1
0
Fork 0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-11-22 10:11:52 +00:00
youtube-dl/youtube_dl/extractor/spreaker.py
2020-11-25 21:39:17 +01:00

176 lines
5.9 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import itertools
from .common import InfoExtractor
from ..compat import compat_str
from ..utils import (
float_or_none,
int_or_none,
str_or_none,
try_get,
unified_timestamp,
url_or_none,
)
def _extract_episode(data, episode_id=None):
title = data['title']
download_url = data['download_url']
series = try_get(data, lambda x: x['show']['title'], compat_str)
uploader = try_get(data, lambda x: x['author']['fullname'], compat_str)
thumbnails = []
for image in ('image_original', 'image_medium', 'image'):
image_url = url_or_none(data.get('%s_url' % image))
if image_url:
thumbnails.append({'url': image_url})
def stats(key):
return int_or_none(try_get(
data,
(lambda x: x['%ss_count' % key],
lambda x: x['stats']['%ss' % key])))
def duration(key):
return float_or_none(data.get(key), scale=1000)
return {
'id': compat_str(episode_id or data['episode_id']),
'url': download_url,
'display_id': data.get('permalink'),
'title': title,
'description': data.get('description'),
'timestamp': unified_timestamp(data.get('published_at')),
'uploader': uploader,
'uploader_id': str_or_none(data.get('author_id')),
'creator': uploader,
'duration': duration('duration') or duration('length'),
'view_count': stats('play'),
'like_count': stats('like'),
'comment_count': stats('message'),
'format': 'MPEG Layer 3',
'format_id': 'mp3',
'container': 'mp3',
'ext': 'mp3',
'thumbnails': thumbnails,
'series': series,
'extractor_key': SpreakerIE.ie_key(),
}
class SpreakerIE(InfoExtractor):
_VALID_URL = r'''(?x)
https?://
api\.spreaker\.com/
(?:
(?:download/)?episode|
v2/episodes
)/
(?P<id>\d+)
'''
_TESTS = [{
'url': 'https://api.spreaker.com/episode/12534508',
'info_dict': {
'id': '12534508',
'display_id': 'swm-ep15-how-to-market-your-music-part-2',
'ext': 'mp3',
'title': 'EP:15 | Music Marketing (Likes) - Part 2',
'description': 'md5:0588c43e27be46423e183076fa071177',
'timestamp': 1502250336,
'upload_date': '20170809',
'uploader': 'SWM',
'uploader_id': '9780658',
'duration': 1063.42,
'view_count': int,
'like_count': int,
'comment_count': int,
'series': 'Success With Music (SWM)',
},
}, {
'url': 'https://api.spreaker.com/download/episode/12534508/swm_ep15_how_to_market_your_music_part_2.mp3',
'only_matching': True,
}, {
'url': 'https://api.spreaker.com/v2/episodes/12534508?export=episode_segments',
'only_matching': True,
}]
def _real_extract(self, url):
episode_id = self._match_id(url)
data = self._download_json(
'https://api.spreaker.com/v2/episodes/%s' % episode_id,
episode_id)['response']['episode']
return _extract_episode(data, episode_id)
class SpreakerPageIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?spreaker\.com/user/[^/]+/(?P<id>[^/?#&]+)'
_TESTS = [{
'url': 'https://www.spreaker.com/user/9780658/swm-ep15-how-to-market-your-music-part-2',
'only_matching': True,
}]
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
episode_id = self._search_regex(
(r'data-episode_id=["\'](?P<id>\d+)',
r'episode_id\s*:\s*(?P<id>\d+)'), webpage, 'episode id')
return self.url_result(
'https://api.spreaker.com/episode/%s' % episode_id,
ie=SpreakerIE.ie_key(), video_id=episode_id)
class SpreakerShowIE(InfoExtractor):
_VALID_URL = r'https?://api\.spreaker\.com/show/(?P<id>\d+)'
_TESTS = [{
'url': 'https://api.spreaker.com/show/4652058',
'info_dict': {
'id': '4652058',
},
'playlist_mincount': 118,
}]
def _entries(self, show_id):
for page_num in itertools.count(1):
episodes = self._download_json(
'https://api.spreaker.com/show/%s/episodes' % show_id,
show_id, note='Downloading JSON page %d' % page_num, query={
'page': page_num,
'max_per_page': 100,
})
pager = try_get(episodes, lambda x: x['response']['pager'], dict)
if not pager:
break
results = pager.get('results')
if not results or not isinstance(results, list):
break
for result in results:
if not isinstance(result, dict):
continue
yield _extract_episode(result)
if page_num == pager.get('last_page'):
break
def _real_extract(self, url):
show_id = self._match_id(url)
return self.playlist_result(self._entries(show_id), playlist_id=show_id)
class SpreakerShowPageIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?spreaker\.com/show/(?P<id>[^/?#&]+)'
_TESTS = [{
'url': 'https://www.spreaker.com/show/success-with-music',
'only_matching': True,
}]
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
show_id = self._search_regex(
r'show_id\s*:\s*(?P<id>\d+)', webpage, 'show id')
return self.url_result(
'https://api.spreaker.com/show/%s' % show_id,
ie=SpreakerShowIE.ie_key(), video_id=show_id)