mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-01 23:12:40 +00:00
54 lines
1.7 KiB
Python
54 lines
1.7 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
from .common import InfoExtractor
|
|
|
|
|
|
class StreamangoIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:www\.)?streamango\.com/(?:f|embed)/(?P<id>.+?)/(?:.+)'
|
|
_TESTS = [{
|
|
'url': 'https://streamango.com/f/clapasobsptpkdfe/20170315_150006_mp4',
|
|
'md5': 'e992787515a182f55e38fc97588d802a',
|
|
'info_dict': {
|
|
'id': 'clapasobsptpkdfe',
|
|
'ext': 'mp4',
|
|
'title': '20170315_150006.mp4',
|
|
'url': r're:https://streamango\.com/v/d/clapasobsptpkdfe~[0-9]{10}~(?:[0-9]+\.){3}[0-9]+~.{8}/720',
|
|
}
|
|
}, {
|
|
'url': 'https://streamango.com/embed/clapasobsptpkdfe/20170315_150006_mp4',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
def extract_url(urltype):
|
|
return self._search_regex(
|
|
r'type\s*:\s*["\']{}["\']\s*,\s*src\s*:\s*["\'](?P<url>.+?)["\'].*'.format(urltype),
|
|
webpage, 'video URL', group='url')
|
|
|
|
video_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
title = self._og_search_title(webpage)
|
|
url = 'https:' + extract_url('video/mp4')
|
|
dashurl = extract_url(r'application/dash\+xml')
|
|
|
|
formats = [{
|
|
'url': url,
|
|
'ext': 'mp4',
|
|
'width': 1280,
|
|
'height': 720,
|
|
'format_id': 'mp4',
|
|
}]
|
|
|
|
formats.extend(self._extract_mpd_formats(
|
|
dashurl, video_id, mpd_id='dash', fatal=False))
|
|
|
|
self._sort_formats(formats)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'url': url,
|
|
'title': title,
|
|
'formats': formats,
|
|
}
|