yt-dlp/yt_dlp/extractor/ondemandkorea.py
Julien Hadley Jack 10e2eb4f81
[extractor/ondemandkorea] Update jw_config regex (#5040)
Authored by: julien-hadleyjack
2022-09-28 14:34:03 +05:30

88 lines
3.3 KiB
Python

import re
from .common import InfoExtractor
from ..utils import (
ExtractorError,
js_to_json,
)
class OnDemandKoreaIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?ondemandkorea\.com/(?P<id>[^/]+)\.html'
_GEO_COUNTRIES = ['US', 'CA']
_TESTS = [{
'url': 'https://www.ondemandkorea.com/ask-us-anything-e351.html',
'info_dict': {
'id': 'ask-us-anything-e351',
'ext': 'mp4',
'title': 'Ask Us Anything : Jung Sung-ho, Park Seul-gi, Kim Bo-min, Yang Seung-won - 09/24/2022',
'description': 'A talk show/game show with a school theme where celebrity guests appear as “transfer students.”',
'thumbnail': r're:^https?://.*\.jpg$',
},
'params': {
'skip_download': 'm3u8 download'
}
}, {
'url': 'https://www.ondemandkorea.com/work-later-drink-now-e1.html',
'info_dict': {
'id': 'work-later-drink-now-e1',
'ext': 'mp4',
'title': 'Work Later, Drink Now : E01',
'description': 'Work Later, Drink First follows three women who find solace in a glass of liquor at the end of the day. So-hee, who gets comfort from a cup of soju af',
'thumbnail': r're:^https?://.*\.png$',
'subtitles': {
'English': 'mincount:1',
},
},
'params': {
'skip_download': 'm3u8 download'
}
}]
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id, fatal=False)
if not webpage:
# Page sometimes returns captcha page with HTTP 403
raise ExtractorError(
'Unable to access page. You may have been blocked.',
expected=True)
if 'msg_block_01.png' in webpage:
self.raise_geo_restricted(
msg='This content is not available in your region',
countries=self._GEO_COUNTRIES)
if 'This video is only available to ODK PLUS members.' in webpage:
raise ExtractorError(
'This video is only available to ODK PLUS members.',
expected=True)
if 'ODK PREMIUM Members Only' in webpage:
raise ExtractorError(
'This video is only available to ODK PREMIUM members.',
expected=True)
title = self._search_regex(
r'class=["\']episode_title["\'][^>]*>([^<]+)',
webpage, 'episode_title', fatal=False) or self._og_search_title(webpage)
jw_config = self._parse_json(
self._search_regex((
r'(?P<options>{\s*[\'"]tracks[\'"].*?})[)\];]+$',
r'playlist\s*=\s*\[(?P<options>.+)];?$',
r'odkPlayer\.init.*?(?P<options>{[^;]+}).*?;',
), webpage, 'jw config', flags=re.MULTILINE | re.DOTALL, group='options'),
video_id, transform_source=js_to_json)
info = self._parse_jwplayer_data(
jw_config, video_id, require_title=False, m3u8_id='hls',
base_url=url)
info.update({
'title': title,
'description': self._og_search_description(webpage),
'thumbnail': self._og_search_thumbnail(webpage)
})
return info