mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-23 09:01:43 +00:00
add96eb9f8
Authored by: seproDev Reviewed-by: bashonly <88596187+bashonly@users.noreply.github.com> Reviewed-by: Simon Sawicki <contact@grub4k.xyz>
94 lines
3.9 KiB
Python
94 lines
3.9 KiB
Python
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
try_get,
|
|
unified_strdate,
|
|
)
|
|
|
|
|
|
class ThreeSpeakIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?3speak\.tv/watch\?v\=[^/]+/(?P<id>[^/$&#?]+)'
|
|
|
|
_TESTS = [{
|
|
'url': 'https://3speak.tv/watch?v=dannyshine/wjgoxyfy',
|
|
'info_dict': {
|
|
'id': 'wjgoxyfy',
|
|
'ext': 'mp4',
|
|
'title': 'Can People who took the Vax think Critically',
|
|
'uploader': 'dannyshine',
|
|
'description': 'md5:181aa7ccb304afafa089b5af3bca7a10',
|
|
'tags': ['sex', 'covid', 'antinatalism', 'comedy', 'vaccines'],
|
|
'thumbnail': 'https://img.3speakcontent.co/wjgoxyfy/thumbnails/default.png',
|
|
'upload_date': '20211021',
|
|
'duration': 2703.867833,
|
|
'filesize': 1620054781,
|
|
},
|
|
'params': {'skip_download': True},
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, video_id)
|
|
json_str = self._html_search_regex(r'JSON\.parse\(\'([^\']+)\'\)', webpage, 'json')
|
|
# The json string itself is escaped. Hence the double parsing
|
|
data_json = self._parse_json(self._parse_json(f'"{json_str}"', video_id), video_id)
|
|
video_json = self._parse_json(data_json['json_metadata'], video_id)
|
|
formats, subtitles = [], {}
|
|
og_m3u8 = self._html_search_regex(r'<meta\s?property=\"ogvideo\"\s?content=\"([^\"]+)\">', webpage, 'og m3u8', fatal=False)
|
|
if og_m3u8:
|
|
https_frmts, https_subs = self._extract_m3u8_formats_and_subtitles(og_m3u8, video_id, fatal=False, m3u8_id='https')
|
|
formats.extend(https_frmts)
|
|
subtitles = self._merge_subtitles(subtitles, https_subs)
|
|
ipfs_m3u8 = try_get(video_json, lambda x: x['video']['info']['ipfs'])
|
|
if ipfs_m3u8:
|
|
ipfs_frmts, ipfs_subs = self._extract_m3u8_formats_and_subtitles(
|
|
f'https://ipfs.3speak.tv/ipfs/{ipfs_m3u8}', video_id, fatal=False, m3u8_id='ipfs')
|
|
formats.extend(ipfs_frmts)
|
|
subtitles = self._merge_subtitles(subtitles, ipfs_subs)
|
|
mp4_file = try_get(video_json, lambda x: x['video']['info']['file'])
|
|
if mp4_file:
|
|
formats.append({
|
|
'url': f'https://threespeakvideo.b-cdn.net/{video_id}/{mp4_file}',
|
|
'ext': 'mp4',
|
|
'format_id': 'https-mp4',
|
|
'duration': try_get(video_json, lambda x: x['video']['info']['duration']),
|
|
'filesize': try_get(video_json, lambda x: x['video']['info']['filesize']),
|
|
'quality': 11,
|
|
'format_note': 'Original file',
|
|
})
|
|
return {
|
|
'id': video_id,
|
|
'title': data_json.get('title') or data_json.get('root_title'),
|
|
'uploader': data_json.get('author'),
|
|
'description': try_get(video_json, lambda x: x['video']['content']['description']),
|
|
'tags': try_get(video_json, lambda x: x['video']['content']['tags']),
|
|
'thumbnail': try_get(video_json, lambda x: x['image'][0]),
|
|
'upload_date': unified_strdate(data_json.get('created')),
|
|
'formats': formats,
|
|
'subtitles': subtitles,
|
|
}
|
|
|
|
|
|
class ThreeSpeakUserIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?3speak\.tv/user/(?P<id>[^/$&?#]+)'
|
|
|
|
_TESTS = [{
|
|
'url': 'https://3speak.tv/user/theycallmedan',
|
|
'info_dict': {
|
|
'id': 'theycallmedan',
|
|
},
|
|
'playlist_mincount': 115,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
playlist_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, playlist_id)
|
|
entries = [
|
|
self.url_result(
|
|
f'https://3speak.tv/watch?v={video}',
|
|
ie=ThreeSpeakIE.ie_key())
|
|
for video in re.findall(r'data-payout\s?\=\s?\"([^\"]+)\"', webpage) if video
|
|
]
|
|
return self.playlist_result(entries, playlist_id)
|