|
|
@ -1,5 +1,7 @@ |
|
|
|
from __future__ import unicode_literals |
|
|
|
|
|
|
|
import re |
|
|
|
|
|
|
|
from ..utils import ( |
|
|
|
int_or_none, |
|
|
|
str_to_int, |
|
|
@ -21,7 +23,13 @@ class Tube8IE(KeezMoviesIE): |
|
|
|
'title': 'Kasia music video', |
|
|
|
'age_limit': 18, |
|
|
|
'duration': 230, |
|
|
|
'categories': ['Teen'], |
|
|
|
'tags': ['dancing'], |
|
|
|
}, |
|
|
|
'params': { |
|
|
|
'proxy': '127.0.0.1:8118', |
|
|
|
} |
|
|
|
|
|
|
|
}, { |
|
|
|
'url': 'http://www.tube8.com/shemale/teen/blonde-cd-gets-kidnapped-by-two-blacks-and-punished-for-being-a-slutty-girl/19569151/', |
|
|
|
'only_matching': True, |
|
|
@ -51,6 +59,17 @@ class Tube8IE(KeezMoviesIE): |
|
|
|
r'<span id="allCommentsCount">(\d+)</span>', |
|
|
|
webpage, 'comment count', fatal=False)) |
|
|
|
|
|
|
|
category = self._search_regex( |
|
|
|
r'Category:\s*</strong>\s*<a[^>]+href=[^>]+>([^<]+)', |
|
|
|
webpage, 'category', fatal=False) |
|
|
|
categories = [category] if category else None |
|
|
|
|
|
|
|
tags_str = self._search_regex( |
|
|
|
r'(?s)Tags:\s*</strong>(.+?)</(?!a)', |
|
|
|
webpage, 'tags', fatal=False) |
|
|
|
tags = [t for t in re.findall( |
|
|
|
r'<a[^>]+href=[^>]+>([^<]+)', tags_str)] if tags_str else None |
|
|
|
|
|
|
|
info.update({ |
|
|
|
'description': description, |
|
|
|
'uploader': uploader, |
|
|
@ -58,6 +77,8 @@ class Tube8IE(KeezMoviesIE): |
|
|
|
'like_count': like_count, |
|
|
|
'dislike_count': dislike_count, |
|
|
|
'comment_count': comment_count, |
|
|
|
'categories': categories, |
|
|
|
'tags': tags, |
|
|
|
}) |
|
|
|
|
|
|
|
return info |