<?php
/**
 * Local-first agent-traffic counters + opt-in daily reporting.
 *
 * @package AEL42A
 */

namespace AEL42A;

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

/**
 * Telemetry.
 *
 * Always-on local counters in one option row (`ael42a_stats`,
 * autoload off): date(Ymd) => { mcp_requests, mcp_tool_calls: {tool: n},
 * markdown_hits, llms_txt_hits }, trimmed to 30 days.
 *
 * Reporting to 42A is opt-in and default OFF (`ael42a_telemetry_opt_in`
 * = '0'). When connected AND opted in, the daily cron posts yesterday's
 * aggregate. The plugin makes no other unprompted network calls, ever.
 */
class Telemetry {

	const OPTION       = 'ael42a_stats';
	const OPT_IN       = 'ael42a_telemetry_opt_in';
	const MAX_DAYS     = 30;
	const SENT_MARKER  = 'ael42a_telemetry_last_sent';
	const MAX_BOT_KEYS = 100;

	/**
	 * Hook the daily reporter and the classified-traffic page-view counter.
	 */
	public function register_hooks() {
		add_action( 'ael42a_daily', array( $this, 'maybe_send_daily' ) );

		// Priority 30: after Markdown_Controller (1, 20) and Llms_Txt's own
		// template_redirect handler, both of which `exit` when they serve a
		// request - so this only ever runs for requests neither intercepted,
		// i.e. regular page views. Avoids double-counting the same request.
		add_action( 'template_redirect', array( $this, 'maybe_record_page_view_bot_hit' ), 30 );
	}

	/**
	 * Whether the user opted into telemetry reporting.
	 *
	 * @return bool
	 */
	public function is_opted_in() {
		return '1' === (string) get_option( self::OPT_IN, '0' );
	}

	/**
	 * Increment a scalar counter for today.
	 *
	 * @param string $key One of mcp_requests|markdown_hits|llms_txt_hits.
	 */
	public function increment( $key ) {
		$stats = $this->get_stats();
		$today = gmdate( 'Ymd' );

		if ( ! isset( $stats[ $today ] ) ) {
			$stats[ $today ] = $this->empty_day();
		}
		if ( ! isset( $stats[ $today ][ $key ] ) ) {
			$stats[ $today ][ $key ] = 0;
		}
		$stats[ $today ][ $key ]++;

		$this->save_stats( $stats );
	}

	/**
	 * Increment the per-tool call counter for today.
	 *
	 * @param string $tool Tool name.
	 */
	public function increment_tool( $tool ) {
		$tool  = substr( preg_replace( '/[^a-zA-Z0-9_\-]/', '', (string) $tool ), 0, 64 );
		$stats = $this->get_stats();
		$today = gmdate( 'Ymd' );

		if ( ! isset( $stats[ $today ] ) ) {
			$stats[ $today ] = $this->empty_day();
		}
		if ( ! isset( $stats[ $today ]['mcp_tool_calls'][ $tool ] ) ) {
			$stats[ $today ]['mcp_tool_calls'][ $tool ] = 0;
		}
		$stats[ $today ]['mcp_tool_calls'][ $tool ]++;

		$this->save_stats( $stats );
	}

	/**
	 * Classify a User-Agent string against the AI bot taxonomy
	 * (Local_Vitals::get_ai_bots()): case-insensitive substring match of
	 * each known bot name within the UA string. When multiple names match,
	 * the LONGEST matching name wins (more specific identifier). Mirrors
	 * classifyUserAgent() in packages/ael/src/bots.ts.
	 *
	 * @param string $user_agent Raw User-Agent header value.
	 * @return array|null Matching bot row ({name, tier, owner, description}), or null.
	 */
	public static function classify_user_agent( $user_agent ) {
		$ua = strtolower( (string) $user_agent );
		if ( '' === $ua ) {
			return null;
		}

		$best = null;
		foreach ( Local_Vitals::get_ai_bots() as $bot ) {
			if ( false === strpos( $ua, strtolower( $bot['name'] ) ) ) {
				continue;
			}
			if ( null === $best || strlen( $bot['name'] ) > strlen( $best['name'] ) ) {
				$best = $bot;
			}
		}

		return $best;
	}

	/**
	 * Increment today's counter for a classified bot name, bounded to
	 * MAX_BOT_KEYS distinct bot names per day (extras silently dropped).
	 *
	 * @param string $bot_name Bot name (from classify_user_agent()).
	 */
	public function record_bot_hit( $bot_name ) {
		$bot_name = substr( preg_replace( '/[^a-zA-Z0-9_\-. ]/', '', (string) $bot_name ), 0, 64 );
		if ( '' === $bot_name ) {
			return;
		}

