mirror of https://github.com/searxng/searxng.git
500 lines
15 KiB
Python
500 lines
15 KiB
Python
# SPDX-License-Identifier: AGPL-3.0-or-later
|
|
"""
|
|
DuckDuckGo Lite
|
|
~~~~~~~~~~~~~~~
|
|
"""
|
|
|
|
from typing import TYPE_CHECKING
|
|
import re
|
|
from urllib.parse import urlencode
|
|
import json
|
|
import babel
|
|
import lxml.html
|
|
|
|
from searx import (
|
|
locales,
|
|
redislib,
|
|
external_bang,
|
|
)
|
|
from searx.utils import (
|
|
eval_xpath,
|
|
extract_text,
|
|
)
|
|
from searx.network import get # see https://github.com/searxng/searxng/issues/762
|
|
from searx import redisdb
|
|
from searx.enginelib.traits import EngineTraits
|
|
from searx.utils import extr
|
|
from searx.exceptions import SearxEngineCaptchaException
|
|
|
|
if TYPE_CHECKING:
|
|
import logging
|
|
|
|
logger: logging.Logger
|
|
|
|
traits: EngineTraits
|
|
|
|
about = {
|
|
"website": 'https://lite.duckduckgo.com/lite/',
|
|
"wikidata_id": 'Q12805',
|
|
"use_official_api": False,
|
|
"require_api_key": False,
|
|
"results": 'HTML',
|
|
}
|
|
|
|
send_accept_language_header = True
|
|
"""DuckDuckGo-Lite tries to guess user's preferred language from the HTTP
|
|
``Accept-Language``. Optional the user can select a region filter (but not a
|
|
language).
|
|
"""
|
|
|
|
# engine dependent config
|
|
categories = ['general', 'web']
|
|
paging = True
|
|
time_range_support = True
|
|
safesearch = True # user can't select but the results are filtered
|
|
|
|
url = "https://html.duckduckgo.com/html"
|
|
|
|
time_range_dict = {'day': 'd', 'week': 'w', 'month': 'm', 'year': 'y'}
|
|
form_data = {'v': 'l', 'api': 'd.js', 'o': 'json'}
|
|
__CACHE = []
|
|
|
|
|
|
def _cache_key(data: dict):
|
|
return 'SearXNG_ddg_web_vqd' + redislib.secret_hash(f"{data['q']}//{data['kl']}")
|
|
|
|
|
|
def cache_vqd(data: dict, value):
|
|
"""Caches a ``vqd`` value from a query."""
|
|
c = redisdb.client()
|
|
if c:
|
|
logger.debug("cache vqd value: %s", value)
|
|
c.set(_cache_key(data), value, ex=600)
|
|
|
|
else:
|
|
logger.debug("MEM cache vqd value: %s", value)
|
|
if len(__CACHE) > 100: # cache vqd from last 100 queries
|
|
__CACHE.pop(0)
|
|
__CACHE.append((_cache_key(data), value))
|
|
|
|
|
|
def get_vqd(data):
|
|
"""Returns the ``vqd`` that fits to the *query* (``data`` from HTTP POST).
|
|
|
|
DDG's bot detection is sensitive to the ``vqd`` value. For some search terms
|
|
(such as extremely long search terms that are often sent by bots), no ``vqd``
|
|
value can be determined.
|
|
|
|
If SearXNG cannot determine a ``vqd`` value, then no request should go out
|
|
to DDG:
|
|
|
|
A request with a wrong ``vqd`` value leads to DDG temporarily putting
|
|
SearXNG's IP on a block list.
|
|
|
|
Requests from IPs in this block list run into timeouts.
|
|
|
|
Not sure, but it seems the block list is a sliding window: to get my IP rid
|
|
from the bot list I had to cool down my IP for 1h (send no requests from
|
|
that IP to DDG).
