mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-21 20:46:36 -05:00
99 lines
3.6 KiB
Python
99 lines
3.6 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import re
|
|
import functools
|
|
|
|
from .common import InfoExtractor
|
|
from ..compat import compat_str
|
|
from ..utils import (
|
|
int_or_none,
|
|
unified_timestamp,
|
|
OnDemandPagedList,
|
|
)
|
|
|
|
|
|
class ACastIE(InfoExtractor):
|
|
IE_NAME = 'acast'
|
|
_VALID_URL = r'https?://(?:www\.)?acast\.com/(?P<channel>[^/]+)/(?P<id>[^/#?]+)'
|
|
_TESTS = [{
|
|
# test with one bling
|
|
'url': 'https://www.acast.com/condenasttraveler/-where-are-you-taipei-101-taiwan',
|
|
'md5': 'ada3de5a1e3a2a381327d749854788bb',
|
|
'info_dict': {
|
|
'id': '57de3baa-4bb0-487e-9418-2692c1277a34',
|
|
'ext': 'mp3',
|
|
'title': '"Where Are You?": Taipei 101, Taiwan',
|
|
'timestamp': 1196172000,
|
|
'upload_date': '20071127',
|
|
'description': 'md5:a0b4ef3634e63866b542e5b1199a1a0e',
|
|
'duration': 211,
|
|
}
|
|
}, {
|
|
# test with multiple blings
|
|
'url': 'https://www.acast.com/sparpodcast/2.raggarmordet-rosterurdetforflutna',
|
|
'md5': 'e87d5b8516cd04c0d81b6ee1caca28d0',
|
|
'info_dict': {
|
|
'id': '2a92b283-1a75-4ad8-8396-499c641de0d9',
|
|
'ext': 'mp3',
|
|
'title': '2. Raggarmordet - Röster ur det förflutna',
|
|
'timestamp': 1477346700,
|
|
'upload_date': '20161024',
|
|
'description': 'md5:4f81f6d8cf2e12ee21a321d8bca32db4',
|
|
'duration': 2766,
|
|
}
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
channel, display_id = re.match(self._VALID_URL, url).groups()
|
|
cast_data = self._download_json(
|
|
'https://play-api.acast.com/splash/%s/%s' % (channel, display_id), display_id)
|
|
e = cast_data['result']['episode']
|
|
return {
|
|
'id': compat_str(e['id']),
|
|
'display_id': display_id,
|
|
'url': e['mediaUrl'],
|
|
'title': e['name'],
|
|
'description': e.get('description'),
|
|
'thumbnail': e.get('image'),
|
|
'timestamp': unified_timestamp(e.get('publishingDate')),
|
|
'duration': int_or_none(e.get('duration')),
|
|
}
|
|
|
|
|
|
class ACastChannelIE(InfoExtractor):
|
|
IE_NAME = 'acast:channel'
|
|
_VALID_URL = r'https?://(?:www\.)?acast\.com/(?P<id>[^/#?]+)'
|
|
_TEST = {
|
|
'url': 'https://www.acast.com/condenasttraveler',
|
|
'info_dict': {
|
|
'id': '50544219-29bb-499e-a083-6087f4cb7797',
|
|
'title': 'Condé Nast Traveler Podcast',
|
|
'description': 'md5:98646dee22a5b386626ae31866638fbd',
|
|
},
|
|
'playlist_mincount': 20,
|
|
}
|
|
_API_BASE_URL = 'https://www.acast.com/api/'
|
|
_PAGE_SIZE = 10
|
|
|
|
@classmethod
|
|
def suitable(cls, url):
|
|
return False if ACastIE.suitable(url) else super(ACastChannelIE, cls).suitable(url)
|
|
|
|
def _fetch_page(self, channel_slug, page):
|
|
casts = self._download_json(
|
|
self._API_BASE_URL + 'channels/%s/acasts?page=%s' % (channel_slug, page),
|
|
channel_slug, note='Download page %d of channel data' % page)
|
|
for cast in casts:
|
|
yield self.url_result(
|
|
'https://www.acast.com/%s/%s' % (channel_slug, cast['url']),
|
|
'ACast', cast['id'])
|
|
|
|
def _real_extract(self, url):
|
|
channel_slug = self._match_id(url)
|
|
channel_data = self._download_json(
|
|
self._API_BASE_URL + 'channels/%s' % channel_slug, channel_slug)
|
|
entries = OnDemandPagedList(functools.partial(
|
|
self._fetch_page, channel_slug), self._PAGE_SIZE)
|
|
return self.playlist_result(entries, compat_str(
|
|
channel_data['id']), channel_data['name'], channel_data.get('description'))
|