X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/fe979149c83b5a935f7d28baf75848a9137316fd..7b54e7278c1ff62e888689ec470f488682311356:/youtube_dl/extractor/anvato.py?ds=sidebyside diff --git a/youtube_dl/extractor/anvato.py b/youtube_dl/extractor/anvato.py index cb29cf1..84e8410 100644 --- a/youtube_dl/extractor/anvato.py +++ b/youtube_dl/extractor/anvato.py @@ -5,6 +5,7 @@ import base64 import hashlib import json import random +import re import time from .common import InfoExtractor @@ -16,6 +17,8 @@ from ..utils import ( intlist_to_bytes, int_or_none, strip_jsonp, + unescapeHTML, + unsmuggle_url, ) @@ -26,6 +29,8 @@ def md5_text(s): class AnvatoIE(InfoExtractor): + _VALID_URL = r'anvato:(?P[^:]+):(?P\d+)' + # Copied from anvplayer.min.js _ANVACK_TABLE = { 'nbcu_nbcd_desktop_web_prod_93d8ead38ce2024f8f544b78306fbd15895ae5e6': 'NNemUkySjxLyPTKvZRiGntBIjEyK8uqicjMakIaQ', @@ -114,8 +119,48 @@ class AnvatoIE(InfoExtractor): 'nbcu_nbcd_desktop_web_prod_93d8ead38ce2024f8f544b78306fbd15895ae5e6_secure': 'NNemUkySjxLyPTKvZRiGntBIjEyK8uqicjMakIaQ' } + _MCP_TO_ACCESS_KEY_TABLE = { + 'qa': 'anvato_mcpqa_demo_web_stage_18b55e00db5a13faa8d03ae6e41f6f5bcb15b922', + 'lin': 'anvato_mcp_lin_web_prod_4c36fbfd4d8d8ecae6488656e21ac6d1ac972749', + 'univison': 'anvato_mcp_univision_web_prod_37fe34850c99a3b5cdb71dab10a417dd5cdecafa', + 'uni': 'anvato_mcp_univision_web_prod_37fe34850c99a3b5cdb71dab10a417dd5cdecafa', + 'dev': 'anvato_mcp_fs2go_web_prod_c7b90a93e171469cdca00a931211a2f556370d0a', + 'sps': 'anvato_mcp_sps_web_prod_54bdc90dd6ba21710e9f7074338365bba28da336', + 'spsstg': 'anvato_mcp_sps_web_prod_54bdc90dd6ba21710e9f7074338365bba28da336', + 'anv': 'anvato_mcp_anv_web_prod_791407490f4c1ef2a4bcb21103e0cb1bcb3352b3', + 'gray': 'anvato_mcp_gray_web_prod_4c10f067c393ed8fc453d3930f8ab2b159973900', + 'hearst': 'anvato_mcp_hearst_web_prod_5356c3de0fc7c90a3727b4863ca7fec3a4524a99', + 'cbs': 'anvato_mcp_cbs_web_prod_02f26581ff80e5bda7aad28226a8d369037f2cbe', + 'telemundo': 'anvato_mcp_telemundo_web_prod_c5278d51ad46fda4b6ca3d0ea44a7846a054f582' + } + + _API_KEY = '3hwbSuqqT690uxjNYBktSQpa5ZrpYYR0Iofx7NcJHyA' + + _ANVP_RE = r']+\bdata-anvp\s*=\s*(["\'])(?P(?:(?!\1).)+)\1' _AUTH_KEY = b'\x31\xc2\x42\x84\x9e\x73\xa0\xce' + _TESTS = [{ + # from https://www.boston25news.com/news/watch-humpback-whale-breaches-right-next-to-fishing-boat-near-nh/817484874 + 'url': 'anvato:8v9BEynrwx8EFLYpgfOWcG1qJqyXKlRM:4465496', + 'info_dict': { + 'id': '4465496', + 'ext': 'mp4', + 'title': 'VIDEO: Humpback whale breaches right next to NH boat', + 'description': 'VIDEO: Humpback whale breaches right next to NH boat. Footage courtesy: Zach Fahey.', + 'duration': 22, + 'timestamp': 1534855680, + 'upload_date': '20180821', + 'uploader': 'ANV', + }, + 'params': { + 'skip_download': True, + }, + }, { + # from https://sanfrancisco.cbslocal.com/2016/06/17/source-oakland-cop-on-leave-for-having-girlfriend-help-with-police-reports/ + 'url': 'anvato:DVzl9QRzox3ZZsP9bNu5Li3X7obQOnqP:3417601', + 'only_matching': True, + }] + def __init__(self, *args, **kwargs): super(AnvatoIE, self).__init__(*args, **kwargs) self.