X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/a070911bf98c751e4425008f1335cc57c86d30fc..c64c42cb1f9487398f8b91a51266ff6e79115b90:/youtube_dl/extractor/condenast.py?ds=sidebyside diff --git a/youtube_dl/extractor/condenast.py b/youtube_dl/extractor/condenast.py index 3db4db4..8d8f605 100644 --- a/youtube_dl/extractor/condenast.py +++ b/youtube_dl/extractor/condenast.py @@ -2,16 +2,20 @@ from __future__ import unicode_literals import re -import json from .common import InfoExtractor from ..compat import ( - compat_urllib_parse, compat_urllib_parse_urlparse, compat_urlparse, ) from ..utils import ( orderedSet, + remove_end, + extract_attributes, + mimetype2ext, + determine_ext, + int_or_none, + parse_iso8601, ) @@ -24,21 +28,33 @@ class CondeNastIE(InfoExtractor): # The keys are the supported sites and the values are the name to be shown # to the user and in the extractor description. _SITES = { - 'wired': 'WIRED', + 'allure': 'Allure', + 'architecturaldigest': 'Architectural Digest', + 'arstechnica': 'Ars Technica', + 'bonappetit': 'Bon Appétit', + 'brides': 'Brides', + 'cnevids': 'Condé Nast', + 'cntraveler': 'Condé Nast Traveler', + 'details': 'Details', + 'epicurious': 'Epicurious', + 'glamour': 'Glamour', + 'golfdigest': 'Golf Digest', 'gq': 'GQ', + 'newyorker': 'The New Yorker', + 'self': 'SELF', + 'teenvogue': 'Teen Vogue', + 'vanityfair': 'Vanity Fair', 'vogue': 'Vogue', - 'glamour': 'Glamour', + 'wired': 'WIRED', 'wmagazine': 'W Magazine', - 'vanityfair': 'Vanity Fair', - 'cnevids': 'Condé Nast', } - _VALID_URL = r'http://(video|www|player)\.(?P%s)\.com/(?Pwatch|series|video|embed)/(?P[^/?#]+)' % '|'.join(_SITES.keys()) + _VALID_URL = r'https?://(?:video|www|player)\.(?P%s)\.com/(?Pwatch|series|video|embed(?:js)?)/(?P[^/?#]+)' % '|'.join(_SITES.keys()) IE_DESC = 'Condé Nast media group: %s' % ', '.join(sorted(_SITES.values())) - EMBED_URL = r'(?:https?:)?//player\.(?P%s)\.com/(?Pembed)/.+?' % '|'.join(_SITES.keys()) + EMBED_URL = r'(?:https?:)?//player\.(?P%s)\.com/(?Pembed(?:js)?)/.+?' % '|'.join(_SITES.keys()) - _TEST = { + _TESTS = [{ 'url': 'http://video.wired.com/watch/3d-printed-speakers-lit-with-led', 'md5': '1921f713ed48aabd715691f774c451f7', 'info_dict': { @@ -46,70 +62,112 @@ class CondeNastIE(InfoExtractor): 'ext': 'mp4', 'title': '3D Printed Speakers Lit With LED', 'description': 'Check out these beautiful 3D printed LED speakers. You can\'t actually buy them, but LumiGeek is working on a board that will let you make you\'re own.', + 'uploader': 'wired', + 'upload_date': '20130314', + 'timestamp': 1363219200, } - } + }, { + # JS embed + 'url': 'http://player.cnevids.com/embedjs/55f9cf8b61646d1acf00000c/5511d76261646d5566020000.js', + 'md5': 'f1a6f9cafb7083bab74a710f65d08999', + 'info_dict': { + 'id': '55f9cf8b61646d1acf00000c', + 'ext': 'mp4', + 'title': '3D printed TSA Travel Sentry keys really do open TSA locks', + 'uploader': 'arstechnica', + 'upload_date': '20150916', + 'timestamp': 1442434955, + } + }] def _extract_series(self, url, webpage): - title = self._html_search_regex(r'
.*?

(.+?)

', - webpage, 'series title', flags=re.DOTALL) + title = self._html_search_regex( + r'(?s)
.*?

(.+?)

', + webpage, 'series title') url_object = compat_urllib_parse_urlparse(url) base_url = '%s://%s' % (url_object.scheme, url_object.netloc) - m_paths = re.finditer(r'

.*?.*?(.+?)

