1
0
Fork 0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-11-28 13:11:49 +00:00
youtube-dl/youtube_dl/extractor/tube8.py
2024-07-06 12:10:32 +02:00

260 lines
9.3 KiB
Python

from __future__ import unicode_literals
import itertools
import re
from time import sleep
from .common import InfoExtractor
from ..utils import (
traverse_obj,
T,
url_or_none,
parse_iso8601,
get_element_by_class,
get_element_by_id,
int_or_none,
parse_qs,
urljoin,
)
class Tube8IE(InfoExtractor):
_VALID_URL = r'https?:\/\/(?:www\.)?tube8\.com\/+porn-video+\/(?P<id>\d+)'
_TESTS = [{
'url': 'https://www.tube8.com/porn-video/189530841/',
'md5': '532408f59e89a32027d873af6289c85a',
'info_dict': {
'id': '189530841',
'ext': 'mp4',
'title': 'Found dildo. She let it cum in her tight ass to keep the secret',
'thumbnail': r're:^https?://.*\.jpg$',
'uploader': 'MaryKrylova',
'timestamp': 1718961736,
'upload_date': '20240621',
}
}]
def _real_extract(self, url):
video_id = self._match_valid_url(url).group('id')
webpage = self._download_webpage(url, video_id)
playervars = self._search_json(
r'\bplayervars\s*:', webpage, 'playervars', video_id)
extra_info = self._search_json(
r'application/ld\+json[\"\']?>\s?', webpage, 'extra_info', video_id)
uploader = traverse_obj(extra_info, (
'@graph', lambda _, v: v.get('author'), 'author'))[0]
thumbnail = traverse_obj(extra_info, (
'@graph', lambda _, v: v.get('thumbnail'), 'thumbnail'))[0]
timestamp = parse_iso8601(traverse_obj(extra_info, (
'@graph', lambda _, v: v.get('datePublished'), 'datePublished'))[0])
# Borrowed from youporn extractor
def get_fmt(x):
v_url = url_or_none(x.get('videoUrl'))
if v_url:
x['videoUrl'] = v_url
return (x['format'], x)
defs_by_format = dict(traverse_obj(playervars, (
'mediaDefinitions', lambda _, v: v.get('format'), T(get_fmt))))
title = traverse_obj(playervars, 'video_title')
if not thumbnail:
thumbnail = traverse_obj(playervars, 'image_url')
# Borrowed from youporn extractor
def get_format_data(f):
if f not in defs_by_format:
return []
return self._download_json(
defs_by_format[f]['videoUrl'], video_id, '{0}-formats'.format(f))
formats = []
for mp4_url in traverse_obj(
get_format_data('mp4'),
(lambda _, v: not isinstance(v['defaultQuality'], bool), 'videoUrl'),
(Ellipsis, 'videoUrl')):
mobj = re.search(r'(?P<height>\d{3,4})[pP]_(?P<bitrate>\d+)[kK]_\d+', mp4_url)
if mobj:
height = int(mobj.group('height'))
tbr = int(mobj.group('bitrate'))
formats.append({
'format_id': '%dp-%dk' % (height, tbr),
'url': mp4_url,
'ext': 'mp4',
'tbr': tbr,
'height': height,
})
self._sort_formats(formats)
return {
'id': video_id,
'title': title,
'thumbnail': thumbnail,
'formats': formats,
'uploader': uploader,
'timestamp': timestamp,
}
# Currently only user channels
class Tube8ListIE(InfoExtractor):
_PLAYLIST_TITLEBAR_RE = r'\s+[Vv]ideos\s*$'
_PAGE_RETRY_COUNT = 0 # ie, no retry
_PAGE_RETRY_DELAY = 2 # seconds
_VALID_URL = r'https?:\/\/(?:www\.)?tube8\.com\.?\/+user-videos\/+(?P<id>\d+)\/+(?P<author>[^\/]+)\/?.*'
_TESTS = [{
'url': 'https://www.tube8.com/user-videos/195075441/MaryKrylova/',
'info_dict': {
'id': '195075441',
},
'playlist_mincount': 29,
}, {
'url': 'https://www.tube8.com/user-videos/195048331/FoxyElf/',
'info_dict': {
'id': '195048331',
},
'playlist_mincount': 86,
}]
# Borrowed from youporn extractor
@classmethod
def _get_title_from_slug(cls, title_slug):
return re.sub(r'[_-]', ' ', title_slug)
# Borrowed from youporn extractor
def _get_next_url(self, url, pl_id, html):