__server_time = None @@ -148,7 +193,8 @@ class AnvatoIE(InfoExtractor): 'api': { 'anvrid': anvrid, 'anvstk': md5_text('%s|%s|%d|%s' % ( - access_key, anvrid, server_time, self._ANVACK_TABLE[access_key])), + access_key, anvrid, server_time, + self._ANVACK_TABLE.get(access_key, self._API_KEY))), 'anvts': server_time, }, } @@ -157,22 +203,16 @@ class AnvatoIE(InfoExtractor): video_data_url, video_id, transform_source=strip_jsonp, data=json.dumps(payload).encode('utf-8')) - def _extract_anvato_videos(self, webpage, video_id): - anvplayer_data = self._parse_json(self._html_search_regex( - r']+data-anvp=\'([^\']+)\'', webpage, - 'Anvato player data'), video_id) - - video_id = anvplayer_data['video'] - access_key = anvplayer_data['accessKey'] - + def _get_anvato_videos(self, access_key, video_id): video_data = self._get_video_json(access_key, video_id) formats = [] for published_url in video_data['published_urls']: video_url = published_url['embed_url'] + media_format = published_url.get('format') ext = determine_ext(video_url) - if ext == 'smil': + if ext == 'smil' or media_format == 'smil': formats.extend(self._extract_smil_formats(video_url, video_id)) continue @@ -183,18 +223,17 @@ class AnvatoIE(InfoExtractor): 'tbr': tbr if tbr != 0 else None, } - if ext == 'm3u8': - # Not using _extract_m3u8_formats here as individual media - # playlists are also included in published_urls. - if tbr is None: - formats.append(self._m3u8_meta_format(video_url, ext='mp4', m3u8_id='hls')) - continue - else: - a_format.update({ - 'format_id': '-'.join(filter(None, ['hls', compat_str(tbr)])), - 'ext': 'mp4', - }) - elif ext == 'mp3': + if media_format == 'm3u8' and tbr is not None: + a_format.update({ + 'format_id': '-'.join(filter(None, ['hls', compat_str(tbr)])), + 'ext': 'mp4', + }) + elif media_format == 'm3u8-variant' or ext == 'm3u8': + formats.extend(self._extract_m3u8_formats( + video_url, video_id, 'mp4', entry_protocol='m3u8_native', + m3u8_id='hls', fatal=False)) + continue + elif ext == 'mp3' or media_format == 'mp3': a_format['vcodec'] = 'none' else: a_format.update({ @@ -218,7 +257,58 @@ class AnvatoIE(InfoExtractor): 'formats': formats, 'title': video_data.get('def_title'), 'description': video_data.get('def_description'), + 'tags': video_data.get('def_tags', '').split(','), 'categories': video_data.get('categories'), 'thumbnail': video_data.get('thumbnail'), + 'timestamp': int_or_none(video_data.get( + 'ts_published') or video_data.get('ts_added')), + 'uploader': video_data.get('mcp_id'), + 'duration': int_or_none(video_data.get('duration')), 'subtitles': subtitles, } + + @staticmethod + def _extract_urls(ie, webpage, video_id): + entries = [] + for mobj in re.finditer(AnvatoIE._ANVP_RE, webpage): + anvplayer_data = ie._parse_json( + mobj.group('anvp'), video_id, transform_source=unescapeHTML, + fatal=False) + if not anvplayer_data: + continue + video = anvplayer_data.get('video') + if not isinstance(video, compat_str) or not video.isdigit(): + continue + access_key = anvplayer_data.get('accessKey') + if not access_key: + mcp = anvplayer_data.get('mcp') + if mcp: + access_key = AnvatoIE._MCP_TO_ACCESS_KEY_TABLE.get( + mcp.lower()) + if not access_key: + continue + entries.append(ie.url_result( + 'anvato:%s:%s' % (access_key, video), ie=AnvatoIE.ie_key(), + video_id=video)) + return entries + + def _extract_anvato_videos(self, webpage, video_id): + anvplayer_data = self._parse_json( + self._html_search_regex( + self._ANVP_RE, webpage, 'Anvato player data', group='anvp'), + video_id) + return self._get_anvato_videos( + anvplayer_data['accessKey'], anvplayer_data['video']) + + def _real_extract(self, url): + url, smuggled_data = unsmuggle_url(url, {}) + self._initialize_geo_bypass({ + 'countries': smuggled_data.get('geo_countries'), + }) + + mobj = re.match(self._VALID_URL, url) + access_key, video_id = mobj.group('access_key_or_mcp', 'id') + if access_key not in self._ANVACK_TABLE: + access_key = self._MCP_TO_ACCESS_KEY_TABLE.get( + access_key) or access_key + return self._get_anvato_videos(access_key, video_id)