[movieclips] fix extraction

pull/9025/merge
remitamine 9 years ago
parent 3e0c3d14d9
commit 3f64379eda

@ -2,39 +2,48 @@
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import sanitized_Request
from ..utils import (
smuggle_url,
float_or_none,
parse_iso8601,
update_url_query,
)
class MovieClipsIE(InfoExtractor):
_VALID_URL = r'https?://(?:www.)?movieclips\.com/videos/(?P<id>[^/?#]+)'
_VALID_URL = r'https?://(?:www.)?movieclips\.com/videos/.+-(?P<id>\d+)(?:\?|$)'
_TEST = {
'url': 'http://www.movieclips.com/videos/warcraft-trailer-1-561180739597?autoPlay=true&playlistId=5',
'url': 'http://www.movieclips.com/videos/warcraft-trailer-1-561180739597',
'md5': '42b5a0352d4933a7bd54f2104f481244',
'info_dict': {
'id': 'pKIGmG83AqD9',
'display_id': 'warcraft-trailer-1-561180739597',
'ext': 'mp4',
'title': 'Warcraft Trailer 1',
'description': 'Watch Trailer 1 from Warcraft (2016). Legendarys WARCRAFT is a 3D epic adventure of world-colliding conflict based.',
'thumbnail': 're:^https?://.*\.jpg$',
'timestamp': 1446843055,
'upload_date': '20151106',
'uploader': 'Movieclips',
},
'add_ie': ['ThePlatform'],
}
def _real_extract(self, url):
display_id = self._match_id(url)
req = sanitized_Request(url)
# it doesn't work if it thinks the browser it's too old
req.add_header('User-Agent', 'Mozilla/5.0 (X11; Linux x86_64; rv:10.0) Gecko/20150101 Firefox/43.0 (Chrome)')
webpage = self._download_webpage(req, display_id)
theplatform_link = self._html_search_regex(r'src="(http://player.theplatform.com/p/.*?)"', webpage, 'theplatform link')
title = self._html_search_regex(r'<title[^>]*>([^>]+)-\s*\d+\s*|\s*Movieclips.com</title>', webpage, 'title')
description = self._html_search_meta('description', webpage)
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
video = next(v for v in self._parse_json(self._search_regex(
r'var\s+__REACT_ENGINE__\s*=\s*({.+});',
webpage, 'react engine'), video_id)['playlist']['videos'] if v['id'] == video_id)
return {
'_type': 'url_transparent',
'url': theplatform_link,
'title': title,
'display_id': display_id,
'description': description,
'ie_key': 'ThePlatform',
'url': smuggle_url(update_url_query(
video['contentUrl'], {'mbr': 'true'}), {'force_smil_url': True}),
'title': self._og_search_title(webpage),
'description': self._html_search_meta('description', webpage),
'duration': float_or_none(video.get('duration')),
'timestamp': parse_iso8601(video.get('dateCreated')),
'thumbnail': video.get('defaultImage'),
'uploader': video.get('provider'),
}

Loading…
Cancel
Save