mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-23 10:37:18 -05:00
428 lines
16 KiB
Python
428 lines
16 KiB
Python
import itertools
|
|
import json
|
|
import math
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..compat import (
|
|
compat_str,
|
|
compat_urllib_parse_unquote,
|
|
compat_urlparse
|
|
)
|
|
from ..utils import (
|
|
extract_attributes,
|
|
ExtractorError,
|
|
InAdvancePagedList,
|
|
int_or_none,
|
|
js_to_json,
|
|
parse_iso8601,
|
|
strip_or_none,
|
|
unified_timestamp,
|
|
unescapeHTML,
|
|
url_or_none,
|
|
)
|
|
|
|
|
|
class PolskieRadioBaseExtractor(InfoExtractor):
|
|
def _extract_webpage_player_entries(self, webpage, playlist_id, base_data):
|
|
media_urls = set()
|
|
|
|
for data_media in re.findall(r'<[^>]+data-media="?({[^>]+})"?', webpage):
|
|
media = self._parse_json(data_media, playlist_id, transform_source=unescapeHTML, fatal=False)
|
|
if not media.get('file') or not media.get('desc'):
|
|
continue
|
|
media_url = self._proto_relative_url(media['file'])
|
|
if media_url in media_urls:
|
|
continue
|
|
media_urls.add(media_url)
|
|
entry = base_data.copy()
|
|
entry.update({
|
|
'id': compat_str(media['id']),
|
|
'url': media_url,
|
|
'duration': int_or_none(media.get('length')),
|
|
'vcodec': 'none' if media.get('provider') == 'audio' else None,
|
|
})
|
|
entry_title = compat_urllib_parse_unquote(media['desc'])
|
|
if entry_title:
|
|
entry['title'] = entry_title
|
|
yield entry
|
|
|
|
|
|
class PolskieRadioIE(PolskieRadioBaseExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?polskieradio(?:24)?\.pl/\d+/\d+/Artykul/(?P<id>[0-9]+)'
|
|
_TESTS = [{ # Old-style single broadcast.
|
|
'url': 'http://www.polskieradio.pl/7/5102/Artykul/1587943,Prof-Andrzej-Nowak-o-historii-nie-da-sie-myslec-beznamietnie',
|
|
'info_dict': {
|
|
'id': '1587943',
|
|
'title': 'Prof. Andrzej Nowak: o historii nie da się myśleć beznamiętnie',
|
|
'description': 'md5:12f954edbf3120c5e7075e17bf9fc5c5',
|
|
},
|
|
'playlist': [{
|
|
'md5': '2984ee6ce9046d91fc233bc1a864a09a',
|
|
'info_dict': {
|
|
'id': '1540576',
|
|
'ext': 'mp3',
|
|
'title': 'md5:d4623290d4ac983bf924061c75c23a0d',
|
|
'timestamp': 1456594200,
|
|
'upload_date': '20160227',
|
|
'duration': 2364,
|
|
'thumbnail': r're:^https?://static\.prsa\.pl/images/.*\.jpg$'
|
|
},
|
|
}],
|
|
}, { # New-style single broadcast.
|
|
'url': 'https://www.polskieradio.pl/8/2382/Artykul/2534482,Zagarysci-Poezja-jak-spoiwo',
|
|
'info_dict': {
|
|
'id': '2534482',
|
|
'title': 'Żagaryści. Poezja jak spoiwo',
|
|
'description': 'md5:f18d95d5dcba747a09b635e21a4c0695',
|
|
},
|
|
'playlist': [{
|
|
'md5': 'd07559829f61d5a93a75755987ded760',
|
|
'info_dict': {
|
|
'id': '2516679',
|
|
'ext': 'mp3',
|
|
'title': 'md5:c6e1234e0b747ad883cb91b7ad06b98c',
|
|
'timestamp': 1592654400,
|
|
'upload_date': '20200620',
|
|
'duration': 1430,
|
|
'thumbnail': r're:^https?://static\.prsa\.pl/images/.*\.jpg$'
|
|
},
|
|
}],
|
|
}, {
|
|
# PR4 audition - other frontend
|
|
'url': 'https://www.polskieradio.pl/10/6071/Artykul/2610977,Poglos-29-pazdziernika-godz-2301',
|
|
'info_dict': {
|
|
'id': '2610977',
|
|
'ext': 'mp3',
|
|
'title': 'Pogłos 29 października godz. 23:01',
|
|
},
|
|
}, {
|
|
'url': 'http://polskieradio.pl/9/305/Artykul/1632955,Bardzo-popularne-slowo-remis',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'http://www.polskieradio.pl/7/5102/Artykul/1587943',
|
|
'only_matching': True,
|
|
}, {
|
|
# with mp4 video
|
|