return urljoin(url, self._search_regex(
r'''<a\s[^>]*?\bhref\s*=\s*("|')(?P<url>(?:(?!\1)[^>])+)\1''',
get_element_by_id('next', html) or '', 'next page',
group='url', default=None))
# Borrowed from youporn extractor
def _entries(self, url, pl_id, html=None, page_num=None):
# separates page sections
PLAYLIST_SECTION_RE = (
r'''<div\s[^>]*\bclass\s*=\s*('|")(?:[\w$-]+\s+|\s)*?title-bar(?:\s+[\w$-]+|\s)*\1[^>]*>'''
)
# contains video link
VIDEO_URL_RE = r'''(?x)
<div\s[^>]*\bdata-video-id\s*=\s*('|")\d+\1[^>]*>\s*
(?:<div\b[\s\S]+?</div>\s*)*
<a\s[^>]*\bhref\s*=\s*('|")(?P<url>(?:(?!\2)[^>])+)\2
'''
def yield_pages(url, html=html, page_num=page_num):
fatal = not html
for pnum in itertools.count(start=page_num or 1):
if not html:
html = self._download_webpage(
url, pl_id, note='Downloading page %d' % pnum,
fatal=fatal)
if not html:
break
fatal = False
yield (url, html, pnum)
# explicit page: extract just that page
if page_num is not None:
break
next_url = self._get_next_url(url, pl_id, html)
if not next_url or next_url == url:
break
url, html = next_url, None
def retry_page(msg, tries_left, page_data):
if tries_left <= 0:
return
self.report_warning(msg, pl_id)
sleep(self._PAGE_RETRY_DELAY)
return next(
yield_pages(page_data[0], page_num=page_data[2]), None)
def yield_entries(html):
for frag in re.split(PLAYLIST_SECTION_RE, html):
if not frag:
continue
t_text = get_element_by_class('title-text', frag or '')
if not (t_text and re.search(self._PLAYLIST_TITLEBAR_RE, t_text)):
continue
for m in re.finditer(VIDEO_URL_RE, frag):
video_url = urljoin(url, m.group('url'))
if video_url:
yield self.url_result(video_url)
last_first_url = None
for page_data in yield_pages(url, html=html, page_num=page_num):
# page_data: url, html, page_num
first_url = None
tries_left = self._PAGE_RETRY_COUNT + 1
while tries_left > 0:
tries_left -= 1
for from_ in yield_entries(page_data[1]):
# may get the same page twice instead of empty page
# or (site bug) intead of actual next page
if not first_url:
first_url = from_['url']
if first_url == last_first_url:
# sometimes (/porntags/) the site serves the previous page
# instead but may provide the correct page after a delay
page_data = retry_page(
'Retrying duplicate page...', tries_left, page_data)
if page_data:
first_url = None
break
continue
yield from_
else:
if not first_url and 'no-result-paragarph1' in page_data[1]:
page_data = retry_page(
'Retrying empty page...', tries_left, page_data)
if page_data:
continue
else:
# success/failure
break
# may get an infinite (?) sequence of empty pages
if not first_url:
break
last_first_url = first_url
# Borrowed from youporn extractor
def _real_extract(self, url, html=None):
# exceptionally, id may be None
m_dict = self._match_valid_url(url).groupdict()
pl_id, page_type, sort = (m_dict.get(k) for k in ('id', 'type', 'sort'))
qs = parse_qs(url)
for q, v in qs.items():
if v:
qs[q] = v[-1]
else:
del qs[q]
base_id = pl_id or 'Tube8'
title = self._get_title_from_slug(base_id)
if page_type:
title = '%s %s' % (page_type.capitalize(), title)
base_id = [base_id.lower()]
if sort is None:
title += ' videos'
else:
title = '%s videos by %s' % (title, re.sub(r'[_-]', ' ', sort))
base_id.append(sort)
if qs:
ps = ['%s=%s' % item for item in sorted(qs.items())]
title += ' (%s)' % ','.join(ps)
base_id.extend(ps)
pl_id = '/'.join(base_id)
return self.playlist_result(
self._entries(url, pl_id, html=html,
page_num=int_or_none(qs.get('page'))),
playlist_id=pl_id)