for key, value in info_dict.items()
if value and key in ('title', 'description', 'uploader', 'upload_date', 'uploader_id', 'location'))
if not all(key in tc.get('info_dict', {}).keys() for key in test_info_dict.keys()):
- sys.stderr.write(u'\n"info_dict": ' + json.dumps(test_info_dict, ensure_ascii=False, indent=2) + u'\n')
+ sys.stderr.write(u'\n"info_dict": ' + json.dumps(test_info_dict, ensure_ascii=False, indent=4) + u'\n')
# Check for the presence of mandatory fields
for key in ('id', 'url', 'title', 'ext'):
IviCompilationIE,
ImdbListIE,
KhanAcademyIE,
+ EveryonesMixtapeIE,
)
self.assertEqual(result['description'], 'How have humans protected their secret messages through history? What has changed today?')
self.assertTrue(len(result['entries']) >= 3)
+ def test_EveryonesMixtape(self):
+ dl = FakeYDL()
+ ie = EveryonesMixtapeIE(dl)
+ result = ie.extract('http://everyonesmixtape.com/#/mix/m7m0jJAbMQi')
+ self.assertIsPlaylist(result)
+ self.assertEqual(result['id'], 'm7m0jJAbMQi')
+ self.assertEqual(result['title'], 'Driving')
+ self.assertEqual(len(result['entries']), 24)
+
if __name__ == '__main__':
unittest.main()
--- /dev/null
+from __future__ import unicode_literals
+
+import re
+
+from .common import InfoExtractor
+from ..utils import (
+ compat_urllib_request,
+ ExtractorError,
+)
+
+
+class EveryonesMixtapeIE(InfoExtractor):
+ _VALID_URL = r'https?://(?:www\.)?everyonesmixtape\.com/#/mix/(?P<id>[0-9a-zA-Z]+)(?:/(?P<songnr>[0-9]))?$'
+
+ _TEST = {
+ 'url': 'http://everyonesmixtape.com/#/mix/m7m0jJAbMQi/5',
+ 'file': '5bfseWNmlds.mp4',
+ "info_dict": {
+ "title": "Passion Pit - \"Sleepyhead\" (Official Music Video)",
+ "uploader": "FKR.TV",
+ "uploader_id": "frenchkissrecords",
+ "description": "Music video for \"Sleepyhead\" from Passion Pit's debut EP Chunk Of Change.\nBuy on iTunes: https://itunes.apple.com/us/album/chunk-of-change-ep/id300087641\n\nDirected by The Wilderness.\n\nhttp://www.passionpitmusic.com\nhttp://www.frenchkissrecords.com",
+ "upload_date": "20081015"
+ },
+ 'params': {
+ 'skip_download': True, # This is simply YouTube
+ }
+ }
+
+ def _real_extract(self, url):
+ mobj = re.match(self._VALID_URL, url)
+ playlist_id = mobj.group('id')
+
+ pllist_url = 'http://everyonesmixtape.com/mixtape.php?a=getMixes&u=-1&linked=%s&explore=' % playlist_id
+ pllist_req = compat_urllib_request.Request(pllist_url)
+ pllist_req.add_header('X-Requested-With', 'XMLHttpRequest')
+
+ playlist_list = self._download_json(
+ pllist_req, playlist_id, note='Downloading playlist metadata')
+ try:
+ playlist_no = next(playlist['id']
+ for playlist in playlist_list
+ if playlist['code'] == playlist_id)
+ except StopIteration:
+ raise ExtractorError('Playlist id not found')
+
+ pl_url = 'http://everyonesmixtape.com/mixtape.php?a=getMix&id=%s&userId=null&code=' % playlist_no
+ pl_req = compat_urllib_request.Request(pl_url)
+ pl_req.add_header('X-Requested-With', 'XMLHttpRequest')
+ playlist = self._download_json(
+ pl_req, playlist_id, note='Downloading playlist info')
+
+ entries = [{
+ '_type': 'url',
+ 'url': t['url'],
+ 'title': t['title'],
+ } for t in playlist['tracks']]
+
+ if mobj.group('songnr'):
+ songnr = int(mobj.group('songnr')) - 1
+ return entries[songnr]
+
+ playlist_title = playlist['mixData']['name']
+ return {
+ '_type': 'playlist',
+ 'id': playlist_id,
+ 'title': playlist_title,
+ 'entries': entries,
+ }