from typing import Optional, List, Dict from app.chain import ChainBase from app.core.config import settings from app.core.context import Context, MediaInfo, TorrentInfo from app.core.meta import MetaBase from app.core.metainfo import MetaInfo from app.helper.sites import SitesHelper from app.log import logger from app.utils.string import StringUtils class SearchChain(ChainBase): """ 站点资源搜索处理链 """ def __init__(self): super().__init__() self.siteshelper = SitesHelper() def process(self, meta: MetaBase, mediainfo: MediaInfo, keyword: str = None, no_exists: Dict[int, List[dict]] = None) -> Optional[List[Context]]: """ 根据媒体信息,执行搜索 :param meta: 元数据 :param mediainfo: 媒体信息 :param keyword: 搜索关键词 :param no_exists: 缺失的媒体信息 """ logger.info(f'开始搜索资源,关键词:{keyword or mediainfo.title} ...') # 未开启的站点不搜索 indexer_sites = [] for indexer in self.siteshelper.get_indexers(): if not settings.INDEXER_SITES \ or any([s in indexer.get("domain") for s in settings.INDEXER_SITES.split(',')]): indexer_sites.append(indexer) if not indexer_sites: logger.warn('未开启任何有效站点,无法搜索资源') return [] # 缺失的媒体信息 if no_exists: # 过滤剧集 season_episodes = {info.get('season'): info.get('episodes') for info in no_exists.get(mediainfo.tmdb_id)} else: season_episodes = None # 执行搜索 torrents: List[TorrentInfo] = self.search_torrents( mediainfo=mediainfo, keyword=keyword, sites=indexer_sites ) if not torrents: logger.warn(f'{keyword or mediainfo.title} 未搜索到资源') return [] # 过滤种子 result: List[TorrentInfo] = self.filter_torrents(torrent_list=torrents, season_episodes=season_episodes) if result is not None: torrents = result if not torrents: logger.warn(f'{keyword or mediainfo.title} 没有符合过滤条件的资源') return [] # 过滤不匹配的资源 _match_torrents = [] if mediainfo: for torrent in torrents: # 比对IMDBID if torrent.imdbid \ and mediainfo.imdb_id \ and torrent.imdbid == mediainfo.imdb_id: logger.info(f'{mediainfo.title} 匹配到资源:{torrent.title}') _match_torrents.append(torrent) continue # 识别 torrent_meta = MetaInfo(torrent.title, torrent.description) # 比对标题 if torrent_meta.get_name() in [mediainfo.title, mediainfo.original_title]: logger.info(f'{mediainfo.title} 匹配到资源:{torrent.title}') _match_torrents.append(torrent) continue # 比对别名和译名 for name in mediainfo.names: if StringUtils.clear(name).strip().upper() == \ StringUtils.clear(torrent_meta.get_name()).strip().upper(): logger.info(f'{mediainfo.title} 匹配到资源:{torrent.title}') _match_torrents.append(torrent) break else: _match_torrents = torrents # 组装上下文返回 return [Context(meta=MetaInfo(torrent.title), mediainfo=mediainfo, torrentinfo=torrent) for torrent in _match_torrents]