X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/d9d7cd0e85dc712461d9185db9df9d6c900a573b..HEAD:/youtube_dl/extractor/vice.py diff --git a/youtube_dl/extractor/vice.py b/youtube_dl/extractor/vice.py index 54e207b..e374995 100644 --- a/youtube_dl/extractor/vice.py +++ b/youtube_dl/extractor/vice.py @@ -1,119 +1,73 @@ # coding: utf-8 from __future__ import unicode_literals -import re -import time +import functools import hashlib import json +import random +import re +import time from .adobepass import AdobePassIE from .common import InfoExtractor -from ..compat import compat_HTTPError +from .youtube import YoutubeIE +from ..compat import ( + compat_HTTPError, + compat_str, +) from ..utils import ( + clean_html, + ExtractorError, int_or_none, + OnDemandPagedList, parse_age_limit, str_or_none, - parse_duration, - ExtractorError, - extract_attributes, + try_get, ) -class ViceBaseIE(AdobePassIE): - def _extract_preplay_video(self, url, locale, webpage): - watch_hub_data = extract_attributes(self._search_regex( - r'(?s)()', webpage, 'watch hub')) - video_id = watch_hub_data['vms-id'] - title = watch_hub_data['video-title'] +class ViceBaseIE(InfoExtractor): + def _call_api(self, resource, resource_key, resource_id, locale, fields, args=''): + return self._download_json( + 'https://video.vice.com/api/v1/graphql', resource_id, query={ + 'query': '''{ + %s(locale: "%s", %s: "%s"%s) { + %s + } +}''' % (resource, locale, resource_key, resource_id, args, fields), + })['data'][resource] - query = {} - is_locked = watch_hub_data.get('video-locked') == '1' - if is_locked: - resource = self._get_mvpd_resource( - 'VICELAND', title, video_id, - watch_hub_data.get('video-rating')) - query['tvetoken'] = self._extract_mvpd_auth( - url, video_id, 'VICELAND', resource) - # signature generation algorithm is reverse engineered from signatureGenerator in - # webpack:///../shared/~/vice-player/dist/js/vice-player.js in - # https://www.viceland.com/assets/common/js/web.vendor.bundle.js - exp = int(time.time()) + 14400 - query.update({ - 'exp': exp, - 'sign': hashlib.sha512(('%s:GET:%d' % (video_id, exp)).encode()).hexdigest(), - }) - - try: - host = 'www.viceland' if is_locked else self._PREPLAY_HOST - preplay = self._download_json( - 'https://%s.com/%s/preplay/%s' % (host, locale, video_id), - video_id, query=query) - except ExtractorError as e: - if isinstance(e.cause, compat_HTTPError) and e.cause.code == 400: - error = json.loads(e.cause.read().decode()) - raise ExtractorError('%s said: %s' % ( - self.IE_NAME, error['details']), expected=True) - raise - - video_data = preplay['video'] - base = video_data['base'] - uplynk_preplay_url = preplay['preplayURL'] - episode = video_data.get('episode', {}) - channel = video_data.get('channel', {}) - - subtitles = {} - cc_url = preplay.get('ccURL') - if cc_url: - subtitles['en'] = [{ - 'url': cc_url, - }] - - return { - '_type': 'url_transparent', - 'url': uplynk_preplay_url, - 'id': video_id, - 'title': title, - 'description': base.get('body') or