|
|
|
|
TL;DR; the ``vqd`` value is needed to pass DDG's bot protection and is used
|
|
by all request to DDG:
|
|
|
|
- DuckDuckGo Lite: ``https://lite.duckduckgo.com/lite`` (POST form data)
|
|
- DuckDuckGo Web: ``https://links.duckduckgo.com/d.js?q=...&vqd=...``
|
|
- DuckDuckGo Images: ``https://duckduckgo.com/i.js??q=...&vqd=...``
|
|
- DuckDuckGo Videos: ``https://duckduckgo.com/v.js??q=...&vqd=...``
|
|
- DuckDuckGo News: ``https://duckduckgo.com/news.js??q=...&vqd=...``
|
|
|
|
"""
|
|
|
|
key = _cache_key(data)
|
|
value = None
|
|
c = redisdb.client()
|
|
if c:
|
|
value = c.get(key)
|
|
if value or value == b'':
|
|
value = value.decode('utf-8')
|
|
logger.debug("re-use CACHED vqd value: %s", value)
|
|
return value
|
|
|
|
else:
|
|
for k, value in __CACHE:
|
|
if k == key:
|
|
logger.debug("MEM re-use CACHED vqd value: %s", value)
|
|
return value
|
|
return None
|
|
|
|
|
|
def get_ddg_lang(eng_traits: EngineTraits, sxng_locale, default='en_US'):
|
|
"""Get DuckDuckGo's language identifier from SearXNG's locale.
|
|
|
|
DuckDuckGo defines its languages by region codes (see
|
|
:py:obj:`fetch_traits`).
|
|
|
|
To get region and language of a DDG service use:
|
|
|
|
.. code: python
|
|
|
|
eng_region = traits.get_region(params['searxng_locale'], traits.all_locale)
|
|
eng_lang = get_ddg_lang(traits, params['searxng_locale'])
|
|
|
|
It might confuse, but the ``l`` value of the cookie is what SearXNG calls
|
|
the *region*:
|
|
|
|
.. code:: python
|
|
|
|
# !ddi paris :es-AR --> {'ad': 'es_AR', 'ah': 'ar-es', 'l': 'ar-es'}
|
|
params['cookies']['ad'] = eng_lang
|
|
params['cookies']['ah'] = eng_region
|
|
params['cookies']['l'] = eng_region
|
|
|
|
.. hint::
|
|
|
|
`DDG-lite <https://lite.duckduckgo.com/lite>`__ and the *no Javascript*
|
|
page https://html.duckduckgo.com/html do not offer a language selection
|
|
to the user, only a region can be selected by the user (``eng_region``
|
|
from the example above). DDG-lite and *no Javascript* store the selected
|
|
region in a cookie::
|
|
|
|
params['cookies']['kl'] = eng_region # 'ar-es'
|
|
|
|
"""
|
|
return eng_traits.custom['lang_region'].get( # type: ignore
|
|
sxng_locale, eng_traits.get_language(sxng_locale, default)
|
|
)
|
|
|
|
|
|
ddg_reg_map = {
|
|
'tw-tzh': 'zh_TW',
|
|
'hk-tzh': 'zh_HK',
|
|
'ct-ca': 'skip', # ct-ca and es-ca both map to ca_ES
|
|
'es-ca': 'ca_ES',
|
|
'id-en': 'id_ID',
|
|
'no-no': 'nb_NO',
|
|
'jp-jp': 'ja_JP',
|
|
'kr-kr': 'ko_KR',
|
|
'xa-ar': 'ar_SA',
|
|
'sl-sl': 'sl_SI',
|
|
'th-en': 'th_TH',
|
|
'vn-en': 'vi_VN',
|
|
}
|
|
|
|
ddg_lang_map = {
|
|
# use ar --> ar_EG (Egypt's arabic)
|
|
"ar_DZ": 'lang_region',
|
|
"ar_JO": 'lang_region',
|
|
"ar_SA": 'lang_region',
|
|
# use bn --> bn_BD
|
|
'bn_IN': 'lang_region',
|
|
# use de --> de_DE
|
|
'de_CH': 'lang_region',
|
|
# use en --> en_US,
|
|
'en_AU': 'lang_region',
|
|
'en_CA': 'lang_region',
|
|
'en_GB': 'lang_region',
|
|
# Esperanto
|
|
'eo_XX': 'eo',
|
|
# use es --> es_ES,
|
|
'es_AR': 'lang_region',
|
|
'es_CL': 'lang_region',
|
|
'es_CO': 'lang_region',
|
|
'es_CR': 'lang_region',
|
|
'es_EC': 'lang_region',
|
|
'es_MX': 'lang_region',
|
|
'es_PE': 'lang_region',
|
|
'es_UY': 'lang_region',
|
|
'es_VE': 'lang_region',
|
|
# use fr --> rf_FR
|
