@clazic/kordoc 2.2.9 → 2.3.0

package/dist/index.d.cts

@@ -119,6 +119,8 @@ interface ParseOptions {
     ocrConcurrency?: number;
     /** 진행률 콜백 — current: 현재 페이지/섹션, total: 전체 수 */
     onProgress?: (current: number, total: number) => void;
+    /** OCR 배치 크기 — CLI 1회 호출당 처리할 페이지 수. gemini/claude/codex 전용. */
+    ocrBatchSize?: number;
     /** PDF 머리글/바닥글 자동 제거 */
     removeHeaderFooter?: boolean;
 }

package/dist/index.d.ts

@@ -119,6 +119,8 @@ interface ParseOptions {
     ocrConcurrency?: number;
     /** 진행률 콜백 — current: 현재 페이지/섹션, total: 전체 수 */
     onProgress?: (current: number, total: number) => void;
+    /** OCR 배치 크기 — CLI 1회 호출당 처리할 페이지 수. gemini/claude/codex 전용. */
+    ocrBatchSize?: number;
     /** PDF 머리글/바닥글 자동 제거 */
     removeHeaderFooter?: boolean;
 }

package/dist/index.js

@@ -2019,7 +2019,7 @@ import { join } from "path";
 import { tmpdir } from "os";
 function getTempDir() {
   if (!_tempDir) {
-    _tempDir = join(process.cwd(), ".kordoc-tmp");
+    _tempDir = join(process.cwd(), "_kordoc_ocr_tmp");
     mkdirSync(_tempDir, { recursive: true });
   }
   return _tempDir;
@@ -2119,8 +2119,8 @@ function buildCliArgs(mode, imagePath) {
   }
 }
 async function callOllamaApi(imagePath) {
-  const { readFileSync: readFileSync2 } = await import("fs");
-  const imageBase64 = readFileSync2(imagePath).toString("base64");
+  const { readFileSync: readFileSync3 } = await import("fs");
+  const imageBase64 = readFileSync3(imagePath).toString("base64");
   const model = process.env.KORDOC_OLLAMA_MODEL || "qwen3-vl:8b";
   const host = process.env.KORDOC_OLLAMA_HOST || "http://localhost:11434";
   const timeoutMs = Number(process.env.KORDOC_OLLAMA_TIMEOUT) || 12e4;
@@ -2216,12 +2216,155 @@ var init_tesseract_provider = __esm({
   }
 });
 
