aboutsummaryrefslogtreecommitdiff
path: root/youtube_dl/extractor
diff options
context:
space:
mode:
Diffstat (limited to 'youtube_dl/extractor')
-rw-r--r--youtube_dl/extractor/vevo.py39
1 files changed, 19 insertions, 20 deletions
diff --git a/youtube_dl/extractor/vevo.py b/youtube_dl/extractor/vevo.py
index a4b26a26f..f0673972c 100644
--- a/youtube_dl/extractor/vevo.py
+++ b/youtube_dl/extractor/vevo.py
@@ -1,3 +1,5 @@
+from __future__ import unicode_literals
+
import re
import json
import xml.etree.ElementTree
@@ -22,16 +24,16 @@ class VevoIE(InfoExtractor):
vevo:)
(?P<id>[^&?#]+)'''
_TESTS = [{
- u'url': u'http://www.vevo.com/watch/hurts/somebody-to-die-for/GB1101300280',
- u'file': u'GB1101300280.mp4',
- u"md5": u"06bea460acb744eab74a9d7dcb4bfd61",
- u'info_dict': {
- u"upload_date": u"20130624",
- u"uploader": u"Hurts",
- u"title": u"Somebody to Die For",
- u"duration": 230.12,
- u"width": 1920,
- u"height": 1080,
+ 'url': 'http://www.vevo.com/watch/hurts/somebody-to-die-for/GB1101300280',
+ 'file': 'GB1101300280.mp4',
+ "md5": "06bea460acb744eab74a9d7dcb4bfd61",
+ 'info_dict': {
+ "upload_date": "20130624",
+ "uploader": "Hurts",
+ "title": "Somebody to Die For",
+ "duration": 230.12,
+ "width": 1920,
+ "height": 1080,
}
}]
_SMIL_BASE_URL = 'http://smil.lvl3.vevo.com/'
@@ -44,7 +46,7 @@ class VevoIE(InfoExtractor):
if version['version'] > last_version['version']:
last_version = version
if last_version['version'] == -1:
- raise ExtractorError(u'Unable to extract last version of the video')
+ raise ExtractorError('Unable to extract last version of the video')
renditions = xml.etree.ElementTree.fromstring(last_version['data'])
formats = []
@@ -85,7 +87,7 @@ class VevoIE(InfoExtractor):
format_url = self._SMIL_BASE_URL + m.group('path')
formats.append({
'url': format_url,
- 'format_id': u'SMIL_' + m.group('cbr'),
+ 'format_id': 'SMIL_' + m.group('cbr'),
'vcodec': m.group('vcodec'),
'acodec': m.group('acodec'),
'vbr': int(m.group('vbr')),
@@ -101,26 +103,25 @@ class VevoIE(InfoExtractor):
video_id = mobj.group('id')
json_url = 'http://videoplayer.vevo.com/VideoService/AuthenticateVideo?isrc=%s' % video_id
- info_json = self._download_webpage(json_url, video_id, u'Downloading json info')
- video_info = json.loads(info_json)['video']
+ video_info = self._download_json(json_url, video_id)['video']
formats = self._formats_from_json(video_info)
try:
smil_url = '%s/Video/V2/VFILE/%s/%sr.smil' % (
self._SMIL_BASE_URL, video_id, video_id.lower())
smil_xml = self._download_webpage(smil_url, video_id,
- u'Downloading SMIL info')
+ 'Downloading SMIL info')
formats.extend(self._formats_from_smil(smil_xml))
except ExtractorError as ee:
if not isinstance(ee.cause, compat_HTTPError):
raise
self._downloader.report_warning(
- u'Cannot download SMIL information, falling back to JSON ..')
+ 'Cannot download SMIL information, falling back to JSON ..')
timestamp_ms = int(self._search_regex(
- r'/Date\((\d+)\)/', video_info['launchDate'], u'launch date'))
+ r'/Date\((\d+)\)/', video_info['launchDate'], 'launch date'))
upload_date = datetime.datetime.fromtimestamp(timestamp_ms // 1000)
- info = {
+ return {
'id': video_id,
'title': video_info['title'],
'formats': formats,
@@ -129,5 +130,3 @@ class VevoIE(InfoExtractor):
'uploader': video_info['mainArtists'][0]['artistName'],
'duration': video_info['duration'],
}
-
- return info