mirror of
https://github.com/ytdl-org/youtube-dl.git
synced 2024-12-05 04:28:02 +01:00
133 lines
4.5 KiB
Python
133 lines
4.5 KiB
Python
# coding: utf-8
|
|
from __future__ import unicode_literals
|
|
|
|
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..compat import (
|
|
compat_urllib_parse_urlencode,
|
|
compat_urlparse,
|
|
)
|
|
from ..utils import (
|
|
get_element_by_attribute,
|
|
int_or_none,
|
|
remove_start,
|
|
)
|
|
|
|
|
|
class MiTeleIE(InfoExtractor):
|
|
IE_DESC = 'mitele.es'
|
|
_VALID_URL = r'https?://www\.mitele\.es/[^/]+/[^/]+/[^/]+/(?P<id>[^/]+)/'
|
|
|
|
_TESTS = [{
|
|
'url': 'http://www.mitele.es/programas-tv/diario-de/la-redaccion/programa-144/',
|
|
# MD5 is unstable
|
|
'info_dict': {
|
|
'id': '0NF1jJnxS1Wu3pHrmvFyw2',
|
|
'display_id': 'programa-144',
|
|
'ext': 'flv',
|
|
'title': 'Tor, la web invisible',
|
|
'description': 'md5:3b6fce7eaa41b2d97358726378d9369f',
|
|
'series': 'Diario de',
|
|
'season': 'La redacción',
|
|
'episode': 'Programa 144',
|
|
'thumbnail': 're:(?i)^https?://.*\.jpg$',
|
|
'duration': 2913,
|
|
},
|
|
}, {
|
|
# no explicit title
|
|
'url': 'http://www.mitele.es/programas-tv/cuarto-milenio/temporada-6/programa-226/',
|
|
'info_dict': {
|
|
'id': 'eLZSwoEd1S3pVyUm8lc6F',
|
|
'display_id': 'programa-226',
|
|
'ext': 'flv',
|
|
'title': 'Cuarto Milenio - Temporada 6 - Programa 226',
|
|
'description': 'md5:50daf9fadefa4e62d9fc866d0c015701',
|
|
'series': 'Cuarto Milenio',
|
|
'season': 'Temporada 6',
|
|
'episode': 'Programa 226',
|
|
'thumbnail': 're:(?i)^https?://.*\.jpg$',
|
|
'duration': 7312,
|
|
},
|
|
'params': {
|
|
'skip_download': True,
|
|
},
|
|
}]
|
|
|
|
def _real_extract(self, url):
|
|
display_id = self._match_id(url)
|
|
|
|
webpage = self._download_webpage(url, display_id)
|
|
|
|
config_url = self._search_regex(
|
|
r'data-config\s*=\s*"([^"]+)"', webpage, 'data config url')
|
|
config_url = compat_urlparse.urljoin(url, config_url)
|
|
|
|
config = self._download_json(
|
|
config_url, display_id, 'Downloading config JSON')
|
|
|
|
mmc = self._download_json(
|
|
config['services']['mmc'], display_id, 'Downloading mmc JSON')
|
|
|
|
formats = []
|
|
for location in mmc['locations']:
|
|
gat = self._proto_relative_url(location.get('gat'), 'http:')
|
|
bas = location.get('bas')
|
|
loc = location.get('loc')
|
|
ogn = location.get('ogn')
|
|
if None in (gat, bas, loc, ogn):
|
|
continue
|
|
token_data = {
|
|
'bas': bas,
|
|
'icd': loc,
|
|
'ogn': ogn,
|
|
'sta': '0',
|
|
}
|
|
media = self._download_json(
|
|
'%s/?%s' % (gat, compat_urllib_parse_urlencode(token_data)),
|
|
display_id, 'Downloading %s JSON' % location['loc'])
|
|
file_ = media.get('file')
|
|
if not file_:
|
|
continue
|
|
formats.extend(self._extract_f4m_formats(
|
|
file_ + '&hdcore=3.2.0&plugin=aasp-3.2.0.77.18',
|
|
display_id, f4m_id=loc))
|
|
self._sort_formats(formats)
|
|
|
|
title = self._search_regex(
|
|
r'class="Destacado-text"[^>]*>\s*<strong>([^<]+)</strong>',
|
|
webpage, 'title', default=None)
|
|
|
|
mobj = re.search(r'''(?sx)
|
|
class="Destacado-text"[^>]*>.*?<h1>\s*
|
|
<span>(?P<series>[^<]+)</span>\s*
|
|
<span>(?P<season>[^<]+)</span>\s*
|
|
<span>(?P<episode>[^<]+)</span>''', webpage)
|
|
series, season, episode = mobj.groups() if mobj else [None] * 3
|
|
|
|
if not title:
|
|
if mobj:
|
|
title = '%s - %s - %s' % (series, season, episode)
|
|
else:
|
|
title = remove_start(self._search_regex(
|
|
r'<title>([^<]+)</title>', webpage, 'title'), 'Ver online ')
|
|
|
|
video_id = self._search_regex(
|
|
r'data-media-id\s*=\s*"([^"]+)"', webpage,
|
|
'data media id', default=None) or display_id
|
|
thumbnail = config.get('poster', {}).get('imageUrl')
|
|
duration = int_or_none(mmc.get('duration'))
|
|
|
|
return {
|
|
'id': video_id,
|
|
'display_id': display_id,
|
|
'title': title,
|
|
'description': get_element_by_attribute('class', 'text', webpage),
|
|
'series': series,
|
|
'season': season,
|
|
'episode': episode,
|
|
'thumbnail': thumbnail,
|
|
'duration': duration,
|
|
'formats': formats,
|
|
}
|