1
0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-09 12:42:28 +01:00
yt-dlp/yt_dlp/extractor/myvideoge.py

54 lines
1.8 KiB
Python
Raw Normal View History

2020-08-08 13:24:02 +02:00
from .common import InfoExtractor
from ..utils import js_to_json
class MyVideoGeIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?myvideo\.ge/v/(?P<id>[0-9]+)'
_TEST = {
'url': 'https://www.myvideo.ge/v/3941048',
'md5': '8c192a7d2b15454ba4f29dc9c9a52ea9',
'info_dict': {
'id': '3941048',
'ext': 'mp4',
'title': 'The best prikol',
'thumbnail': r're:^https?://.*\.jpg$',
'uploader': 'md5:d72addd357b0dd914e704781f7f777d8',
'description': 'md5:5c0371f540f5888d603ebfedd46b6df3'
}
}
def _real_extract(self, url):
video_id = self._match_id(url)
webpage = self._download_webpage(url, video_id)
title = self._html_search_regex(r'<h1[^>]*>([^<]+)</h1>', webpage, 'title')
description = self._og_search_description(webpage)
thumbnail = self._html_search_meta(['og:image'], webpage)
uploader = self._search_regex(r'<a[^>]+class="mv_user_name"[^>]*>([^<]+)<', webpage, 'uploader', fatal=False)
jwplayer_sources = self._parse_json(
self._search_regex(
r"(?s)jwplayer\(\"mvplayer\"\).setup\(.*?sources: (.*?])", webpage, 'jwplayer sources'),
video_id, transform_source=js_to_json)
def _formats_key(f):
if f['label'] == 'SD':
return -1
elif f['label'] == 'HD':
return 1
else:
return 0
jwplayer_sources = sorted(jwplayer_sources, key=_formats_key)
formats = self._parse_jwplayer_formats(jwplayer_sources, video_id)
return {
'id': video_id,
'title': title,
'description': description,
'uploader': uploader,
'formats': formats,
'thumbnail': thumbnail
}