1
0
mirror of https://codeberg.org/polarisfm/youtube-dl synced 2024-11-22 16:44:32 +01:00
youtube-dl/youtube_dl/extractor/cwtv.py
2016-04-28 04:01:21 +06:00

93 lines
3.5 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
int_or_none,
parse_iso8601,
)
class CWTVIE(InfoExtractor):
_VALID_URL = r'https?://(?:www\.)?cw(?:tv|seed)\.com/(?:shows/)?(?:[^/]+/){2}\?.*\bplay=(?P<id>[a-z0-9]{8}-[a-z0-9]{4}-[a-z0-9]{4}-[a-z0-9]{4}-[a-z0-9]{12})'
_TESTS = [{
'url': 'http://cwtv.com/shows/arrow/legends-of-yesterday/?play=6b15e985-9345-4f60-baf8-56e96be57c63',
'info_dict': {
'id': '6b15e985-9345-4f60-baf8-56e96be57c63',
'ext': 'mp4',
'title': 'Legends of Yesterday',
'description': 'Oliver and Barry Allen take Kendra Saunders and Carter Hall to a remote location to keep them hidden from Vandal Savage while they figure out how to defeat him.',
'duration': 2665,
'series': 'Arrow',
'season_number': 4,
'season': '4',
'episode_number': 8,
'upload_date': '20151203',
'timestamp': 1449122100,
},
'params': {
# m3u8 download
'skip_download': True,
}
}, {
'url': 'http://www.cwseed.com/shows/whose-line-is-it-anyway/jeff-davis-4/?play=24282b12-ead2-42f2-95ad-26770c2c6088',
'info_dict': {
'id': '24282b12-ead2-42f2-95ad-26770c2c6088',
'ext': 'mp4',
'title': 'Jeff Davis 4',
'description': 'Jeff Davis is back to make you laugh.',
'duration': 1263,
'series': 'Whose Line Is It Anyway?',
'season_number': 11,
'season': '11',
'episode_number': 20,
'upload_date': '20151006',
'timestamp': 1444107300,
},
'params': {
# m3u8 download
'skip_download': True,
}
}, {
'url': 'http://cwtv.com/thecw/chroniclesofcisco/?play=8adebe35-f447-465f-ab52-e863506ff6d6',
'only_matching': True,
}]
def _real_extract(self, url):
video_id = self._match_id(url)
video_data = self._download_json(
'http://metaframe.digitalsmiths.tv/v2/CWtv/assets/%s/partner/132?format=json' % video_id, video_id)
formats = self._extract_m3u8_formats(
video_data['videos']['variantplaylist']['uri'], video_id, 'mp4')
self._sort_formats(formats)
thumbnails = [{
'url': image['uri'],
'width': image.get('width'),
'height': image.get('height'),
} for image_id, image in video_data['images'].items() if image.get('uri')] if video_data.get('images') else None
video_metadata = video_data['assetFields']
subtitles = {
'en': [{
'url': video_metadata['UnicornCcUrl'],
}],
} if video_metadata.get('UnicornCcUrl') else None
return {
'id': video_id,
'title': video_metadata['title'],
'description': video_metadata.get('description'),
'duration': int_or_none(video_metadata.get('duration')),
'series': video_metadata.get('seriesName'),
'season_number': int_or_none(video_metadata.get('seasonNumber')),
'season': video_metadata.get('seasonName'),
'episode_number': int_or_none(video_metadata.get('episodeNumber')),
'timestamp': parse_iso8601(video_data.get('startTime')),
'thumbnails': thumbnails,
'formats': formats,
'subtitles': subtitles,
}