'url': 'http://www.polskieradio.pl/9/299/Artykul/1634903,Brexit-Leszek-Miller-swiat-sie-nie-zawali-Europa-bedzie-trwac-dalej',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://polskieradio24.pl/130/4503/Artykul/2621876,Narusza-nasza-suwerennosc-Publicysci-o-uzaleznieniu-funduszy-UE-od-praworzadnosci',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
playlist_id = self._match_id(url)
|
|
|
|
webpage = self._download_webpage(url, playlist_id)
|
|
|
|
content = self._search_regex(
|
|
r'(?s)<div[^>]+class="\s*this-article\s*"[^>]*>(.+?)<div[^>]+class="tags"[^>]*>',
|
|
webpage, 'content', default=None)
|
|
|
|
timestamp = unified_timestamp(self._html_search_regex(
|
|
r'(?s)<span[^>]+id="datetime2"[^>]*>(.+?)</span>',
|
|
webpage, 'timestamp', default=None))
|
|
|
|
thumbnail_url = self._og_search_thumbnail(webpage, default=None)
|
|
|
|
title = self._og_search_title(webpage).strip()
|
|
|
|
description = strip_or_none(self._og_search_description(webpage, default=None))
|
|
description = description.replace('\xa0', ' ') if description is not None else None
|
|
|
|
if not content:
|
|
return {
|
|
'id': playlist_id,
|
|
'url': self._proto_relative_url(
|
|
self._search_regex(
|
|
r"source:\s*'(//static\.prsa\.pl/[^']+)'",
|
|
webpage, 'audition record url')),
|
|
'title': title,
|
|
'description': description,
|
|
'timestamp': timestamp,
|
|
'thumbnail': thumbnail_url,
|
|
}
|
|
|
|
entries = self._extract_webpage_player_entries(content, playlist_id, {
|
|
'title': title,
|
|
'timestamp': timestamp,
|
|
'thumbnail': thumbnail_url,
|
|
})
|
|
|
|
return self.playlist_result(entries, playlist_id, title, description)
|
|
|
|
|
|
class PolskieRadioCategoryIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?polskieradio\.pl/\d+(?:,[^/]+)?/(?P<id>\d+)'
|
|
_TESTS = [{
|
|
'url': 'http://www.polskieradio.pl/7/5102,HISTORIA-ZYWA',
|
|
'info_dict': {
|
|
'id': '5102',
|
|
'title': 'HISTORIA ŻYWA',
|
|
},
|
|
'playlist_mincount': 38,
|
|
}, {
|
|
'url': 'http://www.polskieradio.pl/7/4807',
|
|
'info_dict': {
|
|
'id': '4807',
|
|
'title': 'Vademecum 1050. rocznicy Chrztu Polski'
|
|
},
|
|
'playlist_mincount': 5
|
|
}, {
|
|
'url': 'http://www.polskieradio.pl/7/129,Sygnaly-dnia?ref=source',
|
|
'only_matching': True
|
|
}, {
|
|
'url': 'http://www.polskieradio.pl/37,RedakcjaKatolicka/4143,Kierunek-Krakow',
|
|
'info_dict': {
|
|
'id': '4143',
|
|
'title': 'Kierunek Kraków',
|
|
},
|
|
'playlist_mincount': 61
|
|
}, {
|
|
'url': 'http://www.polskieradio.pl/10,czworka/214,muzyka',
|
|
'info_dict': {
|
|
'id': '214',
|
|
'title': 'Muzyka',
|
|
},
|
|
'playlist_mincount': 61
|
|
}, {
|
|
'url': 'http://www.polskieradio.pl/7,Jedynka/5102,HISTORIA-ZYWA',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'http://www.polskieradio.pl/8,Dwojka/196,Publicystyka',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
@classmethod
|
|
def suitable(cls, url):
|
|
return False if PolskieRadioIE.suitable(url) else super(PolskieRadioCategoryIE, cls).suitable(url)
|
|
|
|
def _entries(self, url, page, category_id):
|
|
content = page
|
|
for page_num in itertools.count(2):
|
|
for a_entry, entry_id in re.findall(
|
|
r'(?s)<article[^>]+>.*?(<a[^>]+href=["\']/\d+/\d+/Artykul/(\d+)[^>]+>).*?</article>',
|
|
content):
|
|
entry = extract_attributes(a_entry)
|
|
href = entry.get('href')
|
|
if not href:
|
|
continue
|
|
yield self.url_result(
|
|
compat_urlparse.urljoin(url, href), PolskieRadioIE.ie_key(),
|
|
entry_id, entry.get('title'))
|
|
mobj = re.search(
|
|
r'<div[^>]+class=["\']next["\'][^>]*>\s*<a[^>]+href=(["\'])(?P<url>(?:(?!\1).)+)\1',
