123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566 |
- # SPDX-License-Identifier: AGPL-3.0-or-later
- """Lingva (alternative Google Translate frontend)"""
- from json import loads
- about = {
- "website": 'https://lingva.ml',
- "wikidata_id": None,
- "official_api_documentation": 'https://github.com/thedaviddelta/lingva-translate#public-apis',
- "use_official_api": True,
- "require_api_key": False,
- "results": 'JSON',
- }
- engine_type = 'online_dictionary'
- categories = ['general', 'translate']
- url = "https://lingva.thedaviddelta.com"
- search_url = "{url}/api/v1/{from_lang}/{to_lang}/{query}"
- def request(_query, params):
- params['url'] = search_url.format(
- url=url, from_lang=params['from_lang'][1], to_lang=params['to_lang'][1], query=params['query']
- )
- return params
- def response(resp):
- results = []
- result = loads(resp.text)
- info = result["info"]
- from_to_prefix = "%s-%s " % (resp.search_params['from_lang'][1], resp.search_params['to_lang'][1])
- if "typo" in info:
- results.append({"suggestion": from_to_prefix + info["typo"]})
- if 'definitions' in info: # pylint: disable=too-many-nested-blocks
- for definition in info['definitions']:
- if 'list' in definition:
- for item in definition['list']:
- if 'synonyms' in item:
- for synonym in item['synonyms']:
- results.append({"suggestion": from_to_prefix + synonym})
- infobox = ""
- for translation in info["extraTranslations"]:
- for word in translation["list"]:
- infobox += f"<dl><dt>{word['word']}</dt>"
- for meaning in word["meanings"]:
- infobox += f"<dd>{meaning}</dd>"
- infobox += "</dl>"
- results.append(
- {
- 'infobox': result["translation"],
- 'content': infobox,
- }
- )
- return results
|