base.get('display_body'), - 'thumbnail': watch_hub_data.get('cover-image') or watch_hub_data.get('thumbnail'), - 'duration': int_or_none(video_data.get('video_duration')) or parse_duration(watch_hub_data.get('video-duration')), - 'timestamp': int_or_none(video_data.get('created_at'), 1000), - 'age_limit': parse_age_limit(video_data.get('video_rating')), - 'series': video_data.get('show_title') or watch_hub_data.get('show-title'), - 'episode_number': int_or_none(episode.get('episode_number') or watch_hub_data.get('episode')), - 'episode_id': str_or_none(episode.get('id') or video_data.get('episode_id')), - 'season_number': int_or_none(watch_hub_data.get('season')), - 'season_id': str_or_none(episode.get('season_id')), - 'uploader': channel.get('base', {}).get('title') or watch_hub_data.get('channel-title'), - 'uploader_id': str_or_none(channel.get('id')), - 'subtitles': subtitles, - 'ie_key': 'UplynkPreplay', - } - - -class ViceIE(ViceBaseIE): +class ViceIE(ViceBaseIE, AdobePassIE): IE_NAME = 'vice' - _VALID_URL = r'https?://(?:.+?\.)?vice\.com/(?:(?P[^/]+)/)?videos?/(?P[^/?#&]+)' - + _VALID_URL = r'https?://(?:(?:video|vms)\.vice|(?:www\.)?vice(?:land|tv))\.com/(?P[^/]+)/(?:video/[^/]+|embed)/(?P[\da-f]{24})' _TESTS = [{ - 'url': 'https://news.vice.com/video/experimenting-on-animals-inside-the-monkey-lab', - 'md5': '7d3ae2f9ba5f196cdd9f9efd43657ac2', + 'url': 'https://video.vice.com/en_us/video/pet-cremator/58c69e38a55424f1227dc3f7', 'info_dict': { - 'id': 'N2bzkydjraWDGwnt8jAttCF6Y0PDv4Zj', - 'ext': 'flv', - 'title': 'Monkey Labs of Holland', - 'description': 'md5:92b3c7dcbfe477f772dd4afa496c9149', + 'id': '58c69e38a55424f1227dc3f7', + 'ext': 'mp4', + 'title': '10 Questions You Always Wanted To Ask: Pet Cremator', + 'description': 'md5:fe856caacf61fe0e74fab15ce2b07ca5', + 'uploader': 'vice', + 'uploader_id': '57a204088cb727dec794c67b', + 'timestamp': 1489664942, + 'upload_date': '20170316', + 'age_limit': 14, + }, + 'params': { + # m3u8 download + 'skip_download': True, }, - 'add_ie': ['Ooyala'], }, { + # geo restricted to US 'url': 'https://video.vice.com/en_us/video/the-signal-from-tolva/5816510690b70e6c5fd39a56', 'info_dict': { 'id': '5816510690b70e6c5fd39a56', 'ext': 'mp4', - 'uploader': 'Waypoint', + 'uploader': 'vice', 'title': 'The Signal From Tölva', 'description': 'md5:3927e3c79f9e8094606a2b3c5b5e55d5', - 'uploader_id': '57f7d621e05ca860fa9ccaf9', + 'uploader_id': '57a204088cb727dec794c67b', 'timestamp': 1477941983, 'upload_date': '20161031', }, @@ -121,15 +75,14 @@ class ViceIE(ViceBaseIE): # m3u8 download 'skip_download': True, }, - 'add_ie': ['UplynkPreplay'], }, { 'url': 'https://video.vice.com/alps/video/ulfs-wien-beruchtigste-grafitti-crew-part-1/581b12b60a0e1f4c0fb6ea2f', 'info_dict': { 'id': '581b12b60a0e1f4c0fb6ea2f', 'ext': 'mp4', 'title': 'ULFs - Wien berüchtigste Grafitti Crew - Part 1', - 'description': '

Zwischen Hinterzimmer-Tattoos und U-Bahnschächten erzählen uns die Ulfs, wie es ist, "süchtig nach Sachbeschädigung" zu sein.