|
|
content)
|
|
if not mobj:
|
|
break
|
|
next_url = compat_urlparse.urljoin(url, mobj.group('url'))
|
|
content = self._download_webpage(
|
|
next_url, category_id, 'Downloading page %s' % page_num)
|
|
|
|
def _real_extract(self, url):
|
|
category_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, category_id)
|
|
title = self._html_search_regex(
|
|
r'<title>([^<]+) - [^<]+ - [^<]+</title>',
|
|
webpage, 'title', fatal=False)
|
|
return self.playlist_result(
|
|
self._entries(url, webpage, category_id),
|
|
category_id, title)
|
|
|
|
|
|
class PolskieRadioPlayerIE(InfoExtractor):
|
|
IE_NAME = 'polskieradio:player'
|
|
_VALID_URL = r'https?://player\.polskieradio\.pl/anteny/(?P<id>[^/]+)'
|
|
|
|
_BASE_URL = 'https://player.polskieradio.pl'
|
|
_PLAYER_URL = 'https://player.polskieradio.pl/main.bundle.js'
|
|
_STATIONS_API_URL = 'https://apipr.polskieradio.pl/api/stacje'
|
|
|
|
_TESTS = [{
|
|
'url': 'https://player.polskieradio.pl/anteny/trojka',
|
|
'info_dict': {
|
|
'id': '3',
|
|
'ext': 'm4a',
|
|
'title': 'Trójka',
|
|
},
|
|
'params': {
|
|
'format': 'bestaudio',
|
|
'skip_download': 'endless stream',
|
|
},
|
|
}]
|
|
|
|
def _get_channel_list(self, channel_url='no_channel'):
|
|
player_code = self._download_webpage(
|
|
self._PLAYER_URL, channel_url,
|
|
note='Downloading js player')
|
|
channel_list = js_to_json(self._search_regex(
|
|
r';var r="anteny",a=(\[.+?\])},', player_code, 'channel list'))
|
|
return self._parse_json(channel_list, channel_url)
|
|
|
|
def _real_extract(self, url):
|
|
channel_url = self._match_id(url)
|
|
channel_list = self._get_channel_list(channel_url)
|
|
|
|
channel = next((c for c in channel_list if c.get('url') == channel_url), None)
|
|
|
|
if not channel:
|
|
raise ExtractorError('Channel not found')
|
|
|
|
station_list = self._download_json(self._STATIONS_API_URL, channel_url,
|
|
note='Downloading stream url list',
|
|
headers={
|
|
'Accept': 'application/json',
|
|
'Referer': url,
|
|
'Origin': self._BASE_URL,
|
|
})
|
|
station = next((s for s in station_list
|
|
if s.get('Name') == (channel.get('streamName') or channel.get('name'))), None)
|
|
if not station:
|
|
raise ExtractorError('Station not found even though we extracted channel')
|
|
|
|
formats = []
|
|
for stream_url in station['Streams']:
|
|
stream_url = self._proto_relative_url(stream_url)
|
|
if stream_url.endswith('/playlist.m3u8'):
|
|
formats.extend(self._extract_m3u8_formats(stream_url, channel_url, live=True))
|
|
elif stream_url.endswith('/manifest.f4m'):
|
|
formats.extend(self._extract_mpd_formats(stream_url, channel_url))
|
|
elif stream_url.endswith('/Manifest'):
|
|
formats.extend(self._extract_ism_formats(stream_url, channel_url))
|
|
else:
|
|
formats.append({
|
|
'url': stream_url,
|
|
})
|
|
|
|
return {
|
|
'id': compat_str(channel['id']),
|
|
'formats': formats,
|
|
'title': channel.get('name') or channel.get('streamName'),
|
|
'display_id': channel_url,
|
|
'thumbnail': f'{self._BASE_URL}/images/{channel_url}-color-logo.png',
|
|
'is_live': True,
|
|
}
|
|
|
|
|
|
class PolskieRadioPodcastBaseExtractor(InfoExtractor):
|
|
_API_BASE = 'https://apipodcasts.polskieradio.pl/api'
|
|
|
|
def _parse_episode(self, data):
|
|
return {
|
|
'id': data['guid'],
|
|
'formats': [{
|
|
'url': data['url'],
|
|
'filesize': int_or_none(data.get('fileSize')),
|
|
}],
|
|
'title': data['title'],
|
|
'description': data.get('description'),
|
|
'duration': int_or_none(data.get('length')),
|
|
