> {\n const h = createHash(\"sha256\")\n const stream = createReadStream(p)\n await pipeline(stream, async function* (src) {\n for await (const chunk of src) {\n h.update(chunk)\n // yield 안 함 (소비만)\n }\n })\n return h.digest(\"hex\")\n}\n","/**\n * 폐쇄망(내부망) 배포용 아웃바운드 통신 킬스위치 + 파일 접근 루트 제한.\n *\n * kordoc 자체는 문서 파서라 상시 외부 통신이 없지만, 두 경로만 나간다:\n * 1. OCR/수식 모델 최초 1회 다운로드 (huggingface.co)\n * 2. `kordoc watch --webhook` 사용자가 명시한 URL\n *\n * `KORDOC_OFFLINE=1` 이면 위 경로를 **시도 전에** 차단한다. 보안성 검토에서\n * \"외부 통신 없음\"을 런타임으로 증명하기 위한 단일 관문 — 새 아웃바운드 경로를\n * 추가하면 반드시 `assertNetworkAllowed()` 를 먼저 통과시킬 것.\n *\n * `KORDOC_ROOT` 는 파일 접근을 특정 디렉토리 하위로 제한한다(미설정 시 제한 없음 —\n * 기존 동작 보존). 내부망 배포에서는 두 변수를 함께 고정하는 것을 권장.\n */\n\nimport { realpathSync } from \"fs\"\nimport { isAbsolute, relative, resolve, sep } from \"path\"\nimport { KordocError } from \"../utils.js\"\n\nconst TRUTHY = new Set([\"1\", \"true\", \"yes\", \"on\"])\n\nfunction envFlag(name: string): boolean {\n const v = process.env[name]\n return !!v && TRUTHY.has(v.trim().toLowerCase())\n}\n\n/** 폐쇄망 모드 여부 (`KORDOC_OFFLINE=1|true|yes|on`) */\nexport function isOfflineMode(): boolean {\n return envFlag(\"KORDOC_OFFLINE\")\n}\n\n/**\n * 아웃바운드 통신 관문. 폐쇄망 모드면 요청을 보내기 전에 예외로 끊는다.\n * @param what 차단 대상 설명 (사용자에게 노출됨 — 호스트/URL 은 넣지 말 것)\n * @param hint 대안 안내 (오프라인 사이드로드 방법 등)\n */\nexport function assertNetworkAllowed(what: string, hint?: string): void {\n if (!isOfflineMode()) return\n throw new KordocError(\n `폐쇄망 모드(KORDOC_OFFLINE)에서 차단됨: ${what}` + (hint ? ` — ${hint}` : \"\"),\n )\n}\n\n/** 파일 접근 루트 (`KORDOC_ROOT`). 미설정이면 null = 제한 없음 */\nexport function getAccessRoot(): string | null {\n const raw = process.env.KORDOC_ROOT\n if (!raw || !raw.trim()) return null\n const resolved = resolve(raw.trim())\n try {\n return realpathSync(resolved)\n } catch {\n // 루트가 실재하지 않으면 모든 접근이 막히는 게 안전한 기본값\n return resolved\n }\n}\n\n/** 경로가 루트 하위인지 (루트 자신 포함). 경계 문자열 매칭이 아닌 세그먼트 단위 비교 */\nexport function isWithinRoot(realPath: string, root: string): boolean {\n if (realPath === root) return true\n const rel = relative(root, realPath)\n return rel !== \"\" && !rel.startsWith(\"..\") && !isAbsolute(rel) && !rel.startsWith(`..${sep}`)\n}\n\n/**\n * `KORDOC_ROOT` 가 설정된 경우 경로가 루트 하위인지 검증한다.\n * 인자는 이미 realpath 로 해석된 절대경로여야 한다 (심볼릭 링크 탈출 방지).\n */\nexport function assertWithinRoot(realPath: string): void {\n const root = getAccessRoot()\n if (!root) return\n if (!isWithinRoot(realPath, root)) {\n throw new KordocError(`허용된 작업 디렉토리(KORDOC_ROOT) 밖의 경로입니다: ${realPath}`)\n }\n}\n"]}
