X-Git-Url: https://git.rapsys.eu/youtubedl/blobdiff_plain/90cd207d52d77e075e62cc256cc39dceee05bbf3..382a868cee069f08aacf0b89c9d689ec420d6b2c:/youtube_dl/extractor/myspace.py diff --git a/youtube_dl/extractor/myspace.py b/youtube_dl/extractor/myspace.py index 050f54a..83414a2 100644 --- a/youtube_dl/extractor/myspace.py +++ b/youtube_dl/extractor/myspace.py @@ -1,48 +1,180 @@ +# encoding: utf-8 +from __future__ import unicode_literals + import re import json from .common import InfoExtractor -from ..utils import ( +from ..compat import ( compat_str, ) +from ..utils import ExtractorError class MySpaceIE(InfoExtractor): - _VALID_URL = r'https?://myspace\.com/([^/]+)/video/[^/]+/(?P\d+)' - - _TEST = { - u'url': u'https://myspace.com/coldplay/video/viva-la-vida/100008689', - u'info_dict': { - u'id': u'100008689', - u'ext': u'flv', - u'title': u'Viva La Vida', - u'description': u'The official Viva La Vida video, directed by Hype Williams', - u'uploader': u'Coldplay', - u'uploader_id': u'coldplay', + _VALID_URL = r'https?://myspace\.com/([^/]+)/(?Pvideo/[^/]+/|music/song/.*?)(?P\d+)' + + _TESTS = [ + { + 'url': 'https://myspace.com/fiveminutestothestage/video/little-big-town/109594919', + 'info_dict': { + 'id': '109594919', + 'ext': 'flv', + 'title': 'Little Big Town', + 'description': 'This country quartet was all smiles while playing a sold out show at the Pacific Amphitheatre in Orange County, California.', + 'uploader': 'Five Minutes to the Stage', + 'uploader_id': 'fiveminutestothestage', + }, + 'params': { + # rtmp download + 'skip_download': True, + }, }, - u'params': { - # rtmp download - u'skip_download': True, + # songs + { + 'url': 'https://myspace.com/killsorrow/music/song/of-weakened-soul...-93388656-103880681', + 'info_dict': { + 'id': '93388656', + 'ext': 'flv', + 'title': 'Of weakened soul...', + 'uploader': 'Killsorrow', + 'uploader_id': 'killsorrow', + }, + 'params': { + # rtmp download + 'skip_download': True, + }, + }, { + 'add_ie': ['Vevo'], + 'url': 'https://myspace.com/threedaysgrace/music/song/animal-i-have-become-28400208-28218041', + 'info_dict': { + 'id': 'USZM20600099', + 'ext': 'mp4', + 'title': 'Animal I Have Become', + 'uploader': 'Three Days Grace', + 'timestamp': int, + 'upload_date': '20060502', + }, + 'skip': 'VEVO is only available in some countries', + }, { + 'add_ie': ['Youtube'], + 'url': 'https://myspace.com/starset2/music/song/first-light-95799905-106964426', + 'info_dict': { + 'id': 'ypWvQgnJrSU', + 'ext': 'mp4', + 'title': 'Starset - First Light', + 'description': 'md5:2d5db6c9d11d527683bcda818d332414', + 'uploader': 'Jacob Soren', + 'uploader_id': 'SorenPromotions', + 'upload_date': '20140725', + } }, - } + ] def _real_extract(self, url): mobj = re.match(self._VALID_URL, url) video_id = mobj.group('id') webpage = self._download_webpage(url, video_id) - context = json.loads(self._search_regex(r'context = ({.*?});', webpage, - u'context')) - video = context['video'] - rtmp_url, play_path = video['streamUrl'].split(';', 1) + player_url = self._search_regex( + r'playerSwf":"([^"?]*)', webpage, 'player URL') - return { - 'id': compat_str(video['mediaId']), - 'title': video['title'], + if mobj.group('mediatype').startswith('music/song'): + # songs don't store any useful info in the 'context' variable + song_data = self._search_regex( + r'''.*?)\1''' % name, + song_data, name, default='', group='data') + streamUrl = search_data('stream-url') + if not streamUrl: + vevo_id = search_data('vevo-id') + youtube_id = search_data('youtube-id') + if vevo_id: + self.to_screen('Vevo video detected: %s' % vevo_id) + return self.url_result('vevo:%s' % vevo_id, ie='Vevo') + elif youtube_id: + self.to_screen('Youtube video detected: %s' % youtube_id) + return self.url_result(youtube_id, ie='Youtube') + else: + raise ExtractorError( + 'Found song but don\'t know how to download it') + info = { + 'id': video_id, + 'title': self._og_search_title(webpage), + 'uploader': search_data('artist-name'), + 'uploader_id': search_data('artist-username'), + 'thumbnail': self._og_search_thumbnail(webpage), + } + else: + context = json.loads(self._search_regex( + r'context = ({.*?});', webpage, 'context')) + video = context['video'] + streamUrl = video['streamUrl'] + info = { + 'id': compat_str(video['mediaId']), + 'title': video['title'], + 'description': video['description'], + 'thumbnail': video['imageUrl'], + 'uploader': video['artistName'], + 'uploader_id': video['artistUsername'], + } + + rtmp_url, play_path = streamUrl.split(';', 1) + info.update({ 'url': rtmp_url, 'play_path': play_path, + 'player_url': player_url, 'ext': 'flv', - 'description': video['description'], - 'thumbnail': video['imageUrl'], - 'uploader': video['artistName'], - 'uploader_id': video['artistUsername'], + }) + return info + + +class MySpaceAlbumIE(InfoExtractor): + IE_NAME = 'MySpace:album' + _VALID_URL = r'https?://myspace\.com/([^/]+)/music/album/(?P.*-)(?P<id>\d+)' + + _TESTS = [{ + 'url': 'https://myspace.com/starset2/music/album/transmissions-19455773', + 'info_dict': { + 'title': 'Transmissions', + 'id': '19455773', + }, + 'playlist_count': 14, + 'skip': 'this album is only available in some countries', + }, { + 'url': 'https://myspace.com/killsorrow/music/album/the-demo-18596029', + 'info_dict': { + 'title': 'The Demo', + 'id': '18596029', + }, + 'playlist_count': 5, + }] + + def _real_extract(self, url): + mobj = re.match(self._VALID_URL, url) + playlist_id = mobj.group('id') + display_id = mobj.group('title') + playlist_id + webpage = self._download_webpage(url, display_id) + tracks_paths = re.findall(r'"music:song" content="(.*?)"', webpage) + if not tracks_paths: + raise ExtractorError( + '%s: No songs found, try using proxy' % display_id, + expected=True) + entries = [ + self.url_result(t_path, ie=MySpaceIE.ie_key()) + for t_path in tracks_paths] + return { + '_type': 'playlist', + 'id': playlist_id, + 'display_id': display_id, + 'title': self._og_search_title(webpage), + 'entries': entries, }