from urllib import urlencode
from lxml import html
from urlparse import urljoin
categories = ['images']
base_url = 'https://www.deviantart.com/'
search_url = base_url+'search?'
def request(query, params):
global search_url
params['url'] = search_url + urlencode({'q': query})
return params
def response(resp):
global base_url
results = []
if resp.status_code == 302:
return results
dom = html.fromstring(resp.text)
for result in dom.xpath('//div[contains(@class, "tt-a tt-fh")]'):
link = result.xpath('.//a[contains(@class, "thumb")]')[0]
url = urljoin(base_url, link.attrib.get('href'))
title_links = result.xpath('.//span[@class="details"]//a[contains(@class, "t")]')
title = ''.join(title_links[0].xpath('.//text()'))
img_src = link.xpath('.//img')[0].attrib['src']
results.append({'url': url, 'title': title, 'img_src': img_src, 'template': 'images.html'})
return results