@clazic/kordoc 2.5.1 → 2.5.2

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.
@@ -0,0 +1,234 @@
1
+ #!/usr/bin/env node
2
+ import "./chunk-ZWE3DS7E.js";
3
+
4
+ // src/ocr/batch-provider.ts
5
+ import { spawn, execSync } from "child_process";
6
+ import { writeFileSync, readFileSync, unlinkSync, mkdirSync } from "fs";
7
+ import { join } from "path";
8
+ import { tmpdir } from "os";
9
+ var BATCH_OCR_PROMPT = `\uB2E4\uC74C \uBB38\uC11C \uD398\uC774\uC9C0 \uC774\uBBF8\uC9C0\uB4E4\uC744 OCR\uD558\uC5EC \uC21C\uC218 Markdown\uC73C\uB85C \uBCC0\uD658\uD558\uACE0, \uBA85\uBC31\uD55C OCR \uC624\uC778\uC2DD\uACFC \uD45C \uAD6C\uC870 \uAE68\uC9D0\uB9CC \uAD50\uC815\uD558\uC5EC \uCD5C\uC885 \uACB0\uACFC\uB97C \uCD9C\uB825\uD558\uC138\uC694.
10
+
11
+ [\uD398\uC774\uC9C0 \uAD6C\uBD84 \u2014 \uD544\uC218]
12
+ - \uAC01 \uD398\uC774\uC9C0 \uACB0\uACFC \uC0AC\uC774\uC5D0 \uBC18\uB4DC\uC2DC \uC774 \uAD6C\uBD84\uC790\uB97C \uC0BD\uC785: <!-- PAGE_BREAK -->
13
+
14
+ [\uAE30\uBCF8 \uCD94\uCD9C \uADDC\uCE59]
15
+ - \uBCF8\uBB38, \uD45C, \uC81C\uBAA9, \uB9AC\uC2A4\uD2B8, \uCEA1\uC158\uC744 \uC6D0\uBB38 \uAD6C\uC870 \uADF8\uB300\uB85C Markdown\uC73C\uB85C \uBCC0\uD658
16
+ - \uD5E4\uB529\uC740 \uC2DC\uAC01\uC801 \uD06C\uAE30\xB7\uAD75\uAE30\uC5D0 \uB530\uB77C # ~ ###### \uC0AC\uC6A9
17
+ - \uB9AC\uC2A4\uD2B8\uB294 -, 1. \uC0AC\uC6A9 (\uC6D0\uBB38 \uBC88\uD638 \uCCB4\uACC4\uAC00 \u2460, \uAC00., 1) \uB4F1\uC774\uBA74 \uADF8 \uD45C\uAE30 \uC720\uC9C0)
18
+ - \uC774\uBBF8\uC9C0\xB7\uB3C4\uD615\xB7\uB85C\uACE0\xB7\uD398\uC774\uC9C0\uBC88\uD638\xB7\uBA38\uB9AC\uAE00/\uBC14\uB2E5\uAE00\uC740 \uBB34\uC2DC
19
+ - \uC6D0\uBB38\uC758 \uC77D\uAE30 \uC21C\uC11C(\uC88C\u2192\uC6B0, \uC704\u2192\uC544\uB798, \uB2E4\uB2E8\uC774\uBA74 \uB2E8\uBCC4\uB85C)\uB97C \uC720\uC9C0
20
+
21
+ [\uD45C \uCD94\uCD9C \uADDC\uCE59 \u2014 \uAC00\uC7A5 \uC911\uC694]
22
+ - \uD45C\uB294 \uBC18\uB4DC\uC2DC Markdown \uD30C\uC774\uD504 \uD14C\uC774\uBE14 \uBB38\uBC95 \uC0AC\uC6A9: \uD5E4\uB354 \uD589 + |---| \uAD6C\uBD84\uC120 + \uB370\uC774\uD130 \uD589
23
+ - \uC2DC\uAC01\uC801\uC73C\uB85C \uAD75\uAC70\uB098 \uC74C\uC601(\uD68C\uC0C9/\uC0C9\uC0C1) \uBC30\uACBD\uC774 \uC788\uB294 \uD589\uC744 \uD5E4\uB354\uB85C \uC2DD\uBCC4. \uD5E4\uB354\uAC00 \uC5C6\uB294 \uD45C\uB77C\uB3C4 \uCCAB \uD589\uC744 \uD5E4\uB354\uB85C \uB450\uACE0 |---| \uAD6C\uBD84\uC120 \uCD94\uAC00
24
+ - \uBAA8\uB4E0 \uD589\uC758 \uD30C\uC774\uD504(|) \uAC1C\uC218\uB97C \uD5E4\uB354\uC640 \uB3D9\uC77C\uD558\uAC8C \uB9DE\uCD9C \uAC83 \u2014 \uBE48 \uC140\uC740 \uACF5\uBC31\uC73C\uB85C \uCC44\uC6B0\uACE0 \uC808\uB300 \uC140\uC744 \uC0DD\uB7B5\uD558\uC9C0 \uB9D0 \uAC83
25
+ - \uC140 \uC548\uC758 \uC904\uBC14\uAFC8\uC740 <br>\uB85C \uD45C\uAE30 (\uC2E4\uC81C \uAC1C\uD589 \uBB38\uC790 \uC0AC\uC6A9 \uAE08\uC9C0 \u2014 \uD45C\uAC00 \uAE68\uC9D0)
26
+ - \uBCD1\uD569 \uC140 \uCC98\uB9AC:
27
+ - \uAC00\uB85C \uBCD1\uD569(colspan): \uBCD1\uD569\uB41C \uC140 \uB0B4\uC6A9\uC744 \uCCAB \uCE78\uC5D0 \uC4F0\uACE0 \uB098\uBA38\uC9C0 \uCE78\uC740 \uBE48 \uCE78\uC73C\uB85C \uB458 \uAC83
28
+ - \uC138\uB85C \uBCD1\uD569(rowspan): \uBCD1\uD569\uB41C \uBAA8\uB4E0 \uD589\uC758 \uD574\uB2F9 \uCE78\uC5D0 \uB3D9\uC77C\uD55C \uB0B4\uC6A9\uC744 \uBC18\uBCF5 \uAE30\uC7AC\uD560 \uAC83
29
+ - 2\uB2E8 \uD5E4\uB354(\uD5E4\uB354\uAC00 \uB450 \uC904\uC778 \uACBD\uC6B0): \uC0C1\uC704/\uD558\uC704 \uD5E4\uB354\uB97C " / "\uB85C \uD569\uCCD0 \uB2E8\uC77C \uD5E4\uB354 \uD589\uC73C\uB85C \uB9CC\uB4E4 \uAC83 (\uC608: "2024 / 1\uBD84\uAE30")
30
+ - \uD45C \uC704\xB7\uC544\uB798\uC758 \uCEA1\uC158(\uC608: "<\uD45C 1> ...", "[\uD45C 2-1]")\uC740 \uD45C \uBC14\uB85C \uC704\uC5D0 \uBCC4\uB3C4 \uC904\uB85C \uBCF4\uC874
31
+ - \uD55C \uD398\uC774\uC9C0\uC5D0\uC11C \uD45C\uAC00 \uC911\uAC04\uC5D0 \uB04A\uC5B4\uC838 \uBCF4\uC774\uB354\uB77C\uB3C4, \uD5E4\uB354\uAC00 \uB3D9\uC77C\uD558\uBA74 \uD558\uB098\uC758 \uD45C\uB85C \uC774\uC5B4\uBD99\uC77C \uAC83
32
+ - \uC88C\uCE21 \uCCAB \uCEEC\uB7FC\uC774 \uC138\uB85C\uC4F0\uAE30 \uB77C\uBCA8(\uAD6C\uBD84/\uBD84\uB958/\uC5F0\uB3C4 \uB4F1)\uC774\uB77C\uB3C4 \uC77C\uBC18 \uC140\uB85C \uBCC0\uD658\uD558\uC5EC \uB204\uB77D \uAE08\uC9C0
33
+
34
+ [OCR \uC624\uC778\uC2DD\xB7\uAE68\uC9D0 \uAD50\uC815 \u2014 \uD5C8\uC6A9 \uBC94\uC704\uB9CC]
35
+ - \uC22B\uC790 \uCE78\uC5D0\uC11C 'O'\u2192'0', 'l/I'\u2192'1' \uB4F1 \uB9E5\uB77D\uC0C1 \uBA85\uBC31\uD55C \uC624\uC778\uC2DD\uB9CC \uAD50\uC815
36
+ - \uB2E8\uC5B4 \uC911\uAC04\uC5D0 \uC798\uBABB \uC0BD\uC785\uB41C \uACF5\uBC31 \uC81C\uAC70 (\uC608: "\uC8FC\uAC70 \uC885 \uD569 \uACC4\uD68D" \u2192 "\uC8FC\uAC70\uC885\uD569\uACC4\uD68D")
37
+ - \uC904\uBC14\uAFC8 \uC624\uB958\uB85C \uB04A\uAE34 \uBB38\uC7A5\uC740 \uC758\uBBF8 \uB2E8\uC704\uB85C \uBCD1\uD569 (\uB2E8, \uD45C \uC140 \uB0B4\uBD80\uB294 <br> \uC720\uC9C0)
38
+ - \uD45C\uC5D0\uC11C \uD589 \uC5B4\uAE0B\uB0A8\uC774 \uC758\uC2EC\uB418\uBA74(\uCEEC\uB7FC \uC218 \uBD88\uC77C\uCE58) \uBE48 \uC140\uB85C \uD328\uB529\uD558\uC5EC \uCEEC\uB7FC \uC815\uD569\uC131\uC744 \uC6B0\uC120 \uD655\uBCF4
39
+
40
+ [\uC808\uB300 \uAE08\uC9C0 \uC0AC\uD56D]
41
+ - \uBB38\uC7A5\xB7\uB2E8\uB77D\xB7\uD56D\uBAA9\xB7\uD45C \uD589/\uC5F4\uC744 \uCD94\uAC00\uD558\uAC70\uB098 \uC0AD\uC81C\uD558\uC9C0 \uB9D0 \uAC83
42
+ - \uC22B\uC790, \uD37C\uC13C\uD2B8, \uB0A0\uC9DC, \uB2E8\uC704, \uAE08\uC561\uC744 \uC808\uB300 \uBCC0\uACBD\uD558\uC9C0 \uB9D0 \uAC83
43
+ - \uACE0\uC720\uBA85\uC0AC, \uAE30\uAD00\uBA85, \uBC95\uB839\uBA85, \uC9C0\uBA85, \uC778\uBA85\uC744 \uBCC0\uACBD\uD558\uC9C0 \uB9D0 \uAC83
44
+ - \uD45C\uC758 \uD5E4\uB354 \uD14D\uC2A4\uD2B8, \uCEA1\uC158, \uD589/\uC5F4 \uAC1C\uC218\uB97C \uC784\uC758\uB85C \uBC14\uAFB8\uC9C0 \uB9D0 \uAC83
45
+ - \uCCB4\uD06C\uBC15\uC2A4(\u2611/\u2610/\u25A0/\u25A1), \uD2B9\uC218\uBB38\uC790(\u203B, \u2460\u2461\u2462, \u3260, \uAC00., \uB098.)\uB97C \uC784\uC758\uB85C \uBC14\uAFB8\uC9C0 \uB9D0 \uAC83
46
+ - \uC6D0\uBB38\uC5D0 \uC5C6\uB294 \uB0B4\uC6A9\uC744 \uC694\uC57D\xB7\uBCF4\uC644\xB7\uCD94\uB860\uD558\uC9C0 \uB9D0 \uAC83
47
+ - \`\`\`, \`\`\`markdown \uAC19\uC740 \uCF54\uB4DC\uD39C\uC2A4\uB85C \uAC10\uC2F8\uC9C0 \uB9D0 \uAC83
48
+ - \uC124\uBA85\xB7\uC8FC\uC11D\xB7\uBA54\uD0C0 \uBB38\uC7A5 \uCD94\uAC00 \uAE08\uC9C0
49
+
50
+ [\uBD88\uD655\uC2E4\uD560 \uB54C]
51
+ - \uAE00\uC790\uAC00 \uD750\uB9BF\uD558\uAC70\uB098 \uD310\uB3C5 \uBD88\uAC00\uD558\uBA74 \uBCF4\uC774\uB294 \uADF8\uB300\uB85C \uB450\uAC70\uB098, \uBD88\uAC00\uD53C\uD558\uBA74 \uD55C \uAE00\uC790\uB9CC ?\uB85C \uD45C\uC2DC
52
+ - \uD45C \uAD6C\uC870\uAC00 \uBAA8\uD638\uD558\uBA74 \uCEEC\uB7FC \uC218 \uC77C\uCE58 + \uBE48 \uC140 \uD328\uB529 \uD615\uD0DC\uB85C \uCD9C\uB825
53
+ - \uCD94\uCE21\uBCF4\uB2E4 \uC6D0\uBB38 \uBCF4\uC874\uC744 \uD56D\uC0C1 \uC6B0\uC120`;
54
+ var DEFAULT_BATCH_SIZES = {
55
+ gemini: 5,
56
+ claude: 5,
57
+ codex: 10
58
+ };
59
+ var _batchTempDir = null;
60
+ function getBatchTempDir() {
61
+ if (!_batchTempDir) {
62
+ _batchTempDir = join(process.cwd(), ".kordoc_ocr_tmp");
63
+ mkdirSync(_batchTempDir, { recursive: true });
64
+ if (process.platform === "win32") {
65
+ try {
66
+ execSync(`attrib +h "${_batchTempDir}"`, { stdio: "ignore" });
67
+ } catch {
68
+ }
69
+ }
70
+ }
71
+ return _batchTempDir;
72
+ }
73
+ function createBatchCliProvider(mode, batchSize) {
74
+ return {
75
+ __batch: true,
76
+ batchSize,
77
+ async processBatch(pages) {
78
+ const results = /* @__PURE__ */ new Map();
79
+ const tempDir = getBatchTempDir();
80
+ const tempFiles = [];
81
+ try {
82
+ for (const { image, pageNum } of pages) {
83
+ const path = join(tempDir, `batch-p${pageNum}.png`);
84
+ writeFileSync(path, image);
85
+ tempFiles.push(path);
86
+ }
87
+ let output;
88
+ if (mode === "codex") {
89
+ output = await callBatchCodexCli(tempFiles);
90
+ } else {
91
+ output = await callBatchCli(mode, tempFiles);
92
+ }
93
+ const cleaned = stripCodeFence(output.trim());
94
+ const parts = cleaned.split(/<!--\s*PAGE_BREAK\s*-->/).map((p) => p.trim()).filter((p) => p.length > 0);
95
+ for (let i = 0; i < pages.length; i++) {
