1
0
mirror of https://github.com/searxng/searxng.git synced 2024-11-14 08:40:11 +01:00
searxng/searx/engines/livespace.py
2024-02-20 14:45:07 +01:00

102 lines
2.7 KiB
Python
Executable File

# SPDX-License-Identifier: AGPL-3.0-or-later
# lint: pylint
"""LiveSpace (Videos)
.. hint::
This engine only search for **live streams**!
"""
from urllib.parse import urlencode
from datetime import datetime
from babel import dates
about = {
"website": 'https://live.space',
"wikidata_id": None,
"official_api_documentation": None,
"use_official_api": True,
"require_api_key": False,
"results": 'JSON',
}
categories = ['videos']
base_url = 'https://backend.live.space'
# engine dependent config
paging = True
results_per_page = 10
def request(query, params):
args = {'page': params['pageno'] - 1, 'searchKey': query, 'size': results_per_page}
params['url'] = f"{base_url}/search/public/stream?{urlencode(args)}"
params['headers'] = {'Accept': 'application/json', 'Content-Type': 'application/json'}
return params
def response(resp):
results = []
json = resp.json()
now = datetime.now()
# for live videos
for result in json.get('result', []):
title = result.get("title")
thumbnailUrl = result.get("thumbnailUrl")
category = result.get("category/name")
username = result.get("user", {}).get("userName", "")
url = f'https://live.space/watch/{username}'
# stream tags
# currently the api seems to always return null before the first tag,
# so strip that unless it's not already there
tags = ''
if result.get("tags"):
tags = [x for x in result.get("tags").split(';') if x and x != 'null']
tags = ', '.join(tags)
content = []
if category:
content.append(f'category - {category}')
if tags and len(tags) > 0:
content.append(f'tags - {tags}')
# time & duration
start_time = None
if result.get("startTimeStamp"):
start_time = datetime.fromtimestamp(result.get("startTimeStamp") / 1000)
# for VODs (videos on demand)
end_time = None
if result.get("endTimeStamp"):
end_time = datetime.fromtimestamp(result.get("endTimeStamp") / 1000)
timestring = ""
if start_time:
delta = (now if end_time is None else end_time) - start_time
timestring = dates.format_timedelta(delta, granularity='second')
results.append(
{
'url': url,
'title': title,
'content': "No category or tags." if len(content) == 0 else ' '.join(content),
'author': username,
'length': (">= " if end_time is None else "") + timestring,
'publishedDate': start_time,
'thumbnail': thumbnailUrl,
'template': 'videos.html',
}
)
return results