[ie/murrtube] Fix extractor (#9249)

Closes #7500
Authored by: DrakoCpp
This commit is contained in:
DrakoCpp 2024-07-02 23:49:09 +02:00 committed by GitHub
parent d502f4c6d9
commit 6403530e2d
No known key found for this signature in database
GPG key ID: B5690EEEBB952194

View file

@ -5,39 +5,103 @@
from ..utils import (
ExtractorError,
OnDemandPagedList,
determine_ext,
int_or_none,
try_get,
clean_html,
extract_attributes,
get_element_by_class,
get_element_html_by_id,
parse_count,
remove_end,
update_url,
urlencode_postdata,
)
class MurrtubeIE(InfoExtractor):
_WORKING = False
_VALID_URL = r'''(?x)
(?:
murrtube:|
https?://murrtube\.net/videos/(?P<slug>[a-z0-9\-]+)\-
https?://murrtube\.net/(?:v/|videos/(?P<slug>[a-z0-9-]+?)-)
)
(?P<id>[a-f0-9]{8}\-[a-f0-9]{4}\-[a-f0-9]{4}\-[a-f0-9]{4}\-[a-f0-9]{12})
(?P<id>[A-Z0-9]{4}|[a-f0-9]{8}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{4}-[a-f0-9]{12})
'''
_TEST = {
_TESTS = [{
'url': 'https://murrtube.net/videos/inferno-x-skyler-148b6f2a-fdcc-4902-affe-9c0f41aaaca0',
'md5': '169f494812d9a90914b42978e73aa690',
'md5': '70380878a77e8565d4aea7f68b8bbb35',
'info_dict': {
'id': '148b6f2a-fdcc-4902-affe-9c0f41aaaca0',
'id': 'ca885d8456b95de529b6723b158032e11115d',
'ext': 'mp4',
'title': 'Inferno X Skyler',
'description': 'Humping a very good slutty sheppy (roomate)',
'thumbnail': r're:^https?://.*\.jpg$',
'duration': 284,
'uploader': 'Inferno Wolf',
'age_limit': 18,
'thumbnail': 'https://storage.murrtube.net/murrtube-production/ekbs3zcfvuynnqfx72nn2tkokvsd',
'comment_count': int,
'view_count': int,
'like_count': int,
'tags': ['hump', 'breed', 'Fursuit', 'murrsuit', 'bareback'],
},
}
}, {
'url': 'https://murrtube.net/v/0J2Q',
'md5': '31262f6ac56f0ca75e5a54a0f3fefcb6',
'info_dict': {
'id': '8442998c52134968d9caa36e473e1a6bac6ca',
'ext': 'mp4',
'uploader': 'Hayel',
'title': 'Who\'s in charge now?',
'description': 'md5:795791e97e5b0f1805ea84573f02a997',
'age_limit': 18,
'thumbnail': 'https://storage.murrtube.net/murrtube-production/fb1ojjwiucufp34ya6hxu5vfqi5s',
'comment_count': int,
'view_count': int,
'like_count': int,
},
}]
def _extract_count(self, name, html):
return parse_count(self._search_regex(
rf'([\d,]+)\s+<span[^>]*>{name}</span>', html, name, default=None))
def _real_initialize(self):
homepage = self._download_webpage(
'https://murrtube.net', None, note='Getting session token')
self._request_webpage(
'https://murrtube.net/accept_age_check', None, 'Setting age cookie',
data=urlencode_postdata(self._hidden_inputs(homepage)))
def _real_extract(self, url):
video_id = self._match_id(url)
if video_id.startswith('murrtube:'):
raise ExtractorError('Support for murrtube: prefix URLs is broken')
video_page = self._download_webpage(url, video_id)
video_attrs = extract_attributes(get_element_html_by_id('video', video_page))
playlist = update_url(video_attrs['data-url'], query=None)
video_id = self._search_regex(r'/([\da-f]+)/index.m3u8', playlist, 'video id')
return {
'id': video_id,
'title': remove_end(self._og_search_title(video_page), ' - Murrtube'),
'age_limit': 18,
'formats': self._extract_m3u8_formats(playlist, video_id, 'mp4'),
'description': self._og_search_description(video_page),
'thumbnail': update_url(self._og_search_thumbnail(video_page, default=''), query=None) or None,
'uploader': clean_html(get_element_by_class('pl-1 is-size-6 has-text-lighter', video_page)),
'view_count': self._extract_count('Views', video_page),
'like_count': self._extract_count('Likes', video_page),
'comment_count': self._extract_count('Comments', video_page),
}
class MurrtubeUserIE(InfoExtractor):
_WORKING = False
IE_DESC = 'Murrtube user profile'
_VALID_URL = r'https?://murrtube\.net/(?P<id>[^/]+)$'
_TESTS = [{
'url': 'https://murrtube.net/stormy',
'info_dict': {
'id': 'stormy',
},
'playlist_mincount': 27,
}]
_PAGE_SIZE = 10
def _download_gql(self, video_id, op, note=None, fatal=True):
result = self._download_json(
@ -46,73 +110,6 @@ def _download_gql(self, video_id, op, note=None, fatal=True):
headers={'Content-Type': 'application/json'})
return result['data']
def _real_extract(self, url):
video_id = self._match_id(url)
data = self._download_gql(video_id, {
'operationName': 'Medium',
'variables': {
'id': video_id,
},
'query': '''\
query Medium($id: ID!) {
medium(id: $id) {
title
description
key
duration
commentsCount
likesCount
viewsCount
thumbnailKey
tagList
user {
name
__typename
}
__typename
}
}'''})
meta = data['medium']
storage_url = 'https://storage.murrtube.net/murrtube/'
format_url = storage_url + meta.get('key', '')
thumbnail = storage_url + meta.get('thumbnailKey', '')
if determine_ext(format_url) == 'm3u8':
formats = self._extract_m3u8_formats(
format_url, video_id, 'mp4', entry_protocol='m3u8_native', fatal=False)
else:
formats = [{'url': format_url}]
return {
'id': video_id,
'title': meta.get('title'),
'description': meta.get('description'),
'formats': formats,
'thumbnail': thumbnail,
'duration': int_or_none(meta.get('duration')),
'uploader': try_get(meta, lambda x: x['user']['name']),
'view_count': meta.get('viewsCount'),
'like_count': meta.get('likesCount'),
'comment_count': meta.get('commentsCount'),
'tags': meta.get('tagList'),
'age_limit': 18,
}
class MurrtubeUserIE(MurrtubeIE): # XXX: Do not subclass from concrete IE
_WORKING = False
IE_DESC = 'Murrtube user profile'
_VALID_URL = r'https?://murrtube\.net/(?P<id>[^/]+)$'
_TEST = {
'url': 'https://murrtube.net/stormy',
'info_dict': {
'id': 'stormy',
},
'playlist_mincount': 27,
}
_PAGE_SIZE = 10
def _fetch_page(self, username, user_id, page):
data = self._download_gql(username, {
'operationName': 'Media',