123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105 |
- # SPDX-License-Identifier: AGPL-3.0-or-later
- """Wikimedia Commons (images)
- """
- import datetime
- from urllib.parse import urlencode
- from searx.utils import html_to_text, humanize_bytes
- # about
- about = {
- "website": 'https://commons.wikimedia.org/',
- "wikidata_id": 'Q565',
- "official_api_documentation": 'https://commons.wikimedia.org/w/api.php',
- "use_official_api": True,
- "require_api_key": False,
- "results": 'JSON',
- }
- categories = ['images']
- search_type = 'images'
- base_url = "https://commons.wikimedia.org"
- search_prefix = (
- '?action=query'
- '&format=json'
- '&generator=search'
- '&gsrnamespace=6'
- '&gsrprop=snippet'
- '&prop=info|imageinfo'
- '&iiprop=url|size|mime'
- '&iiurlheight=180' # needed for the thumb url
- )
- paging = True
- number_of_results = 10
- search_types = {
- 'images': 'bitmap|drawing',
- 'videos': 'video',
- 'audio': 'audio',
- 'files': 'multimedia|office|archive|3d',
- }
- def request(query, params):
- language = 'en'
- if params['language'] != 'all':
- language = params['language'].split('-')[0]
- if search_type not in search_types:
- raise ValueError(f"Unsupported search type: {search_type}")
- filetype = search_types[search_type]
- args = {
- 'uselang': language,
- 'gsrlimit': number_of_results,
- 'gsroffset': number_of_results * (params["pageno"] - 1),
- 'gsrsearch': f"filetype:{filetype} {query}",
- }
- params["url"] = f"{base_url}/w/api.php{search_prefix}&{urlencode(args, safe=':|')}"
- return params
- def response(resp):
- results = []
- json = resp.json()
- if not json.get("query", {}).get("pages"):
- return results
- for item in json["query"]["pages"].values():
- imageinfo = item["imageinfo"][0]
- title = item["title"].replace("File:", "").rsplit('.', 1)[0]
- result = {
- 'url': imageinfo["descriptionurl"],
- 'title': title,
- 'content': html_to_text(item["snippet"]),
- }
- if search_type == "images":
- result['template'] = 'images.html'
- result['img_src'] = imageinfo["url"]
- result['thumbnail_src'] = imageinfo["thumburl"]
- result['resolution'] = f'{imageinfo["width"]} x {imageinfo["height"]}'
- else:
- result['thumbnail'] = imageinfo["thumburl"]
- if search_type == "videos":
- result['template'] = 'videos.html'
- if imageinfo.get('duration'):
- result['length'] = datetime.timedelta(seconds=int(imageinfo['duration']))
- result['iframe_src'] = imageinfo['url']
- elif search_type == "files":
- result['template'] = 'files.html'
- result['metadata'] = imageinfo['mime']
- result['size'] = humanize_bytes(imageinfo['size'])
- elif search_type == "audio":
- result['iframe_src'] = imageinfo['url']
- results.append(result)
- return results
|