searxng/searx/engines/deviantart.py
Markus Heiser 3d96a9839a [format.python] initial formatting of the python code
This patch was generated by black [1]::

    make format.python

[1] https://github.com/psf/black

Signed-off-by: Markus Heiser <markus.heiser@darmarit.de>
2021-12-27 09:26:22 +01:00

82 lines
1.9 KiB
Python

# SPDX-License-Identifier: AGPL-3.0-or-later
# lint: pylint
"""
Deviantart (Images)
"""
from urllib.parse import urlencode
from lxml import html
# about
about = {
"website": 'https://www.deviantart.com/',
"wikidata_id": 'Q46523',
"official_api_documentation": 'https://www.deviantart.com/developers/',
"use_official_api": False,
"require_api_key": False,
"results": 'HTML',
}
# engine dependent config
categories = ['images']
paging = True
time_range_support = True
time_range_dict = {
'day': 'popular-24-hours',
'week': 'popular-1-week',
'month': 'popular-1-month',
'year': 'most-recent',
}
# search-url
base_url = 'https://www.deviantart.com'
def request(query, params):
# https://www.deviantart.com/search/deviations?page=5&q=foo
query = {
'page': params['pageno'],
'q': query,
}
if params['time_range'] in time_range_dict:
query['order'] = time_range_dict[params['time_range']]
params['url'] = base_url + '/search/deviations?' + urlencode(query)
return params
def response(resp):
results = []
dom = html.fromstring(resp.text)
for row in dom.xpath('//div[contains(@data-hook, "content_row")]'):
for result in row.xpath('./div'):
a_tag = result.xpath('.//a[@data-hook="deviation_link"]')[0]
noscript_tag = a_tag.xpath('.//noscript')
if noscript_tag:
img_tag = noscript_tag[0].xpath('.//img')
else:
img_tag = a_tag.xpath('.//img')
if not img_tag:
continue
img_tag = img_tag[0]
results.append(
{
'template': 'images.html',
'url': a_tag.attrib.get('href'),
'img_src': img_tag.attrib.get('src'),
'title': img_tag.attrib.get('alt'),
}
)
return results