		$stats = $this->get_stats();
		$today = gmdate( 'Ymd' );

		if ( ! isset( $stats[ $today ] ) ) {
			$stats[ $today ] = $this->empty_day();
		}
		if ( ! isset( $stats[ $today ]['bot_hits'] ) || ! is_array( $stats[ $today ]['bot_hits'] ) ) {
			$stats[ $today ]['bot_hits'] = array();
		}
		if ( ! isset( $stats[ $today ]['bot_hits'][ $bot_name ] ) ) {
			if ( count( $stats[ $today ]['bot_hits'] ) >= self::MAX_BOT_KEYS ) {
				return; // Bounded: drop new distinct bot names beyond the cap.
			}
			$stats[ $today ]['bot_hits'][ $bot_name ] = 0;
		}
		$stats[ $today ]['bot_hits'][ $bot_name ]++;

		$this->save_stats( $stats );
	}

	/**
	 * Classify the current request's User-Agent header and record a bot hit
	 * when it matches a known AI bot. Cheap early-exit when the UA doesn't
	 * match. No paths or PII are ever logged - only the classified bot
	 * name/count. Called from the markdown/llms.txt/MCP handlers and (for
	 * regular page views) from maybe_record_page_view_bot_hit() below.
	 */
	public function record_bot_hit_from_request() {
		$ua = isset( $_SERVER['HTTP_USER_AGENT'] ) ? sanitize_text_field( wp_unslash( $_SERVER['HTTP_USER_AGENT'] ) ) : '';
		if ( '' === $ua ) {
			return;
		}

		$bot = self::classify_user_agent( $ua );
		if ( null === $bot ) {
			return;
		}

		$this->record_bot_hit( $bot['name'] );
	}

	/**
	 * Front-end-only wrapper around record_bot_hit_from_request() for
	 * regular page views (hooked at a low template_redirect priority so it
	 * never double-counts requests the markdown/llms.txt handlers already
	 * served and exited on).
	 */
	public function maybe_record_page_view_bot_hit() {
		if ( is_admin() ) {
			return;
		}

		$this->record_bot_hit_from_request();
	}

	/**
	 * Classified AI-traffic summary for the admin dashboard: tier totals
	 * (training/search/agentic) and the top 5 bots by hit count, over the
	 * last $days days of local counters.
	 *
	 * @param int $days Number of trailing days to aggregate.
	 * @return array {tiers: {training,search,agentic}, top: [{name,tier,hits}]}
	 */
	public function bot_traffic_summary( $days = 7 ) {
		$stats = $this->get_stats();

		$tier_by_name = array();
		foreach ( Local_Vitals::get_ai_bots() as $bot ) {
			$tier_by_name[ $bot['name'] ] = $bot['tier'];
		}

		$tier_totals = array(
			'training' => 0,
			'search'   => 0,
			'agentic'  => 0,
		);
		$totals      = array();

		for ( $i = 0; $i < $days; $i++ ) {
			$key = gmdate( 'Ymd', time() - $i * DAY_IN_SECONDS );
			if ( ! isset( $stats[ $key ]['bot_hits'] ) || ! is_array( $stats[ $key ]['bot_hits'] ) ) {
				continue;
			}
			foreach ( $stats[ $key ]['bot_hits'] as $name => $count ) {
				$count           = (int) $count;
				$totals[ $name ] = isset( $totals[ $name ] ) ? $totals[ $name ] + $count : $count;

				$tier = isset( $tier_by_name[ $name ] ) ? $tier_by_name[ $name ] : 'agentic';
				if ( isset( $tier_totals[ $tier ] ) ) {
					$tier_totals[ $tier ] += $count;
				}
			}
		}

		arsort( $totals );
		$top = array();
		foreach ( $totals as $name => $count ) {
			if ( count( $top ) >= 5 ) {
				break;
			}
			$top[] = array(
				'name' => $name,
				'tier' => isset( $tier_by_name[ $name ] ) ? $tier_by_name[ $name ] : 'agentic',
				'hits' => $count,
			);
		}

		return array(
			'tiers' => $tier_totals,
			'top'   => $top,
		);
	}

	/**
	 * All stored daily aggregates, newest last.
	 *
	 * @return array
	 */
	public function get_stats() {
		$stats = get_option( self::OPTION, array() );
		return is_array( $stats ) ? $stats : array();
	}

