yt-dlp/youtube_dl/extractor/nuvid.py
2014-06-09 20:41:33 +07:00

69 lines
2.5 KiB
Python

from __future__ import unicode_literals
import re
from .common import InfoExtractor
from ..utils import (
parse_duration,
unified_strdate,
compat_urllib_request,
)
class NuvidIE(InfoExtractor):
_VALID_URL = r'^https?://(?:www|m)\.nuvid\.com/video/(?P<id>[0-9]+)'
_TEST = {
'url': 'http://m.nuvid.com/video/1310741/',
'md5': 'eab207b7ac4fccfb4e23c86201f11277',
'info_dict': {
'id': '1310741',
'ext': 'mp4',
'title': 'Horny babes show their awesome bodeis and',
'duration': 129,
'upload_date': '20140508',
'age_limit': 18,
}
}
def _real_extract(self, url):
mobj = re.match(self._VALID_URL, url)
video_id = mobj.group('id')
formats = []
for dwnld_speed, format_id in [(0, '3gp'), (5, 'mp4')]:
request = compat_urllib_request.Request(
'http://m.nuvid.com/play/%s' % video_id)
request.add_header('Cookie', 'skip_download_page=1; dwnld_speed=%d; adv_show=1' % dwnld_speed)
webpage = self._download_webpage(
request, video_id, 'Downloading %s page' % format_id)
video_url = self._html_search_regex(
r'<a href="([^"]+)"\s*>Continue to watch video', webpage, '%s video URL' % format_id, fatal=False)
if not video_url:
continue
formats.append({
'url': video_url,
'format_id': format_id,
})
webpage = self._download_webpage(
'http://m.nuvid.com/video/%s' % video_id, video_id, 'Downloading video page')
title = self._html_search_regex(
r'<div class="title">\s+<h2[^>]*>([^<]+)</h2>', webpage, 'title').strip()
thumbnail = self._html_search_regex(
r'href="(/thumbs/[^"]+)"[^>]*data-link_type="thumbs"',
webpage, 'thumbnail URL', fatal=False)
duration = parse_duration(self._html_search_regex(
r'Length:\s*<span>(\d{2}:\d{2})</span>',webpage, 'duration', fatal=False))
upload_date = unified_strdate(self._html_search_regex(
r'Added:\s*<span>(\d{4}-\d{2}-\d{2})</span>', webpage, 'upload date', fatal=False))
return {
'id': video_id,
'title': title,
'thumbnail': 'http://m.nuvid.com%s' % thumbnail,
'duration': duration,
'upload_date': upload_date,
'age_limit': 18,
'formats': formats,
}