mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-24 02:35:12 +00:00
add96eb9f8
Authored by: seproDev Reviewed-by: bashonly <88596187+bashonly@users.noreply.github.com> Reviewed-by: Simon Sawicki <contact@grub4k.xyz>
85 lines
3.5 KiB
Python
85 lines
3.5 KiB
Python
from .common import InfoExtractor
|
|
from ..utils import (
|
|
clean_html,
|
|
clean_podcast_url,
|
|
get_element_by_class,
|
|
int_or_none,
|
|
parse_iso8601,
|
|
try_get,
|
|
)
|
|
|
|
|
|
class ApplePodcastsIE(InfoExtractor):
|
|
_VALID_URL = r'https?://podcasts\.apple\.com/(?:[^/]+/)?podcast(?:/[^/]+){1,2}.*?\bi=(?P<id>\d+)'
|
|
_TESTS = [{
|
|
'url': 'https://podcasts.apple.com/us/podcast/207-whitney-webb-returns/id1135137367?i=1000482637777',
|
|
'md5': '41dc31cd650143e530d9423b6b5a344f',
|
|
'info_dict': {
|
|
'id': '1000482637777',
|
|
'ext': 'mp3',
|
|
'title': '207 - Whitney Webb Returns',
|
|
'description': 'md5:75ef4316031df7b41ced4e7b987f79c6',
|
|
'upload_date': '20200705',
|
|
'timestamp': 1593932400,
|
|
'duration': 6454,
|
|
'series': 'The Tim Dillon Show',
|
|
'thumbnail': 're:.+[.](png|jpe?g|webp)',
|
|
},
|
|
}, {
|
|
'url': 'https://podcasts.apple.com/podcast/207-whitney-webb-returns/id1135137367?i=1000482637777',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://podcasts.apple.com/podcast/207-whitney-webb-returns?i=1000482637777',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://podcasts.apple.com/podcast/id1135137367?i=1000482637777',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
episode_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, episode_id)
|
|
episode_data = {}
|
|
ember_data = {}
|
|
# new page type 2021-11
|
|
amp_data = self._parse_json(self._search_regex(
|
|
r'(?s)id="shoebox-media-api-cache-amp-podcasts"[^>]*>\s*({.+?})\s*<',
|
|
webpage, 'AMP data', default='{}'), episode_id, fatal=False) or {}
|
|
amp_data = try_get(amp_data,
|
|
lambda a: self._parse_json(
|
|
next(a[x] for x in iter(a) if episode_id in x),
|
|
episode_id),
|
|
dict) or {}
|
|
amp_data = amp_data.get('d') or []
|
|
episode_data = try_get(
|
|
amp_data,
|
|
lambda a: next(x for x in a
|
|
if x['type'] == 'podcast-episodes' and x['id'] == episode_id),
|
|
dict)
|
|
if not episode_data:
|
|
# try pre 2021-11 page type: TODO: consider deleting if no longer used
|
|
ember_data = self._parse_json(self._search_regex(
|
|
r'(?s)id="shoebox-ember-data-store"[^>]*>\s*({.+?})\s*<',
|
|
webpage, 'ember data'), episode_id) or {}
|
|
ember_data = ember_data.get(episode_id) or ember_data
|
|
episode_data = try_get(ember_data, lambda x: x['data'], dict)
|
|
episode = episode_data['attributes']
|
|
description = episode.get('description') or {}
|
|
|
|
series = None
|
|
for inc in (amp_data or ember_data.get('included') or []):
|
|
if inc.get('type') == 'media/podcast':
|
|
series = try_get(inc, lambda x: x['attributes']['name'])
|
|
series = series or clean_html(get_element_by_class('podcast-header__identity', webpage))
|
|
|
|
return {
|
|
'id': episode_id,
|
|
'title': episode.get('name'),
|
|
'url': clean_podcast_url(episode['assetUrl']),
|
|
'description': description.get('standard') or description.get('short'),
|
|
'timestamp': parse_iso8601(episode.get('releaseDateTime')),
|
|
'duration': int_or_none(episode.get('durationInMilliseconds'), 1000),
|
|
'series': series,
|
|
'thumbnail': self._og_search_thumbnail(webpage),
|
|
'vcodec': 'none',
|
|
}
|