|
|
import functools
|
|
|
import json
|
|
|
|
|
|
from .common import InfoExtractor
|
|
|
from ..utils import (
|
|
|
OnDemandPagedList,
|
|
|
float_or_none,
|
|
|
str_or_none,
|
|
|
str_to_int,
|
|
|
traverse_obj,
|
|
|
unified_timestamp,
|
|
|
)
|
|
|
|
|
|
|
|
|
class PodchaserIE(InfoExtractor):
|
|
|
_VALID_URL = r'https?://(?:www\.)?podchaser\.com/podcasts/[\w-]+-(?P<podcast_id>\d+)(?:/episodes/[\w-]+-(?P<id>\d+))?'
|
|
|
_PAGE_SIZE = 100
|
|
|
_TESTS = [{
|
|
|
'url': 'https://www.podchaser.com/podcasts/cum-town-36924/episodes/ep-285-freeze-me-off-104365585',
|
|
|
'info_dict': {
|
|
|
'id': '104365585',
|
|
|
'title': 'Ep. 285 – freeze me off',
|
|
|
'description': 'cam ahn',
|
|
|
'thumbnail': r're:^https?://.*\.jpg$',
|
|
|
'ext': 'mp3',
|
|
|
'categories': ['Comedy'],
|
|
|
'tags': ['comedy', 'dark humor'],
|
|
|
'series': 'Cum Town',
|
|
|
'duration': 3708,
|
|
|
'timestamp': 1636531259,
|
|
|
'upload_date': '20211110',
|
|
|
'rating': 4.0
|
|
|
}
|
|
|
}, {
|
|
|
'url': 'https://www.podchaser.com/podcasts/the-bone-zone-28853',
|
|
|
'info_dict': {
|
|
|
'id': '28853',
|
|
|
'title': 'The Bone Zone',
|
|
|
'description': 'Podcast by The Bone Zone',
|
|
|
},
|
|
|
'playlist_count': 275
|
|
|
}, {
|
|
|
'url': 'https://www.podchaser.com/podcasts/sean-carrolls-mindscape-scienc-699349/episodes',
|
|
|
'info_dict': {
|
|
|
'id': '699349',
|
|
|
'title': 'Sean Carroll\'s Mindscape: Science, Society, Philosophy, Culture, Arts, and Ideas',
|
|
|
'description': 'md5:2cbd8f4749891a84dc8235342e0b5ff1'
|
|
|
},
|
|
|
'playlist_mincount': 225
|
|
|
}]
|
|
|
|
|
|
@staticmethod
|
|
|
def _parse_episode(episode, podcast):
|
|
|
return {
|
|
|
'id': str(episode.get('id')),
|
|
|
'title': episode.get('title'),
|
|
|
'description': episode.get('description'),
|
|
|
'url': episode.get('audio_url'),
|
|
|
'thumbnail': episode.get('image_url'),
|
|
|
'duration': str_to_int(episode.get('length')),
|
|
|
'timestamp': unified_timestamp(episode.get('air_date')),
|
|
|
'rating': float_or_none(episode.get('rating')),
|
|
|
'categories': list(set(traverse_obj(podcast, (('summary', None), 'categories', ..., 'text')))),
|
|
|
'tags': traverse_obj(podcast, ('tags', ..., 'text')),
|
|
|
'series': podcast.get('title'),
|
|
|
}
|
|
|
|
|
|
def _call_api(self, path, *args, **kwargs):
|
|
|
return self._download_json(f'https://api.podchaser.com/{path}', *args, **kwargs)
|
|
|
|
|
|
def _fetch_page(self, podcast_id, podcast, page):
|
|
|
json_response = self._call_api(
|
|
|
'list/episode', podcast_id,
|
|
|
headers={'Content-Type': 'application/json;charset=utf-8'},
|
|
|
data=json.dumps({
|
|
|
'start': page * self._PAGE_SIZE,
|
|
|
'count': self._PAGE_SIZE,
|
|
|
'sort_order': 'SORT_ORDER_RECENT',
|
|
|
'filters': {
|
|
|
'podcast_id': podcast_id
|
|
|
},
|
|
|
'options': {}
|
|
|
}).encode())
|
|
|
|
|
|
for episode in json_response['entities']:
|
|
|
yield self._parse_episode(episode, podcast)
|
|
|
|
|
|
def _real_extract(self, url):
|
|
|
podcast_id, episode_id = self._match_valid_url(url).group('podcast_id', 'id')
|
|
|
podcast = self._call_api(f'podcasts/{podcast_id}', episode_id or podcast_id)
|
|
|
if not episode_id:
|
|
|
return self.playlist_result(
|
|
|
OnDemandPagedList(functools.partial(self._fetch_page, podcast_id, podcast), self._PAGE_SIZE),
|
|
|
str_or_none(podcast.get('id')), podcast.get('title'), podcast.get('description'))
|
|
|
|
|
|
episode = self._call_api(f'episodes/{episode_id}', episode_id)
|
|
|
return self._parse_episode(episode, podcast)
|