|
'fr_CA': 'lang_region',
|
|
'fr_CH': 'lang_region',
|
|
'fr_BE': 'lang_region',
|
|
# use nl --> nl_NL
|
|
'nl_BE': 'lang_region',
|
|
# use pt --> pt_PT
|
|
'pt_BR': 'lang_region',
|
|
# skip these languages
|
|
'od_IN': 'skip',
|
|
'io_XX': 'skip',
|
|
'tokipona_XX': 'skip',
|
|
}
|
|
|
|
|
|
def quote_ddg_bangs(query):
|
|
# quote ddg bangs
|
|
query_parts = []
|
|
|
|
# for val in re.split(r'(\s+)', query):
|
|
for val in re.split(r'(\s+)', query):
|
|
if not val.strip():
|
|
continue
|
|
if val.startswith('!') and external_bang.get_node(external_bang.EXTERNAL_BANGS, val[1:]):
|
|
val = f"'{val}'"
|
|
query_parts.append(val)
|
|
return ' '.join(query_parts)
|
|
|
|
|
|
def request(query, params):
|
|
|
|
query = quote_ddg_bangs(query)
|
|
|
|
if len(query) >= 500:
|
|
# DDG does not accept queries with more than 499 chars
|
|
params["url"] = None
|
|
return
|
|
|
|
# Advanced search syntax ends in CAPTCHA
|
|
# https://duckduckgo.com/duckduckgo-help-pages/results/syntax/
|
|
query = " ".join(
|
|
[
|
|
x.removeprefix("site:").removeprefix("intitle:").removeprefix("inurl:").removeprefix("filetype:")
|
|
for x in query.split()
|
|
]
|
|
)
|
|
eng_region = traits.get_region(params['searxng_locale'], traits.all_locale)
|
|
if eng_region == "wt-wt":
|
|
# https://html.duckduckgo.com/html sets an empty value for "all".
|
|
eng_region = ""
|
|
|
|
params['data']['kl'] = eng_region
|
|
params['cookies']['kl'] = eng_region
|
|
|
|
# eng_lang = get_ddg_lang(traits, params['searxng_locale'])
|
|
|
|
params['url'] = url
|
|
params['method'] = 'POST'
|
|
params['data']['q'] = query
|
|
|
|
# The API is not documented, so we do some reverse engineering and emulate
|
|
# what https://html.duckduckgo.com/html does when you press "next Page" link
|
|
# again and again ..
|
|
|
|
params['headers']['Content-Type'] = 'application/x-www-form-urlencoded'
|
|
|
|
params['headers']['Sec-Fetch-Dest'] = "document"
|
|
params['headers']['Sec-Fetch-Mode'] = "navigate" # at least this one is used by ddg's bot detection
|
|
params['headers']['Sec-Fetch-Site'] = "same-origin"
|
|
params['headers']['Sec-Fetch-User'] = "?1"
|
|
|
|
# Form of the initial search page does have empty values in the form
|
|
if params['pageno'] == 1:
|
|
|
|
params['data']['b'] = ""
|
|
|
|
params['data']['df'] = ''
|
|
if params['time_range'] in time_range_dict:
|
|
|
|
params['data']['df'] = time_range_dict[params['time_range']]
|
|
params['cookies']['df'] = time_range_dict[params['time_range']]
|
|
|
|
if params['pageno'] == 2:
|
|
|
|
# second page does have an offset of 20
|
|
offset = (params['pageno'] - 1) * 20
|
|
params['data']['s'] = offset
|
|
params['data']['dc'] = offset + 1
|
|
|
|
elif params['pageno'] > 2:
|
|
|
|
# third and following pages do have an offset of 20 + n*50
|
|
offset = 20 + (params['pageno'] - 2) * 50
|
|
params['data']['s'] = offset
|
|
params['data']['dc'] = offset + 1
|
|
|
|
if params['pageno'] > 1:
|
|
|
|
# initial page does not have these additional data in the input form
|
|
params['data']['o'] = form_data.get('o', 'json')
|
|
params['data']['api'] = form_data.get('api', 'd.js')
|
|
params['data']['nextParams'] = form_data.get('nextParams', '')
|
|
params['data']['v'] = form_data.get('v', 'l')