+// src/ocr/batch-provider.ts
+var batch_provider_exports = {};
+__export(batch_provider_exports, {
+  DEFAULT_BATCH_SIZES: () => DEFAULT_BATCH_SIZES,
+  createBatchCliProvider: () => createBatchCliProvider
+});
+import { spawnSync as spawnSync2 } from "child_process";
+import { writeFileSync as writeFileSync2, readFileSync as readFileSync2, unlinkSync as unlinkSync2, mkdirSync as mkdirSync2 } from "fs";
+import { join as join2 } from "path";
+import { tmpdir as tmpdir2 } from "os";
+function getBatchTempDir() {
+  if (!_batchTempDir) {
+    _batchTempDir = join2(process.cwd(), "_kordoc_ocr_tmp");
+    mkdirSync2(_batchTempDir, { recursive: true });
+  }
+  return _batchTempDir;
+}
+function createBatchCliProvider(mode, batchSize) {
+  return {
+    __batch: true,
+    batchSize,
+    async processBatch(pages) {
+      const results = /* @__PURE__ */ new Map();
+      const tempDir = getBatchTempDir();
+      const tempFiles = [];
+      try {
+        for (const { image, pageNum } of pages) {
+          const path = join2(tempDir, `batch-p${pageNum}.png`);
+          writeFileSync2(path, image);
+          tempFiles.push(path);
+        }
+        let output;
+        if (mode === "codex") {
+          output = callBatchCodexCli(tempFiles);
+        } else {
+          output = callBatchCli(mode, tempFiles);
+        }
+        const cleaned = stripCodeFence2(output.trim());
+        const parts = cleaned.split(/<!--\s*PAGE_BREAK\s*-->/).map((p) => p.trim()).filter((p) => p.length > 0);
+        for (let i = 0; i < pages.length; i++) {
+          const pageNum = pages[i].pageNum;
+          if (i < parts.length) {
+            results.set(pageNum, { markdown: parts[i] });
+          }
+        }
+      } finally {
+        for (const f of tempFiles) {
+          try {
+            unlinkSync2(f);
+          } catch {
+          }
+        }
+      }
+      return results;
+    }
+  };
+}
+function callBatchCli(mode, imagePaths) {
+  const fileRefs = imagePaths.map((p) => `@${p}`).join("\n");
+  const prompt = `${BATCH_OCR_PROMPT}
+
+${fileRefs}`;
+  let args;
+  if (mode === "gemini") {
+    args = ["--prompt", prompt, "--yolo"];
+    const model = process.env.KORDOC_GEMINI_MODEL;
+    if (model) args.push("--model", model);
+  } else {
+    args = ["--print", prompt];
+    const model = process.env.KORDOC_CLAUDE_MODEL;
+    if (model) args.push("--model", model);
+  }
+  const timeoutMs = 6e4 + imagePaths.length * 2e4;
+  const result = spawnSync2(mode, args, {
+    encoding: "utf-8",
+    timeout: timeoutMs,
+    maxBuffer: 50 * 1024 * 1024,
+    // 50MB (large batch output)
+    ...mode === "claude" ? { cwd: tmpdir2() } : {}
+  });
+  if (result.error) {
+    throw new Error(`${mode} \uBC30\uCE58 OCR \uC2E4\uD328: ${result.error.message}`);
+  }
+  if (result.status !== 0) {
+    const errMsg = result.stderr?.trim() || `exit code ${result.status}`;
+    throw new Error(`${mode} \uBC30\uCE58 OCR \uC2E4\uD328: ${errMsg}`);
+  }
+  return result.stdout || "";
+}
+function callBatchCodexCli(imagePaths) {
+  const outPath = join2(tmpdir2(), `kordoc-codex-batch-${Date.now()}.txt`);
+  try {
+    const args = ["exec", BATCH_OCR_PROMPT];
+    for (const p of imagePaths) {
+      args.push("--image", p);
+    }
+    args.push("--output-last-message", outPath);
+    const model = process.env.KORDOC_CODEX_MODEL;
+    if (model) args.push("--model", model);
+    const timeoutMs = 6e4 + imagePaths.length * 2e4;
+    const result = spawnSync2("codex", args, {
+      encoding: "utf-8",
+      timeout: timeoutMs,
+      maxBuffer: 50 * 1024 * 1024,
+      input: ""
+    });
+    if (result.error) {
+      throw new Error(`codex \uBC30\uCE58 OCR \uC2E4\uD328: ${result.error.message}`);
+    }
+    if (result.status !== 0) {
+      const errMsg = result.stderr?.trim() || `exit code ${result.status}`;
+      throw new Error(`codex \uBC30\uCE58 OCR \uC2E4\uD328: ${errMsg}`);
+    }
+    try {
+      return readFileSync2(outPath, "utf-8");
+    } catch {
+      return result.stdout || "";
+    }
+  } finally {
+    try {
+      unlinkSync2(outPath);
+    } catch {
+    }
+  }
+}
+function stripCodeFence2(text) {
+  const match = text.match(/^```(?:markdown|md)?\s*\n([\s\S]*?)\n```\s*$/m);
+  return match ? match[1].trim() : text;
+}
+var BATCH_OCR_PROMPT, DEFAULT_BATCH_SIZES, _batchTempDir;
+var init_batch_provider = __esm({
+  "src/ocr/batch-provider.ts"() {
+    "use strict";
+    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\uC138\uC694.\n\n\uADDC\uCE59:\n- \uAC01 \uD398\uC774\uC9C0 \uACB0\uACFC \uC0AC\uC774\uC5D0 \uBC18\uB4DC\uC2DC \uC774 \uAD6C\uBD84\uC790\uB97C \uC0BD\uC785: <!-- PAGE_BREAK -->\n- \uD14C\uC774\uBE14\uC740 Markdown \uD14C\uC774\uBE14 \uBB38\uBC95 \uC0AC\uC6A9 (| \uAD6C\uBD84, |---|---| \uD5E4\uB354 \uAD6C\uBD84\uC120 \uD3EC\uD568)\n- \uBCD1\uD569\uB41C \uC140\uC740 \uD574\uB2F9 \uC704\uCE58\uC5D0 \uB0B4\uC6A9 \uAE30\uC7AC\n- \uD5E4\uB529\uC740 \uAE00\uC790 \uD06C\uAE30\uC5D0 \uB530\uB77C ## ~ ###### \uC0AC\uC6A9\n- \uB9AC\uC2A4\uD2B8\uB294 - \uB610\uB294 1. \uC0AC\uC6A9\n- \uC774\uBBF8\uC9C0, \uB3C4\uD615 \uB4F1 \uBE44\uD14D\uC2A4\uD2B8 \uC694\uC18C\uB294 \uBB34\uC2DC\n- \uC6D0\uBB38\uC758 \uC77D\uAE30 \uC21C\uC11C\uC640 \uAD6C\uC870\uB97C \uC720\uC9C0\n- ```\uB85C \uAC10\uC2F8\uC9C0 \uB9D0\uACE0 \uC21C\uC218 Markdown\uB9CC \uCD9C\uB825";