'timestamp': parse_iso8601(data.get('publishDate')),
|
|
'thumbnail': url_or_none(data.get('image')),
|
|
'series': data.get('podcastTitle'),
|
|
'episode': data['title'],
|
|
}
|
|
|
|
|
|
class PolskieRadioPodcastListIE(PolskieRadioPodcastBaseExtractor):
|
|
IE_NAME = 'polskieradio:podcast:list'
|
|
_VALID_URL = r'https?://podcasty\.polskieradio\.pl/podcast/(?P<id>\d+)'
|
|
_TESTS = [{
|
|
'url': 'https://podcasty.polskieradio.pl/podcast/8/',
|
|
'info_dict': {
|
|
'id': '8',
|
|
'title': 'Śniadanie w Trójce',
|
|
'description': 'md5:57abcc27bc4c6a6b25baa3061975b9ef',
|
|
'uploader': 'Beata Michniewicz',
|
|
},
|
|
'playlist_mincount': 714,
|
|
}]
|
|
_PAGE_SIZE = 10
|
|
|
|
def _call_api(self, podcast_id, page):
|
|
return self._download_json(
|
|
f'{self._API_BASE}/Podcasts/{podcast_id}/?pageSize={self._PAGE_SIZE}&page={page}',
|
|
podcast_id, f'Downloading page {page}')
|
|
|
|
def _real_extract(self, url):
|
|
podcast_id = self._match_id(url)
|
|
data = self._call_api(podcast_id, 1)
|
|
|
|
def get_page(page_num):
|
|
page_data = self._call_api(podcast_id, page_num + 1) if page_num else data
|
|
yield from (self._parse_episode(ep) for ep in page_data['items'])
|
|
|
|
return {
|
|
'_type': 'playlist',
|
|
'entries': InAdvancePagedList(
|
|
get_page, math.ceil(data['itemCount'] / self._PAGE_SIZE), self._PAGE_SIZE),
|
|
'id': str(data['id']),
|
|
'title': data['title'],
|
|
'description': data.get('description'),
|
|
'uploader': data.get('announcer'),
|
|
}
|
|
|
|
|
|
class PolskieRadioPodcastIE(PolskieRadioPodcastBaseExtractor):
|
|
IE_NAME = 'polskieradio:podcast'
|
|
_VALID_URL = r'https?://podcasty\.polskieradio\.pl/track/(?P<id>[a-f\d]{8}(?:-[a-f\d]{4}){4}[a-f\d]{8})'
|
|
_TESTS = [{
|
|
'url': 'https://podcasty.polskieradio.pl/track/6eafe403-cb8f-4756-b896-4455c3713c32',
|
|
'info_dict': {
|
|
'id': '6eafe403-cb8f-4756-b896-4455c3713c32',
|
|
'ext': 'mp3',
|
|
'title': 'Theresa May rezygnuje. Co dalej z brexitem?',
|
|
'description': 'md5:e41c409a29d022b70ef0faa61dbded60',
|
|
},
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
podcast_id = self._match_id(url)
|
|
data = self._download_json(
|
|
f'{self._API_BASE}/audio',
|
|
podcast_id, 'Downloading podcast metadata',
|
|
data=json.dumps({
|
|
'guids': [podcast_id],
|
|
}).encode('utf-8'),
|
|
headers={
|
|
'Content-Type': 'application/json',
|
|
})
|
|
return self._parse_episode(data[0])
|
|
|
|
|
|
class PolskieRadioRadioKierowcowIE(PolskieRadioBaseExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?radiokierowcow\.pl/artykul/(?P<id>[0-9]+)'
|
|
IE_NAME = 'polskieradio:kierowcow'
|
|
|
|
_TESTS = [{
|
|
'url': 'https://radiokierowcow.pl/artykul/2694529',
|
|
'info_dict': {
|
|
'id': '2694529',
|
|
'title': 'Zielona fala reliktem przeszłości?',
|
|
'description': 'md5:343950a8717c9818fdfd4bd2b8ca9ff2',
|
|
},
|
|
'playlist_count': 3,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
media_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, media_id)
|
|
nextjs_build = self._search_nextjs_data(webpage, media_id)['buildId']
|
|
article = self._download_json(
|
|
f'https://radiokierowcow.pl/_next/data/{nextjs_build}/artykul/{media_id}.json?articleId={media_id}',
|
|
media_id)
|
|
data = article['pageProps']['data']
|
|
title = data['title']
|
|
entries = self._extract_webpage_player_entries(data['content'], media_id, {
|
|
'title': title,
|
|
})
|
|
|
|
return {
|
|
'_type': 'playlist',
|
|
'id': media_id,
|
|
'entries': entries,
|
|
'title': title,
|
|
'description': data.get('lead'),
|
|
}
|