|
|
params['headers']['Referer'] = url
|
|
|
|
# from here on no more params['data'] shuld be set, since this dict is
|
|
# needed to get a vqd value from the cache ..
|
|
|
|
vqd = get_vqd(params['data'])
|
|
|
|
# Certain conditions must be met in order to call up one of the
|
|
# following pages ...
|
|
|
|
if vqd:
|
|
params['data']['vqd'] = vqd # follow up pages / requests needs a vqd argument
|
|
else:
|
|
# Don't try to call follow up pages without a vqd value. DDG
|
|
# recognizes this as a request from a bot. This lowers the
|
|
# reputation of the SearXNG IP and DDG starts to activate CAPTCHAs.
|
|
params["url"] = None
|
|
return
|
|
|
|
if params['searxng_locale'].startswith("zh"):
|
|
# Some locales (at least China) do not have a "next page" button and ddg
|
|
# will return a HTTP/2 403 Forbidden for a request of such a page.
|
|
params["url"] = None
|
|
return
|
|
|
|
logger.debug("param data: %s", params['data'])
|
|
logger.debug("param cookies: %s", params['cookies'])
|
|
|
|
|
|
def is_ddg_captcha(dom):
|
|
"""In case of CAPTCHA ddg response its own *not a Robot* dialog and is not
|
|
redirected to a CAPTCHA page."""
|
|
|
|
return bool(eval_xpath(dom, "//form[@id='challenge-form']"))
|
|
|
|
|
|
def response(resp):
|
|
|
|
if resp.status_code == 303:
|
|
return []
|
|
|
|
results = []
|
|
doc = lxml.html.fromstring(resp.text)
|
|
|
|
if is_ddg_captcha(doc):
|
|
# set suspend time to zero is OK --> ddg does not block the IP
|
|
raise SearxEngineCaptchaException(suspended_time=0, message=f"CAPTCHA ({resp.search_params['data'].get('kl')})")
|
|
|
|
form = eval_xpath(doc, '//input[@name="vqd"]/..')
|
|
if len(form):
|
|
# some locales (at least China) does not have a "next page" button
|
|
form = form[0]
|
|
form_vqd = eval_xpath(form, '//input[@name="vqd"]/@value')[0]
|
|
|
|
cache_vqd(resp.search_params["data"], form_vqd)
|
|
|
|
# just select "web-result" and ignore results of class "result--ad result--ad--small"
|
|
for div_result in eval_xpath(doc, '//div[@id="links"]/div[contains(@class, "web-result")]'):
|
|
|
|
item = {}
|
|
title = eval_xpath(div_result, './/h2/a')
|
|
if not title:
|
|
# this is the "No results." item in the result list
|
|
continue
|
|
item["title"] = extract_text(title)
|
|
item["url"] = eval_xpath(div_result, './/h2/a/@href')[0]
|
|
item["content"] = extract_text(eval_xpath(div_result, './/a[contains(@class, "result__snippet")]')[0])
|
|
|
|
results.append(item)
|
|
|
|
zero_click_info_xpath = '//div[@id="zero_click_abstract"]'
|
|
zero_click = extract_text(eval_xpath(doc, zero_click_info_xpath)).strip()
|
|
|
|
if zero_click and (
|
|
"Your IP address is" not in zero_click
|
|
and "Your user agent:" not in zero_click
|
|
and "URL Decoded:" not in zero_click
|
|
):
|
|
current_query = resp.search_params["data"].get("q")
|
|
|
|
results.append(
|
|
{
|
|
'answer': zero_click,
|
|
'url': "https://duckduckgo.com/?" + urlencode({"q": current_query}),
|
|
}
|
|
)
|
|
|
|
return results
|
|
|
|
|
|
def fetch_traits(engine_traits: EngineTraits):
|
|
"""Fetch languages & regions from DuckDuckGo.
