forked from zaclys/searxng
67 lines
1.6 KiB
Python
67 lines
1.6 KiB
Python
|
# WolframAlpha (Maths)
|
||
|
#
|
||
|
# @website http://www.wolframalpha.com/
|
||
|
#
|
||
|
# @using-api no
|
||
|
# @results HTML, JS
|
||
|
# @stable no
|
||
|
# @parse answer
|
||
|
|
||
|
import re
|
||
|
import json
|
||
|
from urllib import urlencode
|
||
|
from lxml import html
|
||
|
from searx.engines.xpath import extract_text
|
||
|
|
||
|
# search-url
|
||
|
url = 'http://www.wolframalpha.com/'
|
||
|
search_url = url+'input/?{query}'
|
||
|
|
||
|
|
||
|
# do search-request
|
||
|
def request(query, params):
|
||
|
params['url'] = search_url.format(query=urlencode({'i': query}))
|
||
|
|
||
|
return params
|
||
|
|
||
|
|
||
|
# tries to find answer under the pattern given
|
||
|
def extract_answer(script_list, pattern):
|
||
|
answer = None
|
||
|
|
||
|
# get line that matches the pattern
|
||
|
for script in script_list:
|
||
|
try:
|
||
|
line = re.search(pattern, script.text_content()).group(1)
|
||
|
except AttributeError:
|
||
|
continue
|
||
|
|
||
|
# extract answer from json
|
||
|
answer = line[line.find('{') : line.rfind('}')+1]
|
||
|
answer = json.loads(answer.encode('unicode-escape'))
|
||
|
answer = answer['stringified'].decode('unicode-escape')
|
||
|
|
||
|
return answer
|
||
|
|
||
|
|
||
|
# get response from search-request
|
||
|
def response(resp):
|
||
|
|
||
|
dom = html.fromstring(resp.text)
|
||
|
|
||
|
# the answer is inside a js script
|
||
|
scripts = dom.xpath('//script')
|
||
|
|
||
|
results = []
|
||
|
|
||
|
# answer can be located in different 'pods', although by default it should be in pod_0200
|
||
|
answer = extract_answer(scripts, 'pod_0200\.push(.*)\n')
|
||
|
if not answer:
|
||
|
answer = extract_answer(scripts, 'pod_0100\.push(.*)\n')
|
||
|
if answer:
|
||
|
results.append({'answer': answer})
|
||
|
else:
|
||
|
results.append({'answer': answer})
|
||
|
|
||
|
return results
|