1
0
mirror of https://github.com/yt-dlp/yt-dlp.git synced 2024-11-09 20:52:30 +01:00
yt-dlp/youtube_dl/extractor/streamcz.py
2017-05-21 15:41:52 +07:00

106 lines
3.4 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import hashlib
import time
from .common import InfoExtractor
from ..utils import (
int_or_none,
sanitized_Request,
)
def _get_api_key(api_path):
if api_path.endswith('?'):
api_path = api_path[:-1]
api_key = 'fb5f58a820353bd7095de526253c14fd'
a = '{0:}{1:}{2:}'.format(api_key, api_path, int(round(time.time() / 24 / 3600)))
return hashlib.md5(a.encode('ascii')).hexdigest()
class StreamCZIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?stream\.cz/.+/(?P<id>[0-9]+)'
_API_URL = 'http://www.stream.cz/API'
_TESTS = [{
'url': 'http://www.stream.cz/peklonataliri/765767-ecka-pro-deti',
'md5': '934bb6a6d220d99c010783c9719960d5',
'info_dict': {
'id': '765767',
'ext': 'mp4',
'title': 'Peklo na talíři: Éčka pro děti',
'description': 'Taška s grónskou pomazánkou a další pekelnosti ZDE',
'thumbnail': 're:^http://im.stream.cz/episode/52961d7e19d423f8f06f0100',
'duration': 256,
},
}, {
'url': 'http://www.stream.cz/blanik/10002447-tri-roky-pro-mazanka',
'md5': '849a88c1e1ca47d41403c2ba5e59e261',
'info_dict': {
'id': '10002447',
'ext': 'mp4',
'title': 'Kancelář Blaník: Tři roky pro Mazánka',
'description': 'md5:3862a00ba7bf0b3e44806b544032c859',
'thumbnail': 're:^http://im.stream.cz/episode/537f838c50c11f8d21320000',
'duration': 368,
},
}]
def _real_extract(self, url):
video_id = self._match_id(url)
api_path = '/episode/%s' % video_id
req = sanitized_Request(self._API_URL + api_path)
req.add_header('Api-Password', _get_api_key(api_path))
data = self._download_json(req, video_id)
formats = []
for quality, video in enumerate(data['video_qualities']):
for f in video['formats']:
typ = f['type'].partition('/')[2]
qlabel = video.get('quality_label')
formats.append({
'format_note': '%s-%s' % (qlabel, typ) if qlabel else typ,
'format_id': '%s-%s' % (typ, f['quality']),
'url': f['source'],
'height': int_or_none(f['quality'].rstrip('p')),
'quality': quality,
})
self._sort_formats(formats)
image = data.get('image')
if image:
thumbnail = self._proto_relative_url(
image.replace('{width}', '1240').replace('{height}', '697'),
scheme='http:',
)
else:
thumbnail = None
stream = data.get('_embedded', {}).get('stream:show', {}).get('name')
if stream:
title = '%s: %s' % (stream, data['name'])
else:
title = data['name']
subtitles = {}
srt_url = data.get('subtitles_srt')
if srt_url:
subtitles['cs'] = [{
'ext': 'srt',
'url': srt_url,
}]
return {
'id': video_id,
'title': title,
'thumbnail': thumbnail,
'formats': formats,
'description': data.get('web_site_text'),
'duration': int_or_none(data.get('duration')),
'view_count': int_or_none(data.get('views')),
'subtitles': subtitles,
}