@heripo/pdf-parser 0.1.13 → 0.1.15

package/dist/index.cjs

@@ -31,6 +31,7 @@ var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: tru
 var src_exports = {};
 __export(src_exports, {
   ImagePdfFallbackError: () => ImagePdfFallbackError,
+  InvalidDocumentTypeError: () => InvalidDocumentTypeError,
   PDFParser: () => PDFParser,
   VlmResponseValidator: () => VlmResponseValidator
 });
@@ -1508,6 +1509,32 @@ var PdfTextExtractor = class {
     }
     return result.stdout;
   }
+  /**
+   * Extract text from a range of PDF pages using a single pdftotext invocation.
+   * Returns empty string on failure (logged as warning).
+   *
+   * @param pdfPath - Absolute path to the source PDF file
+   * @param firstPage - First page number (1-based)
+   * @param lastPage - Last page number (1-based, inclusive)
+   */
+  async extractPageRange(pdfPath, firstPage, lastPage) {
+    const result = await spawnAsync("pdftotext", [
+      "-f",
+      firstPage.toString(),
+      "-l",
+      lastPage.toString(),
+      "-layout",
+      pdfPath,
+      "-"
+    ]);
+    if (result.code !== 0) {
+      this.logger.warn(
+        `[PdfTextExtractor] pdftotext failed for pages ${firstPage}-${lastPage}: ${result.stderr || "Unknown error"}`
+      );
+      return "";
+    }
+    return result.stdout;
+  }
   /**
    * Extract text from a single PDF page using pdftotext.
    * Returns empty string on failure (logged as warning).
@@ -2070,8 +2097,9 @@ Note: Hanja are Chinese characters used in Korean documents, different from mode
 
 Answer whether any Hanja characters are present on this page.
 
-Also identify all languages present on this page. Return an array of BCP 47 language tags ordered by prevalence (primary language first).
-Examples: ["ko-KR", "en-US"], ["ja-JP"], ["zh-TW", "en-US"]`;
+Also identify all languages present on this page. Return an array of ocrmac-compatible language tags ordered by prevalence (primary language first).
+Supported tags: ar-SA, ars-SA, cs-CZ, da-DK, de-DE, en-US, es-ES, fr-FR, id-ID, it-IT, ja-JP, ko-KR, ms-MY, nb-NO, nl-NL, nn-NO, no-NO, pl-PL, pt-BR, ro-RO, ru-RU, sv-SE, th-TH, tr-TR, uk-UA, vi-VT, yue-Hans, yue-Hant, zh-Hans, zh-Hant.
+Examples: ["ko-KR", "en-US"], ["ja-JP"], ["zh-Hant", "en-US"]`;
 var OcrStrategySampler = class {
   logger;
   pageRenderer;
@@ -2369,6 +2397,119 @@ var LocalFileServer = class {
   }
 };
 
