searxng/searx/engines/filecrop.py

89 lines
2.5 KiB
Python
Raw Normal View History

2016-11-30 17:43:03 +00:00
from searx.url_utils import urlencode
try:
from HTMLParser import HTMLParser
except:
from html.parser import HTMLParser
url = 'http://www.filecrop.com/'
2014-01-30 00:14:08 +00:00
search_url = url + '/search.php?{query}&size_i=0&size_f=100000000&engine_r=1&engine_d=1&engine_e=1&engine_4=1&engine_m=1&pos={index}' # noqa
paging = True
2014-01-20 01:31:20 +00:00
class FilecropResultParser(HTMLParser):
def __init__(self):
HTMLParser.__init__(self)
self.__start_processing = False
self.results = []
self.result = {}
self.tr_counter = 0
self.data_counter = 0
def handle_starttag(self, tag, attrs):
if tag == 'tr':
2014-01-20 01:31:20 +00:00
if ('bgcolor', '#edeff5') in attrs or\
('bgcolor', '#ffffff') in attrs:
self.__start_processing = True
if not self.__start_processing:
return
if tag == 'label':
2014-01-20 01:31:20 +00:00
self.result['title'] = [attr[1] for attr in attrs
if attr[0] == 'title'][0]
elif tag == 'a' and ('rel', 'nofollow') in attrs\
and ('class', 'sourcelink') in attrs:
if 'content' in self.result:
2014-01-20 01:31:20 +00:00
self.result['content'] += [attr[1] for attr in attrs
if attr[0] == 'title'][0]
else:
2014-01-20 01:31:20 +00:00
self.result['content'] = [attr[1] for attr in attrs
if attr[0] == 'title'][0]
self.result['content'] += ' '
elif tag == 'a':
2014-01-20 01:31:20 +00:00
self.result['url'] = url + [attr[1] for attr in attrs
if attr[0] == 'href'][0]
def handle_endtag(self, tag):
if self.__start_processing is False:
return
if tag == 'tr':
self.tr_counter += 1
if self.tr_counter == 2:
self.__start_processing = False
self.tr_counter = 0
self.data_counter = 0
self.results.append(self.result)
self.result = {}
def handle_data(self, data):
if not self.__start_processing:
return
if 'content' in self.result:
self.result['content'] += data + ' '
else:
self.result['content'] = data + ' '
self.data_counter += 1
2014-01-20 01:31:20 +00:00
def request(query, params):
2014-01-30 00:14:08 +00:00
index = 1 + (params['pageno'] - 1) * 30
2016-11-30 17:43:03 +00:00
params['url'] = search_url.format(query=urlencode({'w': query}), index=index)
return params
2014-01-20 01:31:20 +00:00
def response(resp):
parser = FilecropResultParser()
parser.feed(resp.text)
return parser.results