mirror of
https://codeberg.org/polarisfm/youtube-dl
synced 2024-11-22 08:34:32 +01:00
This commit is contained in:
parent
14bb191634
commit
bfdc8340c9
@ -1,6 +1,7 @@
|
|||||||
from __future__ import unicode_literals
|
from __future__ import unicode_literals
|
||||||
|
|
||||||
from .common import InfoExtractor
|
from .common import InfoExtractor
|
||||||
|
from ..compat import compat_str
|
||||||
from ..utils import (
|
from ..utils import (
|
||||||
parse_duration,
|
parse_duration,
|
||||||
urljoin,
|
urljoin,
|
||||||
@ -8,9 +9,9 @@ from ..utils import (
|
|||||||
|
|
||||||
|
|
||||||
class YourPornIE(InfoExtractor):
|
class YourPornIE(InfoExtractor):
|
||||||
_VALID_URL = r'https?://(?:www\.)?(?:yourporn\.sexy|sxyprn\.com)/post/(?P<id>[^/?#&.]+)'
|
_VALID_URL = r'https?://(?:www\.)?sxyprn\.com/post/(?P<id>[^/?#&.]+)'
|
||||||
_TESTS = [{
|
_TESTS = [{
|
||||||
'url': 'https://yourporn.sexy/post/57ffcb2e1179b.html',
|
'url': 'https://sxyprn.com/post/57ffcb2e1179b.html',
|
||||||
'md5': '6f8682b6464033d87acaa7a8ff0c092e',
|
'md5': '6f8682b6464033d87acaa7a8ff0c092e',
|
||||||
'info_dict': {
|
'info_dict': {
|
||||||
'id': '57ffcb2e1179b',
|
'id': '57ffcb2e1179b',
|
||||||
@ -33,11 +34,19 @@ class YourPornIE(InfoExtractor):
|
|||||||
|
|
||||||
webpage = self._download_webpage(url, video_id)
|
webpage = self._download_webpage(url, video_id)
|
||||||
|
|
||||||
video_url = urljoin(url, self._parse_json(
|
parts = self._parse_json(
|
||||||
self._search_regex(
|
self._search_regex(
|
||||||
r'data-vnfo=(["\'])(?P<data>{.+?})\1', webpage, 'data info',
|
r'data-vnfo=(["\'])(?P<data>{.+?})\1', webpage, 'data info',
|
||||||
group='data'),
|
group='data'),
|
||||||
video_id)[video_id]).replace('/cdn/', '/cdn5/')
|
video_id)[video_id].split('/')
|
||||||
|
|
||||||
|
num = 0
|
||||||
|
for c in parts[6] + parts[7]:
|
||||||
|
if c.isnumeric():
|
||||||
|
num += int(c)
|
||||||
|
parts[5] = compat_str(int(parts[5]) - num)
|
||||||
|
parts[1] += '8'
|
||||||
|
video_url = urljoin(url, '/'.join(parts))
|
||||||
|
|
||||||
title = (self._search_regex(
|
title = (self._search_regex(
|
||||||
r'<[^>]+\bclass=["\']PostEditTA[^>]+>([^<]+)', webpage, 'title',
|
r'<[^>]+\bclass=["\']PostEditTA[^>]+>([^<]+)', webpage, 'title',
|
||||||
@ -54,4 +63,5 @@ class YourPornIE(InfoExtractor):
|
|||||||
'thumbnail': thumbnail,
|
'thumbnail': thumbnail,
|
||||||
'duration': duration,
|
'duration': duration,
|
||||||
'age_limit': 18,
|
'age_limit': 18,
|
||||||
|
'ext': 'mp4',
|
||||||
}
|
}
|
||||||
|
Loading…
Reference in New Issue
Block a user