mirror of
https://codeberg.org/polarisfm/youtube-dl
synced 2024-11-22 16:44:32 +01:00
Merge remote-tracking branch 'peugeot/eporner'
This commit is contained in:
commit
04b4aa4a7b
@ -86,6 +86,7 @@ from .ellentv import (
|
||||
from .elpais import ElPaisIE
|
||||
from .empflix import EmpflixIE
|
||||
from .engadget import EngadgetIE
|
||||
from .eporner import EpornerIE
|
||||
from .escapist import EscapistIE
|
||||
from .everyonesmixtape import EveryonesMixtapeIE
|
||||
from .exfm import ExfmIE
|
||||
|
49
youtube_dl/extractor/eporner.py
Normal file
49
youtube_dl/extractor/eporner.py
Normal file
@ -0,0 +1,49 @@
|
||||
# coding: utf-8
|
||||
from __future__ import unicode_literals
|
||||
|
||||
import re
|
||||
|
||||
from .common import InfoExtractor
|
||||
from ..utils import int_or_none
|
||||
|
||||
class EpornerIE(InfoExtractor):
|
||||
_VALID_URL = r'http?://(?:www\.)?eporner\.com/hd-porn/(?P<id>\d+)/(?P<title_dash>[\w-]+)/?'
|
||||
_TEST = {
|
||||
'url': 'http://www.eporner.com/hd-porn/95008/Infamous-Tiffany-Teen-Strip-Tease-Video/',
|
||||
'md5': '3b427ae4b9d60619106de3185c2987cd',
|
||||
'info_dict': {
|
||||
'id': '95008',
|
||||
'ext': 'flv',
|
||||
'title': 'Infamous Tiffany Teen Strip Tease Video',
|
||||
'duration': 194
|
||||
}
|
||||
}
|
||||
|
||||
def _real_extract(self, url):
|
||||
mobj = re.match(self._VALID_URL, url)
|
||||
video_id = mobj.group('id')
|
||||
webpage = self._download_webpage(url, video_id)
|
||||
title = self._html_search_regex(r'<title>(.*?) - EPORNER', webpage, 'title')
|
||||
|
||||
redirect_code = self._html_search_regex(r'<script type="text/javascript" src="/config5/'+str(video_id)+'/([a-f\d]+)/">', webpage, 'redirect_code')
|
||||
redirect_url = 'http://www.eporner.com/config5/' + str(video_id) +'/'+ redirect_code
|
||||
webpage2 = self._download_webpage(redirect_url, video_id)
|
||||
video_url = self._html_search_regex(r'file: "(.*?)",', webpage2, 'video_url')
|
||||
|
||||
mobj = re.search(r'class="mbtim">(?P<minutes>\d+):(?P<seconds>\d+)</div>', webpage)
|
||||
duration = int(mobj.group('minutes')) * 60 + int(mobj.group('seconds')) if mobj else None
|
||||
|
||||
mobj = re.search(r'id="cinemaviews">((?P<thousands>\d+),)?(?P<units>\d+)<small>views', webpage)
|
||||
try:
|
||||
view_count = int(mobj.group('units'))
|
||||
view_count += int(mobj.group('thousands')) * 1000
|
||||
except:
|
||||
pass
|
||||
|
||||
return {
|
||||
'id': video_id,
|
||||
'url': video_url,
|
||||
'title': title,
|
||||
'duration': int_or_none(duration),
|
||||
'view_count': int_or_none(view_count),
|
||||
}
|
Loading…
Reference in New Issue
Block a user