	/**
	 * The last N days of counters, shaped for the admin traffic table.
	 *
	 * @param int $days Number of days.
	 * @return array[] Rows {date, mcp_requests, markdown_hits, llms_txt_hits, tool_calls_total}.
	 */
	public function recent_days( $days = 7 ) {
		$stats = $this->get_stats();
		$rows  = array();

		for ( $i = $days - 1; $i >= 0; $i-- ) {
			$key = gmdate( 'Ymd', time() - $i * DAY_IN_SECONDS );
			$day = isset( $stats[ $key ] ) ? $stats[ $key ] : $this->empty_day();

			$tool_total = 0;
			if ( ! empty( $day['mcp_tool_calls'] ) && is_array( $day['mcp_tool_calls'] ) ) {
				$tool_total = array_sum( $day['mcp_tool_calls'] );
			}

			$rows[] = array(
				'date'             => gmdate( 'Y-m-d', time() - $i * DAY_IN_SECONDS ),
				'mcp_requests'     => isset( $day['mcp_requests'] ) ? (int) $day['mcp_requests'] : 0,
				'markdown_hits'    => isset( $day['markdown_hits'] ) ? (int) $day['markdown_hits'] : 0,
				'llms_txt_hits'    => isset( $day['llms_txt_hits'] ) ? (int) $day['llms_txt_hits'] : 0,
				'tool_calls_total' => (int) $tool_total,
			);
		}

		return $rows;
	}

	/**
	 * Daily cron: post yesterday's aggregate to 42A when connected + opted in.
	 */
	public function maybe_send_daily() {
		if ( ! $this->is_opted_in() ) {
			return;
		}
		if ( '' === (string) get_option( 'ael42a_site_token', '' ) ) {
			return;
		}

		$yesterday_key  = gmdate( 'Ymd', time() - DAY_IN_SECONDS );
		$yesterday_date = gmdate( 'Y-m-d', time() - DAY_IN_SECONDS );

		if ( get_option( self::SENT_MARKER, '' ) === $yesterday_date ) {
			return; // Already reported.
		}

		$stats = $this->get_stats();
		$day   = isset( $stats[ $yesterday_key ] ) ? $stats[ $yesterday_key ] : $this->empty_day();

		$payload = array(
			// Re-assert consent on every report: the server keeps its own
			// opt-in flag and rejects data unless it has been enabled.
			'opt_in'        => true,
			'date'          => $yesterday_date,
			'mcp_requests'  => isset( $day['mcp_requests'] ) ? (int) $day['mcp_requests'] : 0,
			'markdown_hits' => isset( $day['markdown_hits'] ) ? (int) $day['markdown_hits'] : 0,
			'llms_txt_hits' => isset( $day['llms_txt_hits'] ) ? (int) $day['llms_txt_hits'] : 0,
			'tool_calls'    => isset( $day['mcp_tool_calls'] ) && is_array( $day['mcp_tool_calls'] ) ? $day['mcp_tool_calls'] : new \stdClass(),
		);

		// Additive/optional: only included when there is at least one
		// classified bot hit for the day.
		if ( ! empty( $day['bot_hits'] ) && is_array( $day['bot_hits'] ) ) {
			$payload['bot_hits'] = $day['bot_hits'];
		}

		$client = new Api_Client();
		$result = $client->send_telemetry( $payload );

		if ( ! is_wp_error( $result ) ) {
			update_option( self::SENT_MARKER, $yesterday_date, false );
		}
	}

	/**
	 * Empty per-day structure.
	 *
	 * @return array
	 */
	private function empty_day() {
		return array(
			'mcp_requests'   => 0,
			'mcp_tool_calls' => array(),
			'markdown_hits'  => 0,
			'llms_txt_hits'  => 0,
			'bot_hits'       => array(),
		);
	}

	/**
	 * Persist stats, trimmed to the newest MAX_DAYS entries, autoload off.
	 *
	 * @param array $stats Stats map date(Ymd) => day.
	 */
	private function save_stats( $stats ) {
		ksort( $stats );
		if ( count( $stats ) > self::MAX_DAYS ) {
			$stats = array_slice( $stats, -self::MAX_DAYS, null, true );
		}

		if ( false === add_option( self::OPTION, $stats, '', 'no' ) ) {
			update_option( self::OPTION, $stats, 'no' );
		}
	}
}