', - 'uploader': 'VICE', + 'description': 'Zwischen Hinterzimmer-Tattoos und U-Bahnschächten erzählen uns die Ulfs, wie es ist, "süchtig nach Sachbeschädigung" zu sein.', + 'uploader': 'vice', 'uploader_id': '57a204088cb727dec794c67b', 'timestamp': 1485368119, 'upload_date': '20170125', @@ -139,96 +92,207 @@ class ViceIE(ViceBaseIE): # AES-encrypted m3u8 'skip_download': True, }, - 'add_ie': ['UplynkPreplay'], }, { 'url': 'https://video.vice.com/en_us/video/pizza-show-trailer/56d8c9a54d286ed92f7f30e4', 'only_matching': True, + }, { + 'url': 'https://video.vice.com/en_us/embed/57f41d3556a0a80f54726060', + 'only_matching': True, + }, { + 'url': 'https://vms.vice.com/en_us/video/preplay/58c69e38a55424f1227dc3f7', + 'only_matching': True, + }, { + 'url': 'https://www.viceland.com/en_us/video/thursday-march-1-2018/5a8f2d7ff1cdb332dd446ec1', + 'only_matching': True, }] - _PREPLAY_HOST = 'video.vice' + + @staticmethod + def _extract_urls(webpage): + return re.findall( + r']+\bsrc=["\']((?:https?:)?//video\.vice\.com/[^/]+/embed/[\da-f]{24})', + webpage) + + @staticmethod + def _extract_url(webpage): + urls = ViceIE._extract_urls(webpage) + return urls[0] if urls else None def _real_extract(self, url): locale, video_id = re.match(self._VALID_URL, url).groups() - webpage, urlh = self._download_webpage_handle(url, video_id) - embed_code = self._search_regex( - r'embedCode=([^&\'"]+)', webpage, - 'ooyala embed code', default=None) - if embed_code: - return self.url_result('ooyala:%s' % embed_code, 'Ooyala') - youtube_id = self._search_regex( - r'data-youtube-id="([^"]+)"', webpage, 'youtube id', default=None) - if youtube_id: - return self.url_result(youtube_id, 'Youtube') - return self._extract_preplay_video(urlh.geturl(), locale, webpage) + video = self._call_api('videos', 'id', video_id, locale, '''body + locked + rating + thumbnail_url + title''')[0] + title = video['title'].strip() + rating = video.get('rating') + + query = {} + if video.get('locked'): + resource = self._get_mvpd_resource( + 'VICELAND', title, video_id, rating) + query['tvetoken'] = self._extract_mvpd_auth( + url, video_id, 'VICELAND', resource) + + # signature generation algorithm is reverse engineered from signatureGenerator in + # webpack:///../shared/~/vice-player/dist/js/vice-player.js in + # https://www.viceland.com/assets/common/js/web.vendor.bundle.js + # new JS is located here https://vice-web-statics-cdn.vice.com/vice-player/player-embed.js + exp = int(time.time()) + 1440 + + query.update({ + 'exp': exp, + 'sign': hashlib.sha512(('%s:GET:%d' % (video_id, exp)).encode()).hexdigest(), + 'skipadstitching': 1, + 'platform': 'desktop', + 'rn': random.randint(10000, 100000), + }) + + try: + preplay = self._download_json( + 'https://vms.vice.com/%s/video/preplay/%s' % (locale, video_id), + video_id, query=query) + except ExtractorError as e: + if isinstance(e.cause, compat_HTTPError) and e.cause.code in (400, 401): + error = json.loads(e.cause.read().decode()) + error_message = error.get('error_description') or error['details'] + raise ExtractorError('%s said: %s' % ( + self.IE_NAME, error_message), expected=True) + raise + + video_data = preplay['video'] + formats = self._extract_m3u8_formats( + preplay['playURL'], video_id, 'mp4', 'm3u8_native') + self._sort_formats(formats) + episode = video_data.get('episode') or {} + channel = video_data.get('channel') or {} + season = video_data.get('season') or {} + + subtitles = {} + for subtitle in