mirror of
https://github.com/searxng/searxng.git
synced 2024-12-26 19:20:29 +00:00
[feat] implementation of qwant lite for web search
Related: https://github.com/searxng/searxng/issues/2719 Replace: https://github.com/searxng/searxng/pull/2748 Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
This commit is contained in:
parent
b4e0d2eedc
commit
d9dbcedeb6
3 changed files with 118 additions and 35 deletions
13
docs/dev/engines/online/qwant.rst
Normal file
13
docs/dev/engines/online/qwant.rst
Normal file
|
@ -0,0 +1,13 @@
|
|||
.. _qwant engine:
|
||||
|
||||
=====
|
||||
Qwant
|
||||
=====
|
||||
|
||||
.. contents:: Contents
|
||||
:depth: 2
|
||||
:local:
|
||||
:backlinks: entry
|
||||
|
||||
.. automodule:: searx.engines.qwant
|
||||
:members:
|
|
@ -1,15 +1,29 @@
|
|||
# SPDX-License-Identifier: AGPL-3.0-or-later
|
||||
# lint: pylint
|
||||
"""Qwant (Web, News, Images, Videos)
|
||||
"""This engine uses the Qwant API (https://api.qwant.com/v3) to implement Qwant
|
||||
-Web, -News, -Images and -Videos. The API is undocumented but can be reverse
|
||||
engineered by reading the network log of https://www.qwant.com/ queries.
|
||||
|
||||
This engine uses the Qwant API (https://api.qwant.com/v3). The API is
|
||||
undocumented but can be reverse engineered by reading the network log of
|
||||
https://www.qwant.com/ queries.
|
||||
For Qwant's *web-search* two alternatives are implemented:
|
||||
|
||||
This implementation is used by different qwant engines in the settings.yml::
|
||||
- ``web``: uses the :py:obj:`api_url` which returns a JSON structure
|
||||
- ``web-lite``: uses the :py:obj:`web_lite_url` which returns a HTML page
|
||||
|
||||
|
||||
Configuration
|
||||
=============
|
||||
|
||||
The engine has the following additional settings:
|
||||
|
||||
- :py:obj:`qwant_categ`
|
||||
|
||||
This implementation is used by different qwant engines in the :ref:`settings.yml
|
||||
<settings engine>`:
|
||||
|
||||
.. code:: yaml
|
||||
|
||||
- name: qwant
|
||||
qwant_categ: web
|
||||
qwant_categ: web-lite # alternatively use 'web'
|
||||
...
|
||||
- name: qwant news
|
||||
qwant_categ: news
|
||||
|
@ -21,6 +35,9 @@ This implementation is used by different qwant engines in the settings.yml::
|
|||
qwant_categ: videos
|
||||
...
|
||||
|
||||
Implementations
|
||||
===============
|
||||
|
||||
"""
|
||||
|
||||
from datetime import (
|
||||
|
@ -31,11 +48,18 @@ from json import loads
|
|||
from urllib.parse import urlencode
|
||||
from flask_babel import gettext
|
||||
import babel
|
||||
import lxml
|
||||
|
||||
from searx.exceptions import SearxEngineAPIException
|
||||
from searx.network import raise_for_httperror
|
||||
from searx.enginelib.traits import EngineTraits
|
||||
|
||||
from searx.utils import (
|
||||
eval_xpath,
|
||||
eval_xpath_list,
|
||||
extract_text,
|
||||
)
|
||||
|
||||
traits: EngineTraits
|
||||
|
||||
# about
|
||||
|
@ -51,10 +75,11 @@ about = {
|
|||
# engine dependent config
|
||||
categories = []
|
||||
paging = True
|
||||
qwant_categ = None # web|news|inages|videos
|
||||
qwant_categ = None
|
||||
"""One of ``web``, ``news``, ``images`` or ``videos``"""
|
||||
|
||||
safesearch = True
|
||||
safe_search_map = {0: '&safesearch=0', 1: '&safesearch=1', 2: '&safesearch=2'}
|
||||
# safe_search_map = {0: '&safesearch=0', 1: '&safesearch=1', 2: '&safesearch=2'}
|
||||
|
||||
# fmt: off
|
||||
qwant_news_locales = [
|
||||
|
@ -67,7 +92,12 @@ qwant_news_locales = [
|
|||
# fmt: on
|
||||
|
||||
# search-url
|
||||