96
+ const pageNum = pages[i].pageNum;
97
+ if (i < parts.length) {
98
+ results.set(pageNum, { markdown: parts[i] });
99
+ }
100
+ }
101
+ } finally {
102
+ for (const f of tempFiles) {
103
+ try {
104
+ unlinkSync(f);
105
+ } catch {
106
+ }
107
+ }
108
+ }
109
+ return results;
110
+ }
111
+ };
112
+ }
113
+ function spawnAsync(cmd, args, opts) {
114
+ return new Promise((resolve, reject) => {
115
+ const child = spawn(cmd, args, {
116
+ cwd: opts.cwd,
117
+ env: process.env,
118
+ stdio: ["pipe", "pipe", "pipe"],
119
+ shell: process.platform === "win32"
120
+ });
121
+ let stdout = "";
122
+ let stderr = "";
123
+ let killed = false;
124
+ child.stdout.setEncoding("utf-8");
125
+ child.stderr.setEncoding("utf-8");
126
+ child.stdout.on("data", (d) => {
127
+ stdout += d;
128
+ });
129
+ child.stderr.on("data", (d) => {
130
+ stderr += d;
131
+ });
132
+ const timer = setTimeout(() => {
133
+ killed = true;
134
+ if (process.platform === "win32") {
135
+ child.kill();
136
+ } else {
137
+ child.kill("SIGTERM");
138
+ }
139
+ }, opts.timeoutMs);
140
+ if (opts.stdin !== void 0) {
141
+ child.stdin.end(opts.stdin);
142
+ } else {
143
+ child.stdin.end();
144
+ }
145
+ child.on("close", (code) => {
146
+ clearTimeout(timer);
147
+ if (killed) {
148
+ reject(new Error(`\uD0C0\uC784\uC544\uC6C3 (${Math.round(opts.timeoutMs / 1e3)}\uCD08)`));
149
+ } else {
150
+ resolve({ stdout, stderr, exitCode: code ?? 1 });
151
+ }
152
+ });
153
+ child.on("error", (err) => {
154
+ clearTimeout(timer);
155
+ reject(err);
156
+ });
157
+ });
158
+ }
159
+ async function callBatchCli(mode, imagePaths) {
160
+ const fileRefs = imagePaths.map((p) => `@${p.replace(/\\/g, "/")}`).join("\n");
161
+ const prompt = `${BATCH_OCR_PROMPT}
162
+
163
+ ${fileRefs}`;
164
+ let args;
165
+ if (mode === "gemini") {
166
+ const model = process.env.KORDOC_GEMINI_MODEL ?? "gemini-2.5-flash";
167
+ args = ["--prompt", prompt, "--yolo", "--model", model];
168
+ } else {
169
+ args = ["--print", prompt];
170
+ const model = process.env.KORDOC_CLAUDE_MODEL;
171
+ if (model) args.push("--model", model);
172
+ }
173
+ const timeoutMs = 6e4 + imagePaths.length * 2e4;
174
+ const result = await spawnAsync(mode, args, {
175
+ timeoutMs,
176
+ ...mode === "claude" ? { cwd: tmpdir() } : {}
177
+ });
178
+ if (result.exitCode !== 0) {
179
+ const errMsg = result.stderr?.trim() || `exit code ${result.exitCode}`;
180
+ throw new Error(`${mode} \uBC30\uCE58 OCR \uC2E4\uD328: ${errMsg}`);
181
+ }
182
+ const output = result.stdout || "";
183
+ checkForLimitError(output, mode);
184
+ return output;
185
+ }
186
+ async function callBatchCodexCli(imagePaths) {
187
+ const outPath = join(tmpdir(), `kordoc-codex-batch-${Date.now()}-${Math.random().toString(36).slice(2)}.txt`);
188
+ try {
189
+ const args = ["exec", BATCH_OCR_PROMPT];
190
+ for (const p of imagePaths) {
191
+ args.push("--image", p);
192
+ }
193
+ args.push("--output-last-message", outPath);
194
+ const model = process.env.KORDOC_CODEX_MODEL;
195
+ if (model) args.push("--model", model);
196
+ const timeoutMs = 6e4 + imagePaths.length * 2e4;
197
+ const result = await spawnAsync("codex", args, {
198
+ timeoutMs,
199
+ stdin: ""
200
+ });
201
+ if (result.exitCode !== 0) {
202
+ const errMsg = result.stderr?.trim() || `exit code ${result.exitCode}`;
203
+ throw new Error(`codex \uBC30\uCE58 OCR \uC2E4\uD328: ${errMsg}`);
204
+ }
205
+ let text;
206
+ try {
207
+ text = readFileSync(outPath, "utf-8");
208
+ } catch {
209
+ text = result.stdout || "";
210
+ }
211
+ checkForLimitError(text, "codex");
212
+ return text;
213
+ } finally {
214
+ try {
215
+ unlinkSync(outPath);
216
+ } catch {
217
+ }
218
+ }
219
+ }
220
+ function checkForLimitError(output, mode) {
221
+ const lower = output.toLowerCase();
222
+ if (lower.includes("usage limit") || lower.includes("rate limit")) {
223
+ throw new Error(`${mode} \uC0AC\uC6A9\uB7C9/\uC18D\uB3C4 \uC81C\uD55C: ${output.trim().slice(0, 200)}`);
224
+ }
225
+ }
226
+ function stripCodeFence(text) {
227
+ const match = text.match(/^```(?:markdown|md)?\s*\n([\s\S]*?)\n```\s*$/m);
228
+ return match ? match[1].trim() : text;
229
+ }
230
+ export {
231
+ DEFAULT_BATCH_SIZES,
232
+ createBatchCliProvider
233
+ };
234
+ //# sourceMappingURL=batch-provider-XRF6F26E.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"sources":["../src/ocr/batch-provider.ts"],"sourcesContent":["/**\n * CLI 배치 OCR 프로바이더\n *\n * 여러 페이지 이미지를 단일 CLI 호출로 처리하여 API 호출 수를 대폭 감소.\n * gemini/claude: @file 멀티 참조, codex: --image 멀티 플래그\n *\n * 299페이지 기준:\n * - 기존: CLI 299회 호출 (~30분)\n * - 배치: CLI 3~6회 호출 (~3분)\n */\n\nimport { spawn, execSync } from \"child_process\"\nimport { writeFileSync, readFileSync, unlinkSync, mkdirSync } from \"fs\"\nimport { join } from \"path\"\nimport { tmpdir } from \"os\"\nimport type { StructuredOcrResult, BatchOcrProvider } from \"../types.js\"\n\n/** 배치 OCR 프롬프트 */\nconst BATCH_OCR_PROMPT = `다음 문서 페이지 이미지들을 OCR하여 순수 Markdown으로 변환하고, 명백한 OCR 오인식과 표 구조 깨짐만 교정하여 최종 결과를 출력하세요.\n\n[페이지 구분 — 필수]\n- 각 페이지 결과 사이에 반드시 이 구분자를 삽입: <!-- PAGE_BREAK -->\n\n[기본 추출 규칙]\n- 본문, 표, 제목, 리스트, 캡션을 원문 구조 그대로 Markdown으로 변환\n- 헤딩은 시각적 크기·굵기에 따라 # ~ ###### 사용\n- 리스트는 -, 1. 사용 (원문 번호 체계가 ①, 가., 1) 등이면 그 표기 유지)\n- 이미지·도형·로고·페이지번호·머리글/바닥글은 무시\n- 원문의 읽기 순서(좌→우, 위→아래, 다단이면 단별로)를 유지\n\n[표 추출 규칙 — 가장 중요]\n- 표는 반드시 Markdown 파이프 테이블 문법 사용: 헤더 행 + |---| 구분선 + 데이터 행\n- 시각적으로 굵거나 음영(회색/색상) 배경이 있는 행을 헤더로 식별. 헤더가 없는 표라도 첫 행을 헤더로 두고 |---| 구분선 추가\n- 모든 행의 파이프(|) 개수를 헤더와 동일하게 맞출 것 — 빈 셀은 공백으로 채우고 절대 셀을 생략하지 말 것\n- 셀 안의 줄바꿈은 <br>로 표기 (실제 개행 문자 사용 금지 — 표가 깨짐)\n- 병합 셀 처리:\n - 가로 병합(colspan): 병합된 셀 내용을 첫 칸에 쓰고 나머지 칸은 빈 칸으로 둘 것\n - 세로 병합(rowspan): 병합된 모든 행의 해당 칸에 동일한 내용을 반복 기재할 것\n- 2단 헤더(헤더가 두 줄인 경우): 상위/하위 헤더를 \" / \"로 합쳐 단일 헤더 행으로 만들 것 (예: \"2024 / 1분기\")\n- 표 위·아래의 캡션(예: \"<표 1> ...\", \"[표 2-1]\")은 표 바로 위에 별도 줄로 보존\n- 한 페이지에서 표가 중간에 끊어져 보이더라도, 헤더가 동일하면 하나의 표로 이어붙일 것\n- 좌측 첫 컬럼이 세로쓰기 라벨(구분/분류/연도 등)이라도 일반 셀로 변환하여 누락 금지\n\n[OCR 오인식·깨짐 교정 — 허용 범위만]\n- 숫자 칸에서 'O'→'0', 'l/I'→'1' 등 맥락상 명백한 오인식만 교정\n- 단어 중간에 잘못 삽입된 공백 제거 (예: \"주거 종 합 계획\" → \"주거종합계획\")\n- 줄바꿈 오류로 끊긴 문장은 의미 단위로 병합 (단, 표 셀 내부는 <br> 유지)\n- 표에서 행 어긋남이 의심되면(컬럼 수 불일치) 빈 셀로 패딩하여 컬럼 정합성을 우선 확보\n\n[절대 금지 사항]\n- 문장·단락·항목·표 행/열을 추가하거나 삭제하지 말 것\n- 숫자, 퍼센트, 날짜, 단위, 금액을 절대 변경하지 말 것\n- 고유명사, 기관명, 법령명, 지명, 인명을 변경하지 말 것\n- 표의 헤더 텍스트, 캡션, 행/열 개수를 임의로 바꾸지 말 것\n- 체크박스(☑/☐/■/□), 특수문자(※, ①②③, ㉠, 가., 나.)를 임의로 바꾸지 말 것\n- 원문에 없는 내용을 요약·보완·추론하지 말 것\n- \\`\\`\\`, \\`\\`\\`markdown 같은 코드펜스로 감싸지 말 것\n- 설명·주석·메타 문장 추가 금지\n\n[불확실할 때]\n- 글자가 흐릿하거나 판독 불가하면 보이는 그대로 두거나, 불가피하면 한 글자만 ?로 표시\n- 표 구조가 모호하면 컬럼 수 일치 + 빈 셀 패딩 형태로 출력\n- 추측보다 원문 보존을 항상 우선`\n\n/** 모드별 기본 배치 크기 (CLI 내부 타임아웃 + 실측 기반)\n *\n * gemini CLI: 10장 이상에서 AbortError 발생 (내부 타임아웃).\n * 5장 배치가 안정적으로 동작 확인 (35초/배치).\n * 299페이지 = 60배치 = 기존 299회 대비 80% 감소.\n */\nexport const DEFAULT_BATCH_SIZES: Record<string, number> = {\n gemini: 5,\n claude: 5,\n codex: 10,\n}\n\n/**\n * 임시 디렉토리 — gemini CLI는 cwd 하위 + gitignore 밖만 @참조 가능\n *\n * 숨김 처리:\n * - macOS/Linux: '.' 접두사로 기본 숨김 (ls -a 로만 표시)\n * - Windows: '.' 접두사 + attrib +h 로 숨김 속성 부여\n */\nlet _batchTempDir: string | null = null\nfunction getBatchTempDir(): string {\n if (!_batchTempDir) {\n _batchTempDir = join(process.cwd(), \".kordoc_ocr_tmp\")\n mkdirSync(_batchTempDir, { recursive: true })\n // Windows: dot-prefix만으로 숨김 처리 불충분 → attrib +h 추가\n if (process.platform === \"win32\") {\n try { execSync(`attrib +h \"${_batchTempDir}\"`, { stdio: \"ignore\" }) } catch { /* ignore */ }\n }\n }\n return _batchTempDir\n}\n\n/**\n * 배치 CLI 프로바이더 생성\n */\nexport function createBatchCliProvider(\n mode: \"gemini\" | \"claude\" | \"codex\",\n batchSize: number\n): BatchOcrProvider {\n return {\n __batch: true as const,\n batchSize,\n async processBatch(pages) {\n const results = new Map<number, StructuredOcrResult>()\n const tempDir = getBatchTempDir()\n const tempFiles: string[] = []\n\n try {\n // 1. Write all page images to temp files\n for (const { image, pageNum } of pages) {\n const path = join(tempDir, `batch-p${pageNum}.png`)\n writeFileSync(path, image)\n tempFiles.push(path)\n }\n\n // 2. Call CLI with all file references (비동기 — 병렬 배치 실행 가능)\n let output: string\n if (mode === \"codex\") {\n output = await callBatchCodexCli(tempFiles)\n } else {\n output = await callBatchCli(mode, tempFiles)\n }\n\n // 3. Parse response by PAGE_BREAK separator\n const cleaned = stripCodeFence(output.trim())\n const parts = cleaned.split(/<!