mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-15 07:42:42 +00:00
92 lines
3.1 KiB
Python
92 lines
3.1 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
from ..compat import (
|
|
compat_parse_qs,
|
|
compat_str,
|
|
)
|
|
from ..utils import (
|
|
int_or_none,
|
|
try_get,
|
|
unified_timestamp,
|
|
)
|
|
|
|
|
|
class PornFlipIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?pornflip\.com/(?:v|embed)/(?P<id>[0-9A-Za-z]{11})'
|
|
_TESTS = [{
|
|
'url': 'https://www.pornflip.com/v/wz7DfNhMmep',
|
|
'md5': '98c46639849145ae1fd77af532a9278c',
|
|
'info_dict': {
|
|
'id': 'wz7DfNhMmep',
|
|
'ext': 'mp4',
|
|
'title': '2 Amateurs swallow make his dream cumshots true',
|
|
'thumbnail': r're:^https?://.*\.jpg$',
|
|
'duration': 112,
|
|
'timestamp': 1481655502,
|
|
'upload_date': '20161213',
|
|
'uploader_id': '106786',
|
|
'uploader': 'figifoto',
|
|
'view_count': int,
|
|
'age_limit': 18,
|
|
}
|
|
}, {
|
|
'url': 'https://www.pornflip.com/embed/wz7DfNhMmep',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
|
|
webpage = self._download_webpage(
|
|
'https://www.pornflip.com/v/%s' % video_id, video_id)
|
|
|
|
flashvars = compat_parse_qs(self._search_regex(
|
|
r'<embed[^>]+flashvars=(["\'])(?P<flashvars>(?:(?!\1).)+)\1',
|
|
webpage, 'flashvars', group='flashvars'))
|
|
|
|
title = flashvars['video_vars[title]'][0]
|
|
|
|
def flashvar(kind):
|
|
return try_get(
|
|
flashvars, lambda x: x['video_vars[%s]' % kind][0], compat_str)
|
|
|
|
formats = []
|
|
for key, value in flashvars.items():
|
|
if not (value and isinstance(value, list)):
|
|
continue
|
|
format_url = value[0]
|
|
if key == 'video_vars[hds_manifest]':
|
|
formats.extend(self._extract_mpd_formats(
|
|
format_url, video_id, mpd_id='dash', fatal=False))
|
|
continue
|
|
height = self._search_regex(
|
|
r'video_vars\[video_urls\]\[(\d+)', key, 'height', default=None)
|
|
if not height:
|
|
continue
|
|
formats.append({
|
|
'url': format_url,
|
|
'format_id': 'http-%s' % height,
|
|
'height': int_or_none(height),
|
|
})
|
|
self._sort_formats(formats)
|
|
|
|
uploader = self._html_search_regex(
|
|
(r'<span[^>]+class="name"[^>]*>\s*<a[^>]+>\s*<strong>(?P<uploader>[^<]+)',
|
|
r'<meta[^>]+content=(["\'])[^>]*\buploaded by (?P<uploader>.+?)\1'),
|
|
webpage, 'uploader', fatal=False, group='uploader')
|
|
|
|
return {
|
|
'id': video_id,
|
|
'formats': formats,
|
|
'title': title,
|
|
'thumbnail': flashvar('big_thumb'),
|
|
'duration': int_or_none(flashvar('duration')),
|
|
'timestamp': unified_timestamp(self._html_search_meta(
|
|
'uploadDate', webpage, 'timestamp')),
|
|
'uploader_id': flashvar('author_id'),
|
|
'uploader': uploader,
|
|
'view_count': int_or_none(flashvar('views')),
|
|
'age_limit': 18,
|
|
}
|