<?php
/**
 * Reads AI-provenance signals from image files (IPTC/XMP/C2PA markers).
 *
 * @package AI_Disclosure_For_Chatbots
 */

if ( ! defined( 'ABSPATH' ) ) {
	exit;
}

/**
 * Lightweight metadata inspection: detects whether an uploaded image
 * declares itself AI-generated via industry-standard metadata.
 *
 * Signals checked (in order of reliability):
 * 1. C2PA manifest marker (JUMBF box signature present in file).
 * 2. IPTC/XMP DigitalSourceType = trainedAlgorithmicMedia / compositeWithTrainedAlgorithmicMedia.
 * 3. Known generator strings in XMP/EXIF software tags (DALL-E, Midjourney,
 *    Stable Diffusion, Adobe Firefly, GPT-4o, Gemini/Imagen).
 */
class ADFC_Metadata_Reader {

	const IPTC_AI_SOURCETYPES = array(
		'trainedAlgorithmicMedia',
		'compositeWithTrainedAlgorithmicMedia',
	);

	const GENERATOR_STRINGS = array(
		'DALL-E',
		'DALL·E',
		'OpenAI',
		'Midjourney',
		'Stable Diffusion',
		'StableDiffusion',
		'Adobe Firefly',
		'Firefly',
		'GPT-4o',
		'GPT Image',
		'Imagen',
		'Grok Image',
	);

	/**
	 * Inspect a file for AI-provenance markers.
	 *
	 * @param string $file Absolute path to the image file.
	 * @return array { detected: bool, method: string }
	 */
	public function inspect( $file ) {
		$result = array(
			'detected' => false,
			'method'   => '',
		);

		if ( ! is_readable( $file ) ) {
			return $result;
		}

		$size = filesize( $file );
		if ( false === $size || 0 === $size ) {
			return $result;
		}

		// Read up to the first 512 KB — metadata lives at the top of the file.
		$chunk = $this->read_chunk( $file, min( $size, 512 * 1024 ) );
		if ( '' === $chunk ) {
			return $result;
		}

		// 1. C2PA manifest (JUMBF superbox labelled c2pa).
		if ( false !== strpos( $chunk, 'c2pa' ) && ( false !== strpos( $chunk, 'jumb' ) || false !== strpos( $chunk, 'jumd' ) ) ) {
			$result['detected'] = true;
			$result['method']   = 'c2pa';
			return $result;
		}

		// 2. IPTC/XMP digital source type.
		foreach ( self::IPTC_AI_SOURCETYPES as $type ) {
			if ( false !== stripos( $chunk, $type ) ) {
				$result['detected'] = true;
				$result['method']   = 'iptc-digitalsourcetype';
				return $result;
			}
		}

		// 3. Known generator software strings (weakest signal).
		foreach ( self::GENERATOR_STRINGS as $needle ) {
			if ( false !== stripos( $chunk, $needle ) ) {
				$result['detected'] = true;
				$result['method']   = 'generator-string';
				return $result;
			}
		}

		return $result;
	}

	/**
	 * Read the first N bytes of a file.
	 *
	 * @param string $file  File path.
	 * @param int    $bytes Number of bytes.
	 * @return string
	 */
	private function read_chunk( $file, $bytes ) {
		$handle = fopen( $file, 'rb' ); // phpcs:ignore WordPress.WP.AlternativeFunctions.file_system_operations_fopen
		if ( ! $handle ) {
			return '';
		}
		$data = fread( $handle, $bytes ); // phpcs:ignore WordPress.WP.AlternativeFunctions.file_system_operations_fread
		fclose( $handle ); // phpcs:ignore WordPress.WP.AlternativeFunctions.file_system_operations_fclose
		return false === $data ? '' : $data;
	}
}