--\\s*PAGE_BREAK\\s*-->/)\n .map(p => p.trim())\n .filter(p => p.length > 0)\n\n // 4. Map results to page numbers (best-effort if count mismatch)\n for (let i = 0; i < pages.length; i++) {\n const pageNum = pages[i].pageNum\n if (i < parts.length) {\n results.set(pageNum, { markdown: parts[i] })\n }\n // If fewer parts than pages, remaining pages get no result\n }\n } finally {\n // 5. Clean up temp files\n for (const f of tempFiles) {\n try { unlinkSync(f) } catch { /* ignore */ }\n }\n }\n\n return results\n },\n }\n}\n\n/**\n * 비동기 CLI 실행 헬퍼 — spawn + Promise 래핑.\n * spawnSync는 이벤트 루프를 차단하여 병렬 배치 실행 불가.\n */\nfunction spawnAsync(\n cmd: string,\n args: string[],\n opts: { timeoutMs: number; cwd?: string; stdin?: string }\n): Promise<{ stdout: string; stderr: string; exitCode: number }> {\n return new Promise((resolve, reject) => {\n const child = spawn(cmd, args, {\n cwd: opts.cwd,\n env: process.env,\n stdio: [\"pipe\", \"pipe\", \"pipe\"],\n shell: process.platform === \"win32\",\n })\n\n let stdout = \"\"\n let stderr = \"\"\n let killed = false\n\n child.stdout.setEncoding(\"utf-8\")\n child.stderr.setEncoding(\"utf-8\")\n child.stdout.on(\"data\", (d: string) => { stdout += d })\n child.stderr.on(\"data\", (d: string) => { stderr += d })\n\n const timer = setTimeout(() => {\n killed = true\n if (process.platform === \"win32\") {\n child.kill()\n } else {\n child.kill(\"SIGTERM\")\n }\n }, opts.timeoutMs)\n\n if (opts.stdin !== undefined) {\n child.stdin.end(opts.stdin)\n } else {\n child.stdin.end()\n }\n\n child.on(\"close\", (code) => {\n clearTimeout(timer)\n if (killed) {\n reject(new Error(`타임아웃 (${Math.round(opts.timeoutMs / 1000)}초)`))\n } else {\n resolve({ stdout, stderr, exitCode: code ?? 1 })\n }\n })\n child.on(\"error\", (err) => {\n clearTimeout(timer)\n reject(err)\n })\n })\n}\n\n/** gemini/claude 배치 호출 (비동기) */\nasync function callBatchCli(mode: \"gemini\" | \"claude\", imagePaths: string[]): Promise<string> {\n const fileRefs = imagePaths.map(p => `@${p.replace(/\\\\/g, \"/\")}`).join(\"\\n\")\n const prompt = `${BATCH_OCR_PROMPT}\\n\\n${fileRefs}`\n\n let args: string[]\n if (mode === \"gemini\") {\n const model = process.env.KORDOC_GEMINI_MODEL ?? \"gemini-2.5-flash\"\n args = [\"--prompt\", prompt, \"--yolo\", \"--model\", model]\n } else {\n args = [\"--print\", prompt]\n const model = process.env.KORDOC_CLAUDE_MODEL\n if (model) args.push(\"--model\", model)\n }\n\n const timeoutMs = 60_000 + imagePaths.length * 20_000\n const result = await spawnAsync(mode, args, {\n timeoutMs,\n ...(mode === \"claude\" ? { cwd: tmpdir() } : {}),\n })\n\n if (result.exitCode !== 0) {\n const errMsg = result.stderr?.trim() || `exit code ${result.exitCode}`\n throw new Error(`${mode} 배치 OCR 실패: ${errMsg}`)\n }\n\n const output = result.stdout || \"\"\n checkForLimitError(output, mode)\n return output\n}\n\n/** codex 배치 호출 (비동기) — --image를 여러 번 지정 */\nasync function callBatchCodexCli(imagePaths: string[]): Promise<string> {\n const outPath = join(tmpdir(), `kordoc-codex-batch-${Date.now()}-${Math.random().toString(36).slice(2)}.txt`)\n try {\n const args = [\"exec\", BATCH_OCR_PROMPT]\n for (const p of imagePaths) {\n args.push(\"--image\", p)\n }\n args.push(\"--output-last-message\", outPath)\n const model = process.env.KORDOC_CODEX_MODEL\n if (model) args.push(\"--model\", model)\n\n const timeoutMs = 60_000 + imagePaths.length * 20_000\n const result = await spawnAsync(\"codex\", args, {\n timeoutMs,\n stdin: \"\",\n })\n\n if (result.exitCode !== 0) {\n const errMsg = result.stderr?.trim() || `exit code ${result.exitCode}`\n throw new Error(`codex 배치 OCR 실패: ${errMsg}`)\n }\n\n let text: string\n try {\n text = readFileSync(outPath, \"utf-8\")\n } catch {\n text = result.stdout || \"\"\n }\n checkForLimitError(text, \"codex\")\n return text\n } finally {\n try { unlinkSync(outPath) } catch { /* ignore */ }\n }\n}\n\n/**\n * 출력 텍스트에서 사용량·속도 제한 에러 감지.\n * 해당 메시지가 포함된 경우 throw하여 다음 엔진으로 fallback 트리거.\n */\nfunction checkForLimitError(output: string, mode: string): void {\n const lower = output.toLowerCase()\n if (lower.includes(\"usage limit\") || lower.includes(\"rate limit\")) {\n throw new Error(`${mode} 사용량/속도 제한: ${output.trim().slice(0, 200)}`)\n }\n}\n\n/** LLM 출력에서 코드 펜스 제거 (cli-provider.ts와 동일 로직) */\nfunction stripCodeFence(text: string): string {\n const match = text.match(/^```(?:markdown|md)?\\s*\\n([\\s\\S]*?)\\n```\\s*$/m)\n return match ? match[1].trim() : text\n}\n"],"mappings":";;;;AAWA,SAAS,OAAO,gBAAgB;AAChC,SAAS,eAAe,cAAc,YAAY,iBAAiB;AACnE,SAAS,YAAY;AACrB,SAAS,cAAc;AAIvB,IAAM,mBAAmB;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAoDlB,IAAM,sBAA8C;AAAA,EACzD,QAAQ;AAAA,EACR,QAAQ;AAAA,EACR,OAAO;AACT;AASA,IAAI,gBAA+B;AACnC,SAAS,kBAA0B;AACjC,MAAI,CAAC,eAAe;AAClB,oBAAgB,KAAK,QAAQ,IAAI,GAAG,iBAAiB;AACrD,cAAU,eAAe,EAAE,WAAW,KAAK,CAAC;AAE5C,QAAI,QAAQ,aAAa,SAAS;AAChC,UAAI;AAAE,iBAAS,cAAc,aAAa,KAAK,EAAE,OAAO,SAAS,CAAC;AAAA,MAAE,QAAQ;AAAA,MAAe;AAAA,IAC7F;AAAA,EACF;AACA,SAAO;AACT;AAKO,SAAS,uBACd,MACA,WACkB;AAClB,SAAO;AAAA,IACL,SAAS;AAAA,IACT;AAAA,IACA,MAAM,aAAa,OAAO;AACxB,YAAM,UAAU,oBAAI,IAAiC;AACrD,YAAM,UAAU,gBAAgB;AAChC,YAAM,YAAsB,CAAC;AAE7B,UAAI;AAEF,mBAAW,EAAE,OAAO,QAAQ,KAAK,OAAO;AACtC,gBAAM,OAAO,KAAK,SAAS,UAAU,OAAO,MAAM;AAClD,wBAAc,MAAM,KAAK;AACzB,oBAAU,KAAK,IAAI;AAAA,QACrB;AAGA,YAAI;AACJ,YAAI,SAAS,SAAS;AACpB,mBAAS,MAAM,kBAAkB,SAAS;AAAA,QAC5C,OAAO;AACL,mBAAS,MAAM,aAAa,MAAM,SAAS;AAAA,QAC7C;AAGA,cAAM,UAAU,eAAe,OAAO,KAAK,CAAC;AAC5C,cAAM,QAAQ,QAAQ,MAAM,yBAAyB,EAClD,IAAI,OAAK,EAAE,KAAK,CAAC,EACjB,OAAO,OAAK,EAAE,SAAS,CAAC;AAG3B,iBAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACrC,gBAAM,UAAU,MAAM,CAAC,EAAE;AACzB,cAAI,IAAI,MAAM,QAAQ;AACpB,oBAAQ,IAAI,SAAS,EAAE,UAAU,MAAM,CAAC,EAAE,CAAC;AAAA,UAC7C;AAAA,QAEF;AAAA,MACF,UAAE;AAEA,mBAAW,KAAK,WAAW;AACzB,cAAI;AAAE,uBAAW,CAAC;AAAA,UAAE,QAAQ;AAAA,UAAe;AAAA,QAC7C;AAAA,MACF;AAEA,aAAO;AAAA,IACT;AAAA,EACF;AACF;AAMA,SAAS,WACP,KACA,MACA,MAC+D;AAC/D,SAAO,IAAI,QAAQ,CAAC,SAAS,WAAW;AACtC,UAAM,QAAQ,MAAM,KAAK,MAAM;AAAA,MAC7B,KAAK,KAAK;AAAA,MACV,KAAK,QAAQ;AAAA,MACb,OAAO,CAAC,QAAQ,QAAQ,MAAM;AAAA,MAC9B,OAAO,QAAQ,aAAa;AAAA,IAC9B,CAAC;AAED,QAAI,SAAS;AACb,QAAI,SAAS;AACb,QAAI,SAAS;AAEb,UAAM,OAAO,YAAY,OAAO;AAChC,UAAM,OAAO,YAAY,OAAO;AAChC,UAAM,OAAO,GAAG,QAAQ,CAAC,MAAc;AAAE,gBAAU;AAAA,IAAE,CAAC;AACtD,UAAM,OAAO,GAAG,QAAQ,CAAC,MAAc;AAAE,gBAAU;AAAA,IAAE,CAAC;AAEtD,UAAM,QAAQ,WAAW,MAAM;AAC7B,eAAS;AACT,UAAI,QAAQ,aAAa,SAAS;AAChC,cAAM,KAAK;AAAA,MACb,OAAO;AACL,cAAM,KAAK,SAAS;AAAA,MACtB;AAAA,IACF,GAAG,KAAK,SAAS;AAEjB,QAAI,KAAK,UAAU,QAAW;AAC5B,YAAM,MAAM,IAAI,KAAK,KAAK;AAAA,IAC5B,OAAO;AACL,YAAM,MAAM,IAAI;AAAA,IAClB;AAEA,UAAM,GAAG,SAAS,CAAC,SAAS;AAC1B,mBAAa,KAAK;AAClB,UAAI,QAAQ;AACV,eAAO,IAAI,MAAM,6BAAS,KAAK,MAAM,KAAK,YAAY,GAAI,CAAC,SAAI,CAAC;AAAA,MAClE,OAAO;AACL,gBAAQ,EAAE,QAAQ,QAAQ,UAAU,QAAQ,EAAE,CAAC;AAAA,MACjD;AAAA,IACF,CAAC;AACD,UAAM,GAAG,SAAS,CAAC,QAAQ;AACzB,mBAAa,KAAK;AAClB,aAAO,GAAG;AAAA,IACZ,CAAC;AAAA,EACH,CAAC;AACH;AAGA,eAAe,aAAa,MAA2B,YAAuC;AAC5F,QAAM,WAAW,WAAW,IAAI,OAAK,IAAI,EAAE,QAAQ,OAAO,GAAG,CAAC,EAAE,EAAE,KAAK,IAAI;AAC3E,QAAM,SAAS,GAAG,gBAAgB;AAAA;AAAA,EAAO,QAAQ;AAEjD,MAAI;AACJ,MAAI,SAAS,UAAU;AACrB,UAAM,QAAQ,QAAQ,IAAI,uBAAuB;AACjD,WAAO,CAAC,YAAY,QAAQ,UAAU,WAAW,KAAK;AAAA,EACxD,OAAO;AACL,WAAO,CAAC,WAAW,MAAM;AACzB,UAAM,QAAQ,QAAQ,IAAI;AAC1B,QAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AAAA,EACvC;AAEA,QAAM,YAAY,MAAS,WAAW,SAAS;AAC/C,QAAM,SAAS,MAAM,WAAW,MAAM,MAAM;AAAA,IAC1C;AAAA,IACA,GAAI,SAAS,WAAW,EAAE,KAAK,OAAO,EAAE,IAAI,CAAC;AAAA,EAC/C,CAAC;AAED,MAAI,OAAO,aAAa,GAAG;AACzB,UAAM,SAAS,OAAO,QAAQ,KAAK,KAAK,aAAa,OAAO,QAAQ;AACpE,UAAM,IAAI,MAAM,GAAG,IAAI,mCAAe,MAAM,EAAE;AAAA,EAChD;AAEA,QAAM,SAAS,OAAO,UAAU;AAChC,qBAAmB,QAAQ,IAAI;AAC/B,SAAO;AACT;AAGA,eAAe,kBAAkB,YAAuC;AACtE,QAAM,UAAU,KAAK,OAAO,GAAG,sBAAsB,KAAK,IAAI,CAAC,IAAI,KAAK,OAAO,EAAE,SAAS,EAAE,EAAE,MAAM,CAAC,CAAC,MAAM;AAC5G,MAAI;AACF,UAAM,OAAO,CAAC,QAAQ,gBAAgB;AACtC,eAAW,KAAK,YAAY;AAC1B,WAAK,KAAK,WAAW,CAAC;AAAA,IACxB;AACA,SAAK,KAAK,yBAAyB,OAAO;AAC1C,UAAM,QAAQ,QAAQ,IAAI;AAC1B,QAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AAErC,UAAM,YAAY,MAAS,WAAW,SAAS;AAC/C,UAAM,SAAS,MAAM,WAAW,SAAS,MAAM;AAAA,MAC7C;AAAA,MACA,OAAO;AAAA,IACT,CAAC;AAED,QAAI,OAAO,aAAa,GAAG;AACzB,YAAM,SAAS,OAAO,QAAQ,KAAK,KAAK,aAAa,OAAO,QAAQ;AACpE,YAAM,IAAI,MAAM,wCAAoB,MAAM,EAAE;AAAA,IAC9C;AAEA,QAAI;AACJ,QAAI;AACF,aAAO,aAAa,SAAS,OAAO;AAAA,IACtC,QAAQ;AACN,aAAO,OAAO,UAAU;AAAA,IAC1B;AACA,uBAAmB,MAAM,OAAO;AAChC,WAAO;AAAA,EACT,UAAE;AACA,QAAI;AAAE,iBAAW,OAAO;AAAA,IAAE,QAAQ;AAAA,IAAe;AAAA,EACnD;AACF;AAMA,SAAS,mBAAmB,QAAgB,MAAoB;AAC9D,QAAM,QAAQ,OAAO,YAAY;AACjC,MAAI,MAAM,SAAS,aAAa,KAAK,MAAM,SAAS,YAAY,GAAG;AACjE,UAAM,IAAI,MAAM,GAAG,IAAI,kDAAe,OAAO,KAAK,EAAE,MAAM,GAAG,GAAG,CAAC,EAAE;AAAA,EACrE;AACF;AAGA,SAAS,eAAe,MAAsB;AAC5C,QAAM,QAAQ,KAAK,MAAM,+CAA+C;AACxE,SAAO,QAAQ,MAAM,CAAC,EAAE,KAAK,IAAI;AACnC;","names":[]}
@@ -7,13 +7,13 @@ import {
7
7
  precheckZipSize,
8
8
  sanitizeHref,
9
9
  toArrayBuffer
10
- } from "./chunk-IJGNPAK2.js";
10
+ } from "./chunk-5CILZHRW.js";
11
11
  import {
12
12
  parsePageRange
13
13
  } from "./chunk-MOL7MDBG.js";
14
14
  import {
15
15
  createCliOcrProvider
16
- } from "./chunk-Y4WFKJ5P.js";
16
+ } from "./chunk-S7BHLD2V.js";
17
17
  import {
18
18
  markdownToBlocks
19
19
  } from "./chunk-YW5G6BCJ.js";
@@ -5685,7 +5685,7 @@ async function parsePdfDocument(buffer, options) {
5685
5685
  if (ocrMode === "off") {
5686
5686
  throw Object.assign(new KordocError(`\uC774\uBBF8\uC9C0 \uAE30\uBC18 PDF (${pageCount}\uD398\uC774\uC9C0, ${totalChars}\uC790)`), { isImageBased: true });
5687
5687
  }
5688
- const { resolveOcrProvider } = await import("./resolve-XWYJYKKH.js");
5688
+ const { resolveOcrProvider } = await import("./resolve-ZSUEJK3E.js");
5689
5689
  const { ocrPages } = await import("./provider-T2D5XRTI.js");
5690
5690
  const tryProvider = async (provider, filter) => {
5691
5691
  try {
@@ -10206,4 +10206,4 @@ export {
10206
10206
  cfb/cfb.js:
10207
10207
  (*! crc32.js (C) 2014-present SheetJS -- http://sheetjs.com *)
10208
10208
  */
10209
- //# sourceMappingURL=chunk-QG6BYZMR.js.map
10209
+ //# sourceMappingURL=chunk-25ZYYLVP.js.map
@@ -1,7 +1,7 @@
1
1
  #!/usr/bin/env node
2
2
 
3
3
  // src/utils.ts
4
- var VERSION = true ? "2.5.0" : "0.0.0-dev";
4
+ var VERSION = true ? "2.5.1" : "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;
@@ -105,4 +105,4 @@ export {
105
105
  classifyError,
106
106
  normalizeKordocError
107
107
  };
108
- //# sourceMappingURL=chunk-IJGNPAK2.js.map
108
+ //# sourceMappingURL=chunk-5CILZHRW.js.map
@@ -0,0 +1,200 @@
1
+ #!/usr/bin/env node
2
+
3
+ // src/ocr/cli-provider.ts
4
+ import { spawnSync } from "child_process";
5
+ import { writeFileSync, readFileSync, unlinkSync, mkdirSync } from "fs";
6
+ import { join } from "path";
7
+ import { tmpdir } from "os";
8
+ var OCR_PROMPT = `\uC774 PDF \uD398\uC774\uC9C0 \uC774\uBBF8\uC9C0\uC5D0\uC11C \uD14D\uC2A4\uD2B8\uC640 \uD45C\uB97C \uCD94\uCD9C\uD558\uC5EC \uC21C\uC218 Markdown\uC73C\uB85C \uBCC0\uD658\uD558\uACE0, \uBA85\uBC31\uD55C OCR \uC624\uC778\uC2DD\uACFC \uD45C \uAD6C\uC870 \uAE68\uC9D0\uB9CC \uAD50\uC815\uD558\uC5EC \uCD5C\uC885 \uACB0\uACFC\uB97C \uCD9C\uB825\uD558\uC138\uC694.
9
+
10
+ [\uAE30\uBCF8 \uCD94\uCD9C \uADDC\uCE59]
11
+ - \uBCF8\uBB38, \uD45C, \uC81C\uBAA9, \uB9AC\uC2A4\uD2B8, \uCEA1\uC158\uC744 \uC6D0\uBB38 \uAD6C\uC870 \uADF8\uB300\uB85C Markdown\uC73C\uB85C \uBCC0\uD658
12
+ - \uD5E4\uB529\uC740 \uC2DC\uAC01\uC801 \uD06C\uAE30\xB7\uAD75\uAE30\uC5D0 \uB530\uB77C # ~ ###### \uC0AC\uC6A9
13
+ - \uB9AC\uC2A4\uD2B8\uB294 -, 1. \uC0AC\uC6A9 (\uC6D0\uBB38 \uBC88\uD638 \uCCB4\uACC4\uAC00 \u2460, \uAC00., 1) \uB4F1\uC774\uBA74 \uADF8 \uD45C\uAE30 \uC720\uC9C0)
14
+ - \uC774\uBBF8\uC9C0\xB7\uB3C4\uD615\xB7\uB85C\uACE0\xB7\uD398\uC774\uC9C0\uBC88\uD638\xB7\uBA38\uB9AC\uAE00/\uBC14\uB2E5\uAE00\uC740 \uBB34\uC2DC
15
+ - \uC6D0\uBB38\uC758 \uC77D\uAE30 \uC21C\uC11C(\uC88C\u2192\uC6B0, \uC704\u2192\uC544\uB798, \uB2E4\uB2E8\uC774\uBA74 \uB2E8\uBCC4\uB85C)\uB97C \uC720\uC9C0
16
+
17
+ [\uD45C \uCD94\uCD9C \uADDC\uCE59 \u2014 \uAC00\uC7A5 \uC911\uC694]
18
+ - \uD45C\uB294 \uBC18\uB4DC\uC2DC Markdown \uD30C\uC774\uD504 \uD14C\uC774\uBE14 \uBB38\uBC95 \uC0AC\uC6A9: \uD5E4\uB354 \uD589 + |---| \uAD6C\uBD84\uC120 + \uB370\uC774\uD130 \uD589
19
+ - \uC2DC\uAC01\uC801\uC73C\uB85C \uAD75\uAC70\uB098 \uC74C\uC601(\uD68C\uC0C9/\uC0C9\uC0C1) \uBC30\uACBD\uC774 \uC788\uB294 \uD589\uC744 \uD5E4\uB354\uB85C \uC2DD\uBCC4. \uD5E4\uB354\uAC00 \uC5C6\uB294 \uD45C\uB77C\uB3C4 \uCCAB \uD589\uC744 \uD5E4\uB354\uB85C \uB450\uACE0 |---| \uAD6C\uBD84\uC120 \uCD94\uAC00
20
+ - \uBAA8\uB4E0 \uD589\uC758 \uD30C\uC774\uD504(|) \uAC1C\uC218\uB97C \uD5E4\uB354\uC640 \uB3D9\uC77C\uD558\uAC8C \uB9DE\uCD9C \uAC83 \u2014 \uBE48 \uC140\uC740 \uACF5\uBC31\uC73C\uB85C \uCC44\uC6B0\uACE0 \uC808\uB300 \uC140\uC744 \uC0DD\uB7B5\uD558\uC9C0 \uB9D0 \uAC83
21
+ - \uC140 \uC548\uC758 \uC904\uBC14\uAFC8\uC740 <br>\uB85C \uD45C\uAE30 (\uC2E4\uC81C \uAC1C\uD589 \uBB38\uC790 \uC0AC\uC6A9 \uAE08\uC9C0 \u2014 \uD45C\uAC00 \uAE68\uC9D0)
22
+ - \uBCD1\uD569 \uC140 \uCC98\uB9AC:
23
+ - \uAC00\uB85C \uBCD1\uD569(colspan): \uBCD1\uD569\uB41C \uC140 \uB0B4\uC6A9\uC744 \uCCAB \uCE78\uC5D0 \uC4F0\uACE0 \uB098\uBA38\uC9C0 \uCE78\uC740 \uBE48 \uCE78\uC73C\uB85C \uB458 \uAC83
24
+ - \uC138\uB85C \uBCD1\uD569(rowspan): \uBCD1\uD569\uB41C \uBAA8\uB4E0 \uD589\uC758 \uD574\uB2F9 \uCE78\uC5D0 \uB3D9\uC77C\uD55C \uB0B4\uC6A9\uC744 \uBC18\uBCF5 \uAE30\uC7AC\uD560 \uAC83
25
+ - 2\uB2E8 \uD5E4\uB354(\uD5E4\uB354\uAC00 \uB450 \uC904\uC778 \uACBD\uC6B0): \uC0C1\uC704/\uD558\uC704 \uD5E4\uB354\uB97C " / "\uB85C \uD569\uCCD0 \uB2E8\uC77C \uD5E4\uB354 \uD589\uC73C\uB85C \uB9CC\uB4E4 \uAC83 (\uC608: "2024 / 1\uBD84\uAE30")
26
+ - \uD45C \uC704\xB7\uC544\uB798\uC758 \uCEA1\uC158(\uC608: "<\uD45C 1> ...", "[\uD45C 2-1]")\uC740 \uD45C \uBC14\uB85C \uC704\uC5D0 \uBCC4\uB3C4 \uC904\uB85C \uBCF4\uC874
27
+ - \uD55C \uD398\uC774\uC9C0\uC5D0\uC11C \uD45C\uAC00 \uC911\uAC04\uC5D0 \uB04A\uC5B4\uC838 \uBCF4\uC774\uB354\uB77C\uB3C4, \uD5E4\uB354\uAC00 \uB3D9\uC77C\uD558\uBA74 \uD558\uB098\uC758 \uD45C\uB85C \uC774\uC5B4\uBD99\uC77C \uAC83
28
+ - \uC88C\uCE21 \uCCAB \uCEEC\uB7FC\uC774 \uC138\uB85C\uC4F0\uAE30 \uB77C\uBCA8(\uAD6C\uBD84/\uBD84\uB958/\uC5F0\uB3C4 \uB4F1)\uC774\uB77C\uB3C4 \uC77C\uBC18 \uC140\uB85C \uBCC0\uD658\uD558\uC5EC \uB204\uB77D \uAE08\uC9C0
29
+
30
+ [OCR \uC624\uC778\uC2DD\xB7\uAE68\uC9D0 \uAD50\uC815 \u2014 \uD5C8\uC6A9 \uBC94\uC704\uB9CC]
31
+ - \uC22B\uC790 \uCE78\uC5D0\uC11C 'O'\u2192'0', 'l/I'\u2192'1' \uB4F1 \uB9E5\uB77D\uC0C1 \uBA85\uBC31\uD55C \uC624\uC778\uC2DD\uB9CC \uAD50\uC815
32
+ - \uB2E8\uC5B4 \uC911\uAC04\uC5D0 \uC798\uBABB \uC0BD\uC785\uB41C \uACF5\uBC31 \uC81C\uAC70 (\uC608: "\uC8FC\uAC70 \uC885 \uD569 \uACC4\uD68D" \u2192 "\uC8FC\uAC70\uC885\uD569\uACC4\uD68D")
33
+ - \uC904\uBC14\uAFC8 \uC624\uB958\uB85C \uB04A\uAE34 \uBB38\uC7A5\uC740 \uC758\uBBF8 \uB2E8\uC704\uB85C \uBCD1\uD569 (\uB2E8, \uD45C \uC140 \uB0B4\uBD80\uB294 <br> \uC720\uC9C0)
34
+ - \uD45C\uC5D0\uC11C \uD589 \uC5B4\uAE0B\uB0A8\uC774 \uC758\uC2EC\uB418\uBA74(\uCEEC\uB7FC \uC218 \uBD88\uC77C\uCE58) \uBE48 \uC140\uB85C \uD328\uB529\uD558\uC5EC \uCEEC\uB7FC \uC815\uD569\uC131\uC744 \uC6B0\uC120 \uD655\uBCF4
35
+
36
+ [\uC808\uB300 \uAE08\uC9C0 \uC0AC\uD56D]
37
+ - \uBB38\uC7A5\xB7\uB2E8\uB77D\xB7\uD56D\uBAA9\xB7\uD45C \uD589/\uC5F4\uC744 \uCD94\uAC00\uD558\uAC70\uB098 \uC0AD\uC81C\uD558\uC9C0 \uB9D0 \uAC83
38
+ - \uC22B\uC790, \uD37C\uC13C\uD2B8, \uB0A0\uC9DC, \uB2E8\uC704, \uAE08\uC561\uC744 \uC808\uB300 \uBCC0\uACBD\uD558\uC9C0 \uB9D0 \uAC83
39
