1
0
mirror of https://codeberg.org/polarisfm/youtube-dl synced 2024-12-29 17:37:55 +01:00
youtube-dl/youtube_dl/extractor/nrk.py

297 lines
11 KiB
Python
Raw Normal View History

# encoding: utf-8
from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
ExtractorError,
2014-06-07 16:50:19 +02:00
float_or_none,
2015-01-10 17:46:01 +01:00
parse_duration,
unified_strdate,
)
class NRKIE(InfoExtractor):
_VALID_URL = r'(?:nrk:|https?://(?:www\.)?nrk\.no/video/PS\*)(?P<id>\d+)'
_TESTS = [
{
2015-03-21 13:20:49 +01:00
'url': 'http://www.nrk.no/video/PS*150533',
'md5': 'bccd850baebefe23b56d708a113229c2',
'info_dict': {
'id': '150533',
'ext': 'flv',
'title': 'Dompap og andre fugler i Piip-Show',
2015-03-21 13:20:49 +01:00
'description': 'md5:d9261ba34c43b61c812cb6b0269a5c8f',
2015-03-21 13:21:19 +01:00
'duration': 263,
}
},
{
2015-03-21 13:20:49 +01:00
'url': 'http://www.nrk.no/video/PS*154915',
'md5': '0b1493ba1aae7d9579a5ad5531bc395a',
'info_dict': {
'id': '154915',
'ext': 'flv',
'title': 'Slik høres internett ut når du er blind',
'description': 'md5:a621f5cc1bd75c8d5104cb048c6b8568',
2015-03-21 13:21:19 +01:00
'duration': 20,
}
},
]
def _real_extract(self, url):
2015-03-21 13:20:49 +01:00
video_id = self._match_id(url)
data = self._download_json(
2015-03-21 13:20:49 +01:00
'http://v8.psapi.nrk.no/mediaelement/%s' % video_id,
video_id, 'Downloading media JSON')
if data['usageRights']['isGeoBlocked']:
2015-03-21 13:20:49 +01:00
raise ExtractorError(
'NRK har ikke rettig-heter til å vise dette programmet utenfor Norge',
expected=True)
2015-03-21 13:20:49 +01:00
video_url = data['mediaUrl'] + '?hdcore=3.5.0&plugin=aasp-3.5.0.151.81'
2015-03-21 13:21:19 +01:00
duration = parse_duration(data.get('duration'))
images = data.get('images')
if images:
thumbnails = images['webImages']
thumbnails.sort(key=lambda image: image['pixelWidth'])
thumbnail = thumbnails[-1]['imageUrl']
else:
thumbnail = None
return {
'id': video_id,
'url': video_url,
'ext': 'flv',
'title': data['title'],
'description': data['description'],
2015-03-21 13:21:19 +01:00
'duration': duration,
'thumbnail': thumbnail,
}
class NRKPlaylistIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?nrk\.no/(?!video)(?:[^/]+/)+(?P<id>[^/]+)'
_TESTS = [{
'url': 'http://www.nrk.no/troms/gjenopplev-den-historiske-solformorkelsen-1.12270763',
'info_dict': {
'id': 'gjenopplev-den-historiske-solformorkelsen-1.12270763',
'title': 'Gjenopplev den historiske solformørkelsen',
'description': 'md5:c2df8ea3bac5654a26fc2834a542feed',
},
'playlist_count': 2,
}, {
'url': 'http://www.nrk.no/kultur/bok/rivertonprisen-til-karin-fossum-1.12266449',
'info_dict': {
'id': 'rivertonprisen-til-karin-fossum-1.12266449',
'title': 'Rivertonprisen til Karin Fossum',
'description': 'Første kvinne på 15 år til å vinne krimlitteraturprisen.',
},
'playlist_count': 5,
}]
def _real_extract(self, url):
playlist_id = self._match_id(url)
webpage = self._download_webpage(url, playlist_id)
entries = [
self.url_result('nrk:%s' % video_id, 'NRK')
for video_id in re.findall(
r'class="[^"]*\brich\b[^"]*"[^>]+data-video-id="([^"]+)"',
webpage)
]
playlist_title = self._og_search_title(webpage)
playlist_description = self._og_search_description(webpage)
return self.playlist_result(
entries, playlist_id, playlist_title, playlist_description)
2015-02-18 18:57:01 +01:00
class NRKTVIE(InfoExtractor):
IE_DESC = 'NRK TV and NRK Radio'
_VALID_URL = r'(?P<baseurl>https?://(?:tv|radio)\.nrk(?:super)?\.no/)(?:serie/[^/]+|program)/(?P<id>[a-zA-Z]{4}\d{8})(?:/\d{2}-\d{2}-\d{4})?(?:#del=(?P<part_id>\d+))?'
