1
0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-11-10 19:37:25 +00:00

[expressen] Add extractor

This commit is contained in:
Sergey M․ 2018-06-16 21:14:36 +07:00
parent 81c5df4f2c
commit 734d461ca0
No known key found for this signature in database
GPG Key ID: 2C393E0F18A9236D
2 changed files with 78 additions and 0 deletions

View File

@ -0,0 +1,77 @@
# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
determine_ext,
int_or_none,
unescapeHTML,
unified_timestamp,
)
class ExpressenIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?expressen\.se/tv/(?:[^/]+/)*(?P<id>[^/?#&]+)'
_TESTS = [{
'url': 'https://www.expressen.se/tv/ledare/ledarsnack/ledarsnack-om-arbetslosheten-bland-kvinnor-i-speciellt-utsatta-omraden/',
'md5': '2fbbe3ca14392a6b1b36941858d33a45',
'info_dict': {
'id': '8690962',
'ext': 'mp4',
'title': 'Ledarsnack: Om arbetslösheten bland kvinnor i speciellt utsatta områden',
'description': 'md5:f38c81ff69f3de4d269bbda012fcbbba',
'thumbnail': r're:^https?://.*\.jpg$',
'duration': 788,
'timestamp': 1526639109,
'upload_date': '20180518',
},
}, {
'url': 'https://www.expressen.se/tv/kultur/kulturdebatt-med-expressens-karin-olsson/',
'only_matching': True,
}]
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
def extract_data(name):
return self._parse_json(
self._search_regex(
r'data-%s=(["\'])(?P<value>(?:(?!\1).)+)\1' % name,
webpage, 'info', group='value'),
display_id, transform_source=unescapeHTML)
info = extract_data('video-tracking-info')
video_id = info['videoId']
data = extract_data('article-data')
stream = data['stream']
if determine_ext(stream) == 'm3u8':
formats = self._extract_m3u8_formats(
stream, display_id, 'mp4', entry_protocol='m3u8_native',
m3u8_id='hls')
else:
formats = [{
'url': stream,
}]
self._sort_formats(formats)
title = info.get('titleRaw') or data['title']
description = info.get('descriptionRaw')
thumbnail = info.get('socialMediaImage') or data.get('image')
duration = int_or_none(info.get('videoTotalSecondsDuration') or
data.get('totalSecondsDuration'))
timestamp = unified_timestamp(info.get('publishDate'))
return {
'id': video_id,
'display_id': display_id,
'title': title,
'description': description,
'thumbnail': thumbnail,
'duration': duration,
'timestamp': timestamp,
'formats': formats,
}

View File

@ -335,6 +335,7 @@ from .esri import EsriVideoIE
from .europa import EuropaIE
from .everyonesmixtape import EveryonesMixtapeIE
from .expotv import ExpoTVIE
from .expressen import ExpressenIE
from .extremetube import ExtremeTubeIE
from .eyedotv import EyedoTVIE
from .facebook import (