Rapsys Git
/
youtubedl
/ blobdiff
commit
grep
author
committer
pickaxe
?
search:
re
summary
|
shortlog
|
log
|
commit
|
commitdiff
|
tree
raw
|
inline
| side by side
Prepare to release.
[youtubedl]
/
youtube_dl
/
extractor
/
ellentv.py
diff --git
a/youtube_dl/extractor/ellentv.py
b/youtube_dl/extractor/ellentv.py
index 74bbc5c51c576880465e76453604891b6b5f48ca..e0a13dd76cd4069ff40bcc4c771fd5b9eeb75498 100644
(file)
--- a/
youtube_dl/extractor/ellentv.py
+++ b/
youtube_dl/extractor/ellentv.py
@@
-1,13
+1,9
@@
# coding: utf-8
from __future__ import unicode_literals
# coding: utf-8
from __future__ import unicode_literals
-import json
-
from .common import InfoExtractor
from .common import InfoExtractor
-from ..utils import (
- ExtractorError,
- NO_DEFAULT,
-)
+from .kaltura import KalturaIE
+from ..utils import NO_DEFAULT
class EllenTVIE(InfoExtractor):
class EllenTVIE(InfoExtractor):
@@
-65,7
+61,7
@@
class EllenTVIE(InfoExtractor):
if partner_id and kaltura_id:
break
if partner_id and kaltura_id:
break
- return self.url_result('kaltura:%s:%s' % (partner_id, kaltura_id),
'Kaltura'
)
+ return self.url_result('kaltura:%s:%s' % (partner_id, kaltura_id),
KalturaIE.ie_key()
)
class EllenTVClipsIE(InfoExtractor):
class EllenTVClipsIE(InfoExtractor):
@@
-77,14
+73,14
@@
class EllenTVClipsIE(InfoExtractor):
'id': 'meryl-streep-vanessa-hudgens',
'title': 'Meryl Streep, Vanessa Hudgens',
},
'id': 'meryl-streep-vanessa-hudgens',
'title': 'Meryl Streep, Vanessa Hudgens',
},
- 'playlist_mincount':
7
,
+ 'playlist_mincount':
5
,
}
def _real_extract(self, url):
playlist_id = self._match_id(url)
webpage = self._download_webpage(url, playlist_id)
}
def _real_extract(self, url):
playlist_id = self._match_id(url)
webpage = self._download_webpage(url, playlist_id)
- playlist = self._extract_playlist(webpage)
+ playlist = self._extract_playlist(webpage
, playlist_id
)
return {
'_type': 'playlist',
return {
'_type': 'playlist',
@@
-93,16
+89,13
@@
class EllenTVClipsIE(InfoExtractor):
'entries': self._extract_entries(playlist)
}
'entries': self._extract_entries(playlist)
}
- def _extract_playlist(self, webpage):
+ def _extract_playlist(self, webpage
, playlist_id
):
json_string = self._search_regex(r'playerView.addClips\(\[\{(.*?)\}\]\);', webpage, 'json')
json_string = self._search_regex(r'playerView.addClips\(\[\{(.*?)\}\]\);', webpage, 'json')
- try:
- return json.loads('[{' + json_string + '}]')
- except ValueError as ve:
- raise ExtractorError('Failed to download JSON', cause=ve)
+ return self._parse_json('[{' + json_string + '}]', playlist_id)
def _extract_entries(self, playlist):
return [
self.url_result(
'kaltura:%s:%s' % (item['kaltura_partner_id'], item['kaltura_entry_id']),
def _extract_entries(self, playlist):
return [
self.url_result(
'kaltura:%s:%s' % (item['kaltura_partner_id'], item['kaltura_entry_id']),
-
'Kaltura'
)
+
KalturaIE.ie_key(), video_id=item['kaltura_entry_id']
)
for item in playlist]
for item in playlist]