mirror of
https://github.com/searxng/searxng.git
synced 2024-12-11 20:06:28 +00:00
124 lines
3.5 KiB
Python
124 lines
3.5 KiB
Python
# Yggtorrent (Videos, Music, Files)
|
|
#
|
|
# @website https://www2.yggtorrent.si
|
|
# @provide-api no (nothing found)
|
|
#
|
|
# @using-api no
|
|
# @results HTML (using search portal)
|
|
# @stable no (HTML can change)
|
|
# @parse url, title, seed, leech, publishedDate, filesize
|
|
|
|
from lxml import html
|
|
from operator import itemgetter
|
|
from datetime import datetime
|
|
from urllib.parse import quote
|
|
from searx.engines.xpath import extract_text
|
|
from searx.utils import get_torrent_size
|
|
from searx.poolrequests import get as http_get
|
|
|
|
# engine dependent config
|
|
categories = ['videos', 'music', 'files']
|
|
paging = True
|
|
|
|
# search-url
|
|
url = 'https://www2.yggtorrent.si/'
|
|
search_url = url + 'engine/search?name={search_term}&do=search&page={pageno}&category={search_type}'
|
|
|
|
# yggtorrent specific type-definitions
|
|
search_types = {'files': 'all',
|
|
'music': '2139',
|
|
'videos': '2145'}
|
|
|
|
cookies = dict()
|
|
|
|
|
|
def init(engine_settings=None):
|
|
global cookies
|
|
# initial cookies
|
|
resp = http_get(url, allow_redirects=False)
|
|
if resp.ok:
|
|
for r in resp.history:
|
|
cookies.update(r.cookies)
|
|
cookies.update(resp.cookies)
|
|
|
|
|
|
# do search-request
|
|
def request(query, params):
|
|
search_type = search_types.get(params['category'], 'all')
|
|
pageno = (params['pageno'] - 1) * 50
|
|
|
|
params['url'] = search_url.format(search_term=quote(query),
|
|
search_type=search_type,
|
|
pageno=pageno)
|
|
|
|
params['cookies'] = cookies
|
|
|
|
return params
|
|
|
|
|
|
# get response from search-request
|
|
def response(resp):
|
|
results = []
|
|
dom = html.fromstring(resp.text)
|
|
|
|
search_res = dom.xpath('//section[@id="#torrents"]/div/table/tbody/tr')
|
|
|
|
# return empty array if nothing is found
|
|
if not search_res:
|
|
return []
|
|
|
|
# parse results
|
|
for result in search_res:
|
|
link = result.xpath('.//a[@id="torrent_name"]')[0]
|
|
href = link.attrib.get('href')
|
|
title = extract_text(link)
|
|
seed = result.xpath('.//td[8]/text()')[0]
|
|
leech = result.xpath('.//td[9]/text()')[0]
|
|
|
|
# convert seed to int if possible
|
|
if seed.isdigit():
|
|
seed = int(seed)
|
|
else:
|
|
seed = 0
|
|
|
|
# convert leech to int if possible
|
|
if leech.isdigit():
|
|
leech = int(leech)
|
|
else:
|
|
leech = 0
|
|
|
|
params = {'url': href,
|
|
'title': title,
|
|
'seed': seed,
|
|
'leech': leech,
|
|
'template': 'torrent.html'}
|
|
|
|
# let's try to calculate the torrent size
|
|
try:
|
|
filesize_info = result.xpath('.//td[6]/text()')[0]
|
|
filesize = filesize_info[:-2]
|
|
filesize_multiplier = filesize_info[-2:].lower()
|
|
multiplier_french_to_english = {
|
|
'to': 'TiB',
|
|
'go': 'GiB',
|
|
'mo': 'MiB',
|
|
'ko': 'KiB'
|
|
}
|
|
filesize = get_torrent_size(filesize, multiplier_french_to_english[filesize_multiplier])
|
|
params['filesize'] = filesize
|
|
except:
|
|
pass
|
|
|
|
# extract and convert creation date
|
|
try:
|
|
date_ts = result.xpath('.//td[5]/div/text()')[0]
|
|
date = datetime.fromtimestamp(float(date_ts))
|
|
params['publishedDate'] = date
|
|
except:
|
|
pass
|
|
|
|
# append result
|
|
results.append(params)
|
|
|
|
# return results sorted by seeder
|
|
return sorted(results, key=itemgetter('seed'), reverse=True)
|