|
|
@ -11,6 +11,7 @@ from ..utils import ( |
|
|
|
determine_ext, |
|
|
|
int_or_none, |
|
|
|
str_or_none, |
|
|
|
try_get, |
|
|
|
urljoin, |
|
|
|
) |
|
|
|
|
|
|
@ -24,7 +25,7 @@ class TelecincoIE(InfoExtractor): |
|
|
|
'info_dict': { |
|
|
|
'id': '1876350223', |
|
|
|
'title': 'Bacalao con kokotxas al pil-pil', |
|
|
|
'description': 'md5:1382dacd32dd4592d478cbdca458e5bb', |
|
|
|
'description': 'md5:716caf5601e25c3c5ab6605b1ae71529', |
|
|
|
}, |
|
|
|
'playlist': [{ |
|
|
|
'md5': 'adb28c37238b675dad0f042292f209a7', |
|
|
@ -55,6 +56,26 @@ class TelecincoIE(InfoExtractor): |
|
|
|
'description': 'md5:2771356ff7bfad9179c5f5cd954f1477', |
|
|
|
'duration': 50, |
|
|
|
}, |
|
|
|
}, { |
|
|
|
# video in opening's content |
|
|
|
'url': 'https://www.telecinco.es/vivalavida/fiorella-sobrina-edmundo-arrocet-entrevista_18_2907195140.html', |
|
|
|
'info_dict': { |
|
|
|
'id': '2907195140', |
|
|
|
'title': 'La surrealista entrevista a la sobrina de Edmundo Arrocet: "No puedes venir aquí y tomarnos por tontos"', |
|
|
|
'description': 'md5:73f340a7320143d37ab895375b2bf13a', |
|
|
|
}, |
|
|
|
'playlist': [{ |
|
|
|
'md5': 'adb28c37238b675dad0f042292f209a7', |
|
|
|
'info_dict': { |
|
|
|
'id': 'TpI2EttSDAReWpJ1o0NVh2', |
|
|
|
'ext': 'mp4', |
|
|
|
'title': 'La surrealista entrevista a la sobrina de Edmundo Arrocet: "No puedes venir aquí y tomarnos por tontos"', |
|
|
|
'duration': 1015, |
|
|
|
}, |
|
|
|
}], |
|
|
|
'params': { |
|
|
|
'skip_download': True, |
|
|
|
}, |
|
|
|
}, { |
|
|
|
'url': 'http://www.telecinco.es/informativos/nacional/Pablo_Iglesias-Informativos_Telecinco-entrevista-Pedro_Piqueras_2_1945155182.html', |
|
|
|
'only_matching': True, |
|
|
@ -138,14 +159,25 @@ class TelecincoIE(InfoExtractor): |
|
|
|
r'window\.\$REACTBASE_STATE\.article(?:_multisite)?\s*=\s*({.+})', |
|
|
|
webpage, 'article'), display_id)['article'] |
|
|
|
title = article.get('title') |
|
|
|
description = clean_html(article.get('leadParagraph')) |
|
|
|
description = clean_html(article.get('leadParagraph')) or '' |
|
|
|
if article.get('editorialType') != 'VID': |
|
|
|
entries = [] |
|
|
|
for p in article.get('body', []): |
|
|
|
body = [article.get('opening')] |
|
|
|
body.extend(try_get(article, lambda x: x['body'], list) or []) |
|
|
|
for p in body: |
|
|
|
if not isinstance(p, dict): |
|
|
|
continue |
|
|
|
content = p.get('content') |
|
|
|
if p.get('type') != 'video' or not content: |
|
|
|
if not content: |
|
|
|
continue |
|
|
|
type_ = p.get('type') |
|
|
|
if type_ == 'paragraph': |
|
|
|
content_str = str_or_none(content) |
|
|
|
if content_str: |
|
|
|
description += content_str |
|
|
|
continue |
|
|
|
entries.append(self._parse_content(content, url)) |
|
|
|
if type_ == 'video' and isinstance(content, dict): |
|
|
|
entries.append(self._parse_content(content, url)) |
|
|
|
return self.playlist_result( |
|
|
|
entries, str_or_none(article.get('id')), title, description) |
|
|
|
content = article['opening']['content'] |
|
|
|