preplay.get('subtitleURLs', []): + cc_url = subtitle.get('url') + if not cc_url: + continue + language_code = try_get(subtitle, lambda x: x['languages'][0]['language_code'], compat_str) or 'en' + subtitles.setdefault(language_code, []).append({ + 'url': cc_url, + }) + + return { + 'formats': formats, + 'id': video_id, + 'title': title, + 'description': clean_html(video.get('body')), + 'thumbnail': video.get('thumbnail_url'), + 'duration': int_or_none(video_data.get('video_duration')), + 'timestamp': int_or_none(video_data.get('created_at'), 1000), + 'age_limit': parse_age_limit(video_data.get('video_rating') or rating), + 'series': try_get(video_data, lambda x: x['show']['base']['display_title'], compat_str), + 'episode_number': int_or_none(episode.get('episode_number')), + 'episode_id': str_or_none(episode.get('id') or video_data.get('episode_id')), + 'season_number': int_or_none(season.get('season_number')), + 'season_id': str_or_none(season.get('id') or video_data.get('season_id')), + 'uploader': channel.get('name'), + 'uploader_id': str_or_none(channel.get('id')), + 'subtitles': subtitles, + } -class ViceShowIE(InfoExtractor): - IE_NAME = 'vice:show' - _VALID_URL = r'https?://(?:.+?\.)?vice\.com/(?:[^/]+/)?show/(?P[^/?#&]+)' - _TEST = { - 'url': 'https://munchies.vice.com/en/show/fuck-thats-delicious-2', +class ViceShowIE(ViceBaseIE): + IE_NAME = 'vice:show' + _VALID_URL = r'https?://(?:video\.vice|(?:www\.)?vice(?:land|tv))\.com/(?P[^/]+)/show/(?P[^/?#&]+)' + _PAGE_SIZE = 25 + _TESTS = [{ + 'url': 'https://video.vice.com/en_us/show/fck-thats-delicious', 'info_dict': { - 'id': 'fuck-thats-delicious-2', - 'title': "Fuck, That's Delicious", - 'description': 'Follow the culinary adventures of rapper Action Bronson during his ongoing world tour.', + 'id': '57a2040c8cb727dec794c901', + 'title': 'F*ck, That’s Delicious', + 'description': 'The life and eating habits of rap’s greatest bon vivant, Action Bronson.', }, - 'playlist_count': 17, - } + 'playlist_mincount': 64, + }, { + 'url': 'https://www.vicetv.com/en_us/show/fck-thats-delicious', + 'only_matching': True, + }] - def _real_extract(self, url): - show_id = self._match_id(url) - webpage = self._download_webpage(url, show_id) + def _fetch_page(self, locale, show_id, page): + videos = self._call_api('videos', 'show_id', show_id, locale, '''body + id + url''', ', page: %d, per_page: %d' % (page + 1, self._PAGE_SIZE)) + for video in videos: + yield self.url_result( + video['url'], ViceIE.ie_key(), video.get('id')) - entries = [ - self.url_result(video_url, ViceIE.ie_key()) - for video_url, _ in re.findall( - r']+class="article-title"[^>]+data-id="\d+"[^>]*>\s*]+href="(%s.*?)"' - % ViceIE._VALID_URL, webpage)] + def _real_extract(self, url): + locale, display_id = re.match(self._VALID_URL, url).groups() + show = self._call_api('shows', 'slug', display_id, locale, '''dek + id + title''')[0] + show_id = show['id'] - title = self._search_regex( - r'(.+?)', webpage, 'title', default=None) - if title: - title = re.sub(r'(.+)\s*\|\s*.+$', r'\1', title).strip() - description = self._html_search_meta( - 'description', webpage, 'description') + entries = OnDemandPagedList( + functools.partial(self._fetch_page, locale, show_id), + self._PAGE_SIZE) - return self.playlist_result(entries, show_id, title, description) + return self.playlist_result( + entries, show_id, show.get('title'), show.get('dek')) -class ViceArticleIE(InfoExtractor): +class ViceArticleIE(ViceBaseIE): IE_NAME = 'vice:article' - _VALID_URL = r'https://www.vice.com/[^/]+/article/(?P[^?