2023-09-17 18:14:05 +00:00
|
|
|
# SPDX-License-Identifier: AGPL-3.0-or-later
|
|
|
|
"""Wallhaven_ is a site created by and for people who like wallpapers.
|
|
|
|
|
|
|
|
.. _Wallhaven: https://wallhaven.cc/about#Copyright
|
|
|
|
"""
|
|
|
|
|
|
|
|
from datetime import datetime
|
|
|
|
from urllib.parse import urlencode
|
|
|
|
|
2024-02-20 09:51:58 +00:00
|
|
|
from searx.utils import humanize_bytes
|
|
|
|
|
2023-09-17 18:14:05 +00:00
|
|
|
about = {
|
|
|
|
'website': 'https://wallhaven.cc/',
|
|
|
|
'official_api_documentation': 'https://wallhaven.cc/help/api',
|
|
|
|
'use_official_api': True,
|
|
|
|
'require_api_key': False,
|
|
|
|
'results': 'JSON',
|
|
|
|
}
|
|
|
|
categories = ['images']
|
|
|
|
paging = True
|
|
|
|
|
|
|
|
base_url = "https://wallhaven.cc"
|
|
|
|
|
|
|
|
api_key = ''
|
|
|
|
"""If you own an API key you can add it here, further read `Rate Limiting and
|
|
|
|
Errors`_.
|
|
|
|
|
|
|
|
.. _Rate Limiting and Errors: https://wallhaven.cc/help/api#limits
|
|
|
|
|
|
|
|
"""
|
|
|
|
|
|
|
|
# Possible categories: sfw, sketchy, nsfw
|
|
|
|
safesearch_map = {0: '111', 1: '110', 2: '100'}
|
|
|
|
"""Turn purities on(1) or off(0) NSFW requires a valid API key.
|
|
|
|
|
|
|
|
.. code:: text
|
|
|
|
|
|
|
|
100/110/111 <-- Bits stands for: SFW, Sketchy and NSFW
|
|
|
|
|
|
|
|
`What are SFW, Sketchy and NSFW all about?`_:
|
|
|
|
|
|
|
|
- SFW = "Safe for work" wallpapers. *Grandma approves.*
|
|
|
|
- Sketchy = Not quite SFW not quite NSFW. *Grandma might be uncomfortable.*
|
|
|
|
- NSFW = "Not safe for work". *Grandma isn't sure who you are anymore.*
|
|
|
|
|
|
|
|
.. _What are SFW, Sketchy and NSFW all about?:
|
|
|
|
https://wallhaven.cc/faq#What-are-SFW-Sketchy-and-NSFW-all-about
|
|
|
|
|
|
|
|
"""
|
|
|
|
|
|
|
|
|
|
|
|
def request(query, params):
|
|
|
|
args = {
|
|
|
|
'q': query,
|
|
|
|
'page': params['pageno'],
|
|
|
|
'purity': safesearch_map[params['safesearch']],
|
|
|
|
}
|
|
|
|
|
|
|
|
if api_key:
|
|
|
|
params['api_key'] = api_key
|
|
|
|
|
|
|
|
params['url'] = f"{base_url}/api/v1/search?{urlencode(args)}"
|
|
|
|
return params
|
|
|
|
|
|
|
|
|
|
|
|
def response(resp):
|
|
|
|
results = []
|
|
|
|
|
|
|
|
json = resp.json()
|
|
|
|
|
|
|
|
for result in json['data']:
|
2024-02-20 09:51:58 +00:00
|
|
|
|
2023-09-17 18:14:05 +00:00
|
|
|
results.append(
|
|
|
|
{
|
|
|
|
'template': 'images.html',
|
|
|
|
'title': '',
|
|
|
|
'content': f"{result['category']} / {result['purity']}",
|
|
|
|
'url': result['url'],
|
|
|
|
'img_src': result['path'],
|
|
|
|
'thumbnail_src': result['thumbs']['small'],
|
2024-02-20 09:51:58 +00:00
|
|
|
'resolution': result['resolution'].replace('x', ' x '),
|
2023-09-17 18:14:05 +00:00
|
|
|
'publishedDate': datetime.strptime(result['created_at'], '%Y-%m-%d %H:%M:%S'),
|
2024-02-20 09:51:58 +00:00
|
|
|
'img_format': result['file_type'],
|
|
|
|
'filesize': humanize_bytes(result['file_size']),
|
2023-09-17 18:14:05 +00:00
|
|
|
}
|
|
|
|
)
|
|
|
|
|
|
|
|
return results
|