searxng/searx/engines/google.py
Alexandre Flament 8a69ade875 Revert of #195 when the search language is not english
Sometimes there is two requests to google (depending of the source IP) : one to google.com, the second to google.fr (for instance).

Going to https://www.google.com/ncr and saving the PREF cookie for future use prevent this (there is no redirection).

But, recently (or not ?), by doing this the search returns English results even if the Accept-Language is specified.

There is still a way to prevent this : going to preference, set the search language. I don't know if this can be done by searx.

For now, a quick fix is to disable the use of the PREF cookie when the search language is not English (google engine will slower but returns excepted results).
2015-05-01 21:20:09 +02:00

141 lines
4 KiB
Python

# Google (Web)
#
# @website https://www.google.com
# @provide-api yes (https://developers.google.com/custom-search/)
#
# @using-api no
# @results HTML
# @stable no (HTML can change)
# @parse url, title, content, suggestion
from urllib import urlencode
from urlparse import urlparse, parse_qsl
from lxml import html
from searx.poolrequests import get
from searx.engines.xpath import extract_text, extract_url
# engine dependent config
categories = ['general']
paging = True
language_support = True
# search-url
google_hostname = 'www.google.com'
search_path = '/search'
redirect_path = '/url'
images_path = '/images'
search_url = ('https://' +
google_hostname +
search_path +
'?{query}&start={offset}&gbv=1')
# specific xpath variables
results_xpath = '//li[@class="g"]'
url_xpath = './/h3/a/@href'
title_xpath = './/h3'
content_xpath = './/span[@class="st"]'
suggestion_xpath = '//p[@class="_Bmc"]'
images_xpath = './/div/a'
image_url_xpath = './@href'
image_img_src_xpath = './img/@src'
pref_cookie = ''
# see https://support.google.com/websearch/answer/873?hl=en
def get_google_pref_cookie():
global pref_cookie
if pref_cookie == '':
resp = get('https://www.google.com/ncr', allow_redirects=False)
pref_cookie = resp.cookies["PREF"]
return pref_cookie
# remove google-specific tracking-url
def parse_url(url_string):
parsed_url = urlparse(url_string)
if (parsed_url.netloc in [google_hostname, '']
and parsed_url.path == redirect_path):
query = dict(parse_qsl(parsed_url.query))
return query['q']
else:
return url_string
# do search-request
def request(query, params):
offset = (params['pageno'] - 1) * 10
if params['language'] == 'all':
language = 'en'
else:
language = params['language'].replace('_', '-').lower()
params['url'] = search_url.format(offset=offset,
query=urlencode({'q': query}))
params['headers']['Accept-Language'] = language
if language.startswith('en'):
params['cookies']['PREF'] = get_google_pref_cookie()
return params
# get response from search-request
def response(resp):
results = []
dom = html.fromstring(resp.text)
# parse results
for result in dom.xpath(results_xpath):
title = extract_text(result.xpath(title_xpath)[0])
try:
url = parse_url(extract_url(result.xpath(url_xpath), search_url))
parsed_url = urlparse(url)
if (parsed_url.netloc == google_hostname
and parsed_url.path == search_path):
# remove the link to google news
continue
# images result
if (parsed_url.netloc == google_hostname
and parsed_url.path == images_path):
# only thumbnail image provided,
# so skipping image results
# results = results + parse_images(result)
pass
else:
# normal result
content = extract_text(result.xpath(content_xpath)[0])
# append result
results.append({'url': url,
'title': title,
'content': content})
except:
continue
# parse suggestion
for suggestion in dom.xpath(suggestion_xpath):
# append suggestion
results.append({'suggestion': extract_text(suggestion)})
# return results
return results
def parse_images(result):
results = []
for image in result.xpath(images_xpath):
url = parse_url(extract_text(image.xpath(image_url_xpath)[0]))
img_src = extract_text(image.xpath(image_img_src_xpath)[0])
# append result
results.append({'url': url,
'title': '',
'content': '',
'img_src': img_src,
'template': 'images.html'})
return results