mirror of
https://github.com/searxng/searxng.git
synced 2024-11-25 12:21:00 +00:00
8205f170ff
Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
45 lines
1.2 KiB
Python
45 lines
1.2 KiB
Python
# SPDX-License-Identifier: AGPL-3.0-or-later
|
|
"""Svgrepo (images)
|
|
"""
|
|
|
|
from lxml import html
|
|
from searx.utils import extract_text, eval_xpath, eval_xpath_list
|
|
|
|
about = {
|
|
"website": 'https://www.svgrepo.com',
|
|
"official_api_documentation": 'https://svgapi.com',
|
|
"use_official_api": False,
|
|
"require_api_key": False,
|
|
"results": 'HTML',
|
|
}
|
|
|
|
paging = True
|
|
categories = ['images']
|
|
base_url = "https://www.svgrepo.com"
|
|
|
|
results_xpath = "//div[@class='style_nodeListing__7Nmro']/div"
|
|
url_xpath = ".//a/@href"
|
|
title_xpath = ".//a/@title"
|
|
img_src_xpath = ".//img/@src"
|
|
|
|
|
|
def request(query, params):
|
|
params['url'] = f"{base_url}/vectors/{query}/{params['pageno']}/"
|
|
return params
|
|
|
|
|
|
def response(resp):
|
|
results = []
|
|
|
|
dom = html.fromstring(resp.text)
|
|
for result in eval_xpath_list(dom, results_xpath):
|
|
results.append(
|
|
{
|
|
'template': 'images.html',
|
|
'url': base_url + extract_text(eval_xpath(result, url_xpath)),
|
|
'title': extract_text(eval_xpath(result, title_xpath)).replace(" SVG File", "").replace("Show ", ""),
|
|
'img_src': extract_text(eval_xpath(result, img_src_xpath)),
|
|
}
|
|
)
|
|
|
|
return results
|