mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-25 08:10:36 +01:00
8f1ea7cbb6
Don't rely on the XML being broken (if they fix it, our code wouldn't work anymore).
Instead, use the transform function we already have :)
This partially reverts commit c7bee2a725
.
59 lines
1.8 KiB
Python
59 lines
1.8 KiB
Python
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import fix_xml_ampersands
|
|
|
|
|
|
class EmpflixIE(InfoExtractor):
|
|
_VALID_URL = r'^https?://www\.empflix\.com/videos/.*?-(?P<id>[0-9]+)\.html'
|
|
_TEST = {
|
|
'url': 'http://www.empflix.com/videos/Amateur-Finger-Fuck-33051.html',
|
|
'md5': 'b1bc15b6412d33902d6e5952035fcabc',
|
|
'info_dict': {
|
|
'id': '33051',
|
|
'ext': 'mp4',
|
|
'title': 'Amateur Finger Fuck',
|
|
'description': 'Amateur solo finger fucking.',
|
|
'age_limit': 18,
|
|
}
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
mobj = re.match(self._VALID_URL, url)
|
|
video_id = mobj.group('id')
|
|
|
|
webpage = self._download_webpage(url, video_id)
|
|
age_limit = self._rta_search(webpage)
|
|
|
|
video_title = self._html_search_regex(
|
|
r'name="title" value="(?P<title>[^"]*)"', webpage, 'title')
|
|
video_description = self._html_search_regex(
|
|
r'name="description" value="([^"]*)"', webpage, 'description', fatal=False)
|
|
|
|
cfg_url = self._html_search_regex(
|
|
r'flashvars\.config = escape\("([^"]+)"',
|
|
webpage, 'flashvars.config')
|
|
|
|
cfg_xml = self._download_xml(
|
|
cfg_url, video_id, note='Downloading metadata',
|
|
transform_source=fix_xml_ampersands)
|
|
|
|
formats = [
|
|
{
|
|
'url': item.find('videoLink').text,
|
|
'format_id': item.find('res').text,
|
|
} for item in cfg_xml.findall('./quality/item')
|
|
]
|
|
thumbnail = cfg_xml.find('./startThumb').text
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': video_title,
|
|
'description': video_description,
|
|
'thumbnail': thumbnail,
|
|
'formats': formats,
|
|
'age_limit': age_limit,
|
|
}
|