summaryrefslogtreecommitdiffstats
path: root/yt_dlp/extractor/skylinewebcams.py
blob: 4292bb2ae5ccd31747f3aabd08bb5a8a4037a54e (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
from .common import InfoExtractor


class SkylineWebcamsIE(InfoExtractor):
    _VALID_URL = r'https?://(?:www\.)?skylinewebcams\.com/[^/]+/webcam/(?:[^/]+/)+(?P<id>[^/]+)\.html'
    _TEST = {
        'url': 'https://www.skylinewebcams.com/it/webcam/italia/lazio/roma/scalinata-piazza-di-spagna-barcaccia.html',
        'info_dict': {
            'id': 'scalinata-piazza-di-spagna-barcaccia',
            'ext': 'mp4',
            'title': 're:^Live Webcam Scalinata di Piazza di Spagna - La Barcaccia [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$',
            'description': 'Roma, veduta sulla Scalinata di Piazza di Spagna e sulla Barcaccia',
            'is_live': True,
        },
        'params': {
            'skip_download': True,
        }
    }

    def _real_extract(self, url):
        video_id = self._match_id(url)

        webpage = self._download_webpage(url, video_id)

        stream_url = self._search_regex(
            r'(?:url|source)\s*:\s*(["\'])(?P<url>(?:https?:)?//.+?\.m3u8.*?)\1', webpage,
            'stream url', group='url')

        title = self._og_search_title(webpage)
        description = self._og_search_description(webpage)

        return {
            'id': video_id,
            'url': stream_url,
            'ext': 'mp4',
            'title': title,
            'description': description,
            'is_live': True,
        }