yt-dlp/yt_dlp/extractor/vidio.py

116 lines
4.5 KiB
Python
Raw Normal View History

# coding: utf-8
from __future__ import unicode_literals
import re
2016-06-04 11:47:43 +02:00
from .common import InfoExtractor
from ..utils import (
int_or_none,
parse_iso8601,
str_or_none,
strip_or_none,
try_get,
)
class VidioIE(InfoExtractor):
2016-06-04 11:47:43 +02:00
_VALID_URL = r'https?://(?:www\.)?vidio\.com/watch/(?P<id>\d+)-(?P<display_id>[^/?#&]+)'
_TESTS = [{
'url': 'http://www.vidio.com/watch/165683-dj_ambred-booyah-live-2015',
2016-06-04 11:47:43 +02:00
'md5': 'cd2801394afc164e9775db6a140b91fe',
'info_dict': {
'id': '165683',
2016-06-04 11:47:43 +02:00
'display_id': 'dj_ambred-booyah-live-2015',
'ext': 'mp4',
2016-06-04 11:47:43 +02:00
'title': 'DJ_AMBRED - Booyah (Live 2015)',
'description': 'md5:27dc15f819b6a78a626490881adbadf8',
'thumbnail': r're:^https?://.*\.jpg$',
2016-06-04 11:47:43 +02:00
'duration': 149,
'like_count': int,
'uploader': 'TWELVE Pic',
'timestamp': 1444902800,
'upload_date': '20151015',
'uploader_id': 'twelvepictures',
'channel': 'Cover Music Video',
'channel_id': '280236',
'view_count': int,
'dislike_count': int,
'comment_count': int,
'tags': 'count:4',
},
2016-06-04 11:47:43 +02:00
}, {
'url': 'https://www.vidio.com/watch/77949-south-korea-test-fires-missile-that-can-strike-all-of-the-north',
'only_matching': True,
}, {
# Premier-exclusive video
'url': 'https://www.vidio.com/watch/1550718-stand-by-me-doraemon',
'only_matching': True
2016-06-04 11:47:43 +02:00
}]
def _real_initialize(self):
self._api_key = self._download_json(
'https://www.vidio.com/auth', None, data=b'')['api_key']
2016-06-04 11:47:43 +02:00
def _real_extract(self, url):
video_id, display_id = re.match(self._VALID_URL, url).groups()
data = self._download_json(
'https://api.vidio.com/videos/' + video_id, display_id, headers={
'Content-Type': 'application/vnd.api+json',
'X-API-KEY': self._api_key,
})
video = data['videos'][0]
title = video['title'].strip()
is_premium = video.get('is_premium')
if is_premium:
sources = self._download_json(
'https://www.vidio.com/interactions_stream.json?video_id=%s&type=videos' % video_id,
display_id, note='Downloading premier API JSON')
if not (sources.get('source') or sources.get('source_dash')):
self.raise_login_required(method='cookies')
formats, subs = [], {}
if sources.get('source'):
hls_formats, hls_subs = self._extract_m3u8_formats_and_subtitles(
sources['source'], display_id, 'mp4', 'm3u8_native')
formats.extend(hls_formats)
subs.update(hls_subs)
if sources.get('source_dash'): # TODO: Find video example with source_dash
dash_formats, dash_subs = self._extract_mpd_formats_and_subtitles(
sources['source_dash'], display_id, 'dash')
formats.extend(dash_formats)
subs.update(dash_subs)
else:
hls_url = data['clips'][0]['hls_url']
formats, subs = self._extract_m3u8_formats_and_subtitles(
hls_url, display_id, 'mp4', 'm3u8_native')
2016-06-04 11:47:43 +02:00
2017-04-23 06:48:51 +02:00
self._sort_formats(formats)
2016-06-04 11:47:43 +02:00
get_first = lambda x: try_get(data, lambda y: y[x + 's'][0], dict) or {}
channel = get_first('channel')
user = get_first('user')
username = user.get('username')
get_count = lambda x: int_or_none(video.get('total_' + x))
return {
'id': video_id,
2016-06-04 11:47:43 +02:00
'display_id': display_id,
'title': title,
'description': strip_or_none(video.get('description')),
'thumbnail': video.get('image_url_medium'),
'duration': int_or_none(video.get('duration')),
'like_count': get_count('likes'),
2016-06-04 11:47:43 +02:00
'formats': formats,
'subtitles': subs,
'uploader': user.get('name'),
'timestamp': parse_iso8601(video.get('created_at')),
'uploader_id': username,
'uploader_url': 'https://www.vidio.com/@' + username if username else None,
'channel': channel.get('name'),
'channel_id': str_or_none(channel.get('id')),
'view_count': get_count('view_count'),
'dislike_count': get_count('dislikes'),
'comment_count': get_count('comments'),
'tags': video.get('tag_list'),
}