mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-30 14:50:36 -05:00
a6ae61a4c2
Authored by: nixxo
203 lines
7.3 KiB
Python
203 lines
7.3 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
determine_ext,
|
|
float_or_none,
|
|
parse_duration,
|
|
smuggle_url,
|
|
unified_strdate,
|
|
)
|
|
|
|
|
|
class LA7IE(InfoExtractor):
|
|
IE_NAME = 'la7.it'
|
|
_VALID_URL = r'''(?x)(https?://)?(?:
|
|
(?:www\.)?la7\.it/([^/]+)/(?:rivedila7|video)/|
|
|
tg\.la7\.it/repliche-tgla7\?id=
|
|
)(?P<id>.+)'''
|
|
|
|
_TESTS = [{
|
|
# 'src' is a plain URL
|
|
'url': 'http://www.la7.it/crozza/video/inccool8-02-10-2015-163722',
|
|
'md5': '8b613ffc0c4bf9b9e377169fc19c214c',
|
|
'info_dict': {
|
|
'id': '0_42j6wd36',
|
|
'ext': 'mp4',
|
|
'title': 'Inc.Cool8',
|
|
'description': 'Benvenuti nell\'incredibile mondo della INC. COOL. 8. dove “INC.” sta per “Incorporated” “COOL” sta per “fashion” ed Eight sta per il gesto atletico',
|
|
'thumbnail': 're:^https?://.*',
|
|
'uploader_id': 'kdla7pillole@iltrovatore.it',
|
|
'timestamp': 1443814869,
|
|
'upload_date': '20151002',
|
|
},
|
|
}, {
|
|
'url': 'http://www.la7.it/omnibus/rivedila7/omnibus-news-02-07-2016-189077',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
|
|
if not url.startswith('http'):
|
|
url = '%s//%s' % (self.http_scheme(), url)
|
|
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
player_data = self._search_regex(
|
|
[r'(?s)videoParams\s*=\s*({.+?});', r'videoLa7\(({[^;]+})\);'],
|
|
webpage, 'player data')
|
|
vid = self._search_regex(r'vid\s*:\s*"(.+?)",', player_data, 'vid')
|
|
|
|
return {
|
|
'_type': 'url_transparent',
|
|
'url': smuggle_url('kaltura:103:%s' % vid, {
|
|
'service_url': 'http://nkdam.iltrovatore.it',
|
|
}),
|
|
'id': video_id,
|
|
'title': self._og_search_title(webpage, default=None),
|
|
'description': self._og_search_description(webpage, default=None),
|
|
'thumbnail': self._og_search_thumbnail(webpage, default=None),
|
|
'ie_key': 'Kaltura',
|
|
}
|
|
|
|
|
|
class LA7PodcastEpisodeIE(InfoExtractor):
|
|
IE_NAME = 'la7.it:pod:episode'
|
|
_VALID_URL = r'''(?x)(https?://)?
|
|
(?:www\.)?la7\.it/[^/]+/podcast/([^/]+-)?(?P<id>\d+)'''
|
|
|
|
_TESTS = [{
|
|
'url': 'https://www.la7.it/voicetown/podcast/la-carezza-delle-memoria-di-carlo-verdone-23-03-2021-371497',
|
|
'md5': '7737d4d79b3c1a34b3de3e16297119ed',
|
|
'info_dict': {
|
|
'id': '371497',
|
|
'ext': 'mp3',
|
|
'title': '"La carezza delle memoria" di Carlo Verdone',
|
|
'description': 'md5:5abf07c3c551a687db80af3f9ceb7d52',
|
|
'thumbnail': 'https://www.la7.it/sites/default/files/podcast/371497.jpg',
|
|
'upload_date': '20210323',
|
|
},
|
|
}, {
|
|
# embed url
|
|
'url': 'https://www.la7.it/embed/podcast/371497',
|
|
'only_matching': True,
|
|
}, {
|
|
# date already in the title
|
|
'url': 'https://www.la7.it/propagandalive/podcast/lintervista-di-diego-bianchi-ad-annalisa-cuzzocrea-puntata-del-1932021-20-03-2021-371130',
|
|
'only_matching': True,
|
|
}, {
|
|
# title same as show_title
|
|
'url': 'https://www.la7.it/otto-e-mezzo/podcast/otto-e-mezzo-26-03-2021-372340',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
def _extract_info(self, webpage, video_id=None, ppn=None):
|
|
if not video_id:
|
|
video_id = self._search_regex(
|
|
r'data-nid=([\'"])(?P<vid>\d+)\1',
|
|
webpage, 'video_id', group='vid')
|
