|
|
from __future__ import unicode_literals
import os import re
from .common import InfoExtractor from ..compat import ( compat_urllib_parse, compat_urllib_parse_urlparse, compat_urllib_request, ) from ..utils import ( ExtractorError, str_to_int, ) from ..aes import ( aes_decrypt_text )
class PornHubIE(InfoExtractor): _VALID_URL = r'https?://(?:www\.)?pornhub\.com/view_video\.php\?viewkey=(?P<id>[0-9a-f]+)' _TEST = { 'url': 'http://www.pornhub.com/view_video.php?viewkey=648719015', 'md5': '882f488fa1f0026f023f33576004a2ed', 'info_dict': { 'id': '648719015', 'ext': 'mp4', "uploader": "Babes", "title": "Seductive Indian beauty strips down and fingers her pink pussy", "age_limit": 18 } }
def _extract_count(self, pattern, webpage, name): count = self._html_search_regex(pattern, webpage, '%s count' % name, fatal=False) if count: count = str_to_int(count) return count
def _real_extract(self, url): video_id = self._match_id(url)
req = compat_urllib_request.Request(url) req.add_header('Cookie', 'age_verified=1') webpage = self._download_webpage(req, video_id)
error_msg = self._html_search_regex( r'(?s)<div class="userMessageSection[^"]*".*?>(.*?)</div>', webpage, 'error message', default=None) if error_msg: error_msg = re.sub(r'\s+', ' ', error_msg) raise ExtractorError( 'PornHub said: %s' % error_msg, expected=True, video_id=video_id)
video_title = self._html_search_regex(r'<h1 [^>]+>([^<]+)', webpage, 'title') video_uploader = self._html_search_regex( r'(?s)From: .+?<(?:a href="/users/|a href="/channels/|<span class="username)[^>]+>(.+?)<', webpage, 'uploader', fatal=False) thumbnail = self._html_search_regex(r'"image_url":"([^"]+)', webpage, 'thumbnail', fatal=False) if thumbnail: thumbnail = compat_urllib_parse.unquote(thumbnail)
view_count = self._extract_count(r'<span class="count">([\d,\.]+)</span> views', webpage, 'view') like_count = self._extract_count(r'<span class="votesUp">([\d,\.]+)</span>', webpage, 'like') dislike_count = self._extract_count(r'<span class="votesDown">([\d,\.]+)</span>', webpage, 'dislike') comment_count = self._extract_count( r'All comments \(<var class="videoCommentCount">([\d,\.]+)</var>', webpage, 'comment')
video_urls = list(map(compat_urllib_parse.unquote, re.findall(r'"quality_[0-9]{3}p":"([^"]+)', webpage))) if webpage.find('"encrypted":true') != -1: password = compat_urllib_parse.unquote_plus(self._html_search_regex(r'"video_title":"([^"]+)', webpage, 'password')) video_urls = list(map(lambda s: aes_decrypt_text(s, password, 32).decode('utf-8'), video_urls))
formats = [] for video_url in video_urls: path = compat_urllib_parse_urlparse(video_url).path extension = os.path.splitext(path)[1][1:] format = path.split('/')[5].split('_')[:2] format = "-".join(format)
m = re.match(r'^(?P<height>[0-9]+)P-(?P<tbr>[0-9]+)K$', format) if m is None: height = None tbr = None else: height = int(m.group('height')) tbr = int(m.group('tbr'))
formats.append({ 'url': video_url, 'ext': extension, 'format': format, 'format_id': format, 'tbr': tbr, 'height': height, }) self._sort_formats(formats)
return { 'id': video_id, 'uploader': video_uploader, 'title': video_title, 'thumbnail': thumbnail, 'view_count': view_count, 'like_count': like_count, 'dislike_count': dislike_count, 'comment_count': comment_count, 'formats': formats, 'age_limit': 18, }
|