mirror of
https://github.com/yt-dlp/yt-dlp
synced 2024-12-30 22:24:28 +01:00
cfcaf64a4b
Authored by: pabs3
67 lines
2.7 KiB
Python
67 lines
2.7 KiB
Python
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
|
|
|
|
class RTRFMIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?rtrfm\.com\.au/(?:shows|show-episode)/(?P<id>[^/?\#&]+)'
|
|
_TESTS = [
|
|
{
|
|
'url': 'https://rtrfm.com.au/shows/breakfast/',
|
|
'md5': '46168394d3a5ce237cf47e85d0745413',
|
|
'info_dict': {
|
|
'id': 'breakfast-2021-11-16',
|
|
'ext': 'mp3',
|
|
'series': 'Breakfast with Taylah',
|
|
'title': r're:^Breakfast with Taylah \d{4}-\d{2}-\d{2}$',
|
|
'description': 'md5:0979c3ab1febfbec3f1ccb743633c611',
|
|
},
|
|
'skip': 'ID and md5 changes daily',
|
|
},
|
|
{
|
|
'url': 'https://rtrfm.com.au/show-episode/breakfast-2021-11-11/',
|
|
'md5': '396bedf1e40f96c62b30d4999202a790',
|
|
'info_dict': {
|
|
'id': 'breakfast-2021-11-11',
|
|
'ext': 'mp3',
|
|
'series': 'Breakfast with Taylah',
|
|
'title': 'Breakfast with Taylah 2021-11-11',
|
|
'description': 'md5:0979c3ab1febfbec3f1ccb743633c611',
|
|
},
|
|
},
|
|
{
|
|
'url': 'https://rtrfm.com.au/show-episode/breakfast-2020-06-01/',
|
|
'md5': '594027f513ec36a24b15d65007a24dff',
|
|
'info_dict': {
|
|
'id': 'breakfast-2020-06-01',
|
|
'ext': 'mp3',
|
|
'series': 'Breakfast with Taylah',
|
|
'title': 'Breakfast with Taylah 2020-06-01',
|
|
'description': r're:^Breakfast with Taylah ',
|
|
},
|
|
'skip': 'This audio has expired',
|
|
},
|
|
]
|
|
|
|
def _real_extract(self, url):
|
|
display_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, display_id)
|
|
show, date, title = self._search_regex(
|
|
r'''\.playShow(?:From)?\(['"](?P<show>[^'"]+)['"],\s*['"](?P<date>[0-9]{4}-[0-9]{2}-[0-9]{2})['"],\s*['"](?P<title>[^'"]+)['"]''',
|
|
webpage, 'details', group=('show', 'date', 'title'))
|
|
url = self._download_json(
|
|
'https://restreams.rtrfm.com.au/rzz',
|
|
show, 'Downloading MP3 URL', query={'n': show, 'd': date})['u']
|
|
# This is the only indicator of an error until trying to download the URL and
|
|
# downloads of mp4 URLs always fail (403 for current episodes, 404 for missing).
|
|
if '.mp4' in url:
|
|
url = None
|
|
self.raise_no_formats('Expired or no episode on this date', expected=True)
|
|
return {
|
|
'id': '%s-%s' % (show, date),
|
|
'title': '%s %s' % (title, date),
|
|
'series': title,
|
|
'url': url,
|
|
'release_date': date,
|
|
'description': self._og_search_description(webpage),
|
|
}
|