|
|
1
|
+
{"version":3,"sources":["/Users/mong-e/workspace/kordoc/dist/chunk-FN6NBZ34.cjs","../src/pdf/formula/models.ts","../src/shared/offline.ts"],"names":[],"mappings":"AAAA;AACE;AACF,wDAA6B;AAC7B;AACA;ACMA,gCAA2B;AAC3B,wBAAiC;AACjC,uCAA4C;AAC5C;AACA,wBAAwB;AACxB,4BAA8B;AAC9B,4CAAyB;AACzB,gCAAyB;ADJzB;AACA;AEKA,IAAM,OAAA,kBAAS,IAAI,GAAA,CAAI,CAAC,GAAA,EAAK,MAAA,EAAQ,KAAA,EAAO,IAAI,CAAC,CAAA;AAEjD,SAAS,OAAA,CAAQ,IAAA,EAAuB;AACtC,EAAA,MAAM,EAAA,EAAI,OAAA,CAAQ,GAAA,CAAI,IAAI,CAAA;AAC1B,EAAA,OAAO,CAAC,CAAC,EAAA,GAAK,MAAA,CAAO,GAAA,CAAI,CAAA,CAAE,IAAA,CAAK,CAAA,CAAE,WAAA,CAAY,CAAC,CAAA;AACjD;AAGO,SAAS,aAAA,CAAA,EAAyB;AACvC,EAAA,OAAO,OAAA,CAAQ,gBAAgB,CAAA;AACjC;AAOO,SAAS,oBAAA,CAAqB,IAAA,EAAc,IAAA,EAAqB;AACtE,EAAA,GAAA,CAAI,CAAC,aAAA,CAAc,CAAA,EAAG,MAAA;AACtB,EAAA,MAAM,IAAI,kCAAA;AAAA,IACR,CAAA,gFAAA,EAAiC,IAAI,CAAA,EAAA;AACvC,EAAA;AACF;AFZ4F;AACA;ACDxD;AAC5B,EAAA;AACI,EAAA;AACL,EAAA;AACG,EAAA;AACA,EAAA;AACV;AAE4C;AACpC,EAAA;AACI,EAAA;AACL,EAAA;AACG,EAAA;AACA,EAAA;AACV;AAE4C;AACpC,EAAA;AACI,EAAA;AACL,EAAA;AACG,EAAA;AACA,EAAA;AACV;AAEwC;AAChC,EAAA;AACI,EAAA;AACL,EAAA;AACG,EAAA;AACA,EAAA;AACV;AAE4D;AAC1D,EAAA;AACA,EAAA;AACA,EAAA;AACA,EAAA;AACF;AAOqD;AACtB,EAAA;AACI,EAAA;AACH,IAAA;AAC9B,EAAA;AAC2D,EAAA;AAC7D;AAE8C;AACd,EAAA;AAChC;AAa6G;AAC5E,EAAA;AACL,EAAA;AACiB,IAAA;AAC5B,IAAA;AACT,IAAA;AAC4B,MAAA;AACE,MAAA;AAC1B,IAAA;AACG,MAAA;AACX,IAAA;AACa,IAAA;AACoD,MAAA;AAC/D,MAAA;AACF,IAAA;AACI,IAAA;AACyC,MAAA;AACf,MAAA;AACmC,QAAA;AACxD,MAAA;AACO,QAAA;AACV,UAAA;AACA,UAAA;AACQ,UAAA;AACE,UAAA;AAC4D,UAAA;AACvE,QAAA;AACH,MAAA;AACU,IAAA;AACE,MAAA;AACV,QAAA;AACA,QAAA;AACQ,QAAA;AACE,QAAA;AACgD,QAAA;AAC3D,MAAA;AACH,IAAA;AACF,EAAA;AACO,EAAA;AACT;AAGsE;AACH,EAAA;AACnE;AAoBgI;AAC1F,EAAA;AAEV,EAAA;AACiB,IAAA;AAEU,IAAA;AACpC,sBAAA;AACX,QAAA;AACY,QAAA;AACL,QAAA;AACA,QAAA;AACE,QAAA;AACV,MAAA;AACD,MAAA;AACF,IAAA;AAGI,IAAA;AACoB,MAAA;AAChB,IAAA;AAER,IAAA;AAEgD,IAAA;AAClD,EAAA;AACF;AAGuF;AACV,EAAA;AAC7E;AAGsG;AACpE,EAAA;AACI,EAAA;AACK,EAAA;AACU,EAAA;AACe,oBAAA;AAChE,IAAA;AACF,EAAA;AACI,EAAA;AACoB,IAAA;AAChB,EAAA;AAER,EAAA;AACgD,EAAA;AAClD;AAE4F;AACtF,EAAA;AAC4B,IAAA;AACU,IAAA;AAClC,EAAA;AACC,IAAA;AACT,EAAA;AACI,EAAA;AACyC,IAAA;AACzB,IAAA;AACZ,EAAA;AACC,IAAA;AACT,EAAA;AACF;AAMiB;AAEf,EAAA;AACc,IAAA;AACZ,IAAA;AACF,EAAA;AAG6B,EAAA;AACsB,EAAA;AAEhB,EAAA;AACxB,IAAA;AAAA;AAEO,