+ - \uACE0\uC720\uBA85\uC0AC, \uAE30\uAD00\uBA85, \uBC95\uB839\uBA85, \uC9C0\uBA85, \uC778\uBA85\uC744 \uBCC0\uACBD\uD558\uC9C0 \uB9D0 \uAC83
40
+ - \uD45C\uC758 \uD5E4\uB354 \uD14D\uC2A4\uD2B8, \uCEA1\uC158, \uD589/\uC5F4 \uAC1C\uC218\uB97C \uC784\uC758\uB85C \uBC14\uAFB8\uC9C0 \uB9D0 \uAC83
41
+ - \uCCB4\uD06C\uBC15\uC2A4(\u2611/\u2610/\u25A0/\u25A1), \uD2B9\uC218\uBB38\uC790(\u203B, \u2460\u2461\u2462, \u3260, \uAC00., \uB098.)\uB97C \uC784\uC758\uB85C \uBC14\uAFB8\uC9C0 \uB9D0 \uAC83
42
+ - \uC6D0\uBB38\uC5D0 \uC5C6\uB294 \uB0B4\uC6A9\uC744 \uC694\uC57D\xB7\uBCF4\uC644\xB7\uCD94\uB860\uD558\uC9C0 \uB9D0 \uAC83
43
+ - \`\`\`, \`\`\`markdown \uAC19\uC740 \uCF54\uB4DC\uD39C\uC2A4\uB85C \uAC10\uC2F8\uC9C0 \uB9D0 \uAC83
44
+ - \uC124\uBA85\xB7\uC8FC\uC11D\xB7\uBA54\uD0C0 \uBB38\uC7A5 \uCD94\uAC00 \uAE08\uC9C0
45
+
46
+ [\uBD88\uD655\uC2E4\uD560 \uB54C]
47
+ - \uAE00\uC790\uAC00 \uD750\uB9BF\uD558\uAC70\uB098 \uD310\uB3C5 \uBD88\uAC00\uD558\uBA74 \uBCF4\uC774\uB294 \uADF8\uB300\uB85C \uB450\uAC70\uB098, \uBD88\uAC00\uD53C\uD558\uBA74 \uD55C \uAE00\uC790\uB9CC ?\uB85C \uD45C\uC2DC
48
+ - \uD45C \uAD6C\uC870\uAC00 \uBAA8\uD638\uD558\uBA74 \uCEEC\uB7FC \uC218 \uC77C\uCE58 + \uBE48 \uC140 \uD328\uB529 \uD615\uD0DC\uB85C \uCD9C\uB825
49
+ - \uCD94\uCE21\uBCF4\uB2E4 \uC6D0\uBB38 \uBCF4\uC874\uC744 \uD56D\uC0C1 \uC6B0\uC120`;
50
+ var _tempDir = null;
51
+ function getTempDir() {
52
+ if (!_tempDir) {
53
+ _tempDir = join(process.cwd(), ".kordoc_ocr_tmp");
54
+ mkdirSync(_tempDir, { recursive: true });
55
+ }
56
+ return _tempDir;
57
+ }
58
+ function createCliOcrProvider(mode) {
59
+ return async (pageImage, pageNumber) => {
60
+ const tempPath = join(getTempDir(), `page-${pageNumber}.png`);
61
+ try {
62
+ writeFileSync(tempPath, pageImage);
63
+ let output;
64
+ if (mode === "ollama") {
65
+ output = await callOllamaApi(tempPath);
66
+ } else {
67
+ output = callCli(mode, tempPath);
68
+ }
69
+ return { markdown: stripCodeFence(output.trim()) };
70
+ } finally {
71
+ try {
72
+ unlinkSync(tempPath);
73
+ } catch {
74
+ }
75
+ }
76
+ };
77
+ }
78
+ function checkForLimitError(output, mode) {
79
+ const lower = output.toLowerCase();
80
+ if (lower.includes("usage limit") || lower.includes("rate limit")) {
81
+ throw new Error(`${mode} \uC0AC\uC6A9\uB7C9/\uC18D\uB3C4 \uC81C\uD55C: ${output.trim().slice(0, 200)}`);
82
+ }
83
+ }
84
+ function callCli(mode, imagePath) {
85
+ if (mode === "codex") {
86
+ return callCodexCli(imagePath);
87
+ }
88
+ const args = buildCliArgs(mode, imagePath);
89
+ const result = spawnSync(mode, args, {
90
+ encoding: "utf-8",
91
+ timeout: 6e5,
92
+ maxBuffer: 10 * 1024 * 1024,
93
+ shell: process.platform === "win32",
94
+ // claude: /tmp에서 실행하여 프로젝트 CLAUDE.md의 규칙 간섭 방지
95
+ ...mode === "claude" ? { cwd: tmpdir() } : {}
96
+ });
97
+ if (result.error) {
98
+ throw new Error(`${mode} CLI \uC2E4\uD589 \uC2E4\uD328: ${result.error.message}`);
99
+ }
100
+ if (result.status !== 0) {
101
+ const errMsg = result.stderr?.trim() || `exit code ${result.status}`;
102
+ throw new Error(`${mode} OCR \uC2E4\uD328: ${errMsg}`);
103
+ }
104
+ const output = result.stdout || "";
105
+ checkForLimitError(output, mode);
106
+ return output;
107
+ }
108
+ function callCodexCli(imagePath) {
109
+ const outPath = join(tmpdir(), `kordoc-codex-out-${Date.now()}.txt`);
110
+ try {
111
+ const args = ["exec", OCR_PROMPT, "--image", imagePath, "--output-last-message", outPath];
112
+ const model = process.env.KORDOC_CODEX_MODEL;
113
+ if (model) args.push("--model", model);
114
+ const result = spawnSync("codex", args, {
115
+ encoding: "utf-8",
116
+ timeout: 18e4,
117
+ maxBuffer: 10 * 1024 * 1024,
118
+ input: "",
119
+ // stdin EOF 즉시 전달 (대화형 입력 차단)
120
+ shell: process.platform === "win32"
121
+ });
122
+ if (result.error) {
123
+ throw new Error(`codex CLI \uC2E4\uD589 \uC2E4\uD328: ${result.error.message}`);
124
+ }
125
+ if (result.status !== 0) {
126
+ const errMsg = result.stderr?.trim() || `exit code ${result.status}`;
127
+ throw new Error(`codex OCR \uC2E4\uD328: ${errMsg}`);
128
+ }
129
+ let text;
130
+ try {
131
+ text = readFileSync(outPath, "utf-8");
132
+ } catch {
133
+ text = result.stdout || "";
134
+ }
135
+ checkForLimitError(text, "codex");
136
+ return text;
137
+ } finally {
138
+ try {
139
+ unlinkSync(outPath);
140
+ } catch {
141
+ }
142
+ }
143
+ }
144
+ function buildCliArgs(mode, imagePath) {
145
+ const normalizedPath = imagePath.replace(/\\/g, "/");
146
+ const promptWithImage = `${OCR_PROMPT}
147
+
148
+ \uC774\uBBF8\uC9C0: @${normalizedPath}`;
149
+ switch (mode) {
150
+ case "gemini": {
151
+ const args = ["--prompt", promptWithImage, "--yolo"];
152
+ const model = process.env.KORDOC_GEMINI_MODEL;
153
+ if (model) args.push("--model", model);
154
+ return args;
155
+ }
156
+ case "claude": {
157
+ const args = ["--print", promptWithImage];
158
+ const model = process.env.KORDOC_CLAUDE_MODEL;
159
+ if (model) args.push("--model", model);
160
+ return args;
161
+ }
162
+ default:
163
+ throw new Error(`\uC9C0\uC6D0\uD558\uC9C0 \uC54A\uB294 CLI: ${mode}`);
164
+ }
165
+ }
166
+ async function callOllamaApi(imagePath) {
167
+ const { readFileSync: readFileSync2 } = await import("fs");
168
+ const imageBase64 = readFileSync2(imagePath).toString("base64");
169
+ const model = process.env.KORDOC_OLLAMA_MODEL || "qwen3-vl:8b";
170
+ const host = process.env.KORDOC_OLLAMA_HOST || "http://localhost:11434";
171
+ const timeoutMs = Number(process.env.KORDOC_OLLAMA_TIMEOUT) || 12e4;
172
+ const response = await fetch(`${host}/api/chat`, {
173
+ method: "POST",
174
+ headers: { "Content-Type": "application/json" },
175
+ body: JSON.stringify({
176
+ model,
177
+ messages: [{
178
+ role: "user",
179
+ content: OCR_PROMPT,
180
+ images: [imageBase64]
181
+ }],
182
+ stream: false
183
+ }),
184
+ signal: AbortSignal.timeout(timeoutMs)
185
+ });
186
+ if (!response.ok) {
187
+ throw new Error(`Ollama API \uC624\uB958: ${response.status} ${response.statusText}`);
188
+ }
189
+ const data = await response.json();
190
+ return data.message?.content || "";
191
+ }
192
+ function stripCodeFence(text) {
193
+ const match = text.match(/^```(?:markdown|md)?\s*([\s\S]*?)```\s*$/m);
194
+ return match ? match[1].trim() : text;
195
+ }
196
+
197
+ export {
198
+ createCliOcrProvider
199
+ };
200
+ //# sourceMappingURL=chunk-S7BHLD2V.js.map
@@ -1 +1 @@
1
- {"version":3,"sources":["../src/ocr/cli-provider.ts"],"sourcesContent":["/**\n * CLI 기반 OCR 프로바이더\n *\n * gemini / claude / codex / ollama CLI를 subprocess로 호출하여\n * PDF 페이지 이미지를 Markdown으로 변환.\n *\n * 이미지 전달 방식:\n * - gemini: -p \"프롬프트 @이미지경로\" (@ 파일 참조)\n * - claude: -p \"프롬프트 @이미지경로\" (@ 파일 참조, --print 모드)\n * - codex: exec -i 이미지경로 \"프롬프트\" (-i/--image 플래그)\n * - ollama: REST API (localhost:11434) — CLI는 이미지 입력 미지원\n */\n\nimport { spawnSync } from \"child_process\"\nimport { writeFileSync, readFileSync, unlinkSync, mkdirSync } from \"fs\"\nimport { join } from \"path\"\nimport { tmpdir } from \"os\"\nimport type { OcrMode, StructuredOcrResult } from \"../types.js\"\n\n/** OCR 프롬프트 — 모든 CLI 공통 */\nconst OCR_PROMPT = `이 PDF 페이지 이미지에서 텍스트와 테이블을 추출하여 순수 Markdown으로 변환하세요.\n규칙:\n- 테이블은 Markdown 테이블 문법 사용 (| 구분, |---|---| 헤더 구분선 포함)\n- 병합된 셀은 해당 위치에 내용 기재\n- 헤딩은 글자 크기에 따라 ## ~ ###### 사용\n- 리스트는 - 또는 1. 사용\n- 이미지, 도형 등 비텍스트 요소는 무시\n- 원문의 읽기 순서와 구조를 유지\n- \\`\\`\\`로 감싸지 말고 순수 Markdown만 출력`\n\n/** 임시 디렉토리 (프로세스당 1회 생성)\n *\n * gemini CLI는 /tmp/ 등 시스템 임시 디렉토리를 워크스페이스 외부로 간주하여\n * @파일참조 시 접근을 거부할 수 있음. cwd 하위 폴더를 사용하면 모든 CLI에서 접근 가능.\n *\n * ⚠️ .gitignore에 포함된 경로(예: .kordoc-tmp/)는 gemini CLI가 무시하므로\n * 반드시 gitignore되지 않는 이름 사용. 파일은 try/finally로 즉시 정리.\n */\nlet _tempDir: string | null = null\nfunction getTempDir(): string {\n if (!_tempDir) {\n _tempDir = join(process.cwd(), \".kordoc_ocr_tmp\")\n mkdirSync(_tempDir, { recursive: true })\n }\n return _tempDir\n}\n\n/**\n * CLI OcrProvider 생성.