|
|
|
@ -1,7 +1,5 @@
|
|
|
|
|
from __future__ import unicode_literals
|
|
|
|
|
|
|
|
|
|
import re
|
|
|
|
|
|
|
|
|
|
from .common import InfoExtractor
|
|
|
|
|
from ..utils import ExtractorError
|
|
|
|
|
|
|
|
|
@ -13,21 +11,22 @@ class MacGameStoreIE(InfoExtractor):
|
|
|
|
|
|
|
|
|
|
_TEST = {
|
|
|
|
|
'url': 'http://www.macgamestore.com/mediaviewer.php?trailer=2450',
|
|
|
|
|
'file': '2450.m4v',
|
|
|
|
|
'md5': '8649b8ea684b6666b4c5be736ecddc61',
|
|
|
|
|
'info_dict': {
|
|
|
|
|
'id': '2450',
|
|
|
|
|
'ext': 'm4v',
|
|
|
|
|
'title': 'Crow',
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
def _real_extract(self, url):
|
|
|
|
|
mobj = re.match(self._VALID_URL, url)
|
|
|
|
|
video_id = mobj.group('id')
|
|
|
|
|
|
|
|
|
|
webpage = self._download_webpage(url, video_id, 'Downloading trailer page')
|
|
|
|
|
video_id = self._match_id(url)
|
|
|
|
|
webpage = self._download_webpage(
|
|
|
|
|
url, video_id, 'Downloading trailer page')
|
|
|
|
|
|
|
|
|
|
if re.search(r'>Missing Media<', webpage) is not None:
|
|
|
|
|
raise ExtractorError('Trailer %s does not exist' % video_id, expected=True)
|
|
|
|
|
if '>Missing Media<' in webpage:
|
|
|
|
|
raise ExtractorError(
|
|
|
|
|
'Trailer %s does not exist' % video_id, expected=True)
|
|
|
|
|
|
|
|
|
|
video_title = self._html_search_regex(
|
|
|
|
|
r'<title>MacGameStore: (.*?) Trailer</title>', webpage, 'title')
|
|
|
|
|