MAAA;AAChB,IAAA;AACD,EAAA;AAC2B,EAAA;AAChB,IAAA;AAC0C,MAAA;AACpD,IAAA;AACF,EAAA;AAEmD,EAAA;AACQ,EAAA;AAC1C,EAAA;AAEoB,EAAA;AACjC,EAAA;AACyF,IAAA;AACzC,IAAA;AAC5B,MAAA;AACP,sBAAA;AACX,QAAA;AACA,QAAA;AACA,QAAA;AACO,QAAA;AACR,MAAA;AACF,IAAA;AACwB,IAAA;AACf,EAAA;AACN,IAAA;AACmB,MAAA;AACf,IAAA;AAER,IAAA;AAC+D,IAAA;AACjE,EAAA;AAEa,kBAAA;AACX,IAAA;AACA,IAAA;AACA,IAAA;AACO,IAAA;AACR,EAAA;AAGG,EAAA;AACA,EAAA;AACkC,IAAA;AAC1B,EAAA;AACN,IAAA;AACmB,MAAA;AACf,IAAA;AAER,IAAA;AACiE,IAAA;AACnE,EAAA;AAE4B,EAAA;AACtB,IAAA;AACmB,MAAA;AACf,IAAA;AAER,IAAA;AACU,IAAA;AAC4D,MAAA;AACtE,IAAA;AACF,EAAA;AAEgC,EAAA;AACnB,kBAAA;AACX,IAAA;AACA,IAAA;AACA,IAAA;AACO,IAAA;AACR,EAAA;AACH;AAEwD;AACzB,EAAA;AACI,EAAA;AACY,EAAA;AACZ,IAAA;AACf,MAAA;AAEhB,IAAA;AACD,EAAA;AACoB,EAAA;AACvB;AD1E4F;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA","file":"/Users/mong-e/workspace/kordoc/dist/chunk-FN6NBZ34.cjs","sourcesContent":[null,"/**\n * 수식 OCR 모델 자동 다운로드 + SHA-256 검증\n *\n * 캐시 위치: `~/.cache/kordoc/models/pix2text/`\n * 총 ~155MB (MFD 44MB + MFR encoder 87MB + MFR decoder 30MB + tokenizer 40KB)\n *\n * HF URL + SHA 는 Docufinder archive/rust-formula-ocr-wip 에서 검증 완료됨.\n * 모델 버전 갱신 시 URL + SHA 를 함께 갱신해야 한다.\n */\n\nimport { createHash } from \"crypto\"\nimport { createReadStream } from \"fs\"\nimport { mkdir, stat, unlink, rename } from \"fs/promises\"\nimport { createWriteStream } from \"fs\"\nimport { homedir } from \"os\"\nimport { join, dirname } from \"path\"\nimport { pipeline } from \"stream/promises\"\nimport { Readable } from \"stream\"\nimport { assertNetworkAllowed } from \"../../shared/offline.js\"\n\nexport interface ModelSpec {\n name: string\n filename: string\n url: string\n sha256: string\n sizeMb: number\n}\n\n/** 모델 스펙 (SHA-256 은 실제 다운로드로 검증됨 — 변경 금지) */\nexport const MFD_MODEL: ModelSpec = {\n name: \"Pix2Text MFD\",\n filename: \"mfd.onnx\",\n url: \"https://huggingface.co/breezedeus/pix2text-mfd/resolve/main/mfd-v20240618.onnx\",\n sha256: \"51a8854743b17ae654729af8db82a630c1ccfa06debf4856c8b28055f87d02c1\",\n sizeMb: 42,\n}\n\nexport const MFR_ENCODER_MODEL: ModelSpec = {\n name: \"Pix2Text MFR encoder\",\n filename: \"encoder_model.onnx\",\n url: \"https://huggingface.co/breezedeus/pix2text-mfr/resolve/main/encoder_model.onnx\",\n sha256: \"bd8d5c322792e9ec45793af5569e9748f82a3d728a9e00213dbfc56c1486f37d\",\n sizeMb: 87,\n}\n\nexport const MFR_DECODER_MODEL: ModelSpec = {\n name: \"Pix2Text MFR decoder\",\n filename: \"decoder_model.onnx\",\n url: \"https://huggingface.co/breezedeus/pix2text-mfr/resolve/main/decoder_model.onnx\",\n sha256: \"fd0f92d7a012f3dae41e1ac79421aea0ea888b5a66cb3f9a004e424f82f3daed\",\n sizeMb: 30,\n}\n\nexport const MFR_TOKENIZER: ModelSpec = {\n name: \"Pix2Text MFR tokenizer\",\n filename: \"tokenizer.json\",\n url: \"https://huggingface.co/breezedeus/pix2text-mfr/resolve/main/tokenizer.json\",\n sha256: \"3e2ab757277d22639bec28c9d7972e352d3d1dba223051fa674002dc5ab64df3\",\n sizeMb: 1,\n}\n\nexport const ALL_FORMULA_MODELS: ReadonlyArray<ModelSpec> = [\n MFD_MODEL,\n MFR_ENCODER_MODEL,\n MFR_DECODER_MODEL,\n MFR_TOKENIZER,\n]\n\n/**\n * 모델 캐시 하위 디렉토리 경로. 환경 변수 `KORDOC_MODEL_CACHE` 가 설정되면\n * 해당 경로를 우선 사용, 없으면 `~/.cache/kordoc/models/<subdir>/`.\n * (수식 OCR 외 텍스트 OCR 등 다른 모델군도 공용 — subdir 로 격리)\n */\nexport function getModelsDir(subdir: string): string {\n const override = process.env.KORDOC_MODEL_CACHE\n if (override && override.trim()) {\n return join(override, subdir)\n }\n return join(homedir(), \".cache\", \"kordoc\", \"models\", subdir)\n}\n\nexport function getFormulaModelsDir(): string {\n return getModelsDir(\"pix2text\")\n}\n\nexport interface ModelStatus {\n spec: ModelSpec\n localPath: string\n exists: boolean\n /** SHA 검증까지 끝났는가 (exists && valid) */\n verified: boolean\n /** 검증 실패 시 이유 */\n invalidReason?: string\n}\n\n/** 지정 모델군의 존재/유효성 상태 반환 (다운로드 없이 확인만) */\nexport async function getModelStatusIn(dir: string, specs: ReadonlyArray<ModelSpec>): Promise<ModelStatus[]> {\n const result: ModelStatus[] = []\n for (const spec of specs) {\n const localPath = join(dir, spec.filename)\n let exists = false\n try {\n const s = await stat(localPath)\n exists = s.isFile() && s.size > 0\n } catch {\n exists = false\n }\n if (!exists) {\n result.push({ spec, localPath, exists: false, verified: false })\n continue\n }\n try {\n const actual = await sha256OfFile(localPath)\n if (actual === spec.sha256) {\n result.push({ spec, localPath, exists: true, verified: true })\n } else {\n result.push({\n spec,\n localPath,\n exists: true,\n verified: false,\n invalidReason: `SHA256 mismatch: expected ${spec.sha256}, got ${actual}`,\n })\n }\n } catch (e) {\n result.push({\n spec,\n localPath,\n exists: true,\n verified: false,\n invalidReason: `SHA compute failed: ${(e as Error).message}`,\n })\n }\n }\n return result\n}\n\n/** 모든 