|
|
|
media_url = self._search_regex(
|
|
(r'src:\s*([\'"])(?P<url>.+?mp3.+?)\1',
|
|
r'data-podcast=([\'"])(?P<url>.+?mp3.+?)\1'),
|
|
webpage, 'media_url', group='url')
|
|
ext = determine_ext(media_url)
|
|
formats = [{
|
|
'url': media_url,
|
|
'format_id': ext,
|
|
'ext': ext,
|
|
}]
|
|
self._sort_formats(formats)
|
|
|
|
title = self._html_search_regex(
|
|
(r'<div class="title">(?P<title>.+?)</',
|
|
r'<title>(?P<title>[^<]+)</title>',
|
|
r'title:\s*([\'"])(?P<title>.+?)\1'),
|
|
webpage, 'title', group='title')
|
|
|
|
description = (
|
|
self._html_search_regex(
|
|
(r'<div class="description">(.+?)</div>',
|
|
r'<div class="description-mobile">(.+?)</div>',
|
|
r'<div class="box-txt">([^<]+?)</div>',
|
|
r'<div class="field-content"><p>(.+?)</p></div>'),
|
|
webpage, 'description', default=None)
|
|
or self._html_search_meta('description', webpage))
|
|
|
|
thumb = self._html_search_regex(
|
|
(r'<div class="podcast-image"><img src="(.+?)"></div>',
|
|
r'<div class="container-embed"[^<]+url\((.+?)\);">',
|
|
r'<div class="field-content"><img src="(.+?)"'),
|
|
webpage, 'thumbnail', fatal=False, default=None)
|
|
|
|
duration = parse_duration(self._html_search_regex(
|
|
r'<span class="(?:durata|duration)">([\d:]+)</span>',
|
|
webpage, 'duration', fatal=False, default=None))
|
|
|
|
date = self._html_search_regex(
|
|
r'class="data">\s*(?:<span>)?([\d\.]+)\s*</',
|
|
webpage, 'date', default=None)
|
|
|
|
date_alt = self._search_regex(
|
|
r'(\d+[\./]\d+[\./]\d+)', title, 'date_alt', default=None)
|
|
ppn = ppn or self._search_regex(
|
|
r'ppN:\s*([\'"])(?P<ppn>.+?)\1',
|
|
webpage, 'ppn', group='ppn', default=None)
|
|
# if the date is not in the title
|
|
# and title is the same as the show_title
|
|
# add the date to the title
|
|
if date and not date_alt and ppn and ppn.lower() == title.lower():
|
|
title += ' del %s' % date
|
|
return {
|
|
'id': video_id,
|
|
'title': title,
|
|
'description': description,
|
|
'duration': float_or_none(duration),
|
|
'formats': formats,
|
|
'thumbnail': thumb,
|
|
'upload_date': unified_strdate(date),
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
video_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, video_id)
|
|
|
|
return self._extract_info(webpage, video_id)
|
|
|
|
|
|
class LA7PodcastIE(LA7PodcastEpisodeIE):
|
|
IE_NAME = 'la7.it:podcast'
|
|
_VALID_URL = r'(https?://)?(www\.)?la7\.it/(?P<id>[^/]+)/podcast/?(?:$|[#?])'
|
|
|
|
_TESTS = [{
|
|
'url': 'https://www.la7.it/propagandalive/podcast',
|
|
'info_dict': {
|
|
'id': 'propagandalive',
|
|
'title': "Propaganda Live",
|
|
},
|
|
'playlist_count': 10,
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
playlist_id = self._match_id(url)
|
|
webpage = self._download_webpage(url, playlist_id)
|
|
|
|
title = (
|
|
self._html_search_regex(
|
|
r'<h1.*?>(.+?)</h1>', webpage, 'title', fatal=False, default=None)
|
|
or self._og_search_title(webpage))
|
|
ppn = self._search_regex(
|
|
r'window\.ppN\s*=\s*([\'"])(?P<ppn>.+?)\1',
|
|
webpage, 'ppn', group='ppn', default=None)
|
|
|
|
entries = []
|
|
for episode in re.finditer(
|
|
r'<div class="container-podcast-property">([\s\S]+?)(?:</div>\s*){3}',
|
|
webpage):
|
|
entries.append(self._extract_info(episode.group(1), ppn=ppn))
|
|
|
|
return self.playlist_result(entries, playlist_id, title)
|