mirror of
https://github.com/searxng/searxng
synced 2024-10-30 21:20:28 +00:00
8205f170ff
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
134 lines
3.7 KiB
Python
134 lines
3.7 KiB
Python
# SPDX-License-Identifier: AGPL-3.0-or-later
|
|
"""
|
|
Wolfram|Alpha (Science)
|
|
"""
|
|
|
|
from json import loads
|
|
from time import time
|
|
from urllib.parse import urlencode
|
|
|
|
from searx.network import get as http_get
|
|
|
|
# about
|
|
about = {
|
|
"website": 'https://www.wolframalpha.com/',
|
|
"wikidata_id": 'Q207006',
|
|
"official_api_documentation": 'https://products.wolframalpha.com/api/',
|
|
"use_official_api": False,
|
|
"require_api_key": False,
|
|
"results": 'JSON',
|
|
}
|
|
|
|
# search-url
|
|
url = 'https://www.wolframalpha.com/'
|
|
|
|
search_url = (
|
|
url + 'input/json.jsp'
|
|
'?async=false'
|
|
'&banners=raw'
|
|
'&debuggingdata=false'
|
|
'&format=image,plaintext,imagemap,minput,moutput'
|
|
'&formattimeout=2'
|
|
'&{query}'
|
|
'&output=JSON'
|
|
'&parsetimeout=2'
|
|
'&proxycode={token}'
|
|
'&scantimeout=0.5'
|
|
'&sponsorcategories=true'
|
|
'&statemethod=deploybutton'
|
|
)
|
|
|
|
referer_url = url + 'input/?{query}'
|
|
|
|
token = {'value': '', 'last_updated': None}
|
|
|
|
# pods to display as image in infobox
|
|
# this pods do return a plaintext, but they look better and are more useful as images
|
|
image_pods = {'VisualRepresentation', 'Illustration', 'Symbol'}
|
|
|
|
|
|
# seems, wolframalpha resets its token in every hour
|
|
def obtain_token():
|
|
update_time = time() - (time() % 3600)
|
|
try:
|
|
token_response = http_get('https://www.wolframalpha.com/input/api/v1/code?ts=9999999999999999999', timeout=2.0)
|
|
token['value'] = loads(token_response.text)['code']
|
|
token['last_updated'] = update_time
|
|
except: # pylint: disable=bare-except
|
|
pass
|
|
return token
|
|
|
|
|
|
def init(engine_settings=None): # pylint: disable=unused-argument
|
|
obtain_token()
|
|
|
|
|
|
# do search-request
|
|
def request(query, params):
|
|
# obtain token if last update was more than an hour
|
|
if time() - (token['last_updated'] or 0) > 3600:
|
|
obtain_token()
|
|
params['url'] = search_url.format(query=urlencode({'input': query}), token=token['value'])
|
|
params['headers']['Referer'] = referer_url.format(query=urlencode({'i': query}))
|
|
|
|
return params
|
|
|
|
|
|
# get response from search-request
|
|
def response(resp):
|
|
results = []
|
|
|
|
resp_json = loads(resp.text)
|
|
|
|
if not resp_json['queryresult']['success']:
|
|
return []
|
|
|
|
# handle resp_json['queryresult']['assumptions']?
|
|
result_chunks = []
|
|
infobox_title = ""
|
|
result_content = ""
|
|
for pod in resp_json['queryresult']['pods']:
|
|
pod_id = pod.get('id', '')
|
|
pod_title = pod.get('title', '')
|
|
pod_is_result = pod.get('primary', None)
|
|
|
|
if 'subpods' not in pod:
|
|
continue
|
|
|
|
if pod_id == 'Input' or not infobox_title:
|
|
infobox_title = pod['subpods'][0]['plaintext']
|
|
|
|
for subpod in pod['subpods']:
|
|
if subpod['plaintext'] != '' and pod_id not in image_pods:
|
|
# append unless it's not an actual answer
|
|
if subpod['plaintext'] != '(requires interactivity)':
|
|
result_chunks.append({'label': pod_title, 'value': subpod['plaintext']})
|
|
|
|
if pod_is_result or not result_content:
|
|
if pod_id != "Input":
|
|
result_content = pod_title + ': ' + subpod['plaintext']
|
|
|
|
elif 'img' in subpod:
|
|
result_chunks.append({'label': pod_title, 'image': subpod['img']})
|
|
|
|
if not result_chunks:
|
|
return []
|
|
|
|
results.append(
|
|
{
|
|
'infobox': infobox_title,
|
|
'attributes': result_chunks,
|
|
'urls': [{'title': 'Wolfram|Alpha', 'url': resp.request.headers['Referer']}],
|
|
}
|
|
)
|
|
|
|
results.append(
|
|
{
|
|
'url': resp.request.headers['Referer'],
|
|
'title': 'Wolfram|Alpha (' + infobox_title + ')',
|
|
'content': result_content,
|
|
}
|
|
)
|
|
|
|
return results
|