mirror of
https://github.com/searxng/searxng.git
synced 2024-11-05 12:50:11 +01:00
56399cf1ea
Allow to search for images on 500px. It doesn't use the official API, but the page result. Less stable, but less API key to possess... Two engines were necessary for Searchcode because there are to search mode : search for documentation or search for code example. Both use open APIs.
58 lines
1.5 KiB
Python
58 lines
1.5 KiB
Python
## 500px (Images)
|
|
#
|
|
# @website https://500px.com
|
|
# @provide-api yes (https://developers.500px.com/)
|
|
#
|
|
# @using-api no
|
|
# @results HTML
|
|
# @stable no (HTML can change)
|
|
# @parse url, title, thumbnail, img_src, content
|
|
#
|
|
# @todo rewrite to api
|
|
|
|
|
|
from urllib import urlencode
|
|
from urlparse import urljoin
|
|
from lxml import html
|
|
|
|
# engine dependent config
|
|
categories = ['images']
|
|
paging = True
|
|
|
|
# search-url
|
|
base_url = 'https://500px.com'
|
|
search_url = base_url+'/search?search?page={pageno}&type=photos&{query}'
|
|
|
|
|
|
# do search-request
|
|
def request(query, params):
|
|
params['url'] = search_url.format(pageno=params['pageno'],
|
|
query=urlencode({'q': query}))
|
|
|
|
return params
|
|
|
|
|
|
# get response from search-request
|
|
def response(resp):
|
|
results = []
|
|
|
|
dom = html.fromstring(resp.text)
|
|
|
|
# parse results
|
|
for result in dom.xpath('//div[@class="photo"]'):
|
|
link = result.xpath('.//a')[0]
|
|
url = urljoin(base_url, link.attrib.get('href'))
|
|
title = result.xpath('.//div[@class="title"]//text()')[0]
|
|
img_src = link.xpath('.//img')[0].attrib['src']
|
|
content = result.xpath('.//div[@class="info"]//text()')[0]
|
|
|
|
# append result
|
|
results.append({'url': url,
|
|
'title': title,
|
|
'img_src': img_src,
|
|
'content': content,
|
|
'template': 'images.html'})
|
|
|
|
# return results
|
|
return results
|