Browse Source

[discoverygo] Add extractor (Closes #10245)

master
Sergey M․ 8 years ago
parent
commit
d92cb46305
No known key found for this signature in database GPG Key ID: 2C393E0F18A9236D
2 changed files with 99 additions and 0 deletions
  1. 98
      youtube_dl/extractor/discoverygo.py
  2. 1
      youtube_dl/extractor/extractors.py

98
youtube_dl/extractor/discoverygo.py

@ -0,0 +1,98 @@
from __future__ import unicode_literals
from .common import InfoExtractor
from ..compat import compat_str
from ..utils import (
extract_attributes,
int_or_none,
parse_age_limit,
unescapeHTML,
)
class DiscoveryGoIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?discoverygo\.com/(?:[^/]+/)*(?P<id>[^/?#&]+)'
_TEST = {
'url': 'https://www.discoverygo.com/love-at-first-kiss/kiss-first-ask-questions-later/',
'info_dict': {
'id': '57a33c536b66d1cd0345eeb1',
'ext': 'mp4',
'title': 'Kiss First, Ask Questions Later!',
'description': 'md5:fe923ba34050eae468bffae10831cb22',
'duration': 2579,
'series': 'Love at First Kiss',
'season_number': 1,
'episode_number': 1,
'age_limit': 14,
},
}
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
container = extract_attributes(
self._search_regex(
r'(<div[^>]+class=["\']video-player-container[^>]+>)',
webpage, 'video container'))
video = self._parse_json(
unescapeHTML(container.get('data-video') or container.get('data-json')),
display_id)
title = video['name']
stream = video['stream']
STREAM_URL_SUFFIX = 'streamUrl'
formats = []
for stream_kind in ('', 'hds'):
suffix = STREAM_URL_SUFFIX.capitalize() if stream_kind else STREAM_URL_SUFFIX
stream_url = stream.get('%s%s' % (stream_kind, suffix))
if not stream_url:
continue
if stream_kind == '':
formats.extend(self._extract_m3u8_formats(
stream_url, display_id, 'mp4', entry_protocol='m3u8_native',
m3u8_id='hls', fatal=False))
elif stream_kind == 'hds':
formats.extend(self._extract_f4m_formats(
stream_url, display_id, f4m_id=stream_kind, fatal=False))
self._sort_formats(formats)
video_id = video.get('id') or display_id
description = video.get('description', {}).get('detailed')
duration = int_or_none(video.get('duration'))
series = video.get('show', {}).get('name')
season_number = int_or_none(video.get('season', {}).get('number'))
episode_number = int_or_none(video.get('episodeNumber'))
tags = video.get('tags')
age_limit = parse_age_limit(video.get('parental', {}).get('rating'))
subtitles = {}
captions = stream.get('captions')
if isinstance(captions, list):
for caption in captions:
subtitle_url = caption.get('fileUrl')
if (not subtitle_url or not isinstance(subtitle_url, compat_str) or
not subtitle_url.startswith('http')):
continue
lang = caption.get('fileLang', 'en')
subtitles.setdefault(lang, []).append({'url': subtitle_url})
return {
'id': video_id,
'display_id': display_id,
'title': title,
'description': description,
'duration': duration,
'series': series,
'season_number': season_number,
'episode_number': episode_number,
'tags': tags,
'age_limit': age_limit,
'formats': formats,
'subtitles': subtitles,
}

1
youtube_dl/extractor/extractors.py

@ -221,6 +221,7 @@ from .dvtv import DVTVIE
from .dumpert import DumpertIE
from .defense import DefenseGouvFrIE
from .discovery import DiscoveryIE
from .discoverygo import DiscoveryGoIE
from .dispeak import DigitallySpeakingIE
from .dropbox import DropboxIE
from .dw import (

Loading…
Cancel
Save