\n *\n * @param mode - 사용할 CLI (gemini, claude, codex, ollama)\n * @returns OcrProvider 함수 (StructuredOcrResult 반환)\n */\nexport function createCliOcrProvider(\n mode: Exclude<OcrMode, \"auto\" | \"off\">\n): (pageImage: Uint8Array, pageNumber: number, mimeType: \"image/png\") => Promise<StructuredOcrResult> {\n return async (pageImage: Uint8Array, pageNumber: number): Promise<StructuredOcrResult> => {\n const tempPath = join(getTempDir(), `page-${pageNumber}.png`)\n\n try {\n writeFileSync(tempPath, pageImage)\n\n let output: string\n if (mode === \"ollama\") {\n output = await callOllamaApi(tempPath)\n } else {\n output = callCli(mode, tempPath)\n }\n\n return { markdown: stripCodeFence(output.trim()) }\n } finally {\n try { unlinkSync(tempPath) } catch { /* 임시 파일 정리 실패 무시 */ }\n }\n }\n}\n\n/**\n * 출력 텍스트에서 사용량·속도 제한 에러 감지.\n * 해당 메시지가 포함된 경우 throw하여 다음 엔진으로 fallback 트리거.\n */\nfunction checkForLimitError(output: string, mode: string): void {\n const lower = output.toLowerCase()\n if (lower.includes(\"usage limit\") || lower.includes(\"rate limit\")) {\n throw new Error(`${mode} 사용량/속도 제한: ${output.trim().slice(0, 200)}`)\n }\n}\n\n/**\n * CLI 실행 — gemini / claude / codex\n *\n * @throws CLI 실행 실패 또는 타임아웃(180초) 시 Error\n */\nfunction callCli(mode: string, imagePath: string): string {\n // codex는 --output-last-message로 대화 헤더 없는 깔끔한 출력 사용\n if (mode === \"codex\") {\n return callCodexCli(imagePath)\n }\n\n const args = buildCliArgs(mode, imagePath)\n\n const result = spawnSync(mode, args, {\n encoding: \"utf-8\",\n timeout: 600_000,\n maxBuffer: 10 * 1024 * 1024,\n shell: process.platform === \"win32\",\n // claude: /tmp에서 실행하여 프로젝트 CLAUDE.md의 규칙 간섭 방지\n ...(mode === \"claude\" ? { cwd: tmpdir() } : {}),\n })\n\n if (result.error) {\n throw new Error(`${mode} CLI 실행 실패: ${result.error.message}`)\n }\n if (result.status !== 0) {\n const errMsg = result.stderr?.trim() || `exit code ${result.status}`\n throw new Error(`${mode} OCR 실패: ${errMsg}`)\n }\n\n const output = result.stdout || \"\"\n checkForLimitError(output, mode)\n return output\n}\n\n/**\n * codex exec 실행 — --output-last-message로 대화 헤더 없는 깔끔한 출력.\n * 인자 순서: `codex exec <prompt> --image <file> --output-last-message <outfile>`\n */\nfunction callCodexCli(imagePath: string): string {\n // 출력 파일은 /tmp/ 사용 — codex sandbox는 cwd 내 쓰기를 막을 수 있음\n const outPath = join(tmpdir(), `kordoc-codex-out-${Date.now()}.txt`)\n try {\n const args = [\"exec\", OCR_PROMPT, \"--image\", imagePath, \"--output-last-message\", outPath]\n const model = process.env.KORDOC_CODEX_MODEL\n if (model) args.push(\"--model\", model)\n\n const result = spawnSync(\"codex\", args, {\n encoding: \"utf-8\",\n timeout: 180_000,\n maxBuffer: 10 * 1024 * 1024,\n input: \"\", // stdin EOF 즉시 전달 (대화형 입력 차단)\n shell: process.platform === \"win32\",\n })\n\n if (result.error) {\n throw new Error(`codex CLI 실행 실패: ${result.error.message}`)\n }\n if (result.status !== 0) {\n const errMsg = result.stderr?.trim() || `exit code ${result.status}`\n throw new Error(`codex OCR 실패: ${errMsg}`)\n }\n\n // --output-last-message 파일에서 읽기 (없으면 stdout 폴백)\n let text: string\n try {\n text = readFileSync(outPath, \"utf-8\")\n } catch {\n text = result.stdout || \"\"\n }\n checkForLimitError(text, \"codex\")\n return text\n } finally {\n try { unlinkSync(outPath) } catch { /* 무시 */ }\n }\n}\n\n/**\n * CLI별 인자 배열 생성.\n *\n * gemini: [\"--prompt\", \"프롬프트 @이미지경로\", \"--yolo\"]\n * - -y/--yolo: 자동 승인 (OCR은 도구 사용 없으므로 실질적 영향 없음)\n * - @ 파일 참조로 이미지를 컨텍스트에 포함\n *\n * claude: [\"--print\", \"프롬프트 @이미지경로\"]\n * - --print(-p): 비대화형 출력 모드\n * - @ 파일 참조로 이미지를 컨텍스트에 포함\n *\n * codex: callCodexCli()에서 별도 처리\n * - `codex exec <prompt> --image <file> --output-last-message <outfile>`\n * - 프롬프트가 --image보다 앞에 위치해야 함 (인자 순서 중요)\n *\n * ⚠️ CLI 버전에 따라 문법이 다를 수 있음. 업데이트 시 --help 재확인 필요.\n */\nfunction buildCliArgs(mode: string, imagePath: string): string[] {\n const normalizedPath = imagePath.replace(/\\\\/g, \"/\")\n const promptWithImage = `${OCR_PROMPT}\n\n이미지: @${normalizedPath}`\n\n switch (mode) {\n case \"gemini\": {\n const args = [\"--prompt\", promptWithImage, \"--yolo\"]\n const model = process.env.KORDOC_GEMINI_MODEL\n if (model) args.push(\"--model\", model)\n return args\n }\n\n case \"claude\": {\n const args = [\"--print\", promptWithImage]\n const model = process.env.KORDOC_CLAUDE_MODEL\n if (model) args.push(\"--model\", model)\n return args\n }\n\n default:\n throw new Error(`지원하지 않는 CLI: ${mode}`)\n }\n}\n\n/**\n * Ollama REST API 호출 — CLI는 이미지 입력을 지원하지 않으므로 API 직접 사용.\n *\n * 기본 모델: KORDOC_OLLAMA_MODEL 환경변수 또는 \"gemma4:27b\"\n * 기본 호스트: KORDOC_OLLAMA_HOST 환경변수 또는 \"http://localhost:11434\"\n *\n * @throws Ollama 서버 미실행 또는 응답 오류 시 Error\n */\nasync function callOllamaApi(imagePath: string): Promise<string> {\n const { readFileSync } = await import(\"fs\")\n const imageBase64 = readFileSync(imagePath).toString(\"base64\")\n\n const model = process.env.KORDOC_OLLAMA_MODEL || \"qwen3-vl:8b\"\n const host = process.env.KORDOC_OLLAMA_HOST || \"http://localhost:11434\"\n const timeoutMs = Number(process.env.KORDOC_OLLAMA_TIMEOUT) || 120_000\n\n const response = await fetch(`${host}/api/chat`, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({\n model,\n messages: [{\n role: \"user\",\n content: OCR_PROMPT,\n images: [imageBase64],\n }],\n stream: false,\n }),\n signal: AbortSignal.timeout(timeoutMs),\n })\n\n if (!response.ok) {\n throw new Error(`Ollama API 오류: ${response.status} ${response.statusText}`)\n }\n\n const data = await response.json() as { message?: { content?: string } }\n return data.message?.content || \"\"\n}\n\n/**\n * LLM 출력에서 코드 펜스 제거.\n * LLM이 가끔 결과를 ```markdown ... ``` 으로 감싸는 경우 처리.\n */\nfunction stripCodeFence(text: string): string {\n const match = text.match(/^```(?:markdown|md)?\\s*([\\s\\S]*?)```\\s*$/m)\n return match ? match[1].trim() : text\n}\n"],"mappings":";;;AAaA,SAAS,iBAAiB;AAC1B,SAAS,eAAe,cAAc,YAAY,iBAAiB;AACnE,SAAS,YAAY;AACrB,SAAS,cAAc;AAIvB,IAAM,aAAa;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAkBnB,IAAI,WAA0B;AAC9B,SAAS,aAAqB;AAC5B,MAAI,CAAC,UAAU;AACb,eAAW,KAAK,QAAQ,IAAI,GAAG,iBAAiB;AAChD,cAAU,UAAU,EAAE,WAAW,KAAK,CAAC;AAAA,EACzC;AACA,SAAO;AACT;AAQO,SAAS,qBACd,MACoG;AACpG,SAAO,OAAO,WAAuB,eAAqD;AACxF,UAAM,WAAW,KAAK,WAAW,GAAG,QAAQ,UAAU,MAAM;AAE5D,QAAI;AACF,oBAAc,UAAU,SAAS;AAEjC,UAAI;AACJ,UAAI,SAAS,UAAU;AACrB,iBAAS,MAAM,cAAc,QAAQ;AAAA,MACvC,OAAO;AACL,iBAAS,QAAQ,MAAM,QAAQ;AAAA,MACjC;AAEA,aAAO,EAAE,UAAU,eAAe,OAAO,KAAK,CAAC,EAAE;AAAA,IACnD,UAAE;AACA,UAAI;AAAE,mBAAW,QAAQ;AAAA,MAAE,QAAQ;AAAA,MAAuB;AAAA,IAC5D;AAAA,EACF;AACF;AAMA,SAAS,mBAAmB,QAAgB,MAAoB;AAC9D,QAAM,QAAQ,OAAO,YAAY;AACjC,MAAI,MAAM,SAAS,aAAa,KAAK,MAAM,SAAS,YAAY,GAAG;AACjE,UAAM,IAAI,MAAM,GAAG,IAAI,kDAAe,OAAO,KAAK,EAAE,MAAM,GAAG,GAAG,CAAC,EAAE;AAAA,EACrE;AACF;AAOA,SAAS,QAAQ,MAAc,WAA2B;AAExD,MAAI,SAAS,SAAS;AACpB,WAAO,aAAa,SAAS;AAAA,EAC/B;AAEA,QAAM,OAAO,aAAa,MAAM,SAAS;AAEzC,QAAM,SAAS,UAAU,MAAM,MAAM;AAAA,IACnC,UAAU;AAAA,IACV,SAAS;AAAA,IACT,WAAW,KAAK,OAAO;AAAA,IACvB,OAAO,QAAQ,aAAa;AAAA;AAAA,IAE5B,GAAI,SAAS,WAAW,EAAE,KAAK,OAAO,EAAE,IAAI,CAAC;AAAA,EAC/C,CAAC;AAED,MAAI,OAAO,OAAO;AAChB,UAAM,IAAI,MAAM,GAAG,IAAI,mCAAe,OAAO,MAAM,OAAO,EAAE;AAAA,EAC9D;AACA,MAAI,OAAO,WAAW,GAAG;AACvB,UAAM,SAAS,OAAO,QAAQ,KAAK,KAAK,aAAa,OAAO,MAAM;AAClE,UAAM,IAAI,MAAM,GAAG,IAAI,sBAAY,MAAM,EAAE;AAAA,EAC7C;AAEA,QAAM,SAAS,OAAO,UAAU;AAChC,qBAAmB,QAAQ,IAAI;AAC/B,SAAO;AACT;AAMA,SAAS,aAAa,WAA2B;AAE/C,QAAM,UAAU,KAAK,OAAO,GAAG,oBAAoB,KAAK,IAAI,CAAC,MAAM;AACnE,MAAI;AACF,UAAM,OAAO,CAAC,QAAQ,YAAY,WAAW,WAAW,yBAAyB,OAAO;AACxF,UAAM,QAAQ,QAAQ,IAAI;AAC1B,QAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AAErC,UAAM,SAAS,UAAU,SAAS,MAAM;AAAA,MACtC,UAAU;AAAA,MACV,SAAS;AAAA,MACT,WAAW,KAAK,OAAO;AAAA,MACvB,OAAO;AAAA;AAAA,MACP,OAAO,QAAQ,aAAa;AAAA,IAC9B,CAAC;AAED,QAAI,OAAO,OAAO;AAChB,YAAM,IAAI,MAAM,wCAAoB,OAAO,MAAM,OAAO,EAAE;AAAA,IAC5D;AACA,QAAI,OAAO,WAAW,GAAG;AACvB,YAAM,SAAS,OAAO,QAAQ,KAAK,KAAK,aAAa,OAAO,MAAM;AAClE,YAAM,IAAI,MAAM,2BAAiB,MAAM,EAAE;AAAA,IAC3C;AAGA,QAAI;AACJ,QAAI;AACF,aAAO,aAAa,SAAS,OAAO;AAAA,IACtC,QAAQ;AACN,aAAO,OAAO,UAAU;AAAA,IAC1B;AACA,uBAAmB,MAAM,OAAO;AAChC,WAAO;AAAA,EACT,UAAE;AACA,QAAI;AAAE,iBAAW,OAAO;AAAA,IAAE,QAAQ;AAAA,IAAW;AAAA,EAC/C;AACF;AAmBA,SAAS,aAAa,MAAc,WAA6B;AAC/D,QAAM,iBAAiB,UAAU,QAAQ,OAAO,GAAG;AACnD,QAAM,kBAAkB,GAAG,UAAU;AAAA;AAAA,uBAE/B,cAAc;AAEpB,UAAQ,MAAM;AAAA,IACZ,KAAK,UAAU;AACb,YAAM,OAAO,CAAC,YAAY,iBAAiB,QAAQ;AACnD,YAAM,QAAQ,QAAQ,IAAI;AAC1B,UAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AACrC,aAAO;AAAA,IACT;AAAA,IAEA,KAAK,UAAU;AACb,YAAM,OAAO,CAAC,WAAW,eAAe;AACxC,YAAM,QAAQ,QAAQ,IAAI;AAC1B,UAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AACrC,aAAO;AAAA,IACT;AAAA,IAEA;AACE,YAAM,IAAI,MAAM,8CAAgB,IAAI,EAAE;AAAA,EAC1C;AACF;AAUA,eAAe,cAAc,WAAoC;AAC/D,QAAM,EAAE,cAAAA,cAAa,IAAI,MAAM,OAAO,IAAI;AAC1C,QAAM,cAAcA,cAAa,SAAS,EAAE,SAAS,QAAQ;AAE7D,QAAM,QAAQ,QAAQ,IAAI,uBAAuB;AACjD,QAAM,OAAO,QAAQ,IAAI,sBAAsB;AAC/C,QAAM,YAAY,OAAO,QAAQ,IAAI,qBAAqB,KAAK;AAE/D,QAAM,WAAW,MAAM,MAAM,GAAG,IAAI,aAAa;AAAA,IAC/C,QAAQ;AAAA,IACR,SAAS,EAAE,gBAAgB,mBAAmB;AAAA,IAC9C,MAAM,KAAK,UAAU;AAAA,MACnB;AAAA,MACA,UAAU,CAAC;AAAA,QACT,MAAM;AAAA,QACN,SAAS;AAAA,QACT,QAAQ,CAAC,WAAW;AAAA,MACtB,CAAC;AAAA,MACD,QAAQ;AAAA,IACV,CAAC;AAAA,IACD,QAAQ,YAAY,QAAQ,SAAS;AAAA,EACvC,CAAC;AAED,MAAI,CAAC,SAAS,IAAI;AAChB,UAAM,IAAI,MAAM,4BAAkB,SAAS,MAAM,IAAI,SAAS,UAAU,EAAE;AAAA,EAC5E;AAEA,QAAM,OAAO,MAAM,SAAS,KAAK;AACjC,SAAO,KAAK,SAAS,WAAW;AAClC;AAMA,SAAS,eAAe,MAAsB;AAC5C,QAAM,QAAQ,KAAK,MAAM,2CAA2C;AACpE,SAAO,QAAQ,MAAM,CAAC,EAAE,KAAK,IAAI;AACnC;","names":["readFileSync"]}
1
+ {"version":3,"sources":["../src/ocr/cli-provider.ts"],"sourcesContent":["/**\n * CLI 기반 OCR 프로바이더\n *\n * gemini / claude / codex / ollama CLI를 subprocess로 호출하여\n * PDF 페이지 이미지를 Markdown으로 변환.\n *\n * 이미지 전달 방식:\n * - gemini: -p \"프롬프트 @이미지경로\" (@ 파일 참조)\n * - claude: -p \"프롬프트 @이미지경로\" (@ 파일 참조, --print 모드)\n * - codex: exec -i 이미지경로 \"프롬프트\" (-i/--image 플래그)\n * - ollama: REST API (localhost:11434) — CLI는 이미지 입력 미지원\n */\n\nimport { spawnSync } from \"child_process\"\nimport { writeFileSync, readFileSync, unlinkSync, mkdirSync } from \"fs\"\nimport { join } from \"path\"\nimport { tmpdir } from \"os\"\nimport type { OcrMode, StructuredOcrResult } from \"../types.js\"\n\n/** OCR 프롬프트 — 모든 CLI 공통 */\nconst OCR_PROMPT = `이 PDF 페이지 이미지에서 텍스트와 표를 추출하여 순수 Markdown으로 변환하고, 명백한 OCR 오인식과 표 구조 깨짐만 교정하여 최종 결과를 출력하세요.\n\n[기본 추출 규칙]\n- 본문, 표, 제목, 리스트, 캡션을 원문 구조 그대로 Markdown으로 변환\n- 헤딩은 시각적 크기·굵기에 따라 # ~ ###### 사용\n- 리스트는 -, 1. 사용 (원문 번호 체계가 ①, 가., 1) 등이면 그 표기 유지)\n- 이미지·도형·로고·페이지번호·머리글/바닥글은 무시\n- 원문의 읽기 순서(좌→우, 위→아래, 다단이면 단별로)를 유지\n\n[표 추출 규칙 — 가장 중요]\n- 표는 반드시 Markdown 파이프 테이블 문법 사용: 헤더 행 + |---| 구분선 + 데이터 행\n- 시각적으로 굵거나 음영(회색/색상) 배경이 있는 행을 헤더로 식별. 헤더가 없는 표라도 첫 행을 헤더로 두고 |---| 구분선 추가\n- 모든 행의 파이프(|) 개수를 헤더와 동일하게 맞출 것 — 빈 셀은 공백으로 채우고 절대 셀을 생략하지 말 것\n- 셀 안의 줄바꿈은 <br>로 표기 (실제 개행 문자 사용 금지 — 표가 깨짐)\n- 병합 셀 처리:\n - 가로 병합(colspan): 병합된 셀 내용을 첫 칸에 쓰고 나머지 칸은 빈 칸으로 둘 것\n - 세로 병합(rowspan): 병합된 모든 행의 해당 칸에 동일한 내용을 반복 기재할 것\n- 2단 헤더(헤더가 두 줄인 경우): 상위/하위 헤더를 \" / \"로 합쳐 단일 헤더 행으로 만들 것 (예: \"2024 / 1분기\")\n- 표 위·아래의 캡션(예: \"<표 1> ...\", \"[표 2-1]\")은 표 바로 위에 별도 줄로 보존\n- 한 페이지에서 표가 중간에 끊어져 보이더라도, 헤더가 동일하면 하나의 표로 이어붙일 것\n- 좌측 첫 컬럼이 세로쓰기 라벨(구분/분류/연도 등)이라도 일반 셀로 변환하여 누락 금지\n\n[OCR 오인식·깨짐 교정 — 허용 범위만]\n- 숫자 칸에서 'O'→'0', 'l/I'→'1' 등 맥락상 명백한 오인식만 교정\n- 단어 중간에 잘못 삽입된 공백 제거 (예: \"주거 종 합 계획\" → \"주거종합계획\")\n- 줄바꿈 오류로 끊긴 문장은 의미 단위로 병합 (단, 표 셀 내부는 <br> 유지)\n- 표에서 행 어긋남이 의심되면(컬럼 수 불일치) 빈 셀로 패딩하여 컬럼 정합성을 우선 확보\n\n[절대 금지 사항]\n- 문장·단락·항목·표 행/열을 추가하거나 삭제하지 말 것\n- 숫자, 퍼센트, 날짜, 단위, 금액을 절대 변경하지 말 것\n- 고유명사, 기관명, 법령명, 지명, 인명을 변경하지 말 것\n- 표의 헤더 텍스트, 캡션, 행/열 개수를 임의로 바꾸지 말 것\n- 체크박스(☑/☐/■/□), 특수문자(※, ①②③, ㉠, 가., 나.)를 임의로 바꾸지 말 것\n- 원문에 없는 내용을 요약·보완·추론하지 말 것\n- \\`\\`\\`, \\`\\`\\`markdown 같은 코드펜스로 감싸지 말 것\n- 설명·주석·메타 문장 추가 금지\n\n[불확실할 때]\n- 글자가 흐릿하거나 판독 불가하면 보이는 그대로 두거나, 불가피하면 한 글자만 ?로 표시\n- 표 구조가 모호하면 컬럼 수 일치 + 빈 셀 패딩 형태로 출력\n- 추측보다 원문 보존을 항상 우선`\n\n/** 임시 디렉토리 (프로세스당 1회 생성)\n *\n * gemini CLI는 /tmp/ 등 시스템 임시 디렉토리를 워크스페이스 외부로 간주하여\n * @파일참조 시 접근을 거부할 수 있음. cwd 하위 폴더를 사용하면 모든 CLI에서 접근 가능.\n *\n * ⚠️ .gitignore에 포함된 경로(예: .kordoc-tmp/)는 gemini CLI가 무시하므로\n * 반드시 gitignore되지 않는 이름 사용. 파일은 try/finally로 즉시 정리.\n */\nlet _tempDir: string | null = null\nfunction getTempDir(): string {\n if (!_tempDir) {\n _tempDir = join(process.cwd(), \".kordoc_ocr_tmp\")\n mkdirSync(_tempDir, { recursive: true })\n }\n return _tempDir\n}\n\n/**\n * CLI OcrProvider 생성.\n *\n * @param mode - 사용할 CLI (gemini, claude, codex, ollama)\n * @returns OcrProvider 함수 (StructuredOcrResult 반환)\n */\nexport function createCliOcrProvider(\n mode: Exclude<OcrMode, \"auto\" | \"off\">\n): (pageImage: Uint8Array, pageNumber: number, mimeType: \"image/png\") => Promise<StructuredOcrResult> {\n return async (pageImage: Uint8Array, pageNumber: number): Promise<StructuredOcrResult> => {\n const tempPath = join(getTempDir(), `page-${pageNumber}.png`)\n\n try {\n writeFileSync(tempPath, pageImage)\n\n let output: string\n if (mode === \"ollama\") {\n output = await callOllamaApi(tempPath)\n } else {\n output = callCli(mode, tempPath)\n }\n\n return { markdown: stripCodeFence(output.trim()) }\n } finally {\n try { unlinkSync(tempPath) } catch { /* 임시 파일 정리 실패 무시 */ }\n }\n }\n}\n\n/**\n * 출력 텍스트에서 사용량·속도 제한 에러 감지.\n * 해당 메시지가 포함된 경우 throw하여 다음 엔진으로 fallback 트리거.\n */\nfunction checkForLimitError(output: string, mode: string): void {\n const lower = output.toLowerCase()\n if (lower.includes(\"usage limit\") || lower.includes(\"rate limit\")) {\n throw new Error(`${mode} 사용량/속도 제한: ${output.trim().slice(0, 200)}`)\n }\n}\n\n/**\n * CLI 실행 — gemini / claude / codex\n *\n * @throws CLI 실행 실패 또는 타임아웃(180초) 시 Error\n */\nfunction callCli(mode: string, imagePath: string): string {\n // codex는 --output-last-message로 대화 헤더 없는 깔끔한 출력 사용\n if (mode === \"codex\") {\n return callCodexCli(imagePath)\n }\n\n const args = buildCliArgs(mode, imagePath)\n\n const result = spawnSync(mode, args, {\n encoding: \"utf-8\",\n timeout: 600_000,\n maxBuffer: 10 * 1024 * 1024,\n shell: process.platform === \"win32\",\n // claude: /tmp에서 실행하여 프로젝트 CLAUDE.md의 규칙 간섭 방지\n ...(mode === \"claude\" ? { cwd: tmpdir() } : {}),\n })\n\n if (result.error) {\n throw new Error(`${mode} CLI 실행 실패: ${result.error.message}`)\n }\n if (result.status !== 0) {\n const errMsg = result.stderr?.trim() || `exit code ${result.status}`\n throw new Error(`${mode} OCR 실패: ${errMsg}`)\n }\n\n const output = result.stdout || \"\"\n checkForLimitError(output, mode)\n return output\n}\n\n/**\n * codex exec 실행 — --output-last-message로 대화 헤더 없는 깔끔한 출력.\n * 인자 순서: `codex exec <prompt> --image <file> --output-last-message <outfile>`\n */\nfunction callCodexCli(imagePath: string): string {\n // 출력 파일은 /tmp/ 사용 — codex sandbox는 cwd 내 쓰기를 막을 수 있음\n const outPath = join(tmpdir(), `kordoc-codex-out-${Date.now()}.txt`)\n try {\n const args = [\"exec\", OCR_PROMPT, \"--image\", imagePath, \"--output-last-message\", outPath]\n const model = process.env.KORDOC_CODEX_MODEL\n if (model) args.push(\"--model\", model)\n\n const result = spawnSync(\"codex\", args, {\n encoding: \"utf-8\",\n timeout: 180_000,\n maxBuffer: 10 * 1024 * 1024,\n input: \"\", // stdin EOF 즉시 전달 (대화형 입력 차단)\n shell: process.platform === \"win32\",\n })\n\n if (result.error) {\n throw new Error(`codex CLI 실행 실패: ${result.error.message}`)\n }\n if (result.status !== 0) {\n const errMsg = result.stderr?.trim() || `exit code ${result.status}`\n throw new Error(`codex OCR 실패: ${errMsg}`)\n }\n\n // --output-last-message 파일에서 읽기 (없으면 stdout 폴백)\n let text: string\n try {\n text = readFileSync(outPath, \"utf-8\")\n } catch {\n text = result.stdout || \"\"\n }\n checkForLimitError(text, \"codex\")\n return text\n } finally {\n try { unlinkSync(outPath) } catch { /* 무시 */ }\n }\n}\n\n/**\n * CLI별 인자 배열 생성.\n *\n * gemini: [\"--prompt\", \"프롬프트 @이미지경로\", \"--yolo\"]\n * - -y/--yolo: 자동 승인 (OCR은 도구 사용 없으므로 실질적 영향 없음)\n * - @ 파일 참조로 이미지를 컨텍스트에 포함\n *\n * claude: [\"--print\", \"프롬프트 @이미지경로\"]\n * - --print(-p): 비대화형 출력 모드\n * - @ 파일 참조로 이미지를 컨텍스트에 포함\n *\n * codex: callCodexCli()에서 별도 처리\n * - `codex exec <prompt> --image <file> --output-last-message <outfile>`\n * - 프롬프트가 --image보다 앞에 위치해야 함 (인자 순서 중요)\n *\n * ⚠️ CLI 버전에 따라 문법이 다를 수 있음. 