mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-11-16 15:46:09 +01:00
72 lines
2.2 KiB
Python
72 lines
2.2 KiB
Python
|
from __future__ import unicode_literals
|
||
|
|
||
|
from .common import InfoExtractor
|
||
|
from ..compat import compat_str
|
||
|
from ..utils import (
|
||
|
int_or_none,
|
||
|
qualities,
|
||
|
)
|
||
|
|
||
|
|
||
|
class KamcordIE(InfoExtractor):
|
||
|
_VALID_URL = r'https?://(?:www\.)?kamcord\.com/v/(?P<id>[^/?#&]+)'
|
||
|
_TEST = {
|
||
|
'url': 'https://www.kamcord.com/v/hNYRduDgWb4',
|
||
|
'md5': 'c3180e8a9cfac2e86e1b88cb8751b54c',
|
||
|
'info_dict': {
|
||
|
'id': 'hNYRduDgWb4',
|
||
|
'ext': 'mp4',
|
||
|
'title': 'Drinking Madness',
|
||
|
'uploader': 'jacksfilms',
|
||
|
'uploader_id': '3044562',
|
||
|
'view_count': int,
|
||
|
'like_count': int,
|
||
|
'comment_count': int,
|
||
|
},
|
||
|
}
|
||
|
|
||
|
def _real_extract(self, url):
|
||
|
video_id = self._match_id(url)
|
||
|
|
||
|
webpage = self._download_webpage(url, video_id)
|
||
|
|
||
|
video = self._parse_json(
|
||
|
self._search_regex(
|
||
|
r'window\.__props\s*=\s*({.+?});?(?:\n|\s*</script)',
|
||
|
webpage, 'video'),
|
||
|
video_id)['video']
|
||
|
|
||
|
title = video['title']
|
||
|
|
||
|
formats = self._extract_m3u8_formats(
|
||
|
video['play']['hls'], video_id, 'mp4', entry_protocol='m3u8_native')
|
||
|
self._sort_formats(formats)
|
||
|
|
||
|
uploader = video.get('user', {}).get('username')
|
||
|
uploader_id = video.get('user', {}).get('id')
|
||
|
|
||
|
view_count = int_or_none(video.get('viewCount'))
|
||
|
like_count = int_or_none(video.get('heartCount'))
|
||
|
comment_count = int_or_none(video.get('messageCount'))
|
||
|
|
||
|
preference_key = qualities(('small', 'medium', 'large'))
|
||
|
|
||
|
thumbnails = [{
|
||
|
'url': thumbnail_url,
|
||
|
'id': thumbnail_id,
|
||
|
'preference': preference_key(thumbnail_id),
|
||
|
} for thumbnail_id, thumbnail_url in (video.get('thumbnail') or {}).items()
|
||
|
if isinstance(thumbnail_id, compat_str) and isinstance(thumbnail_url, compat_str)]
|
||
|
|
||
|
return {
|
||
|
'id': video_id,
|
||
|
'title': title,
|
||
|
'uploader': uploader,
|
||
|
'uploader_id': uploader_id,
|
||
|
'view_count': view_count,
|
||
|
'like_count': like_count,
|
||
|
'comment_count': comment_count,
|
||
|
'thumbnails': thumbnails,
|
||
|
'formats': formats,
|
||
|
}
|