_TESTS = [
{
'url': 'https://tv.nrk.no/serie/20-spoersmaal-tv/MUHH48000314/23-05-2014',
2014-12-31 19:37:57 +01:00
'md5': 'adf2c5454fa2bf032f47a9f8fb351342',
'info_dict': {
'id': 'MUHH48000314',
'ext': 'flv',
'title': '20 spørsmål',
'description': 'md5:bdea103bc35494c143c6a9acdd84887a',
'upload_date': '20140523',
'duration': 1741.52,
},
},
{
'url': 'https://tv.nrk.no/program/mdfp15000514',
2014-12-31 19:37:57 +01:00
'md5': '383650ece2b25ecec996ad7b5bb2a384',
'info_dict': {
'id': 'mdfp15000514',
'ext': 'flv',
'title': 'Kunnskapskanalen: Grunnlovsjubiléet - Stor ståhei for ingenting',
'description': 'md5:654c12511f035aed1e42bdf5db3b206a',
'upload_date': '20140524',
'duration': 4605.0,
},
},
{
# single playlist video
'url': 'https://tv.nrk.no/serie/tour-de-ski/MSPO40010515/06-01-2015#del=2',
'md5': 'adbd1dbd813edaf532b0a253780719c2',
'info_dict': {
'id': 'MSPO40010515-part2',
'ext': 'flv',
'title': 'Tour de Ski: Sprint fri teknikk, kvinner og menn 06.01.2015 (del 2:2)',
'description': 'md5:238b67b97a4ac7d7b4bf0edf8cc57d26',
'upload_date': '20150106',
},
'skip': 'Only works from Norway',
},
{
'url': 'https://tv.nrk.no/serie/tour-de-ski/MSPO40010515/06-01-2015',
'playlist': [
{
'md5': '9480285eff92d64f06e02a5367970a7a',
'info_dict': {
'id': 'MSPO40010515-part1',
'ext': 'flv',
'title': 'Tour de Ski: Sprint fri teknikk, kvinner og menn 06.01.2015 (del 1:2)',
'description': 'md5:238b67b97a4ac7d7b4bf0edf8cc57d26',
'upload_date': '20150106',
},
},
{
'md5': 'adbd1dbd813edaf532b0a253780719c2',
'info_dict': {
'id': 'MSPO40010515-part2',
'ext': 'flv',
'title': 'Tour de Ski: Sprint fri teknikk, kvinner og menn 06.01.2015 (del 2:2)',
'description': 'md5:238b67b97a4ac7d7b4bf0edf8cc57d26',
'upload_date': '20150106',
},
},
],
'info_dict': {
'id': 'MSPO40010515',
'title': 'Tour de Ski: Sprint fri teknikk, kvinner og menn',
'description': 'md5:238b67b97a4ac7d7b4bf0edf8cc57d26',
'upload_date': '20150106',
'duration': 6947.5199999999995,
},
'skip': 'Only works from Norway',
},
{
'url': 'https://radio.nrk.no/serie/dagsnytt/NPUB21019315/12-07-2015#',
'only_matching': True,
}
]
2015-01-10 17:46:01 +01:00
def _debug_print(self, txt):
if self._downloader.params.get('verbose', False):
self.to_screen('[debug] %s' % txt)
2015-02-18 18:57:01 +01:00
def _get_subtitles(self, subtitlesurl, video_id, baseurl):
2015-01-10 17:46:01 +01:00
url = "%s%s" % (baseurl, subtitlesurl)
self._debug_print('%s: Subtitle url: %s' % (video_id, url))
captions = self._download_xml(
url, video_id, 'Downloading subtitles')
2015-01-10 17:46:01 +01:00
lang = captions.get('lang', 'no')
2015-02-18 18:57:01 +01:00
return {lang: [
{'ext': 'ttml', 'url': url},
]}
2015-01-10 17:46:01 +01:00
def _extract_f4m(self, manifest_url, video_id):
2015-07-12 16:42:18 +02:00
return self._extract_f4m_formats(
manifest_url + '?hdcore=3.1.1&plugin=aasp-3.1.1.69.124', video_id, f4m_id='hds')
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group('id')
part_id = mobj.group('part_id')
2015-01-10 17:46:01 +01:00
baseurl = mobj.group('baseurl')
webpage = self._download_webpage(url, video_id)
title = self._html_search_meta(
'title', webpage, 'title')
description = self._html_search_meta(
'description', webpage, 'description')
thumbnail = self._html_search_regex(
r'data-posterimage="([^"]+)"',
webpage, 'thumbnail', fatal=False)
upload_date = unified_strdate(self._html_search_meta(
'rightsfrom', webpage, 'upload date', fatal=False))
duration = float_or_none(self._html_search_regex(
r'data-duration="([^"]+)"',
webpage, 'duration', fatal=False))
# playlist
parts = re.findall(
r'<a href="#del=(\d+)"[^>]+data-argument="([^"]+)">([^<]+)</a>', webpage)
if parts:
entries = []
for current_part_id, stream_url, part_title in parts:
if part_id and current_part_id != part_id:
continue
video_part_id = '%s-part%s' % (video_id, current_part_id)
formats = self._extract_f4m(stream_url, video_part_id)
entries.append({
'id': video_part_id,
'title': part_title,
'description': description,
'thumbnail': thumbnail,
'upload_date': upload_date,
'formats': formats,
})
if part_id:
if entries:
return entries[0]
else:
playlist = self.playlist_result(entries, video_id, title, description)
playlist.update({
'thumbnail': thumbnail,
'upload_date': upload_date,
'duration': duration,
})
return playlist
formats = []
f4m_url = re.search(r'data-media="([^"]+)"', webpage)
if f4m_url:
formats.extend(self._extract_f4m(f4m_url.group(1), video_id))
m3u8_url = re.search(r'data-hls-media="([^"]+)"', webpage)
if m3u8_url:
2015-07-12 16:42:18 +02:00
formats.extend(self._extract_m3u8_formats(m3u8_url.group(1), video_id, 'mp4', m3u8_id='hls'))
self._sort_formats(formats)
2015-01-10 17:46:01 +01:00
subtitles_url = self._html_search_regex(
r'data-subtitlesurl[ ]*=[ ]*"([^"]+)"',
webpage, 'subtitle URL', default=None)
subtitles = None
if subtitles_url:
2015-02-18 18:57:01 +01:00
subtitles = self.extract_subtitles(subtitles_url, video_id, baseurl)
2015-01-10 17:46:01 +01:00
return {
'id': video_id,
'title': title,
'description': description,
'thumbnail': thumbnail,
'upload_date': upload_date,
'duration': duration,
'formats': formats,
2015-01-10 17:46:01 +01:00
'subtitles': subtitles,
}