+    DEFAULT_BATCH_SIZES = {
+      gemini: 5,
+      claude: 5,
+      codex: 10
+    };
+    _batchTempDir = null;
+  }
+});
+
 // src/ocr/resolve.ts
 var resolve_exports = {};
 __export(resolve_exports, {
   resolveOcrProvider: () => resolveOcrProvider
 });
-async function resolveOcrProvider(mode, warnings, concurrency) {
+async function resolveOcrProvider(mode, warnings, concurrency, batchSize) {
   if (mode === "off") {
     throw new Error("OCR\uC774 \uBE44\uD65C\uC131\uD654\uB418\uC5B4 \uC788\uC2B5\uB2C8\uB2E4 (--ocr off).");
   }
@@ -2234,6 +2377,14 @@ async function resolveOcrProvider(mode, warnings, concurrency) {
       }
       return createTesseractProvider2();
     }
+    if (mode === "gemini" || mode === "claude" || mode === "codex") {
+      const { createBatchCliProvider: createBatchCliProvider2, DEFAULT_BATCH_SIZES: DEFAULT_BATCH_SIZES2 } = await Promise.resolve().then(() => (init_batch_provider(), batch_provider_exports));
+      const effectiveBatch = batchSize ?? DEFAULT_BATCH_SIZES2[mode];
+      if (effectiveBatch > 1) {
+        return createBatchCliProvider2(mode, effectiveBatch);
+      }
+      return createCliOcrProvider(mode);
+    }
     return createCliOcrProvider(mode);
   }
   const detected = detectAvailableOcr();
@@ -2257,6 +2408,14 @@ async function resolveOcrProvider(mode, warnings, concurrency) {
     }
     return createTesseractProvider2();
   }
+  if (detected === "gemini" || detected === "codex" || detected === "claude") {
+    const { createBatchCliProvider: createBatchCliProvider2, DEFAULT_BATCH_SIZES: DEFAULT_BATCH_SIZES2 } = await Promise.resolve().then(() => (init_batch_provider(), batch_provider_exports));
+    const effectiveBatch = batchSize ?? DEFAULT_BATCH_SIZES2[detected];
+    if (effectiveBatch > 1) {
+      return createBatchCliProvider2(detected, effectiveBatch);
+    }
+    return createCliOcrProvider(detected);
+  }
   return createCliOcrProvider(detected);
 }
 var init_resolve = __esm({
@@ -2419,8 +2578,14 @@ function ocrResultToBlocks(result, pageNum) {
   }
   return pageBlocks;
 }
-async function ocrPages(doc, provider, pageFilter, effectivePageCount, warnings, concurrency = 1) {
+function isBatchProvider(p) {
+  return !!p && typeof p === "object" && "__batch" in p && p.__batch === true;
+}
+async function ocrPages(doc, provider, pageFilter, effectivePageCount, warnings, concurrency = 1, onProgress) {
   const blocks = [];
+  if (isBatchProvider(provider)) {
+    return ocrPagesBatch(doc, provider, pageFilter, effectivePageCount, warnings, onProgress);
+  }
   if (concurrency <= 1) {
     for (let i = 1; i <= effectivePageCount; i++) {
       if (pageFilter && !pageFilter.has(i)) continue;
@@ -2466,6 +2631,47 @@ async function ocrPages(doc, provider, pageFilter, effectivePageCount, warnings,
   }
   return blocks;
 }
+async function ocrPagesBatch(doc, provider, pageFilter, effectivePageCount, warnings, onProgress) {
+  const blocks = [];
+  const pageNumbers = [];
+  for (let i = 1; i <= effectivePageCount; i++) {
+    if (pageFilter && !pageFilter.has(i)) continue;
+    pageNumbers.push(i);
+  }
+  const pageImages = [];
+  for (const pageNum of pageNumbers) {
+    const page = await doc.getPage(pageNum);
+    const image = await renderPageToPng(page);
+    pageImages.push({ image, pageNum });
+  }
+  const batches = [];
+  for (let i = 0; i < pageImages.length; i += provider.batchSize) {
+    batches.push(pageImages.slice(i, i + provider.batchSize));
+  }
+  let processed = 0;
+  for (const batch of batches) {
+    try {
+      const results = await provider.processBatch(batch);
+      for (const { pageNum } of batch) {
+        const result = results.get(pageNum);
+        if (result) {
+          for (const b of ocrResultToBlocks(result, pageNum)) blocks.push(b);
+        }
+        processed++;
+        onProgress?.(processed, pageNumbers.length);
+      }
+    } catch (err) {
+      const range = `${batch[0].pageNum}-${batch[batch.length - 1].pageNum}`;
+      warnings?.push({
+        message: `\uBC30\uCE58 OCR \uC2E4\uD328 (\uD398\uC774\uC9C0 ${range}): ${err instanceof Error ? err.message : "\uC54C \uC218 \uC5C6\uB294 \uC624\uB958"}`,
+        code: "OCR_PAGE_FAILED"
+      });
+      processed += batch.length;
+      onProgress?.(processed, pageNumbers.length);
+    }
+  }
+  return blocks;
+}
 async function renderPageToPng(page) {
   const { createCanvas } = await import("@napi-rs/canvas");
   const scale = 2;
@@ -2531,7 +2737,7 @@ import JSZip2 from "jszip";
 import { DOMParser } from "@xmldom/xmldom";
 