+// src/utils/task-failure-details.ts
+var MAX_RESULT_RETRIES = 3;
+var RESULT_RETRY_DELAY_MS = 2e3;
+async function getTaskFailureDetails(task, logger, logPrefix) {
+  for (let attempt = 0; attempt < MAX_RESULT_RETRIES; attempt++) {
+    try {
+      if (attempt > 0) {
+        await new Promise((r) => setTimeout(r, RESULT_RETRY_DELAY_MS));
+      }
+      const result = await task.getResult();
+      if (result.errors?.length) {
+        return result.errors.map((e) => e.message).join("; ");
+      }
+      return `status: ${result.status ?? "unknown"}`;
+    } catch (err) {
+      if (attempt === MAX_RESULT_RETRIES - 1) {
+        logger.error(
+          `${logPrefix} Failed to retrieve task result after ${MAX_RESULT_RETRIES} attempts:`,
+          err
+        );
+        return "unable to retrieve error details";
+      }
+      logger.warn(
+        `${logPrefix} Result not available yet, retrying (${attempt + 1}/${MAX_RESULT_RETRIES})...`
+      );
+    }
+  }
+  return "unable to retrieve error details";
+}
+
+// src/validators/document-type-validator.ts
+var import_zod = require("zod");
+
+// src/errors/invalid-document-type-error.ts
+var InvalidDocumentTypeError = class extends Error {
+  constructor(reason) {
+    super(
+      `The uploaded PDF does not appear to be a Korean archaeological investigation report. Reason: ${reason}`
+    );
+    this.reason = reason;
+  }
+  name = "InvalidDocumentTypeError";
+  code = "INVALID_DOCUMENT_TYPE";
+};
+
+// src/validators/document-type-validator.ts
+var SYSTEM_PROMPT = `You are given text extracted from the first and last pages of a PDF document.
+Determine if this document is a Korean archaeological investigation report (\uACE0\uACE0\uD559 \uC870\uC0AC \uBCF4\uACE0\uC11C).
+
+Valid types include:
+- \uBC1C\uAD74\uC870\uC0AC\uBCF4\uACE0\uC11C (excavation investigation report)
+- \uC2DC\uAD74\uC870\uC0AC\uBCF4\uACE0\uC11C (trial excavation report)
+- \uC9C0\uD45C\uC870\uC0AC\uBCF4\uACE0\uC11C (surface survey report)
+- \uC815\uBC00\uBC1C\uAD74\uC870\uC0AC\uBCF4\uACE0\uC11C (detailed excavation report)
+- \uC218\uC911\uBC1C\uAD74\uC870\uC0AC\uBCF4\uACE0\uC11C (underwater excavation report)
+
+NOT valid (these are NOT archaeological investigation reports):
+- \uC218\uB9AC\uBCF4\uACE0\uC11C (repair/restoration report)
+- \uB2E8\uC21C \uC2E4\uCE21 \uBCF4\uACE0\uC11C (simple measurement report)
+- \uAC74\uCD95\uC870\uC0AC\uBCF4\uACE0\uC11C (architectural investigation report)
+- \uD559\uC220\uC870\uC0AC\uBCF4\uACE0\uC11C (academic research report)
+- \uD658\uACBD\uC601\uD5A5\uD3C9\uAC00 (environmental impact assessment)
+- General academic papers or textbooks about archaeology`;
+var documentTypeSchema = import_zod.z.object({
+  isValid: import_zod.z.boolean().describe("Whether this is a Korean archaeological investigation report"),
+  reason: import_zod.z.string().describe("Brief reason for the decision")
+});
+var DocumentTypeValidator = class {
+  textExtractor;
+  constructor(textExtractor) {
+    this.textExtractor = textExtractor;
+  }
+  /**
+   * Validate that the PDF at the given path is an archaeological investigation report.
+   *
+   * @throws {InvalidDocumentTypeError} if the document is not a valid report type
+   */
+  async validate(pdfPath, model, options) {
+    const totalPages = await this.textExtractor.getPageCount(pdfPath);
+    if (totalPages === 0) return;
+    const frontText = await this.textExtractor.extractPageRange(
+      pdfPath,
+      1,
+      Math.min(10, totalPages)
+    );
+    let backText = "";
+    if (totalPages > 20) {
+      backText = await this.textExtractor.extractPageRange(
+        pdfPath,
+        Math.max(1, totalPages - 9),
+        totalPages
+      );
+    }
+    const combinedText = (frontText + "\n" + backText).trim();
+    if (combinedText.length === 0) return;
+    const result = await LLMCaller.call({
+      schema: documentTypeSchema,
+      systemPrompt: SYSTEM_PROMPT,
+      userPrompt: `--- Document text (first and last pages) ---
+${combinedText}`,
+      primaryModel: model,
+      maxRetries: 2,
+      temperature: 0,
+      abortSignal: options?.abortSignal,
+      component: "DocumentTypeValidator",
+      phase: "validation"
+    });
+    if (!result.output.isValid) {
+      throw new InvalidDocumentTypeError(result.output.reason);
+    }
+  }
+};
+
 // src/core/chunked-pdf-converter.ts
 var import_node_fs7 = require("fs");
 var import_promises4 = require("fs/promises");
@@ -2714,14 +2855,15 @@ var ChunkedPDFConverter = class {
       const status = await task.poll();
       if (status.task_status === "success") return;
