mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-07 09:20:36 +00:00
114 lines
4.4 KiB
Python
114 lines
4.4 KiB
Python
# coding: utf-8
|
||
from __future__ import unicode_literals
|
||
|
||
import re
|
||
|
||
from .turner import TurnerBaseIE
|
||
from ..utils import (
|
||
float_or_none,
|
||
int_or_none,
|
||
strip_or_none,
|
||
)
|
||
|
||
|
||
class TBSIE(TurnerBaseIE):
|
||
_VALID_URL = r'https?://(?:www\.)?(?P<site>tbs|tntdrama)\.com/(?:movies|shows/[^/]+/(?:clips|season-\d+/episode-\d+))/(?P<id>[^/?#]+)'
|
||
_TESTS = [{
|
||
'url': 'http://www.tntdrama.com/shows/the-alienist/clips/monster',
|
||
'info_dict': {
|
||
'id': '8d384cde33b89f3a43ce5329de42903ed5099887',
|
||
'ext': 'mp4',
|
||
'title': 'Monster',
|
||
'description': 'Get a first look at the theatrical trailer for TNT’s highly anticipated new psychological thriller The Alienist, which premieres January 22 on TNT.',
|
||
'timestamp': 1508175329,
|
||
'upload_date': '20171016',
|
||
},
|
||
'params': {
|
||
# m3u8 download
|
||
'skip_download': True,
|
||
}
|
||
}, {
|
||
'url': 'http://www.tbs.com/shows/search-party/season-1/episode-1/explicit-the-mysterious-disappearance-of-the-girl-no-one-knew',
|
||
'only_matching': True,
|
||
}, {
|
||
'url': 'http://www.tntdrama.com/movies/star-wars-a-new-hope',
|
||
'only_matching': True,
|
||
}]
|
||
|
||
def _real_extract(self, url):
|
||
domain, display_id = re.match(self._VALID_URL, url).groups()
|
||
site = domain[:3]
|
||
webpage = self._download_webpage(url, display_id)
|
||
video_data = self._parse_json(self._search_regex(
|
||
r'<script[^>]+?data-drupal-selector="drupal-settings-json"[^>]*?>({.+?})</script>',
|
||
webpage, 'drupal setting'), display_id)['turner_playlist'][0]
|
||
|
||
media_id = video_data['mediaID']
|
||
title = video_data['title']
|
||
|
||
streams_data = self._download_json(
|
||
'http://medium.ngtv.io/media/%s/tv' % media_id,
|
||
media_id)['media']['tv']
|
||
duration = None
|
||
chapters = []
|
||
formats = []
|
||
for supported_type in ('unprotected', 'bulkaes'):
|
||
stream_data = streams_data.get(supported_type, {})
|
||
m3u8_url = stream_data.get('secureUrl') or stream_data.get('url')
|
||
if not m3u8_url:
|
||
continue
|
||
if stream_data.get('playlistProtection') == 'spe':
|
||
m3u8_url = self._add_akamai_spe_token(
|
||
'http://www.%s.com/service/token_spe' % site,
|
||
m3u8_url, media_id, {
|
||
'url': url,
|
||
'site_name': site.upper(),
|
||
'auth_required': video_data.get('authRequired') == '1',
|
||
})
|
||
formats.extend(self._extract_m3u8_formats(
|
||
m3u8_url, media_id, 'mp4', m3u8_id='hls', fatal=False))
|
||
|
||
duration = float_or_none(stream_data.get('totalRuntime') or video_data.get('duration'))
|
||
|
||
if not chapters:
|
||
for chapter in stream_data.get('contentSegments', []):
|
||
start_time = float_or_none(chapter.get('start'))
|
||
duration = float_or_none(chapter.get('duration'))
|
||
if start_time is None or duration is None:
|
||
continue
|
||
chapters.append({
|
||
'start_time': start_time,
|
||
'end_time': start_time + duration,
|
||
})
|
||
self._sort_formats(formats)
|
||
|
||
thumbnails = []
|
||
for image_id, image in video_data.get('images', {}).items():
|
||
image_url = image.get('url')
|
||
if not image_url or image.get('type') != 'video':
|
||
continue
|
||
i = {
|
||
'id': image_id,
|
||
'url': image_url,
|
||
}
|
||
mobj = re.search(r'(\d+)x(\d+)', image_url)
|
||
if mobj:
|
||
i.update({
|
||
'width': int(mobj.group(1)),
|
||
'height': int(mobj.group(2)),
|
||
})
|
||
thumbnails.append(i)
|
||
|
||
return {
|
||
'id': media_id,
|
||
'title': title,
|
||
'description': strip_or_none(video_data.get('descriptionNoTags') or video_data.get('shortDescriptionNoTags')),
|
||
'duration': duration,
|
||
'timestamp': int_or_none(video_data.get('created')),
|
||
'season_number': int_or_none(video_data.get('season')),
|
||
'episode_number': int_or_none(video_data.get('episode')),
|
||
'cahpters': chapters,
|
||
'thumbnails': thumbnails,
|
||
'formats': formats,
|
||
}
|