 // src/utils.ts
-var VERSION = true ? "2.2.9" : "0.0.0-dev";
+var VERSION = true ? "2.3.0" : "0.0.0-dev";
 function toArrayBuffer(buf) {
   if (buf.byteOffset === 0 && buf.byteLength === buf.buffer.byteLength) {
     return buf.buffer;
@@ -6049,7 +6255,8 @@ async function parsePdfDocument(buffer, options) {
         try {
           const { resolveOcrProvider: resolveOcrProvider2 } = await Promise.resolve().then(() => (init_resolve(), resolve_exports));
           const concurrency = options?.ocrConcurrency ?? 1;
-          ocrProvider = await resolveOcrProvider2(ocrMode, warnings, concurrency);
+          const batchSize = options?.ocrBatchSize;
+          ocrProvider = await resolveOcrProvider2(ocrMode, warnings, concurrency, batchSize);
         } catch (resolveErr) {
           if (ocrMode !== "auto") {
             throw Object.assign(
@@ -6064,7 +6271,7 @@ async function parsePdfDocument(buffer, options) {
         try {
           const { ocrPages: ocrPages2 } = await Promise.resolve().then(() => (init_provider(), provider_exports));
           const concurrency = options?.ocrConcurrency ?? 1;
-          ocrBlocks = await ocrPages2(doc, ocrProvider, pageFilter, effectivePageCount, warnings, concurrency);
+          ocrBlocks = await ocrPages2(doc, ocrProvider, pageFilter, effectivePageCount, warnings, concurrency, options?.onProgress);
         } catch {
         } finally {
           const terminable = ocrProvider;