Function extractPdfText

src/pdf.js:22–76

(buf)

 * @returns {string} Extracted text, or empty string if no text layer
 */
export function extractPdfText(buf) {
  const pages = [];
  let streamCount = 0;
  let totalDecoded = 0;

  // Find all stream...endstream blocks
  let pos = 0;
  while (pos < buf.length) {
    const streamStart = buf.indexOf('stream\n', pos);
    if (streamStart === -1) break;

    const dataStart = streamStart + 7; // skip "stream\n"
    // Handle \r\n after "stream"
    const actualStart = buf[streamStart + 6] === 0x0d ? dataStart + 1 : dataStart;

    const endStream = buf.indexOf('\nendstream', actualStart);
    if (endStream === -1) break;

    const streamData = buf.subarray(actualStart, endStream);
    streamCount++;
    if (streamCount > MAX_STREAMS) throw new Error('PDF stream count exceeds safety limit');

    // Check if this stream has FlateDecode by looking back at the dictionary
    const dictStart = Math.max(0, streamStart - 500);
    const dictText = buf.subarray(dictStart, streamStart).toString('latin1');
    const isFlate = dictText.includes('FlateDecode');

    let decoded;
    try {
      if (isFlate) {
        const inflated = inflateSync(streamData, { maxOutputLength: MAX_STREAM_DECODED });
        if (inflated.length > MAX_STREAM_DECODED) throw new Error('PDF decoded content exceeds safety limit');
        totalDecoded += inflated.length;
        if (totalDecoded > MAX_TOTAL_DECODED) throw new Error('PDF decoded content exceeds safety limit');
        decoded = inflated.toString('latin1');
      } else {
        totalDecoded += streamData.length;
        if (streamData.length > MAX_STREAM_DECODED || totalDecoded > MAX_TOTAL_DECODED) throw new Error('PDF decoded content exceeds safety limit');
        decoded = streamData.toString('latin1');
      }
    } catch (e) {
      if (/limit|exceed|maxOutputLength|Buffer larger/i.test(e.message) || e.code === 'ERR_BUFFER_TOO_LARGE') throw e;
      pos = endStream + 10;
      continue;
    }

    // Extract text from PDF operators
    const text = extractTextOps(decoded);
    if (text.trim()) pages.push(text.trim());

    pos = endStream + 10;
  }

  return pages.join('\n\n');
}

/**
 * Extract text from PDF content stream operators.