|
|
- # coding: utf-8
- from __future__ import unicode_literals
-
- import re
-
- from .common import InfoExtractor
- from ..compat import compat_str
- from ..utils import (
- determine_ext,
- ExtractorError,
- int_or_none,
- unescapeHTML,
- )
-
-
- class MSNIE(InfoExtractor):
- _VALID_URL = r'https?://(?:www\.)?msn\.com/(?:[^/]+/)+(?P<display_id>[^/]+)/[a-z]{2}-(?P<id>[\da-zA-Z]+)'
- _TESTS = [{
- 'url': 'http://www.msn.com/en-ae/foodanddrink/joinourtable/criminal-minds-shemar-moore-shares-a-touching-goodbye-message/vp-BBqQYNE',
- 'md5': '8442f66c116cbab1ff7098f986983458',
- 'info_dict': {
- 'id': 'BBqQYNE',
- 'display_id': 'criminal-minds-shemar-moore-shares-a-touching-goodbye-message',
- 'ext': 'mp4',
- 'title': 'Criminal Minds - Shemar Moore Shares A Touching Goodbye Message',
- 'description': 'md5:e8e89b897b222eb33a6b5067a8f1bc25',
- 'duration': 104,
- 'uploader': 'CBS Entertainment',
- 'uploader_id': 'IT0X5aoJ6bJgYerJXSDCgFmYPB1__54v',
- },
- }, {
- 'url': 'http://www.msn.com/en-ae/news/offbeat/meet-the-nine-year-old-self-made-millionaire/ar-BBt6ZKf',
- 'only_matching': True,
- }, {
- 'url': 'http://www.msn.com/en-ae/video/watch/obama-a-lot-of-people-will-be-disappointed/vi-AAhxUMH',
- 'only_matching': True,
- }, {
- # geo restricted
- 'url': 'http://www.msn.com/en-ae/foodanddrink/joinourtable/the-first-fart-makes-you-laugh-the-last-fart-makes-you-cry/vp-AAhzIBU',
- 'only_matching': True,
- }, {
- 'url': 'http://www.msn.com/en-ae/entertainment/bollywood/watch-how-salman-khan-reacted-when-asked-if-he-would-apologize-for-his-‘raped-woman’-comment/vi-AAhvzW6',
- 'only_matching': True,
- }]
-
- def _real_extract(self, url):
- mobj = re.match(self._VALID_URL, url)
- video_id, display_id = mobj.group('id', 'display_id')
-
- webpage = self._download_webpage(url, display_id)
-
- video = self._parse_json(
- self._search_regex(
- r'data-metadata\s*=\s*(["\'])(?P<data>.+?)\1',
- webpage, 'video data', default='{}', group='data'),
- display_id, transform_source=unescapeHTML)
-
- if not video:
- error = unescapeHTML(self._search_regex(
- r'data-error=(["\'])(?P<error>.+?)\1',
- webpage, 'error', group='error'))
- raise ExtractorError('%s said: %s' % (self.IE_NAME, error), expected=True)
-
- title = video['title']
-
- formats = []
- for file_ in video.get('videoFiles', []):
- format_url = file_.get('url')
- if not format_url:
- continue
- ext = determine_ext(format_url)
- if ext == 'ism':
- formats.extend(self._extract_ism_formats(
- format_url + '/Manifest', display_id, 'mss', fatal=False))
- if 'm3u8' in format_url:
- # m3u8_native should not be used here until
- # https://github.com/rg3/youtube-dl/issues/9913 is fixed
- m3u8_formats = self._extract_m3u8_formats(
- format_url, display_id, 'mp4',
- m3u8_id='hls', fatal=False)
- formats.extend(m3u8_formats)
- else:
- formats.append({
- 'url': format_url,
- 'ext': 'mp4',
- 'format_id': 'http',
- 'width': int_or_none(file_.get('width')),
- 'height': int_or_none(file_.get('height')),
- })
- self._sort_formats(formats)
-
- subtitles = {}
- for file_ in video.get('files', []):
- format_url = file_.get('url')
- format_code = file_.get('formatCode')
- if not format_url or not format_code:
- continue
- if compat_str(format_code) == '3100':
- subtitles.setdefault(file_.get('culture', 'en'), []).append({
- 'ext': determine_ext(format_url, 'ttml'),
- 'url': format_url,
- })
-
- return {
- 'id': video_id,
- 'display_id': display_id,
- 'title': title,
- 'description': video.get('description'),
- 'thumbnail': video.get('headlineImage', {}).get('url'),
- 'duration': int_or_none(video.get('durationSecs')),
- 'uploader': video.get('sourceFriendly'),
- 'uploader_id': video.get('providerId'),
- 'creator': video.get('creator'),
- 'subtitles': subtitles,
- 'formats': formats,
- }
|