diff options
Diffstat (limited to 'youtube_dl/extractor/izlesene.py')
-rw-r--r-- | youtube_dl/extractor/izlesene.py | 18 |
1 files changed, 9 insertions, 9 deletions
diff --git a/youtube_dl/extractor/izlesene.py b/youtube_dl/extractor/izlesene.py index a83dd249f..d16d483ee 100644 --- a/youtube_dl/extractor/izlesene.py +++ b/youtube_dl/extractor/izlesene.py @@ -5,11 +5,11 @@ import re from .common import InfoExtractor from ..utils import ( - get_element_by_id, - parse_iso8601, determine_ext, - int_or_none, float_or_none, + get_element_by_id, + int_or_none, + parse_iso8601, str_to_int, ) @@ -30,7 +30,7 @@ class IzleseneIE(InfoExtractor): 'description': 'md5:253753e2655dde93f59f74b572454f6d', 'thumbnail': 're:^http://.*\.jpg', 'uploader_id': 'pelikzzle', - 'timestamp': 1404298698, + 'timestamp': 1404302298, 'upload_date': '20140702', 'duration': 95.395, 'age_limit': 0, @@ -46,7 +46,7 @@ class IzleseneIE(InfoExtractor): 'description': 'Tarkan Dortmund 2006 Konseri', 'thumbnail': 're:^http://.*\.jpg', 'uploader_id': 'parlayankiz', - 'timestamp': 1163318593, + 'timestamp': 1163322193, 'upload_date': '20061112', 'duration': 253.666, 'age_limit': 0, @@ -55,15 +55,15 @@ class IzleseneIE(InfoExtractor): ] def _real_extract(self, url): - mobj = re.match(self._VALID_URL, url) - video_id = mobj.group('id') - url = 'http://www.izlesene.com/video/%s' % video_id + video_id = self._match_id(url) + url = 'http://www.izlesene.com/video/%s' % video_id webpage = self._download_webpage(url, video_id) title = self._og_search_title(webpage) description = self._og_search_description(webpage) - thumbnail = self._og_search_thumbnail(webpage) + thumbnail = self._proto_relative_url( + self._og_search_thumbnail(webpage), scheme='http:') uploader = self._html_search_regex( r"adduserUsername\s*=\s*'([^']+)';", |