yt-dlc/youtube_dl/extractor/imdb.py
Jaime Marquínez Ferrándiz acf37ca151 [imdb] Fix the resolution values (fixes #1847)
We were using the size of the player, it was the same for all the formats
2013-11-29 07:56:14 +01:00

60 lines
2.1 KiB
Python

import re
import json
from .common import InfoExtractor
from ..utils import (
compat_urlparse,
get_element_by_attribute,
)
class ImdbIE(InfoExtractor):
IE_NAME = u'imdb'
IE_DESC = u'Internet Movie Database trailers'
_VALID_URL = r'http://www\.imdb\.com/video/imdb/vi(?P<id>\d+)'
_TEST = {
u'url': u'http://www.imdb.com/video/imdb/vi2524815897',
u'md5': u'9f34fa777ade3a6e57a054fdbcb3a068',
u'info_dict': {
u'id': u'2524815897',
u'ext': u'mp4',
u'title': u'Ice Age: Continental Drift Trailer (No. 2) - IMDb',
u'description': u'md5:9061c2219254e5d14e03c25c98e96a81',
u'duration': 151,
}
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group('id')
webpage = self._download_webpage(url,video_id)
descr = get_element_by_attribute('itemprop', 'description', webpage)
available_formats = re.findall(
r'case \'(?P<f_id>.*?)\' :$\s+url = \'(?P<path>.*?)\'', webpage,
flags=re.MULTILINE)
formats = []
for f_id, f_path in available_formats:
format_page = self._download_webpage(
compat_urlparse.urljoin(url, f_path),
u'Downloading info for %s format' % f_id)
json_data = self._search_regex(
r'<script[^>]+class="imdb-player-data"[^>]*?>(.*?)</script>',
format_page, u'json data', flags=re.DOTALL)
info = json.loads(json_data)
format_info = info['videoPlayerObject']['video']
formats.append({
'format_id': f_id,
'url': format_info['url'],
'height': int(info['titleObject']['encoding']['selected'][:-1]),
})
return {
'id': video_id,
'title': self._og_search_title(webpage),
'formats': formats,
'description': descr,
'thumbnail': format_info['slate'],
'duration': int(info['titleObject']['title']['duration_seconds']),
}