mirror of https://github.com/searxng/searxng.git
parent
2a6d84dab5
commit
a6673a1a94
|
@ -1,10 +1,12 @@
|
|||
# SPDX-License-Identifier: AGPL-3.0-or-later
|
||||
"""
|
||||
1x (Images)
|
||||
# lint: pylint
|
||||
"""1x (Images)
|
||||
|
||||
"""
|
||||
|
||||
from lxml import html, etree
|
||||
from urllib.parse import urlencode, urljoin
|
||||
from lxml import html, etree
|
||||
|
||||
from searx.utils import extract_text, eval_xpath_list, eval_xpath_getindex
|
||||
|
||||
# about
|
||||
|
@ -38,13 +40,14 @@ def request(query, params):
|
|||
def response(resp):
|
||||
results = []
|
||||
xmldom = etree.fromstring(resp.content)
|
||||
xmlsearchresult = eval_xpath_getindex(xmldom, '//searchresult', 0)
|
||||
xmlsearchresult = eval_xpath_getindex(xmldom, '//data', 0)
|
||||
dom = html.fragment_fromstring(xmlsearchresult.text, create_parent='div')
|
||||
for link in eval_xpath_list(dom, '/div/table/tr/td/div[2]//a'):
|
||||
for link in eval_xpath_list(dom, '//a'):
|
||||
url = urljoin(base_url, link.attrib.get('href'))
|
||||
title = extract_text(link)
|
||||
thumbnail_src = urljoin(gallery_url, eval_xpath_getindex(link, './/img', 0).attrib['src'])
|
||||
|
||||
thumbnail_src = urljoin(
|
||||
gallery_url, (eval_xpath_getindex(link, './/img', 0).attrib['src']).replace(base_url, '')
|
||||
)
|
||||
# append result
|
||||
results.append(
|
||||
{
|
||||
|
|
Loading…
Reference in New Issue