|
|
|
|
SearXNG's ``all`` locale maps DuckDuckGo's "Alle regions" (``wt-wt``).
|
|
DuckDuckGo's language "Browsers preferred language" (``wt_WT``) makes no
|
|
sense in a SearXNG request since SearXNG's ``all`` will not add a
|
|
``Accept-Language`` HTTP header. The value in ``engine_traits.all_locale``
|
|
is ``wt-wt`` (the region).
|
|
|
|
Beside regions DuckDuckGo also defines its languages by region codes. By
|
|
example these are the english languages in DuckDuckGo:
|
|
|
|
- en_US
|
|
- en_AU
|
|
- en_CA
|
|
- en_GB
|
|
|
|
The function :py:obj:`get_ddg_lang` evaluates DuckDuckGo's language from
|
|
SearXNG's locale.
|
|
|
|
"""
|
|
# pylint: disable=too-many-branches, too-many-statements, disable=import-outside-toplevel
|
|
from searx.utils import js_variable_to_python
|
|
|
|
# fetch regions
|
|
|
|
engine_traits.all_locale = 'wt-wt'
|
|
|
|
# updated from u661.js to u.7669f071a13a7daa57cb / should be updated automatically?
|
|
resp = get('https://duckduckgo.com/dist/util/u.7669f071a13a7daa57cb.js')
|
|
|
|
if not resp.ok: # type: ignore
|
|
print("ERROR: response from DuckDuckGo is not OK.")
|
|
|
|
js_code = extr(resp.text, 'regions:', ',snippetLengths')
|
|
|
|
regions = json.loads(js_code)
|
|
for eng_tag, name in regions.items():
|
|
|
|
if eng_tag == 'wt-wt':
|
|
engine_traits.all_locale = 'wt-wt'
|
|
continue
|
|
|
|
region = ddg_reg_map.get(eng_tag)
|
|
if region == 'skip':
|
|
continue
|
|
|
|
if not region:
|
|
eng_territory, eng_lang = eng_tag.split('-')
|
|
region = eng_lang + '_' + eng_territory.upper()
|
|
|
|
try:
|
|
sxng_tag = locales.region_tag(babel.Locale.parse(region))
|
|
except babel.UnknownLocaleError:
|
|
print("ERROR: %s (%s) -> %s is unknown by babel" % (name, eng_tag, region))
|
|
continue
|
|
|
|
conflict = engine_traits.regions.get(sxng_tag)
|
|
if conflict:
|
|
if conflict != eng_tag:
|
|
print("CONFLICT: babel %s --> %s, %s" % (sxng_tag, conflict, eng_tag))
|
|
continue
|
|
engine_traits.regions[sxng_tag] = eng_tag
|
|
|
|
# fetch languages
|
|
|
|
engine_traits.custom['lang_region'] = {}
|
|
|
|
js_code = extr(resp.text, 'languages:', ',regions')
|
|
|
|
languages = js_variable_to_python(js_code)
|
|
for eng_lang, name in languages.items():
|
|
|
|
if eng_lang == 'wt_WT':
|
|
continue
|
|
|
|
babel_tag = ddg_lang_map.get(eng_lang, eng_lang)
|
|
if babel_tag == 'skip':
|
|
continue
|
|
|
|
try:
|
|
|
|
if babel_tag == 'lang_region':
|
|
sxng_tag = locales.region_tag(babel.Locale.parse(eng_lang))
|
|
engine_traits.custom['lang_region'][sxng_tag] = eng_lang
|
|
continue
|
|
|
|
sxng_tag = locales.language_tag(babel.Locale.parse(babel_tag))
|
|
|
|
except babel.UnknownLocaleError:
|
|
print("ERROR: language %s (%s) is unknown by babel" % (name, eng_lang))
|
|
continue
|
|
|
|
conflict = engine_traits.languages.get(sxng_tag)
|
|
if conflict:
|
|
if conflict != eng_lang:
|
|
print("CONFLICT: babel %s --> %s, %s" % (sxng_tag, conflict, eng_lang))
|
|
continue
|
|
engine_traits.languages[sxng_tag] = eng_lang
|