', - r'
(.+?)
', - ], - webpage, 'description', fatal=False, flags=re.DOTALL) + query = {} + params = self._search_regex( + r'(?s)var params = {(.+?)}[;,]', webpage, 'player params', default=None) + if params: + query.update({ + 'videoId': self._search_regex(r'videoId: [\'"](.+?)[\'"]', params, 'video id'), + 'playerId': self._search_regex(r'playerId: [\'"](.+?)[\'"]', params, 'player id'), + 'target': self._search_regex(r'target: [\'"](.+?)[\'"]', params, 'target'), + }) else: - description = None - params = self._search_regex(r'var params = {(.+?)}[;,]', webpage, - 'player params', flags=re.DOTALL) - video_id = self._search_regex(r'videoId: [\'"](.+?)[\'"]', params, 'video id') - player_id = self._search_regex(r'playerId: [\'"](.+?)[\'"]', params, 'player id') - target = self._search_regex(r'target: [\'"](.+?)[\'"]', params, 'target') - data = compat_urllib_parse.urlencode({'videoId': video_id, - 'playerId': player_id, - 'target': target, - }) - base_info_url = self._search_regex(r'url = [\'"](.+?)[\'"][,;]', - webpage, 'base info url', - default='http://player.cnevids.com/player/loader.js?') - info_url = base_info_url + data - info_page = self._download_webpage(info_url, video_id, - 'Downloading video info') - video_info = self._search_regex(r'var video = ({.+?});', info_page, 'video info') - video_info = json.loads(video_info) - - formats = [{ - 'format_id': '%s-%s' % (fdata['type'].split('/')[-1], fdata['quality']), - 'url': fdata['src'], - 'ext': fdata['type'].split('/')[-1], - 'quality': 1 if fdata['quality'] == 'high' else 0, - } for fdata in video_info['sources'][0]] + params = extract_attributes(self._search_regex( + r'(<[^>]+data-js="video-player"[^>]+>)', + webpage, 'player params element')) + query.update({ + 'videoId': params['data-video'], + 'playerId': params['data-player'], + 'target': params['id'], + }) + video_id = query['videoId'] + video_info = None + info_page = self._download_webpage( + 'http://player.cnevids.com/player/video.js', + video_id, 'Downloading video info', query=query, fatal=False) + if info_page: + video_info = self._parse_json(self._search_regex( + r'loadCallback\(({.+})\)', info_page, 'video info'), video_id)['video'] + else: + info_page = self._download_webpage( + 'http://player.cnevids.com/player/loader.js', + video_id, 'Downloading loader info', query=query) + video_info = self._parse_json(self._search_regex( + r'var\s+video\s*=\s*({.+?});', info_page, 'video info'), video_id) + title = video_info['title'] + + formats = [] + for fdata in video_info.get('sources', [{}])[0]: + src = fdata.get('src') + if not src: + continue + ext = mimetype2ext(fdata.get('type')) or determine_ext(src) + quality = fdata.get('quality') + formats.append({ + 'format_id': ext + ('-%s' % quality if quality else ''), + 'url': src, + 'ext': ext, + 'quality': 1 if quality == 'high' else 0, + }) self._sort_formats(formats) - return { + info = self._search_json_ld( + webpage, video_id, fatal=False) if url_type != 'embed' else {} + info.update({ 'id': video_id, 'formats': formats, - 'title': video_info['title'], - 'thumbnail': video_info['poster_frame'], - 'description': description, - } + 'title': title, + 'thumbnail': video_info.get('poster_frame'), + 'uploader': video_info.get('brand'), + 'duration': int_or_none(video_info.get('duration')), + 'tags': video_info.get('tags'), + 'series': video_info.get('series_title'), + 'season': video_info.get('season_title'), + 'timestamp': parse_iso8601(video_info.get('premiere_date')), + }) + return info def _real_extract(self, url): - mobj = re.match(self._VALID_URL, url) - site = mobj.group('site') - url_type = mobj.group('type') - item_id = mobj.group('id') + site, url_type, item_id = re.match(self._VALID_URL, url).groups() + + # Convert JS embed to regular embed + if url_type == 'embedjs': + parsed_url = compat_urlparse.urlparse(url) + url = compat_urlparse.urlunparse(parsed_url._replace( + path=remove_end(parsed_url.path, '.js').replace('/embedjs/', '/embed/'))) + url_type = 'embed' self.to_screen('Extracting from %s with the Condé Nast extractor' % self._SITES[site]) webpage = self._download_webpage(url, item_id)