       if (status.task_status === "failure") {
-        let details = "unknown";
-        try {
-          const result = await task.getResult();
-          if (result.errors?.length) {
-            details = result.errors.map((e) => e.message).join("; ");
-          }
-        } catch {
-        }
+        const elapsed = ((Date.now() - startTime) / 1e3).toFixed(1);
+        this.logger.error(
+          `[ChunkedPDFConverter] Task ${task.taskId} failed after ${elapsed}s`
+        );
+        const details = await getTaskFailureDetails(
+          task,
+          this.logger,
+          "[ChunkedPDFConverter]"
+        );
         throw new Error(`[ChunkedPDFConverter] Chunk task failed: ${details}`);
       }
       await new Promise(
@@ -2986,8 +3128,27 @@ var PDFConverter = class {
     this.enableImagePdfFallback = enableImagePdfFallback;
     this.timeout = timeout;
   }
+  documentTypeValidated = false;
+  /**
+   * Validate that the PDF is a Korean archaeological investigation report.
+   * Skipped when no documentValidationModel is configured or for non-local URLs.
+   * Only runs once per converter instance (flag prevents duplicate checks on recursive calls).
+   */
+  async validateDocumentType(url, options, abortSignal) {
+    if (this.documentTypeValidated) return;
+    this.documentTypeValidated = true;
+    if (!options.documentValidationModel) return;
+    const pdfPath = url.startsWith("file://") ? url.slice(7) : null;
+    if (!pdfPath) return;
+    const textExtractor = new PdfTextExtractor(this.logger);
+    const validator = new DocumentTypeValidator(textExtractor);
+    await validator.validate(pdfPath, options.documentValidationModel, {
+      abortSignal
+    });
+  }
   async convert(url, reportId, onComplete, cleanupAfterCallback, options, abortSignal) {
     this.logger.info("[PDFConverter] Converting:", url);
+    await this.validateDocumentType(url, options, abortSignal);
     if (options.chunkedConversion && url.startsWith("file://")) {
       const chunked = new ChunkedPDFConverter(
         this.logger,
@@ -3042,6 +3203,7 @@ var PDFConverter = class {
     const aggregator = options.aggregator ?? new LLMTokenUsageAggregator();
     const trackedOptions = { ...options, aggregator };
     const pdfPath = url.startsWith("file://") ? url.slice(7) : null;
+    await this.validateDocumentType(url, trackedOptions, abortSignal);
     const strategy = await this.determineStrategy(
       pdfPath,
       reportId,
@@ -3358,6 +3520,7 @@ var PDFConverter = class {
     return {
       ...(0, import_es_toolkit.omit)(options, [
         "num_threads",
+        "document_timeout",
         "forceImagePdf",
         "strategySamplerModel",
         "vlmProcessorModel",
@@ -3367,7 +3530,8 @@ var PDFConverter = class {
         "onTokenUsage",
         "chunkedConversion",
         "chunkSize",
-        "chunkMaxRetries"
+        "chunkMaxRetries",
+        "documentValidationModel"
       ]),
       to_formats: ["json", "html"],
       image_export_mode: "embedded",
@@ -3379,6 +3543,8 @@ var PDFConverter = class {
         framework: "livetext"
       },
       generate_picture_images: true,
+      do_picture_classification: true,
+      do_picture_description: true,
       generate_page_images: false,
       // Page images are rendered by PageRenderer (ImageMagick) after conversion
       images_scale: 2,
@@ -3393,6 +3559,9 @@ var PDFConverter = class {
       accelerator_options: {
         device: "mps",
         num_threads: options.num_threads
+      },
+      ...options.document_timeout !== void 0 && {
+        document_timeout: options.document_timeout
       }
     };
   }
@@ -3479,16 +3648,7 @@ var PDFConverter = class {
    * Fetch detailed error information from a failed task result.
    */
   async getTaskFailureDetails(task) {
-    try {
-      const result = await task.getResult();
-      if (result.errors?.length) {
-        return result.errors.map((e) => e.message).join("; ");
-      }
-      return `status: ${result.status ?? "unknown"}`;
-    } catch (err) {
-      this.logger.error("[PDFConverter] Failed to retrieve task result:", err);
-      return "unable to retrieve error details";
-    }
+    return getTaskFailureDetails(task, this.logger, "[PDFConverter]");
   }
   async downloadResult(taskId) {
     this.logger.info(
@@ -4073,6 +4233,7 @@ var VlmResponseValidator = class {
 // Annotate the CommonJS export names for ESM import in node:
 0 && (module.exports = {
   ImagePdfFallbackError,
+  InvalidDocumentTypeError,
   PDFParser,
   VlmResponseValidator
 });