diff options
Diffstat (limited to 'youtube_dl')
-rw-r--r-- | youtube_dl/extractor/karaoketv.py | 32 |
1 files changed, 13 insertions, 19 deletions
diff --git a/youtube_dl/extractor/karaoketv.py b/youtube_dl/extractor/karaoketv.py index 4d50308cc..4856200d8 100644 --- a/youtube_dl/extractor/karaoketv.py +++ b/youtube_dl/extractor/karaoketv.py @@ -1,12 +1,12 @@ # coding: utf-8 from __future__ import unicode_literals -import re -import json -import sys - from .common import InfoExtractor -from ..utils import compat_urllib_parse, ExtractorError +from ..compat import compat_urllib_parse +from ..utils import ( + ExtractorError, + js_to_json, +) class KaraoketvIE(InfoExtractor): @@ -21,22 +21,16 @@ class KaraoketvIE(InfoExtractor): } def _real_extract(self, url): - - # BUG: SSL23_GET_SERVER_HELLO:unknown protocol - if sys.hexversion < 0x03000000: - raise ExtractorError("Only python 3 supported.\n") - - mobj = re.match(self._VALID_URL, url) - - video_id = mobj.group('id') - + video_id = self._match_id(url) webpage = self._download_webpage(url, video_id) - settings_json = compat_urllib_parse.unquote_plus(self._search_regex(r'config=(.*)', self._og_search_video_url(webpage ,video_id), '')) - - urls_info_webpage = self._download_webpage(settings_json, 'Downloading settings json') + page_video_url = self._og_search_video_url(webpage, video_id) + config_json = compat_urllib_parse.unquote_plus(self._search_regex( + r'config=(.*)', page_video_url, 'configuration')) - urls_info_json = json.loads(urls_info_webpage.replace('\'', '"')) + urls_info_json = self._download_json( + config_json, video_id, 'Downloading configuration', + transform_source=js_to_json) url = urls_info_json['playlist'][0]['url'] @@ -44,4 +38,4 @@ class KaraoketvIE(InfoExtractor): 'id': video_id, 'title': self._og_search_title(webpage), 'url': url, - }
\ No newline at end of file + } |