수식 모델의 존재/유효성 상태 반환 (다운로드 없이 확인만) */\nexport async function getFormulaModelStatus(): Promise<ModelStatus[]> {\n return getModelStatusIn(getFormulaModelsDir(), ALL_FORMULA_MODELS)\n}\n\nexport interface DownloadProgress {\n spec: ModelSpec\n /** 현재까지 다운로드한 바이트 */\n downloaded: number\n /** 전체 바이트 (알 수 있으면) */\n total: number | null\n /** \"download\" | \"verify\" | \"done\" | \"skip\" */\n phase: \"download\" | \"verify\" | \"done\" | \"skip\" | \"error\"\n message?: string\n}\n\nexport type ProgressHandler = (p: DownloadProgress) => void\n\n/**\n * 지정 모델군을 다운로드/검증한다.\n * - 이미 있고 SHA 일치 → skip\n * - 없음 → 다운로드 후 SHA 검증. 실패 시 파일 삭제.\n */\nexport async function ensureModelsIn(dir: string, specs: ReadonlyArray<ModelSpec>, onProgress?: ProgressHandler): Promise<void> {\n await mkdir(dir, { recursive: true })\n\n for (const spec of specs) {\n const localPath = join(dir, spec.filename)\n\n if (await isExistingValid(localPath, spec.sha256)) {\n onProgress?.({\n spec,\n downloaded: 0,\n total: null,\n phase: \"skip\",\n message: \"이미 존재 + SHA 일치\",\n })\n continue\n }\n\n // 기존 파일 있지만 SHA 불일치 → 삭제\n try {\n await unlink(localPath)\n } catch {\n // 없을 수 있음\n }\n\n await downloadToFile(spec, localPath, onProgress)\n }\n}\n\n/** 필요한 모든 수식 모델을 다운로드/검증한다. */\nexport async function ensureFormulaModels(onProgress?: ProgressHandler): Promise<void> {\n return ensureModelsIn(getFormulaModelsDir(), ALL_FORMULA_MODELS, onProgress)\n}\n\n/** 단일 모델만 확인/다운로드 (진단 UI 에서 개별 상태 새로고침용) */\nexport async function ensureSingleModel(spec: ModelSpec, onProgress?: ProgressHandler): Promise<void> {\n const dir = getFormulaModelsDir()\n await mkdir(dir, { recursive: true })\n const localPath = join(dir, spec.filename)\n if (await isExistingValid(localPath, spec.sha256)) {\n onProgress?.({ spec, downloaded: 0, total: null, phase: \"skip\" })\n return\n }\n try {\n await unlink(localPath)\n } catch {\n // 재다운로드 전 손상/잔존 파일 정리 best-effort — 애초에 없으면 실패해도 무방\n }\n await downloadToFile(spec, localPath, onProgress)\n}\n\nasync function isExistingValid(localPath: string, sha256Expected: string): Promise<boolean> {\n try {\n const s = await stat(localPath)\n if (!s.isFile() || s.size === 0) return false\n } catch {\n return false\n }\n try {\n const actual = await sha256OfFile(localPath)\n return actual === sha256Expected\n } catch {\n return false\n }\n}\n\nasync function downloadToFile(\n spec: ModelSpec,\n