mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-18 07:01:45 +00:00
parent
069c6ccf02
commit
d984a98def
@ -12,6 +12,7 @@ from ..compat import (
|
||||
)
|
||||
from ..utils import (
|
||||
ExtractorError,
|
||||
float_or_none,
|
||||
unified_strdate,
|
||||
int_or_none,
|
||||
qualities,
|
||||
@ -96,6 +97,14 @@ class OdnoklassnikiIE(InfoExtractor):
|
||||
'skip_download': True,
|
||||
},
|
||||
'skip': 'Video has not been found',
|
||||
}, {
|
||||
'note': 'Only available in mobile webpage',
|
||||
'url': 'https://m.ok.ru/video/2361249957145',
|
||||
'info_dict': {
|
||||
'id': '2361249957145',
|
||||
'title': 'Быковское крещение',
|
||||
'duration': 3038.181,
|
||||
},
|
||||
}, {
|
||||
'url': 'http://ok.ru/web-api/video/moviePlayer/20079905452',
|
||||
'only_matching': True,
|
||||
@ -131,13 +140,24 @@ class OdnoklassnikiIE(InfoExtractor):
|
||||
return mobj.group('url')
|
||||
|
||||
def _real_extract(self, url):
|
||||
try:
|
||||
return self._extract_desktop(url)
|
||||
except ExtractorError as e:
|
||||
try:
|
||||
return self._extract_mobile(url)
|
||||
except ExtractorError:
|
||||
# error message of desktop webpage is in English
|
||||
raise e
|
||||
|
||||
def _extract_desktop(self, url):
|
||||
start_time = int_or_none(compat_parse_qs(
|
||||
compat_urllib_parse_urlparse(url).query).get('fromTime', [None])[0])
|
||||
|
||||
video_id = self._match_id(url)
|
||||
|
||||
webpage = self._download_webpage(
|
||||
'http://ok.ru/video/%s' % video_id, video_id)
|
||||
'http://ok.ru/video/%s' % video_id, video_id,
|
||||
note='Downloading desktop webpage')
|
||||
|
||||
error = self._search_regex(
|
||||
r'[^>]+class="vp_video_stub_txt"[^>]*>([^<]+)<',
|
||||
@ -265,3 +285,32 @@ class OdnoklassnikiIE(InfoExtractor):
|
||||
|
||||
info['formats'] = formats
|
||||
return info
|
||||
|
||||
def _extract_mobile(self, url):
|
||||
video_id = self._match_id(url)
|
||||
|
||||
webpage = self._download_webpage(
|
||||
'http://m.ok.ru/video/%s' % video_id, video_id,
|
||||
note='Downloading mobile webpage')
|
||||
|
||||
error = self._search_regex(
|
||||
r'видео</a>\s*<div\s+class="empty">(.+?)</div>',
|
||||
webpage, 'error', default=None)
|
||||
if error:
|
||||
raise ExtractorError(error, expected=True)
|
||||
|
||||
json_data = self._search_regex(
|
||||
r'data-video="(.+?)"', webpage, 'json data')
|
||||
json_data = self._parse_json(unescapeHTML(json_data), video_id) or {}
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': json_data.get('videoName'),
|
||||
'duration': float_or_none(json_data.get('videoDuration'), scale=1000),
|
||||
'thumbnail': json_data.get('videoPosterSrc'),
|
||||
'formats': [{
|
||||
'format_id': 'mobile',
|
||||
'url': json_data.get('videoSrc'),
|
||||
'ext': 'mp4',
|
||||
}]
|
||||
}
|
||||
|
Loading…
Reference in New Issue
Block a user