mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-12-30 00:26:24 +00:00
[pornovoisines] Add extractor
This commit is contained in:
parent
93f787070f
commit
575dad3c98
@ -382,6 +382,7 @@ from .pornhub import (
|
|||||||
PornHubPlaylistIE,
|
PornHubPlaylistIE,
|
||||||
)
|
)
|
||||||
from .pornotube import PornotubeIE
|
from .pornotube import PornotubeIE
|
||||||
|
from .pornovoisines import PornoVoisinesIE
|
||||||
from .pornoxo import PornoXOIE
|
from .pornoxo import PornoXOIE
|
||||||
from .primesharetv import PrimeShareTVIE
|
from .primesharetv import PrimeShareTVIE
|
||||||
from .promptfile import PromptFileIE
|
from .promptfile import PromptFileIE
|
||||||
|
101
youtube_dl/extractor/pornovoisines.py
Normal file
101
youtube_dl/extractor/pornovoisines.py
Normal file
@ -0,0 +1,101 @@
|
|||||||
|
# coding: utf-8
|
||||||
|
from __future__ import unicode_literals
|
||||||
|
|
||||||
|
import re
|
||||||
|
import datetime
|
||||||
|
import random
|
||||||
|
|
||||||
|
from ..compat import compat_urllib_parse
|
||||||
|
from .common import InfoExtractor
|
||||||
|
|
||||||
|
class PornoVoisinesIE(InfoExtractor):
|
||||||
|
_VALID_URL = r'^((?:http://)?(?:www\.)?pornovoisines.com)/showvideo/(\d+)/([^/]+)'
|
||||||
|
|
||||||
|
VIDEO_URL_TEMPLATE = 'http://stream%d.pornovoisines.com' \
|
||||||
|
'/static/media/video/transcoded/%s-640x360-1000-trscded.mp4'
|
||||||
|
|
||||||
|
SERVER_NUMBERS = (1, 2)
|
||||||
|
|
||||||
|
_TEST = {
|
||||||
|
'url': 'http://www.pornovoisines.com/showvideo/1285/recherche-appartement/',
|
||||||
|
'md5': '5ac670803bc12e9e7f9f662ce64cf1d1',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '1285',
|
||||||
|
'display_id': 'recherche-appartement',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': "Recherche appartement",
|
||||||
|
'upload_date': '20140925',
|
||||||
|
'view_count': int,
|
||||||
|
'duration': 120,
|
||||||
|
'categories': ["Débutante", "Scénario", "Sodomie"],
|
||||||
|
'description': 're:^Pour la .+ original...$',
|
||||||
|
'thumbnail': 're:^http://',
|
||||||
|
'uploader': "JMTV",
|
||||||
|
'average_rating': float,
|
||||||
|
'comment_count': int,
|
||||||
|
'age_limit': 18,
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
@classmethod
|
||||||
|
def build_video_url(cls, id):
|
||||||
|
server_nr = random.choice(cls.SERVER_NUMBERS)
|
||||||
|
return cls.VIDEO_URL_TEMPLATE % (server_nr, id)
|
||||||
|
|
||||||
|
@staticmethod
|
||||||
|
def parse_upload_date(str):
|
||||||
|
return datetime.datetime.strptime(str, "%d-%m-%Y").strftime("%Y%m%d")
|
||||||
|
|
||||||
|
@staticmethod
|
||||||
|
def parse_categories(str):
|
||||||
|
return map(lambda s: s.strip(), str.split(','))
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
mobj = re.match(self._VALID_URL, url)
|
||||||
|
url_prefix = mobj.group(1)
|
||||||
|
id = mobj.group(2)
|
||||||
|
display_id = mobj.group(3)
|
||||||
|
|
||||||
|
webpage = self._download_webpage(url, id)
|
||||||
|
|
||||||
|
title = self._html_search_regex(r'<h1>(.+?)</h1>', webpage, 'title',
|
||||||
|
flags=re.DOTALL)
|
||||||
|
url = self.build_video_url(id)
|
||||||
|
upload_date = self.parse_upload_date(
|
||||||
|
self._search_regex(r'Publié le (\d\d-\d\d-\d{4})', webpage,
|
||||||
|
'upload date'))
|
||||||
|
view_count = int(self._search_regex(r'(\d+) vues', webpage, 'view count'))
|
||||||
|
duration = int(self._search_regex('Durée (\d+)', webpage, 'duration'))
|
||||||
|
categories = self.parse_categories(self._html_search_regex(
|
||||||
|
r'<li class="categorie">(.+?)</li>', webpage, "categories",
|
||||||
|
flags=re.DOTALL))
|
||||||
|
description = self._html_search_regex(
|
||||||
|
r'<article id="descriptif">(.+?)</article>', webpage, "description",
|
||||||
|
flags=re.DOTALL)
|
||||||
|
thumbnail = url_prefix + self._html_search_regex(re.compile(
|
||||||
|
'<div id="mediaspace' + id + '">.*?<img src="(.+?)"', re.DOTALL),
|
||||||
|
webpage, "thumbnail")
|
||||||
|
uploader = re.sub(r' *\| *$', '',
|
||||||
|
self._html_search_regex(r'<li class="auteur">(.+?)</li>', webpage,
|
||||||
|
"uploader", flags=re.DOTALL))
|
||||||
|
average_rating = float(self._search_regex(r'Note : (\d+,\d+)',
|
||||||
|
webpage, "average rating").replace(',', '.'))
|
||||||
|
comment_count = int(self._search_regex(r'\((\d+)\)', webpage,
|
||||||
|
"comment count"))
|
||||||
|
|
||||||
|
return {
|
||||||
|
'id': id,
|
||||||
|
'display_id': display_id,
|
||||||
|
'url': url,
|
||||||
|
'title': title,
|
||||||
|
'upload_date': upload_date,
|
||||||
|
'view_count': view_count,
|
||||||
|
'duration': duration,
|
||||||
|
'categories': categories,
|
||||||
|
'description': description,
|
||||||
|
'thumbnail': thumbnail,
|
||||||
|
'uploader': uploader,
|
||||||
|
'average_rating': average_rating,
|
||||||
|
'comment_count': comment_count,
|
||||||
|
'age_limit': 18,
|
||||||
|
}
|
Loading…
Reference in New Issue
Block a user