localPath: string,\n onProgress?: ProgressHandler,\n): Promise<void> {\n // 폐쇄망 모드에서는 요청 자체를 보내지 않는다 (사이드로드 안내로 대체)\n assertNetworkAllowed(\n `${spec.name} 모델 다운로드`,\n \"온라인 PC에서 `kordoc models --export <디렉토리>` 로 내보낸 뒤 이 PC에서 `kordoc models --import <디렉토리>` 하거나, KORDOC_MODEL_CACHE 로 모델 캐시 경로를 지정하세요\",\n )\n\n // 먼저 .part 로 받고 검증 후 rename — 중단된 다운로드가 \"정상 파일\"로 오인되는 걸 방지\n const partPath = `${localPath}.part`\n await mkdir(dirname(localPath), { recursive: true })\n\n const resp = await fetch(spec.url, {\n headers: {\n // HF CDN 은 UA 없으면 가끔 403 을 뱉는다\n \"User-Agent\": \"kordoc-formula-ocr/1.0 (+https://github.com/chrisryugj/kordoc)\",\n },\n })\n if (!resp.ok || !resp.body) {\n throw new Error(\n `${spec.name} 다운로드 실패: HTTP ${resp.status} ${resp.statusText} (${spec.url})`,\n )\n }\n\n const lenHeader = resp.headers.get(\"content-length\")\n const total = lenHeader ? Number.parseInt(lenHeader, 10) : null\n let downloaded = 0\n\n const ws = createWriteStream(partPath)\n try {\n const reader = Readable.fromWeb(resp.body as unknown as import(\"stream/web\").ReadableStream)\n reader.on(\"data\", (chunk: Buffer | Uint8Array) => {\n downloaded += chunk.length\n onProgress?.({\n spec,\n downloaded,\n total,\n phase: \"download\",\n })\n })\n await pipeline(reader, ws)\n } catch (e) {\n try {\n await unlink(partPath)\n } catch {\n // .part 임시파일 정리 best-effort — 본 에러(스트리밍 실패)를 가리지 않도록 무시\n }\n throw new Error(`${spec.name} 스트리밍 실패: ${(e as Error).message}`)\n }\n\n onProgress?.({\n spec,\n downloaded,\n total,\n phase: \"verify\",\n })\n\n // SHA 검증\n let actual: string\n try {\n actual = await sha256OfFile(partPath)\n } catch (e) {\n try {\n await unlink(partPath)\n } catch {\n // .part 임시파일 정리 best-effort — 본 에러(SHA 계산 실패)를 가리지 않도록 무시\n }\n throw new Error(`${spec.name} SHA 계산 실패: ${(e as Error).message}`)\n }\n\n if (actual !== spec.sha256) {\n try {\n await unlink(partPath)\n } catch {\n // 오염된 .part 정리 best-effort — 본 에러(SHA mismatch)를 가리지 않도록 무시\n }\n throw new Error(\n `${spec.name} SHA256 mismatch: expected ${spec.sha256}, got ${actual} — 모델 URL 이 오염되었거나 전송 중 손상되었습니다.`,\n )\n }\n\n await rename(partPath, localPath)\n onProgress?.({\n spec,\n downloaded,\n total,\n phase: \"done\",\n })\n}\n\nasync function sha256OfFile(p: string): Promise<string> {\n const h = createHash(\"sha256\")\n const stream = createReadStream(p)\n await pipeline(stream, async function* (src) {\n for await (const chunk of src) {\n h.update(chunk)\n // yield 안 함 (소비만)\n }\n })\n return h.digest(\"hex\")\n}\n","/**\n * 폐쇄망(내부망) 배포용 아웃바운드 통신 킬스위치 + 파일 접근 루트 제한.