mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-12-30 15:15:51 +01:00
50 lines
1.6 KiB
Python
50 lines
1.6 KiB
Python
|
from __future__ import unicode_literals
|
||
|
|
||
|
import re
|
||
|
|
||
|
from .brightcove import BrightcoveIE
|
||
|
from .common import InfoExtractor
|
||
|
from ..utils import (
|
||
|
ExtractorError,
|
||
|
)
|
||
|
|
||
|
|
||
|
class NownessIE(InfoExtractor):
|
||
|
_VALID_URL = r'https?://(?:www\.)?nowness\.com/[^?#]*?/(?P<id>[0-9]+)/(?P<slug>[^/]+?)(?:$|[?#])'
|
||
|
|
||
|
_TEST = {
|
||
|
'url': 'http://www.nowness.com/day/2013/6/27/3131/candor--the-art-of-gesticulation',
|
||
|
'file': '2520295746001.mp4',
|
||
|
'md5': '0ece2f70a7bd252c7b00f3070182d418',
|
||
|
'info_dict': {
|
||
|
'description': 'Candor: The Art of Gesticulation',
|
||
|
'uploader': 'Nowness',
|
||
|
'title': 'Candor: The Art of Gesticulation',
|
||
|
}
|
||
|
}
|
||
|
|
||
|
def _real_extract(self, url):
|
||
|
mobj = re.match(self._VALID_URL, url)
|
||
|
video_id = mobj.group('slug')
|
||
|
|
||
|
webpage = self._download_webpage(url, video_id)
|
||
|
player_url = self._search_regex(
|
||
|
r'"([^"]+/content/issue-[0-9.]+.js)"', webpage, 'player URL')
|
||
|
real_id = self._search_regex(
|
||
|
r'\sdata-videoId="([0-9]+)"', webpage, 'internal video ID')
|
||
|
|
||
|
player_code = self._download_webpage(
|
||
|
player_url, video_id,
|
||
|
note='Downloading player JavaScript',
|
||
|
errnote='Player download failed')
|
||
|
player_code = player_code.replace("'+d+'", real_id)
|
||
|
|
||
|
bc_url = BrightcoveIE._extract_brightcove_url(player_code)
|
||
|
if bc_url is None:
|
||
|
raise ExtractorError('Could not find player definition')
|
||
|
return {
|
||
|
'_type': 'url',
|
||
|
'url': bc_url,
|
||
|
'ie_key': 'Brightcove',
|
||
|
}
|