업데이트 시 --help 재확인 필요.\n */\nfunction buildCliArgs(mode: string, imagePath: string): string[] {\n const normalizedPath = imagePath.replace(/\\\\/g, \"/\")\n const promptWithImage = `${OCR_PROMPT}\n\n이미지: @${normalizedPath}`\n\n switch (mode) {\n case \"gemini\": {\n const args = [\"--prompt\", promptWithImage, \"--yolo\"]\n const model = process.env.KORDOC_GEMINI_MODEL\n if (model) args.push(\"--model\", model)\n return args\n }\n\n case \"claude\": {\n const args = [\"--print\", promptWithImage]\n const model = process.env.KORDOC_CLAUDE_MODEL\n if (model) args.push(\"--model\", model)\n return args\n }\n\n default:\n throw new Error(`지원하지 않는 CLI: ${mode}`)\n }\n}\n\n/**\n * Ollama REST API 호출 — CLI는 이미지 입력을 지원하지 않으므로 API 직접 사용.\n *\n * 기본 모델: KORDOC_OLLAMA_MODEL 환경변수 또는 \"gemma4:27b\"\n * 기본 호스트: KORDOC_OLLAMA_HOST 환경변수 또는 \"http://localhost:11434\"\n *\n * @throws Ollama 서버 미실행 또는 응답 오류 시 Error\n */\nasync function callOllamaApi(imagePath: string): Promise<string> {\n const { readFileSync } = await import(\"fs\")\n const imageBase64 = readFileSync(imagePath).toString(\"base64\")\n\n const model = process.env.KORDOC_OLLAMA_MODEL || \"qwen3-vl:8b\"\n const host = process.env.KORDOC_OLLAMA_HOST || \"http://localhost:11434\"\n const timeoutMs = Number(process.env.KORDOC_OLLAMA_TIMEOUT) || 120_000\n\n const response = await fetch(`${host}/api/chat`, {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/json\" },\n body: JSON.stringify({\n model,\n messages: [{\n role: \"user\",\n content: OCR_PROMPT,\n images: [imageBase64],\n }],\n stream: false,\n }),\n signal: AbortSignal.timeout(timeoutMs),\n })\n\n if (!response.ok) {\n throw new Error(`Ollama API 오류: ${response.status} ${response.statusText}`)\n }\n\n const data = await response.json() as { message?: { content?: string } }\n return data.message?.content || \"\"\n}\n\n/**\n * LLM 출력에서 코드 펜스 제거.\n * LLM이 가끔 결과를 ```markdown ... ``` 으로 감싸는 경우 처리.\n */\nfunction stripCodeFence(text: string): string {\n const match = text.match(/^```(?:markdown|md)?\\s*([\\s\\S]*?)```\\s*$/m)\n return match ? match[1].trim() : text\n}\n"],"mappings":";;;AAaA,SAAS,iBAAiB;AAC1B,SAAS,eAAe,cAAc,YAAY,iBAAiB;AACnE,SAAS,YAAY;AACrB,SAAS,cAAc;AAIvB,IAAM,aAAa;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAmDnB,IAAI,WAA0B;AAC9B,SAAS,aAAqB;AAC5B,MAAI,CAAC,UAAU;AACb,eAAW,KAAK,QAAQ,IAAI,GAAG,iBAAiB;AAChD,cAAU,UAAU,EAAE,WAAW,KAAK,CAAC;AAAA,EACzC;AACA,SAAO;AACT;AAQO,SAAS,qBACd,MACoG;AACpG,SAAO,OAAO,WAAuB,eAAqD;AACxF,UAAM,WAAW,KAAK,WAAW,GAAG,QAAQ,UAAU,MAAM;AAE5D,QAAI;AACF,oBAAc,UAAU,SAAS;AAEjC,UAAI;AACJ,UAAI,SAAS,UAAU;AACrB,iBAAS,MAAM,cAAc,QAAQ;AAAA,MACvC,OAAO;AACL,iBAAS,QAAQ,MAAM,QAAQ;AAAA,MACjC;AAEA,aAAO,EAAE,UAAU,eAAe,OAAO,KAAK,CAAC,EAAE;AAAA,IACnD,UAAE;AACA,UAAI;AAAE,mBAAW,QAAQ;AAAA,MAAE,QAAQ;AAAA,MAAuB;AAAA,IAC5D;AAAA,EACF;AACF;AAMA,SAAS,mBAAmB,QAAgB,MAAoB;AAC9D,QAAM,QAAQ,OAAO,YAAY;AACjC,MAAI,MAAM,SAAS,aAAa,KAAK,MAAM,SAAS,YAAY,GAAG;AACjE,UAAM,IAAI,MAAM,GAAG,IAAI,kDAAe,OAAO,KAAK,EAAE,MAAM,GAAG,GAAG,CAAC,EAAE;AAAA,EACrE;AACF;AAOA,SAAS,QAAQ,MAAc,WAA2B;AAExD,MAAI,SAAS,SAAS;AACpB,WAAO,aAAa,SAAS;AAAA,EAC/B;AAEA,QAAM,OAAO,aAAa,MAAM,SAAS;AAEzC,QAAM,SAAS,UAAU,MAAM,MAAM;AAAA,IACnC,UAAU;AAAA,IACV,SAAS;AAAA,IACT,WAAW,KAAK,OAAO;AAAA,IACvB,OAAO,QAAQ,aAAa;AAAA;AAAA,IAE5B,GAAI,SAAS,WAAW,EAAE,KAAK,OAAO,EAAE,IAAI,CAAC;AAAA,EAC/C,CAAC;AAED,MAAI,OAAO,OAAO;AAChB,UAAM,IAAI,MAAM,GAAG,IAAI,mCAAe,OAAO,MAAM,OAAO,EAAE;AAAA,EAC9D;AACA,MAAI,OAAO,WAAW,GAAG;AACvB,UAAM,SAAS,OAAO,QAAQ,KAAK,KAAK,aAAa,OAAO,MAAM;AAClE,UAAM,IAAI,MAAM,GAAG,IAAI,sBAAY,MAAM,EAAE;AAAA,EAC7C;AAEA,QAAM,SAAS,OAAO,UAAU;AAChC,qBAAmB,QAAQ,IAAI;AAC/B,SAAO;AACT;AAMA,SAAS,aAAa,WAA2B;AAE/C,QAAM,UAAU,KAAK,OAAO,GAAG,oBAAoB,KAAK,IAAI,CAAC,MAAM;AACnE,MAAI;AACF,UAAM,OAAO,CAAC,QAAQ,YAAY,WAAW,WAAW,yBAAyB,OAAO;AACxF,UAAM,QAAQ,QAAQ,IAAI;AAC1B,QAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AAErC,UAAM,SAAS,UAAU,SAAS,MAAM;AAAA,MACtC,UAAU;AAAA,MACV,SAAS;AAAA,MACT,WAAW,KAAK,OAAO;AAAA,MACvB,OAAO;AAAA;AAAA,MACP,OAAO,QAAQ,aAAa;AAAA,IAC9B,CAAC;AAED,QAAI,OAAO,OAAO;AAChB,YAAM,IAAI,MAAM,wCAAoB,OAAO,MAAM,OAAO,EAAE;AAAA,IAC5D;AACA,QAAI,OAAO,WAAW,GAAG;AACvB,YAAM,SAAS,OAAO,QAAQ,KAAK,KAAK,aAAa,OAAO,MAAM;AAClE,YAAM,IAAI,MAAM,2BAAiB,MAAM,EAAE;AAAA,IAC3C;AAGA,QAAI;AACJ,QAAI;AACF,aAAO,aAAa,SAAS,OAAO;AAAA,IACtC,QAAQ;AACN,aAAO,OAAO,UAAU;AAAA,IAC1B;AACA,uBAAmB,MAAM,OAAO;AAChC,WAAO;AAAA,EACT,UAAE;AACA,QAAI;AAAE,iBAAW,OAAO;AAAA,IAAE,QAAQ;AAAA,IAAW;AAAA,EAC/C;AACF;AAmBA,SAAS,aAAa,MAAc,WAA6B;AAC/D,QAAM,iBAAiB,UAAU,QAAQ,OAAO,GAAG;AACnD,QAAM,kBAAkB,GAAG,UAAU;AAAA;AAAA,uBAE/B,cAAc;AAEpB,UAAQ,MAAM;AAAA,IACZ,KAAK,UAAU;AACb,YAAM,OAAO,CAAC,YAAY,iBAAiB,QAAQ;AACnD,YAAM,QAAQ,QAAQ,IAAI;AAC1B,UAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AACrC,aAAO;AAAA,IACT;AAAA,IAEA,KAAK,UAAU;AACb,YAAM,OAAO,CAAC,WAAW,eAAe;AACxC,YAAM,QAAQ,QAAQ,IAAI;AAC1B,UAAI,MAAO,MAAK,KAAK,WAAW,KAAK;AACrC,aAAO;AAAA,IACT;AAAA,IAEA;AACE,YAAM,IAAI,MAAM,8CAAgB,IAAI,EAAE;AAAA,EAC1C;AACF;AAUA,eAAe,cAAc,WAAoC;AAC/D,QAAM,EAAE,cAAAA,cAAa,IAAI,MAAM,OAAO,IAAI;AAC1C,QAAM,cAAcA,cAAa,SAAS,EAAE,SAAS,QAAQ;AAE7D,QAAM,QAAQ,QAAQ,IAAI,uBAAuB;AACjD,QAAM,OAAO,QAAQ,IAAI,sBAAsB;AAC/C,QAAM,YAAY,OAAO,QAAQ,IAAI,qBAAqB,KAAK;AAE/D,QAAM,WAAW,MAAM,MAAM,GAAG,IAAI,aAAa;AAAA,IAC/C,QAAQ;AAAA,IACR,SAAS,EAAE,gBAAgB,mBAAmB;AAAA,IAC9C,MAAM,KAAK,UAAU;AAAA,MACnB;AAAA,MACA,UAAU,CAAC;AAAA,QACT,MAAM;AAAA,QACN,SAAS;AAAA,QACT,QAAQ,CAAC,WAAW;AAAA,MACtB,CAAC;AAAA,MACD,QAAQ;AAAA,IACV,CAAC;AAAA,IACD,QAAQ,YAAY,QAAQ,SAAS;AAAA,EACvC,CAAC;AAED,MAAI,CAAC,SAAS,IAAI;AAChB,UAAM,IAAI,MAAM,4BAAkB,SAAS,MAAM,IAAI,SAAS,UAAU,EAAE;AAAA,EAC5E;AAEA,QAAM,OAAO,MAAM,SAAS,KAAK;AACjC,SAAO,KAAK,SAAS,WAAW;AAClC;AAMA,SAAS,eAAe,MAAsB;AAC5C,QAAM,QAAQ,KAAK,MAAM,2CAA2C;AACpE,SAAO,QAAQ,MAAM,CAAC,EAAE,KAAK,IAAI;AACnC;","names":["readFileSync"]}
package/dist/cli.js CHANGED
@@ -4,13 +4,13 @@ import {
4
4
  markdownToHwpx,
5
5
  markdownToXlsx,
6
6
  parse
7
- } from "./chunk-QG6BYZMR.js";
7
+ } from "./chunk-25ZYYLVP.js";
8
8
  import {
9
9
  VERSION,
10
10
  toArrayBuffer
11
- } from "./chunk-IJGNPAK2.js";
11
+ } from "./chunk-5CILZHRW.js";
12
12
  import "./chunk-MOL7MDBG.js";
13
- import "./chunk-Y4WFKJ5P.js";
13
+ import "./chunk-S7BHLD2V.js";
14
14
  import "./chunk-YW5G6BCJ.js";
15
15
  import {
16
16
  createLoggerFromEnv,
@@ -173,7 +173,7 @@ async function runParse(files, opts) {
173
173
  saveImages(absPath);
174
174
  }
175
175
  } catch (err) {
176
- const { sanitizeError } = await import("./utils-RBXHHCLI.js");
176
+ const { sanitizeError } = await import("./utils-H2BL5GNR.js");
177
177
  process.stderr.write(`
178
178
  [kordoc] ERROR: ${fileName} \u2014 ${sanitizeError(err)}
179
179
  `);
@@ -255,7 +255,7 @@ program.command("convert <input>").description("\uB9C8\uD06C\uB2E4\uC6B4 \uD30C\
255
255
  `));
256
256
  }
257
257
  } catch (err) {
258
- const { sanitizeError } = await import("./utils-RBXHHCLI.js");
258
+ const { sanitizeError } = await import("./utils-H2BL5GNR.js");
259
259
  process.stderr.write(` FAIL
260
260
  `);
261
261
  process.stderr.write(` \u2192 ${sanitizeError(err)}
@@ -287,7 +287,7 @@ program.command("init-env").description("kordoc\uC6A9 .env \uD15C\uD50C\uB9BF \u
287
287
  }
288
288
  });
289
289
  program.command("watch <dir>").description("\uB514\uB809\uD1A0\uB9AC \uAC10\uC2DC \u2014 \uC0C8 \uBB38\uC11C \uC790\uB3D9 \uBCC0\uD658").option("--webhook <url>", "\uACB0\uACFC \uC804\uC1A1 \uC6F9\uD6C5 URL").option("-d, --out-dir <dir>", "\uBCC0\uD658 \uACB0\uACFC \uCD9C\uB825 \uB514\uB809\uD1A0\uB9AC").option("-p, --pages <range>", "\uD398\uC774\uC9C0/\uC139\uC158 \uBC94\uC704").option("--format <type>", "\uCD9C\uB825 \uD615\uC2DD: markdown \uB610\uB294 json", "markdown").option("--silent", "\uC9C4\uD589 \uBA54\uC2DC\uC9C0 \uC228\uAE30\uAE30").action(async (dir, opts) => {
290
- const { watchDirectory } = await import("./watch-5CCMTZ7F.js");
290
+ const { watchDirectory } = await import("./watch-D6ODQLPJ.js");
291
291
  await watchDirectory({
292
292
  dir,
293
293
  outDir: opts.outDir,