yt-dlp/youtube_dl/extractor/polskieradio.py
Jakub Adam Wieczorek 8d3737cda7 [polskieradio] Add support for downloading whole programmes.
This extends the Polskie Radio (the Polish national radio) extractor to
enable the user to download all the broadcasts of a single programme.
2016-09-06 21:34:44 +02:00

179 lines
6.6 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..compat import (
compat_str,
compat_urllib_parse_unquote,
compat_urlparse
)
from ..utils import (
int_or_none,
strip_or_none,
unified_timestamp,
)
class PolskieRadioProgrammeIE(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': 34,
}, {
'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/7,Jedynka/5102,HISTORIA-ZYWA',
'only_matching': True
}]
def _get_entries_from_page_content(self, base_url, content):
entries = []
articles = re.findall(
r'<article class="ID-(\d+) article">\s+<a href="([^"]+)"( data-layer="[^"]*")? class="[^"]*" title="([^"]+)">',
content)
for article_id, article_url, _, article_title in articles:
resolved_article_url = compat_urlparse.urljoin(base_url, article_url)
entries.append(self.url_result(
resolved_article_url,
ie='PolskieRadio',
video_id=article_id,
video_title=article_title))
return entries
@classmethod
def suitable(cls, url):
return False if PolskieRadioIE.suitable(url) else super(PolskieRadioProgrammeIE, cls).suitable(url)
def _real_extract(self, url):
programme_id = self._match_id(url)
webpage = self._download_webpage(url, programme_id)
title = self._html_search_regex(
r'<a href="[^"]+" id=".*_linkCategory" title="[^"]+">(.+?)</a>',
webpage, 'title', fatal=False)
description = None
entries = self._get_entries_from_page_content(url, webpage)
pages = re.findall(r'<a( href="([^"]+/Strona/)\d+")? id="[^"]+" title="strona&#32;(\d+)"', webpage)
page_count = max(int(page_number) for _, _, page_number in pages) if pages else 1
if page_count > 1:
page_url_root = next(url for _, url, _ in pages if len(url) > 0)
for page_number in range(2, page_count + 1):
page_url = page_url_root + str(page_number)
resolved_page_url = compat_urlparse.urljoin(url, page_url)
page_content = self._download_webpage(
resolved_page_url, programme_id,
note="Downloading page number %d" % page_number)
entries.extend(self._get_entries_from_page_content(url, page_content))
return self.playlist_result(entries, programme_id, title, description)
class PolskieRadioIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?polskieradio\.pl/\d+/\d+/Artykul/(?P<id>[0-9]+)'
_TESTS = [{
'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': 're:^https?://static\.prsa\.pl/images/.*\.jpg$'
},
}],
}, {
'url': 'http://www.polskieradio.pl/265/5217/Artykul/1635803,Euro-2016-nie-ma-miejsca-na-blad-Polacy-graja-ze-Szwajcaria-o-cwiercfinal',
'info_dict': {
'id': '1635803',
'title': 'Euro 2016: nie ma miejsca na błąd. Polacy grają ze Szwajcarią o ćwierćfinał',
'description': 'md5:01cb7d0cad58664095d72b51a1ebada2',
},
'playlist_mincount': 12,
}, {
'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,
}]
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="audio atarticle"[^>]*>(.+?)<script>',
webpage, 'content')
timestamp = unified_timestamp(self._html_search_regex(
r'(?s)<span[^>]+id="datetime2"[^>]*>(.+?)</span>',
webpage, 'timestamp', fatal=False))
thumbnail_url = self._og_search_thumbnail(webpage)
entries = []
media_urls = set()
for data_media in re.findall(r'<[^>]+data-media=({[^>]+})', content):
media = self._parse_json(data_media, playlist_id, fatal=False)
if not media.get('file') or not media.get('desc'):
continue
media_url = self._proto_relative_url(media['file'], 'http:')
if media_url in media_urls:
continue
media_urls.add(media_url)
entries.append({
'id': compat_str(media['id']),
'url': media_url,
'title': compat_urllib_parse_unquote(media['desc']),
'duration': int_or_none(media.get('length')),
'vcodec': 'none' if media.get('provider') == 'audio' else None,
'timestamp': timestamp,
'thumbnail': thumbnail_url
})
title = self._og_search_title(webpage).strip()
description = strip_or_none(self._og_search_description(webpage))
return self.playlist_result(entries, playlist_id, title, description)