url = 'https://api.qwant.com/v3/search/{keyword}?{query}&count={count}&offset={offset}'
|
||||
|
||||
api_url = 'https://api.qwant.com/v3/search/'
|
||||
"""URL of Qwant's API (JSON)"""
|
||||
|
||||
web_lite_url = 'https://lite.qwant.com/'
|
||||
"""URL of Qwant-Lite (HTML)"""
|
||||
|
||||
|
||||
def request(query, params):
|
||||
|
@ -76,38 +106,78 @@ def request(query, params):
|
|||
if not query:
|
||||
return None
|
||||
|
||||
count = 10 # web: count must be equal to 10
|
||||
|
||||
if qwant_categ == 'images':
|
||||
count = 50
|
||||
offset = (params['pageno'] - 1) * count
|
||||
# count + offset must be lower than 250
|
||||
offset = min(offset, 199)
|
||||
else:
|
||||
offset = (params['pageno'] - 1) * count
|
||||
# count + offset must be lower than 50
|
||||
offset = min(offset, 40)
|
||||
|
||||
params['url'] = url.format(
|
||||
keyword=qwant_categ,
|
||||
query=urlencode({'q': query}),
|
||||
offset=offset,
|
||||
count=count,
|
||||
)
|
||||
|
||||
# add quant's locale
|
||||
q_locale = traits.get_region(params["searxng_locale"], default='en_US')
|
||||
params['url'] += '&locale=' + q_locale
|
||||
|
||||
# add safesearch option
|
||||
params['url'] += safe_search_map.get(params['safesearch'], '')
|
||||
|
||||
url = api_url + f'{qwant_categ}?'
|
||||
args = {'q': query}
|
||||
params['raise_for_httperror'] = False
|
||||
|
||||
if qwant_categ == 'web-lite':
|
||||
|
||||
# qwant-lite delivers only 5 pages maximum
|
||||
if params['pageno'] > 5:
|
||||
return None
|
||||
|
||||
url = web_lite_url + '?'
|
||||
args['locale'] = q_locale.lower()
|
||||
args['l'] = q_locale.split('_')[0]
|
||||
args['s'] = params['safesearch']
|
||||
args['p'] = params['pageno']
|
||||
|
||||
params['raise_for_httperror'] = True
|
||||
|
||||
elif qwant_categ == 'images':
|
||||
|
||||
args['locale'] = q_locale
|
||||
args['safesearch'] = params['safesearch']
|
||||
|
||||
args['count'] = 50
|
||||
offset = (params['pageno'] - 1) * args['count']
|
||||
# count + offset must be lower than 250
|
||||
args['offset'] = min(offset, 199)
|
||||
|
||||
else: # web, news, videos
|
||||
|
||||
args['locale'] = q_locale
|
||||
args['safesearch'] = params['safesearch']
|
||||
|
||||
args['count'] = 10
|
||||
offset = (params['pageno'] - 1) * args['count']
|
||||
# count + offset must be lower than 50
|
||||
args['offset'] = min(offset, 39)
|
||||
|
||||
params['url'] = url + urlencode(args)
|
||||
|
||||
return params
|
||||
|
||||
|
||||
def response(resp):
|
||||
"""Get response from Qwant's search request"""
|
||||
|
||||
if qwant_categ == 'web-lite':
|
||||
return parse_web_lite(resp)
|
||||
return parse_web_api
|
||||
|
||||
|
||||
def parse_web_lite(resp):
|
||||
"""Parse results from Qwant-Lite"""
|
||||
|
||||
results = []
|
||||
dom = lxml.html.fromstring(resp.text)
|
||||
|
||||
for item in eval_xpath_list(dom, '//section/article'):
|
||||
results.append(
|
||||
{
|
||||
'url': extract_text(eval_xpath(item, './span')),
|
||||
'title': extract_text(eval_xpath(item, './h2/a')),
|
||||
'content': extract_text(eval_xpath(item, './p')),
|
||||
}
|
||||
)
|
||||
|
||||
return results
|
||||
|
||||
|
||||
def parse_web_api(resp):
|
||||
"""Parse results from Qwant's API"""
|
||||
# pylint: disable=too-many-locals, too-many-branches, too-many-statements
|
||||
|
||||
results = []
|
||||
|
|
|
@ -1256,7 +1256,7 @@ engines:
|
|||
results: HTML
|
||||
|
||||
- name: qwant
|
||||
qwant_categ: web
|
||||
qwant_categ: web-lite
|
||||
engine: qwant
|
||||
shortcut: qw
|
||||
categories: [general, web]
|
||||
|
|
Loading…
Reference in a new issue