file2md/dist/parsers/pdf-parser.js

A TypeScript library for converting various document types (PDF, DOCX, XLSX, PPTX, HWP, HWPX) into Markdown with image and layout preservation

import pdfParse from 'pdf-parse';
import { PDFExtractor } from '../utils/pdf-extractor.js';
import { ParseError, InvalidFileError } from '../types/errors.js';
/**
 * Parse PDF buffer and convert to markdown with enhanced layout preservation
 */
export async function parsePdf(buffer, imageExtractor, options = {}) {
    try {
        const data = await pdfParse(buffer);
        const pdfExtractor = new PDFExtractor(imageExtractor);
        let markdown = '';
        const images = [];
        let pageCount = data.numpages || 1;
        // Apply maxPages limit if specified
        if (options.maxPages && options.maxPages > 0) {
            pageCount = Math.min(pageCount, options.maxPages);
        }
        // Try to extract text with enhanced layout
        if (data.text && data.text.trim()) {
            console.log('📄 Extracting text with layout enhancement...');
            try {
                const enhancedText = await pdfExtractor.enhanceTextWithLayout(data.text, data);
                markdown += enhancedText;
            }
            catch {
                console.warn('Layout enhancement failed, falling back to basic text extraction');
                // Fall back to basic text extraction
                markdown = extractBasicText(data.text);
            }
        }
        // If text is minimal or extraction failed, convert pages to images
        if (!data.text || data.text.trim().length < 100) {
            console.log('📸 Converting PDF pages to images for better preservation...');
            try {
                const pageImages = await pdfExtractor.extractImagesFromPDF(buffer);
                if (pageImages.length > 0) {
                    // Convert page images to ImageData format
                    for (const page of pageImages) {
                        images.push({
                            originalPath: `page_${page.pageNumber}`,
                            savedPath: page.imagePath,
                            basePath: '',
                            format: 'png',
                            dimensions: page.dimensions
                        });
                    }
                    if (markdown.trim()) {
                        markdown += '\n\n---\n\n## Visual Content\n\n';
                    }
                    markdown += await pdfExtractor.createPageBreaks(pageImages);
                }
            }
            catch (imageError) {
                console.warn('Failed to extract PDF as images:', imageError instanceof Error ? imageError.message : 'Unknown error');
            }
        }
        // Fallback to basic text if everything else fails
        if (!markdown.trim()) {
            if (data.text && data.text.trim()) {
                markdown = extractBasicText(data.text);
            }
            else {
                throw new InvalidFileError('PDF file appears to be empty or contains no extractable text');
            }
        }
        return {
            markdown,
            images,
            pageCount,
            metadata: {
                version: data.version || 'unknown',
                info: data.info || {},
                textLength: data.text?.length || 0
            }
        };
    }
    catch (error) {
        if (error instanceof InvalidFileError) {
            throw error;
        }
        const message = error instanceof Error ? error.message : 'Unknown error';
        if (message.includes('Invalid PDF') || message.includes('PDF')) {
            throw new InvalidFileError('Invalid or corrupted PDF file', error);
        }
        throw new ParseError('PDF', message, error);
    }
}
/**
 * Extract basic text from PDF with minimal formatting
 */
function extractBasicText(text) {
    const lines = text.split('\n');
    const cleanedLines = lines
        .map(line => line.trim())
        .filter(line => line.length > 0);
    return cleanedLines.join('\n');
}
//# sourceMappingURL=pdf-parser.js.map