mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-04 04:07:26 +00:00
38 lines
1.1 KiB
Python
38 lines
1.1 KiB
Python
# encoding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
|
|
from ..utils import (
|
|
float_or_none,
|
|
xpath_text
|
|
)
|
|
|
|
|
|
class NuevoBaseIE(InfoExtractor):
|
|
def _extract_nuevo(self, config_url, video_id):
|
|
config = self._download_xml(
|
|
config_url, video_id, transform_source=lambda s: s.strip())
|
|
|
|
title = xpath_text(config, './title', 'title', fatal=True).strip()
|
|
video_id = xpath_text(config, './mediaid', default=video_id)
|
|
thumbnail = xpath_text(config, ['./image', './thumb'])
|
|
duration = float_or_none(xpath_text(config, './duration'))
|
|
|
|
formats = []
|
|
for element_name, format_id in (('file', 'sd'), ('filehd', 'hd')):
|
|
video_url = xpath_text(config, element_name)
|
|
if video_url:
|
|
formats.append({
|
|
'url': video_url,
|
|
'format_id': format_id,
|
|
})
|
|
self._check_formats(formats, video_id)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': title,
|
|
'thumbnail': thumbnail,
|
|
'duration': duration,
|
|
'formats': formats
|
|
}
|