mirror of
https://codeberg.org/polarisfm/youtube-dl
synced 2024-11-23 00:54:31 +01:00
76 lines
2.5 KiB
Python
76 lines
2.5 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .turner import TurnerBaseIE
|
|
from ..utils import (
|
|
int_or_none,
|
|
parse_iso8601,
|
|
)
|
|
|
|
|
|
class TruTVIE(TurnerBaseIE):
|
|
_VALID_URL = r'https?://(?:www\.)?trutv\.com/(?:shows|full-episodes)/(?P<series_slug>[0-9A-Za-z-]+)/(?:videos/(?P<clip_slug>[0-9A-Za-z-]+)|(?P<id>\d+))'
|
|
_TEST = {
|
|
'url': 'https://www.trutv.com/shows/the-carbonaro-effect/videos/sunlight-activated-flower.html',
|
|
'info_dict': {
|
|
'id': 'f16c03beec1e84cd7d1a51f11d8fcc29124cc7f1',
|
|
'ext': 'mp4',
|
|
'title': 'Sunlight-Activated Flower',
|
|
'description': "A customer is stunned when he sees Michael's sunlight-activated flower.",
|
|
},
|
|
'params': {
|
|
# m3u8 download
|
|
'skip_download': True,
|
|
},
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
series_slug, clip_slug, video_id = re.match(self._VALID_URL, url).groups()
|
|
|
|
if video_id:
|
|
path = 'episode'
|
|
display_id = video_id
|
|
else:
|
|
path = 'series/clip'
|
|
display_id = clip_slug
|
|
|
|
data = self._download_json(
|
|
'https://api.trutv.com/v2/web/%s/%s/%s' % (path, series_slug, display_id),
|
|
display_id)
|
|
video_data = data['episode'] if video_id else data['info']
|
|
media_id = video_data['mediaId']
|
|
title = video_data['title'].strip()
|
|
|
|
info = self._extract_ngtv_info(
|
|
media_id, {}, {
|
|
'url': url,
|
|
'site_name': 'truTV',
|
|
'auth_required': video_data.get('isAuthRequired'),
|
|
})
|
|
|
|
thumbnails = []
|
|
for image in video_data.get('images', []):
|
|
image_url = image.get('srcUrl')
|
|
if not image_url:
|
|
continue
|
|
thumbnails.append({
|
|
'url': image_url,
|
|
'width': int_or_none(image.get('width')),
|
|
'height': int_or_none(image.get('height')),
|
|
})
|
|
|
|
info.update({
|
|
'id': media_id,
|
|
'display_id': display_id,
|
|
'title': title,
|
|
'description': video_data.get('description'),
|
|
'thumbnails': thumbnails,
|
|
'timestamp': parse_iso8601(video_data.get('publicationDate')),
|
|
'series': video_data.get('showTitle'),
|
|
'season_number': int_or_none(video_data.get('seasonNum')),
|
|
'episode_number': int_or_none(video_data.get('episodeNum')),
|
|
})
|
|
return info
|