haruhi-dl/haruhi_dl/extractor/lurker.py
2021-02-05 13:00:36 +01:00

96 lines
3.5 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import json
from .common import InfoExtractor
from ..utils import (
ExtractorError,
clean_html,
compat_str,
parse_iso8601,
)
class LurkerIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?lurker\.(?:pl|land)/post/(?P<id>[A-Za-z\d]{9})'
_TESTS = [{
# no title (microblog/comment)
'url': 'https://www.lurker.pl/post/HMPQwlZrm',
'info_dict': {
'id': 'HMPQwlZrm',
'title': 'md5:0cb2f2e431ad7c604017a3dcdc1c3dcd',
'timestamp': 1610548036,
'uploader': 'paladyn',
'uploader_id': 'paladyn',
'age_limit': 18,
},
'playlist_count': 1,
}, {
# with title
'url': 'https://www.lurker.pl/post/Aqump8Lbm',
'info_dict': {
'id': 'Aqump8Lbm',
'title': 'Trzech mężczyzn trzymało się za ręce, tworząc ludzki łańcuch,, aby dotrzeć do bezpańskiego psa uwięzionego w zamarzniętym jeziorze.',
'timestamp': 1610805850,
'description': 'https://www.liveleak.com/view?t=al95a_1610628122\n#ludzie #pomocy #ratunek #zwierzeta #psy #pies',
'uploader': 'R20_swap',
'uploader_id': 'r20_swap',
'age_limit': 0,
},
'playlist_count': 1,
}]
def _real_extract(self, url):
post_id = self._match_id(url)
meta = self._download_json('https://lurker.land/api/v3?action=POST_GET', post_id,
data=bytes(compat_str(json.dumps({
'id': post_id,
'subsort': 'best',
})).encode('utf-8')),
headers={
'Accept': 'application/json',
'Content-Type': 'application/json',
'Origin': 'https://www.lurker.land',
'Referer': 'https://www.lurker.land/post/%s' % post_id,
})
post = None
for post_i in meta['posts']:
if post_i.get('id') == post_id:
post = post_i
break
if not post:
raise ExtractorError('Post not found')
entries = []
for link in post['links']:
entries.append({
'_type': 'url',
'url': link['url'],
})
# ids are username-like but lowercased, possibly because of nickname changes - https://www.wykop.pl/link/5895209/
uploader = post.get('userId')
for user_i in meta['users']:
if user_i.get('id') == uploader:
uploader = user_i['username']
break
return {
'_type': 'playlist',
'entries': entries,
'id': post_id,
'title': post['title'] if post['hasTitle'] else clean_html(post.get('body')),
'description': clean_html(post.get('body')) if post['hasTitle'] else None,
'uploader': uploader,
'uploader_id': post.get('userId'),
'uploader_url': 'https://lurker.land/user/%s' % uploader,
'timestamp': parse_iso8601(post.get('createdAt')),
'like_count': post.get('upvotes'),
'dislike_count': post.get('downvotes'),
'average_rating': post.get('value'),
'age_limit': 18 if 'nsfw' in (post.get('tags') or ()) else 0,
}