\n *\n * kordoc 자체는 문서 파서라 상시 외부 통신이 없지만, 두 경로만 나간다:\n * 1. OCR/수식 모델 최초 1회 다운로드 (huggingface.co)\n * 2. `kordoc watch --webhook` 사용자가 명시한 URL\n *\n * `KORDOC_OFFLINE=1` 이면 위 경로를 **시도 전에** 차단한다. 보안성 검토에서\n * \"외부 통신 없음\"을 런타임으로 증명하기 위한 단일 관문 — 새 아웃바운드 경로를\n * 추가하면 반드시 `assertNetworkAllowed()` 를 먼저 통과시킬 것.\n *\n * `KORDOC_ROOT` 는 파일 접근을 특정 디렉토리 하위로 제한한다(미설정 시 제한 없음 —\n * 기존 동작 보존). 내부망 배포에서는 두 변수를 함께 고정하는 것을 권장.\n */\n\nimport { realpathSync } from \"fs\"\nimport { isAbsolute, relative, resolve, sep } from \"path\"\nimport { KordocError } from \"../utils.js\"\n\nconst TRUTHY = new Set([\"1\", \"true\", \"yes\", \"on\"])\n\nfunction envFlag(name: string): boolean {\n const v = process.env[name]\n return !!v && TRUTHY.has(v.trim().toLowerCase())\n}\n\n/** 폐쇄망 모드 여부 (`KORDOC_OFFLINE=1|true|yes|on`) */\nexport function isOfflineMode(): boolean {\n return envFlag(\"KORDOC_OFFLINE\")\n}\n\n/**\n * 아웃바운드 통신 관문. 폐쇄망 모드면 요청을 보내기 전에 예외로 끊는다.\n * @param what 차단 대상 설명 (사용자에게 노출됨 — 호스트/URL 은 넣지 말 것)\n * @param hint 대안 안내 (오프라인 사이드로드 방법 등)\n */\nexport function assertNetworkAllowed(what: string, hint?: string): void {\n if (!isOfflineMode()) return\n throw new KordocError(\n `폐쇄망 모드(KORDOC_OFFLINE)에서 차단됨: ${what}` + (hint ? ` — ${hint}` : \"\"),\n )\n}\n\n/** 파일 접근 루트 (`KORDOC_ROOT`). 미설정이면 null = 제한 없음 */\nexport function getAccessRoot(): string | null {\n const raw = process.env.KORDOC_ROOT\n if (!raw || !raw.trim()) return null\n const resolved = resolve(raw.trim())\n try {\n return realpathSync(resolved)\n } catch {\n // 루트가 실재하지 않으면 모든 접근이 막히는 게 안전한 기본값\n return resolved\n }\n}\n\n/** 경로가 루트 하위인지 (루트 자신 포함). 경계 문자열 매칭이 아닌 세그먼트 단위 비교 */\nexport function isWithinRoot(realPath: string, root: string): boolean {\n if (realPath === root) return true\n const rel = relative(root, realPath)\n return rel !== \"\" && !rel.startsWith(\"..\") && !isAbsolute(rel) && !rel.startsWith(`..${sep}`)\n}\n\n/**\n * `KORDOC_ROOT` 가 설정된 경우 경로가 루트 하위인지 검증한다.\n * 인자는 이미 realpath 로 해석된 절대경로여야 한다 (심볼릭 링크 탈출 방지).\n */\nexport function assertWithinRoot(realPath: string): void {\n const root = getAccessRoot()\n if (!root) return\n if (!isWithinRoot(realPath, root)) {\n throw new KordocError(`허용된 작업 디렉토리(KORDOC_ROOT) 밖의 경로입니다: ${realPath}`)\n }\n}\n"]}
|