#]+)' + _VALID_URL = r'https://(?:www\.)?vice\.com/(?P[^/]+)/article/(?:[0-9a-z]{6}/)?(?P[^?#]+)' _TESTS = [{ 'url': 'https://www.vice.com/en_us/article/on-set-with-the-woman-making-mormon-porn-in-utah', 'info_dict': { 'id': '58dc0a3dee202d2a0ccfcbd8', 'ext': 'mp4', - 'title': 'Mormon War on Porn ', - 'description': 'md5:ad396a2481e7f8afb5ed486878421090', - 'uploader': 'VICE', - 'uploader_id': '57a204088cb727dec794c693', - 'timestamp': 1489160690, - 'upload_date': '20170310', + 'title': 'Mormon War on Porn', + 'description': 'md5:1c5d91fe25fa8aa304f9def118b92dbf', + 'uploader': 'vice', + 'uploader_id': '57a204088cb727dec794c67b', + 'timestamp': 1491883129, + 'upload_date': '20170411', + 'age_limit': 17, }, 'params': { # AES-encrypted m3u8 'skip_download': True, }, - 'add_ie': ['UplynkPreplay'], + 'add_ie': [ViceIE.ie_key()], }, { 'url': 'https://www.vice.com/en_us/article/how-to-hack-a-car', - 'md5': 'a7ecf64ee4fa19b916c16f4b56184ae2', + 'md5': '13010ee0bc694ea87ec40724397c2349', 'info_dict': { 'id': '3jstaBeXgAs', 'ext': 'mp4', 'title': 'How to Hack a Car: Phreaked Out (Episode 2)', 'description': 'md5:ee95453f7ff495db8efe14ae8bf56f30', - 'uploader_id': 'MotherboardTV', 'uploader': 'Motherboard', + 'uploader_id': 'MotherboardTV', 'upload_date': '20140529', }, - 'add_ie': ['Youtube'], + 'add_ie': [YoutubeIE.ie_key()], + }, { + 'url': 'https://www.vice.com/en_us/article/znm9dx/karley-sciortino-slutever-reloaded', + 'md5': 'a7ecf64ee4fa19b916c16f4b56184ae2', + 'info_dict': { + 'id': '57f41d3556a0a80f54726060', + 'ext': 'mp4', + 'title': "Making The World's First Male Sex Doll", + 'description': 'md5:19b00b215b99961cf869c40fbe9df755', + 'uploader': 'vice', + 'uploader_id': '57a204088cb727dec794c67b', + 'timestamp': 1476919911, + 'upload_date': '20161019', + 'age_limit': 17, + }, + 'params': { + 'skip_download': True, + 'format': 'bestvideo', + }, + 'add_ie': [ViceIE.ie_key()], }, { 'url': 'https://www.vice.com/en_us/article/cowboy-capitalists-part-1', 'only_matching': True, @@ -238,14 +302,11 @@ class ViceArticleIE(InfoExtractor): }] def _real_extract(self, url): - display_id = self._match_id(url) - - webpage = self._download_webpage(url, display_id) + locale, display_id = re.match(self._VALID_URL, url).groups() - prefetch_data = self._parse_json(self._search_regex( - r'window\.__PREFETCH_DATA\s*=\s*({.*});', - webpage, 'prefetch data'), display_id) - body = prefetch_data['body'] + article = self._call_api('articles', 'slug', display_id, locale, '''body + embed_code''')[0] + body = article['body'] def _url_res(video_url, ie_key): return { @@ -255,20 +316,22 @@ class ViceArticleIE(InfoExtractor): 'ie_key': ie_key, } + vice_url = ViceIE._extract_url(body) + if vice_url: + return _url_res(vice_url, ViceIE.ie_key()) + embed_code = self._search_regex( r'embedCode=([^&\'"]+)', body, 'ooyala embed code', default=None) if embed_code: return _url_res('ooyala:%s' % embed_code, 'Ooyala') - youtube_url = self._html_search_regex( - r']+src="(.*youtube\.com/.*)"', - body, 'YouTube URL', default=None) + youtube_url = YoutubeIE._extract_url(body) if youtube_url: - return _url_res(youtube_url, 'Youtube') + return _url_res(youtube_url, YoutubeIE.ie_key()) video_url = self._html_search_regex( r'data-video-url="([^"]+)"', - prefetch_data['embed_code'], 'video URL') + article['embed_code'], 'video URL') return _url_res(video_url, ViceIE.ie_key())