mirror of
https://github.com/ytdl-org/youtube-dl
synced 2024-12-05 03:02:56 +01:00
127 lines
4.5 KiB
Python
127 lines
4.5 KiB
Python
# encoding: utf-8
|
||
from __future__ import unicode_literals
|
||
|
||
import re
|
||
|
||
from .common import InfoExtractor
|
||
from ..utils import (
|
||
determine_ext,
|
||
int_or_none,
|
||
float_or_none,
|
||
parse_iso8601,
|
||
remove_end,
|
||
)
|
||
|
||
|
||
class TV2IE(InfoExtractor):
|
||
_VALID_URL = 'http://(?:www\.)?tv2\.no/v/(?P<id>\d+)'
|
||
_TEST = {
|
||
'url': 'http://www.tv2.no/v/916509/',
|
||
'md5': '9cb9e3410b18b515d71892f27856e9b1',
|
||
'info_dict': {
|
||
'id': '916509',
|
||
'ext': 'flv',
|
||
'title': 'Se Gryttens hyllest av Steven Gerrard',
|
||
'description': 'TV 2 Sportens huspoet tar avskjed med Liverpools kaptein Steven Gerrard.',
|
||
'timestamp': 1431715610,
|
||
'upload_date': '20150515',
|
||
'duration': 156.967,
|
||
'view_count': int,
|
||
'categories': list,
|
||
}
|
||
}
|
||
|
||
def _real_extract(self, url):
|
||
video_id = self._match_id(url)
|
||
|
||
formats = []
|
||
format_urls = []
|
||
for protocol in ('HDS', 'HLS'):
|
||
data = self._download_json(
|
||
'http://sumo.tv2.no/api/web/asset/%s/play.json?protocol=%s&videoFormat=SMIL+ISMUSP' % (video_id, protocol),
|
||
video_id, 'Downloading play JSON')['playback']
|
||
for item in data['items']['item']:
|
||
video_url = item.get('url')
|
||
if not video_url or video_url in format_urls:
|
||
continue
|
||
format_id = '%s-%s' % (protocol.lower(), item.get('mediaFormat'))
|
||
if not self._is_valid_url(video_url, video_id, format_id):
|
||
continue
|
||
format_urls.append(video_url)
|
||
ext = determine_ext(video_url)
|
||
if ext == 'f4m':
|
||
formats.extend(self._extract_f4m_formats(
|
||
video_url, video_id, f4m_id=format_id))
|
||
elif ext == 'm3u8':
|
||
formats.extend(self._extract_m3u8_formats(
|
||
video_url, video_id, 'mp4', m3u8_id=format_id))
|
||
elif ext == 'ism' or video_url.endswith('.ism/Manifest'):
|
||
pass
|
||
else:
|
||
formats.append({
|
||
'url': video_url,
|
||
'format_id': format_id,
|
||
'tbr': int_or_none(item.get('bitrate')),
|
||
'filesize': int_or_none(item.get('fileSize')),
|
||
})
|
||
self._sort_formats(formats)
|
||
|
||
asset = self._download_json(
|
||
'http://sumo.tv2.no/api/web/asset/%s.json' % video_id,
|
||
video_id, 'Downloading metadata JSON')['asset']
|
||
|
||
title = asset['title']
|
||
description = asset.get('description')
|
||
timestamp = parse_iso8601(asset.get('createTime'))
|
||
duration = float_or_none(asset.get('accurateDuration') or asset.get('duration'))
|
||
view_count = int_or_none(asset.get('views'))
|
||
categories = asset.get('keywords', '').split(',')
|
||
|
||
thumbnails = [{
|
||
'id': thumbnail.get('@type'),
|
||
'url': thumbnail.get('url'),
|
||
} for _, thumbnail in asset.get('imageVersions', {}).items()]
|
||
|
||
return {
|
||
'id': video_id,
|
||
'url': video_url,
|
||
'title': title,
|
||
'description': description,
|
||
'thumbnails': thumbnails,
|
||
'timestamp': timestamp,
|
||
'duration': duration,
|
||
'view_count': view_count,
|
||
'categories': categories,
|
||
'formats': formats,
|
||
}
|
||
|
||
|
||
class TV2ArticleIE(InfoExtractor):
|
||
_VALID_URL = 'http://(?:www\.)?tv2\.no/(?:a|\d{4}/\d{2}/\d{2}(/[^/]+)+)/(?P<id>\d+)'
|
||
_TESTS = [{
|
||
'url': 'http://www.tv2.no/2015/05/16/nyheter/alesund/krim/pingvin/6930542',
|
||
'info_dict': {
|
||
'id': '6930542',
|
||
'title': 'Russen hetses etter pingvintyveri – innrømmer å ha åpnet luken på buret',
|
||
'description': 'md5:339573779d3eea3542ffe12006190954',
|
||
},
|
||
'playlist_count': 2,
|
||
}, {
|
||
'url': 'http://www.tv2.no/a/6930542',
|
||
'only_matching': True,
|
||
}]
|
||
|
||
def _real_extract(self, url):
|
||
playlist_id = self._match_id(url)
|
||
|
||
webpage = self._download_webpage(url, playlist_id)
|
||
|
||
entries = [
|
||
self.url_result('http://www.tv2.no/v/%s' % video_id, 'TV2')
|
||
for video_id in re.findall(r'data-assetid="(\d+)"', webpage)]
|
||
|
||
title = remove_end(self._og_search_title(webpage), ' - TV2.no')
|
||
description = remove_end(self._og_search_description(webpage), ' - TV2.no')
|
||
|
||
return self.playlist_result(entries, playlist_id, title, description)
|