mirror of
https://github.com/searxng/searxng.git
synced 2024-11-17 18:00:12 +01:00
49bbd250d9
Another website and new documentation
123 lines
3.2 KiB
Python
123 lines
3.2 KiB
Python
# SPDX-License-Identifier: AGPL-3.0-or-later
|
|
"""
|
|
Invidious (Videos)
|
|
"""
|
|
|
|
from urllib.parse import quote_plus
|
|
from dateutil import parser
|
|
import time
|
|
import random
|
|
|
|
# about
|
|
about = {
|
|
"website": 'https://api.invidious.io/',
|
|
"wikidata_id": 'Q79343316',
|
|
"official_api_documentation": 'https://github.com/iv-org/documentation/blob/master/API.md',
|
|
"use_official_api": True,
|
|
"require_api_key": False,
|
|
"results": 'JSON',
|
|
}
|
|
|
|
# engine dependent config
|
|
categories = ["videos", "music"]
|
|
paging = True
|
|
time_range_support = True
|
|
|
|
|
|
# search-url
|
|
|
|
base_url = ''
|
|
base_url_rand = ''
|
|
|
|
|
|
# do search-request
|
|
def request(query, params):
|
|
global base_url_rand
|
|
time_range_dict = {
|
|
"day": "today",
|
|
"week": "week",
|
|
"month": "month",
|
|
"year": "year",
|
|
}
|
|
|
|
if isinstance(base_url, list):
|
|
base_url_rand = random.choice(base_url)
|
|
else:
|
|
base_url_rand = base_url
|
|
|
|
search_url = base_url_rand + "api/v1/search?q={query}"
|
|
params["url"] = search_url.format(
|
|
query=quote_plus(query)
|
|
) + "&page={pageno}".format(pageno=params["pageno"])
|
|
|
|
if params["time_range"] in time_range_dict:
|
|
params["url"] += "&date={timerange}".format(
|
|
timerange=time_range_dict[params["time_range"]]
|
|
)
|
|
|
|
if params["language"] != "all":
|
|
lang = params["language"].split("-")
|
|
if len(lang) == 2:
|
|
params["url"] += "&range={lrange}".format(lrange=lang[1])
|
|
|
|
return params
|
|
|
|
|
|
# get response from search-request
|
|
def response(resp):
|
|
results = []
|
|
|
|
search_results = resp.json()
|
|
embedded_url = (
|
|
'<iframe width="540" height="304" '
|
|
+ 'data-src="'
|
|
+ base_url_rand
|
|
+ 'embed/{videoid}" '
|
|
+ 'frameborder="0" allowfullscreen></iframe>'
|
|
)
|
|
|
|
base_invidious_url = base_url_rand + "watch?v="
|
|
|
|
for result in search_results:
|
|
rtype = result.get("type", None)
|
|
if rtype == "video":
|
|
videoid = result.get("videoId", None)
|
|
if not videoid:
|
|
continue
|
|
|
|
url = base_invidious_url + videoid
|
|
embedded = embedded_url.format(videoid=videoid)
|
|
thumbs = result.get("videoThumbnails", [])
|
|
thumb = next(
|
|
(th for th in thumbs if th["quality"] == "sddefault"), None
|
|
)
|
|
if thumb:
|
|
thumbnail = thumb.get("url", "")
|
|
else:
|
|
thumbnail = ""
|
|
|
|
publishedDate = parser.parse(
|
|
time.ctime(result.get("published", 0))
|
|
)
|
|
length = time.gmtime(result.get("lengthSeconds"))
|
|
if length.tm_hour:
|
|
length = time.strftime("%H:%M:%S", length)
|
|
else:
|
|
length = time.strftime("%M:%S", length)
|
|
|
|
results.append(
|
|
{
|
|
"url": url,
|
|
"title": result.get("title", ""),
|
|
"content": result.get("description", ""),
|
|
'length': length,
|
|
"template": "videos.html",
|
|
"author": result.get("author"),
|
|
"publishedDate": publishedDate,
|
|
"embedded": embedded,
|
|
"thumbnail": thumbnail,
|
|
}
|
|
)
|
|
|
|
return results
|