mirror of
				https://github.com/searxng/searxng.git
				synced 2025-10-31 02:27:06 -04:00 
			
		
		
		
	Added seepeer to config and added seepeer search parser
This commit is contained in:
		
							parent
							
								
									aa85045a7d
								
							
						
					
					
						commit
						2c2123b2e8
					
				
							
								
								
									
										78
									
								
								searx/engines/seedpeer.py
									
									
									
									
									
										Normal file
									
								
							
							
						
						
									
										78
									
								
								searx/engines/seedpeer.py
									
									
									
									
									
										Normal file
									
								
							| @ -0,0 +1,78 @@ | ||||
| #  Seedpeer (Videos, Music, Files) | ||||
| # | ||||
| # @website     http://seedpeer.eu | ||||
| # @provide-api no (nothing found) | ||||
| # | ||||
| # @using-api   no | ||||
| # @results     HTML (using search portal) | ||||
| # @stable      yes (HTML can change) | ||||
| # @parse       url, title, content, seed, leech, magnetlink | ||||
| 
 | ||||
| from urlparse import urljoin | ||||
| from cgi import escape | ||||
| from urllib import quote | ||||
| from lxml import html | ||||
| from operator import itemgetter | ||||
| from searx.engines.xpath import extract_text | ||||
| 
 | ||||
| 
 | ||||
| url = 'http://www.seedpeer.eu/' | ||||
| search_url = url + 'search/{search_term}/7/{page_no}.html' | ||||
| # specific xpath variables | ||||
| torrent_xpath = '//*[@id="body"]/center/center/table[2]/tr/td/a' | ||||
| alternative_torrent_xpath = '//*[@id="body"]/center/center/table[1]/tr/td/a' | ||||
| title_xpath = '//*[@id="body"]/center/center/table[2]/tr/td/a/text()' | ||||
| alternative_title_xpath = '//*[@id="body"]/center/center/table/tr/td/a' | ||||
| seeds_xpath = '//*[@id="body"]/center/center/table[2]/tr/td[4]/font/text()' | ||||
| alternative_seeds_xpath = '//*[@id="body"]/center/center/table/tr/td[4]/font/text()' | ||||
| peers_xpath = '//*[@id="body"]/center/center/table[2]/tr/td[5]/font/text()' | ||||
| alternative_peers_xpath = '//*[@id="body"]/center/center/table/tr/td[5]/font/text()' | ||||
| age_xpath = '//*[@id="body"]/center/center/table[2]/tr/td[2]/text()' | ||||
| alternative_age_xpath = '//*[@id="body"]/center/center/table/tr/td[2]/text()' | ||||
| size_xpath = '//*[@id="body"]/center/center/table[2]/tr/td[3]/text()' | ||||
| alternative_size_xpath = '//*[@id="body"]/center/center/table/tr/td[3]/text()' | ||||
| 
 | ||||
| 
 | ||||
| # do search-request | ||||
| def request(query, params): | ||||
|     params['url'] = search_url.format(search_term=quote(query), | ||||
|                                       page_no=params['pageno'] - 1) | ||||
|     return params | ||||
| 
 | ||||
| 
 | ||||
| # get response from search-request | ||||
| def response(resp): | ||||
|     results = [] | ||||
|     dom = html.fromstring(resp.text) | ||||
|     torrent_links = dom.xpath(torrent_xpath) | ||||
|     if len(torrent_links) > 0: | ||||
|         seeds = dom.xpath(seeds_xpath) | ||||
|         peers = dom.xpath(peers_xpath) | ||||
|         titles = dom.xpath(title_xpath) | ||||
|         sizes = dom.xpath(size_xpath) | ||||
|         ages = dom.xpath(age_xpath) | ||||
|     else:  # under ~5 results uses a different xpath | ||||
|         torrent_links = dom.xpath(alternative_torrent_xpath) | ||||
|         seeds = dom.xpath(alternative_seeds_xpath) | ||||
|         peers = dom.xpath(alternative_peers_xpath) | ||||
|         titles = dom.xpath(alternative_title_xpath) | ||||
|         sizes = dom.xpath(alternative_size_xpath) | ||||
|         ages = dom.xpath(alternative_age_xpath) | ||||
|     # return empty array if nothing is found | ||||
|     if not torrent_links: | ||||
|         return [] | ||||
| 
 | ||||
|     # parse results | ||||
|     for index, result in enumerate(torrent_links): | ||||
|         link = result.attrib.get('href') | ||||
|         href = urljoin(url, link) | ||||
|         results.append({'url': href, | ||||
|                         'title': titles[index].text_content(), | ||||
|                         'content': '{}, {}'.format(sizes[index], ages[index]), | ||||
|                         'seed': seeds[index], | ||||
|                         'leech': peers[index], | ||||
| 
 | ||||
|                         'template': 'torrent.html'}) | ||||
| 
 | ||||
|     # return results sorted by seeder | ||||
|     return sorted(results, key=itemgetter('seed'), reverse=True) | ||||
| @ -495,6 +495,10 @@ engines: | ||||
|     timeout: 6.0 | ||||
|     categories : science | ||||
| 
 | ||||
|   - name : seedpeer | ||||
|     engine : seedpeer | ||||
|     shortcut: speu | ||||
| 
 | ||||
| #The blekko technology and team have joined IBM Watson! -> https://blekko.com/ | ||||
| #  - name : blekko images | ||||
| #    engine : blekko_images | ||||
|  | ||||
		Loading…
	
	
			
			x
			
			
		
	
		Reference in New Issue
	
	Block a user