# coding: utf-8
from __future__ import unicode_literals
-import re
-
from .turner import TurnerBaseIE
+from ..utils import int_or_none
class CartoonNetworkIE(TurnerBaseIE):
_VALID_URL = r'https?://(?:www\.)?cartoonnetwork\.com/video/(?:[^/]+/)+(?P<id>[^/?#]+)-(?:clip|episode)\.html'
_TEST = {
- 'url': 'http://www.cartoonnetwork.com/video/teen-titans-go/starfire-the-cat-lady-clip.html',
+ 'url': 'https://www.cartoonnetwork.com/video/ben-10/how-to-draw-upgrade-episode.html',
'info_dict': {
- 'id': '8a250ab04ed07e6c014ef3f1e2f9016c',
+ 'id': '6e3375097f63874ebccec7ef677c1c3845fa850e',
'ext': 'mp4',
- 'title': 'Starfire the Cat Lady',
- 'description': 'Robin decides to become a cat so that Starfire will finally love him.',
+ 'title': 'How to Draw Upgrade',
+ 'description': 'md5:2061d83776db7e8be4879684eefe8c0f',
},
'params': {
# m3u8 download
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
- id_type, video_id = re.search(r"_cnglobal\.cvp(Video|Title)Id\s*=\s*'([^']+)';", webpage).groups()
- query = ('id' if id_type == 'Video' else 'titleId') + '=' + video_id
- return self._extract_cvp_info(
- 'http://www.cartoonnetwork.com/video-seo-svc/episodeservices/getCvpPlaylist?networkName=CN2&' + query, video_id, {
- 'secure': {
- 'media_src': 'http://androidhls-secure.cdn.turner.com/toon/big',
- 'tokenizer_src': 'https://token.vgtf.net/token/token_mobile',
- },
- }, {
+
+ def find_field(global_re, name, content_re=None, value_re='[^"]+', fatal=False):
+ metadata_re = ''
+ if content_re:
+ metadata_re = r'|video_metadata\.content_' + content_re
+ return self._search_regex(
+ r'(?:_cnglobal\.currentVideo\.%s%s)\s*=\s*"(%s)";' % (global_re, metadata_re, value_re),
+ webpage, name, fatal=fatal)
+
+ media_id = find_field('mediaId', 'media id', 'id', '[0-9a-f]{40}', True)
+ title = find_field('episodeTitle', 'title', '(?:episodeName|name)', fatal=True)
+
+ info = self._extract_ngtv_info(
+ media_id, {'networkId': 'cartoonnetwork'}, {
'url': url,
'site_name': 'CartoonNetwork',
- 'auth_required': self._search_regex(
- r'_cnglobal\.cvpFullOrPreviewAuth\s*=\s*(true|false);',
- webpage, 'auth required', default='false') == 'true',
+ 'auth_required': find_field('authType', 'auth type') != 'unauth',
})
+
+ series = find_field(
+ 'propertyName', 'series', 'showName') or self._html_search_meta('partOfSeries', webpage)
+ info.update({
+ 'id': media_id,
+ 'display_id': display_id,
+ 'title': title,
+ 'description': self._html_search_meta('description', webpage),
+ 'series': series,
+ 'episode': title,
+ })
+
+ for field in ('season', 'episode'):
+ field_name = field + 'Number'
+ info[field + '_number'] = int_or_none(find_field(
+ field_name, field + ' number', value_re=r'\d+') or self._html_search_meta(field_name, webpage))
+
+ return info