1
0
mirror of https://github.com/searxng/searxng.git synced 2024-11-10 23:20:12 +01:00
searxng/searx/engines/solidtorrents.py
Markus Heiser 8205f170ff [mod] pylint all engines without PYLINT_SEARXNG_DISABLE_OPTION
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
2024-03-11 14:55:38 +01:00

81 lines
2.4 KiB
Python

# SPDX-License-Identifier: AGPL-3.0-or-later
"""SolidTorrents
"""
from datetime import datetime
from urllib.parse import urlencode
import random
from lxml import html
from searx.utils import (
extract_text,
eval_xpath,
eval_xpath_getindex,
eval_xpath_list,
get_torrent_size,
)
about = {
"website": 'https://www.solidtorrents.to/',
"wikidata_id": None,
"official_api_documentation": None,
"use_official_api": False,
"require_api_key": False,
"results": 'HTML',
}
categories = ['files']
paging = True
# base_url can be overwritten by a list of URLs in the settings.yml
base_url = 'https://solidtorrents.to'
def request(query, params):
if isinstance(base_url, list):
params['base_url'] = random.choice(base_url)
else:
params['base_url'] = base_url
search_url = params['base_url'] + '/search?{query}'
query = urlencode({'q': query, 'page': params['pageno']})
params['url'] = search_url.format(query=query)
return params
def response(resp):
results = []
dom = html.fromstring(resp.text)
for result in eval_xpath(dom, '//li[contains(@class, "search-result")]'):
torrentfile = eval_xpath_getindex(result, './/a[contains(@class, "dl-torrent")]/@href', 0, None)
magnet = eval_xpath_getindex(result, './/a[contains(@class, "dl-magnet")]/@href', 0, None)
if torrentfile is None or magnet is None:
continue # ignore anime results that which aren't actually torrents
title = eval_xpath_getindex(result, './/h5[contains(@class, "title")]', 0, None)
url = eval_xpath_getindex(result, './/h5[contains(@class, "title")]/a/@href', 0, None)
categ = eval_xpath(result, './/a[contains(@class, "category")]')
stats = eval_xpath_list(result, './/div[contains(@class, "stats")]/div', min_len=5)
params = {
'seed': extract_text(stats[3]),
'leech': extract_text(stats[2]),
'title': extract_text(title),
'url': resp.search_params['base_url'] + url,
'filesize': get_torrent_size(*extract_text(stats[1]).split()),
'magnetlink': magnet,
'torrentfile': torrentfile,
'metadata': extract_text(categ),
'template': "torrent.html",
}
try:
params['publishedDate'] = datetime.strptime(extract_text(stats[4]), '%b %d, %Y')
except ValueError:
pass
results.append(params)
return results