from .common import InfoExtractor from ..utils import try_get, unified_strdate, unified_timestamp class NFHSNetworkIE(InfoExtractor): IE_NAME = 'NFHSNetwork' _VALID_URL = r'https?://(?:www\.)?nfhsnetwork\.com/events/[\w-]+/(?P(?:gam|evt|dd|)?[\w\d]{0,10})' _TESTS = [{ # Auto-generated two-team sport (pixellot) 'url': 'https://www.nfhsnetwork.com/events/rockford-high-school-rockford-mi/gamcf7e54cfbc', 'info_dict': { 'id': 'gamcf7e54cfbc', 'ext': 'mp4', 'title': 'Rockford vs Spring Lake - Girls Varsity Lacrosse 03/27/2021', 'uploader': 'MHSAA - Michigan: Rockford High School, Rockford, MI', 'uploader_id': 'cd2622cf76', 'uploader_url': 'https://www.nfhsnetwork.com/schools/rockford-high-school-rockford-mi', 'location': 'Rockford, Michigan', 'timestamp': 1616859000, 'upload_date': '20210327', }, 'params': { # m3u8 download 'skip_download': True, }, }, { # Non-sport activity with description 'url': 'https://www.nfhsnetwork.com/events/limon-high-school-limon-co/evt4a30e3726c', 'info_dict': { 'id': 'evt4a30e3726c', 'ext': 'mp4', 'title': 'Drama Performance Limon High School vs. Limon High School - 12/13/2020', 'description': 'Join the broadcast of the Limon High School Musical Performance at 2 PM.', 'uploader': 'CHSAA: Limon High School, Limon, CO', 'uploader_id': '7d2d121332', 'uploader_url': 'https://www.nfhsnetwork.com/schools/limon-high-school-limon-co', 'location': 'Limon, Colorado', 'timestamp': 1607893200, 'upload_date': '20201213', }, 'params': { # m3u8 download 'skip_download': True, }, }, { # Postseason game 'url': 'https://www.nfhsnetwork.com/events/nfhs-network-special-events/dd8de71d45', 'info_dict': { 'id': 'dd8de71d45', 'ext': 'mp4', 'title': '2015 UA Holiday Classic Tournament: National Division - 12/26/2015', 'uploader': 'SoCal Sports Productions', 'uploader_id': '063dba0150', 'uploader_url': 'https://www.nfhsnetwork.com/affiliates/socal-sports-productions', 'location': 'San Diego, California', 'timestamp': 1451187000, 'upload_date': '20151226', }, 'params': { # m3u8 download 'skip_download': True, }, }, { # Video with no broadcasts object 'url': 'https://www.nfhsnetwork.com/events/wiaa-wi/9aa2f92f82', 'info_dict': { 'id': '9aa2f92f82', 'ext': 'mp4', 'title': 'Competitive Equity - 01/21/2015', 'description': 'Committee members discuss points of their research regarding a competitive equity plan', 'uploader': 'WIAA - Wisconsin: Wisconsin Interscholastic Athletic Association', 'uploader_id': 'a49f7d1002', 'uploader_url': 'https://www.nfhsnetwork.com/associations/wiaa-wi', 'location': 'Stevens Point, Wisconsin', 'timestamp': 1421856000, 'upload_date': '20150121', }, 'params': { # m3u8 download 'skip_download': True, }, }, ] def _real_extract(self, url): video_id = self._match_id(url) webpage = self._download_webpage(url, video_id) data = self._download_json( 'https://cfunity.nfhsnetwork.com/v2/game_or_event/' + video_id, video_id) publisher = data.get('publishers')[0] # always exists broadcast = (publisher.get('broadcasts') or publisher.get('vods'))[0] # some (older) videos don't have a broadcasts object uploader = publisher.get('formatted_name') or publisher.get('name') uploader_id = publisher.get('publisher_key') pub_type = publisher.get('type') uploader_prefix = ( 'schools' if pub_type == 'school' else 'associations' if 'association' in pub_type else 'affiliates' if (pub_type == 'publisher' or pub_type == 'affiliate') else 'schools') uploader_page = 'https://www.nfhsnetwork.com/{}/{}'.format(uploader_prefix, publisher.get('slug')) location = '{}, {}'.format(data.get('city'), data.get('state_name')) description = broadcast.get('description') is_live = broadcast.get('on_air') or broadcast.get('status') == 'on_air' or False timestamp = unified_timestamp(data.get('local_start_time')) upload_date = unified_strdate(data.get('local_start_time')) title = ( self._og_search_title(webpage) or self._html_search_regex(r'

(.*?)

', webpage, 'title')) title = title.split('|')[0].strip() video_type = 'broadcasts' if is_live else 'vods' key = broadcast.get('key') if is_live else try_get(publisher, lambda x: x['vods'][0]['key']) m3u8_url = self._download_json( f'https://cfunity.nfhsnetwork.com/v2/{video_type}/{key}/url', video_id).get('video_url') formats = self._extract_m3u8_formats(m3u8_url, video_id, 'mp4', live=is_live) return { 'id': video_id, 'title': title, 'formats': formats, 'description': description, 'timestamp': timestamp, 'uploader': uploader, 'uploader_id': uploader_id, 'uploader_url': uploader_page, 'location': location, 'upload_date': upload_date, 'is_live': is_live, '_format_sort_fields': ('res', 'tbr'), }