mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-27 10:52:34 +00:00
54 lines
1.9 KiB
Python
54 lines
1.9 KiB
Python
import re
|
|
import operator
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
fix_xml_all_ampersand,
|
|
)
|
|
|
|
|
|
class MetacriticIE(InfoExtractor):
|
|
_VALID_URL = r'https?://www\.metacritic\.com/.+?/trailers/(?P<id>\d+)'
|
|
|
|
_TEST = {
|
|
u'url': u'http://www.metacritic.com/game/playstation-4/infamous-second-son/trailers/3698222',
|
|
u'file': u'3698222.mp4',
|
|
u'info_dict': {
|
|
u'title': u'inFamous: Second Son - inSide Sucker Punch: Smoke & Mirrors',
|
|
u'description': u'Take a peak behind-the-scenes to see how Sucker Punch brings smoke into the universe of inFAMOUS Second Son on the PS4.',
|
|
u'duration': 221,
|
|
},
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
mobj = re.match(self._VALID_URL, url)
|
|
video_id = mobj.group('id')
|
|
webpage = self._download_webpage(url, video_id)
|
|
# The xml is not well formatted, there are raw '&'
|
|
info = self._download_xml('http://www.metacritic.com/video_data?video=' + video_id,
|
|
video_id, u'Downloading info xml', transform_source=fix_xml_all_ampersand)
|
|
|
|
clip = next(c for c in info.findall('playList/clip') if c.find('id').text == video_id)
|
|
formats = []
|
|
for videoFile in clip.findall('httpURI/videoFile'):
|
|
rate_str = videoFile.find('rate').text
|
|
video_url = videoFile.find('filePath').text
|
|
formats.append({
|
|
'url': video_url,
|
|
'ext': 'mp4',
|
|
'format_id': rate_str,
|
|
'rate': int(rate_str),
|
|
})
|
|
formats.sort(key=operator.itemgetter('rate'))
|
|
|
|
description = self._html_search_regex(r'<b>Description:</b>(.*?)</p>',
|
|
webpage, u'description', flags=re.DOTALL)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': clip.find('title').text,
|
|
'formats': formats,
|
|
'description': description,
|
|
'duration': int(clip.find('duration').text),
|
|
}
|