mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-30 03:33:02 +00:00
add96eb9f8
Authored by: seproDev Reviewed-by: bashonly <88596187+bashonly@users.noreply.github.com> Reviewed-by: Simon Sawicki <contact@grub4k.xyz>
67 lines
2.3 KiB
Python
67 lines
2.3 KiB
Python
from .common import InfoExtractor
|
||
from ..utils import (
|
||
determine_ext,
|
||
unescapeHTML,
|
||
)
|
||
|
||
|
||
class CJSWIE(InfoExtractor):
|
||
_VALID_URL = r'https?://(?:www\.)?cjsw\.com/program/(?P<program>[^/]+)/episode/(?P<id>\d+)'
|
||
_TESTS = [{
|
||
'url': 'http://cjsw.com/program/freshly-squeezed/episode/20170620',
|
||
'md5': 'cee14d40f1e9433632c56e3d14977120',
|
||
'info_dict': {
|
||
'id': '91d9f016-a2e7-46c5-8dcb-7cbcd7437c41',
|
||
'ext': 'mp3',
|
||
'title': 'Freshly Squeezed – Episode June 20, 2017',
|
||
'description': 'md5:c967d63366c3898a80d0c7b0ff337202',
|
||
'series': 'Freshly Squeezed',
|
||
'episode_id': '20170620',
|
||
},
|
||
}, {
|
||
# no description
|
||
'url': 'http://cjsw.com/program/road-pops/episode/20170707/',
|
||
'only_matching': True,
|
||
}]
|
||
|
||
def _real_extract(self, url):
|
||
mobj = self._match_valid_url(url)
|
||
program, episode_id = mobj.group('program', 'id')
|
||
audio_id = f'{program}/{episode_id}'
|
||
|
||
webpage = self._download_webpage(url, episode_id)
|
||
|
||
title = unescapeHTML(self._search_regex(
|
||
(r'<h1[^>]+class=["\']episode-header__title["\'][^>]*>(?P<title>[^<]+)',
|
||
r'data-audio-title=(["\'])(?P<title>(?:(?!\1).)+)\1'),
|
||
webpage, 'title', group='title'))
|
||
|
||
audio_url = self._search_regex(
|
||
r'<button[^>]+data-audio-src=(["\'])(?P<url>(?:(?!\1).)+)\1',
|
||
webpage, 'audio url', group='url')
|
||
|
||
audio_id = self._search_regex(
|
||
r'/([\da-f]{8}-[\da-f]{4}-[\da-f]{4}-[\da-f]{4}-[\da-f]{12})\.mp3',
|
||
audio_url, 'audio id', default=audio_id)
|
||
|
||
formats = [{
|
||
'url': audio_url,
|
||
'ext': determine_ext(audio_url, 'mp3'),
|
||
'vcodec': 'none',
|
||
}]
|
||
|
||
description = self._html_search_regex(
|
||
r'<p>(?P<description>.+?)</p>', webpage, 'description',
|
||
default=None)
|
||
series = self._search_regex(
|
||
r'data-showname=(["\'])(?P<name>(?:(?!\1).)+)\1', webpage,
|
||
'series', default=program, group='name')
|
||
|
||
return {
|
||
'id': audio_id,
|
||
'title': title,
|
||
'description': description,
|
||
'formats': formats,
|
||
'series': series,
|
||
'episode_id': episode_id,
|
||
}
|