1
0
mirror of https://github.com/ytdl-org/youtube-dl.git synced 2024-11-16 06:17:27 +00:00
youtube-dl/youtube_dl/extractor/rtve.py
Alba Mendez d7939e2c07 improve metadata
grab metadata from the new endpoint (videos/%d.json)
instead of the legacy endpoint
(videos/%d/config/alacarta_videos.json),
which is what is actually used for the UI now.

- gives more up to date titles
- adds description / URL
- makes supporting audios easy (see next commit)
2022-09-17 20:24:15 +02:00

327 lines
12 KiB
Python

# coding: utf-8
from __future__ import unicode_literals
import base64
import io
import re
import sys
from .common import InfoExtractor
from ..compat import (
compat_b64decode,
compat_struct_unpack,
)
from ..utils import (
clean_html,
determine_ext,
ExtractorError,
float_or_none,
qualities,
std_headers,
)
_bytes_to_chr = (lambda x: x) if sys.version_info[0] == 2 else (lambda x: map(chr, x))
class RTVEPlayIE(InfoExtractor):
IE_NAME = 'rtve.es:play'
IE_DESC = 'RTVE Play'
_VALID_URL = r'https?://(?:www\.)?rtve\.es/(playz?/videos|filmoteca)/[^/]+/[^/]+/(?P<id>\d+)'
_TESTS = [{
'url': 'http://www.rtve.es/alacarta/videos/balonmano/o-swiss-cup-masculina-final-espana-suecia/2491869/',
'md5': '2c70aacf8a415d1b4e7fcc0525951162',
'info_dict': {
'id': '2491869',
'ext': 'mp4',
'title': 'Final de la Swiss Cup masculina: España-Suecia',
'description': 'Swiss Cup masculina, Final: España-Suecia.',
'duration': 5024.566,
'series': 'Balonmano',
},
'expected_warnings': ['Failed to download MPD manifest', 'Failed to download m3u8 information'],
}, {
'note': 'Live stream',
'url': 'http://www.rtve.es/alacarta/videos/television/24h-live/1694255/',
'info_dict': {
'id': '1694255',
'ext': 'mp4',
'title': 're:^24H LIVE [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$',
'description': '24H LIVE',
'is_live': True,
},
'params': {
'skip_download': 'live stream',
},
}, {
'url': 'http://www.rtve.es/alacarta/videos/servir-y-proteger/servir-proteger-capitulo-104/4236788/',
'md5': '30b8827cba25f39d1af5a7c482cc8ac5',
'info_dict': {
'id': '4236788',
'ext': 'mp4',
'title': 'Capítulo 104',
'description': 'md5:caae29ae04291875e611dd667fe84641',
'duration': 3222.0,
},
'expected_warnings': ['Failed to download MPD manifest', 'Failed to download m3u8 information'],
}, {
'url': 'http://www.rtve.es/m/alacarta/videos/cuentame-como-paso/cuentame-como-paso-t16-ultimo-minuto-nuestra-vida-capitulo-276/2969138/?media=tve',
'only_matching': True,
}, {
'url': 'http://www.rtve.es/filmoteca/no-do/not-1-introduccion-primer-noticiario-espanol/1465256/',
'only_matching': True,
}]
def _real_initialize(self):
user_agent_b64 = base64.b64encode(std_headers['User-Agent'].encode('utf-8')).decode('utf-8')
self._manager = self._download_json(
'http://www.rtve.es/odin/loki/' + user_agent_b64,
None, 'Fetching manager info')['manager']
@staticmethod
def _decrypt_url(png):
encrypted_data = io.BytesIO(compat_b64decode(png)[8:])
while True:
length = compat_struct_unpack('!I', encrypted_data.read(4))[0]
chunk_type = encrypted_data.read(4)
if chunk_type == b'IEND':
break
data = encrypted_data.read(length)
if chunk_type == b'tEXt':
alphabet_data, text = data.replace(b'\0', b'').split(b'#')
components = text.split(b'%%')
if len(components) < 2:
components.insert(0, b'')
quality, url_data = components
alphabet = []
e = 0
d = 0
for l in _bytes_to_chr(alphabet_data):
if d == 0:
alphabet.append(l)
d = e = (e + 1) % 4
else:
d -= 1
url = ''
f = 0
e = 3
b = 1
for letter in _bytes_to_chr(url_data):
if f == 0:
l = int(letter) * 10
f = 1
else:
if e == 0:
l += int(letter)
url += alphabet[l]
e = (b + 3) % 4
f = 0
b += 1
else:
e -= 1
yield quality.decode(), url
encrypted_data.read(4) # CRC
def _extract_png_formats(self, video_id):
png = self._download_webpage(
'http://ztnr.rtve.es/ztnr/movil/thumbnail/%s/videos/%s.png' % (self._manager, video_id),
video_id, 'Downloading url information', query={'q': 'v2'})
q = qualities(['Media', 'Alta', 'HQ', 'HD_READY', 'HD_FULL'])
formats = []
for quality, video_url in self._decrypt_url(png):
ext = determine_ext(video_url)
if ext == 'm3u8':
formats.extend(self._extract_m3u8_formats(
video_url, video_id, 'mp4', 'm3u8_native',
m3u8_id='hls', fatal=False))
elif ext == 'mpd':
formats.extend(self._extract_mpd_formats(
video_url, video_id, 'dash', fatal=False))
else:
formats.append({
'format_id': quality,
'quality': q(quality),
'url': video_url,
})
self._sort_formats(formats)
return formats
def _real_extract(self, url):
return self._real_extract_from_id(self._match_id(url))
def _real_extract_from_id(self, video_id):
info = self._download_json(
'http://www.rtve.es/api/videos/%s.json' % video_id,
video_id)['page']['items'][0]
if (info.get('pubState') or {}).get('code') == 'DESPU':
raise ExtractorError('The video is no longer available', expected=True)
title = info['title'].strip()
formats = self._extract_png_formats(video_id)
subtitles = None
sbt_file = info.get('sbtFile')
if sbt_file:
subtitles = self.extract_subtitles(video_id, sbt_file)
is_live = info.get('consumption') == 'live'
return {
'id': video_id,
'title': self._live_title(title) if is_live else title,
'formats': formats,
'url': info.get('htmlUrl'),
'description': clean_html(info.get('description')),
'thumbnail': info.get('thumbnail'),
'subtitles': subtitles,
'duration': float_or_none(info.get('duration'), 1000),
'is_live': is_live,
'series': (info.get('programInfo') or {}).get('title'),
}
def _get_subtitles(self, video_id, sub_file):
subs = self._download_json(
sub_file + '.json', video_id,
'Downloading subtitles info')['page']['items']
return dict(
(s['lang'], [{'ext': 'vtt', 'url': s['src']}])
for s in subs)
class RTVEInfantilIE(RTVEPlayIE):
IE_NAME = 'rtve.es:infantil'
IE_DESC = 'RTVE infantil'
_VALID_URL = r'https?://(?:www\.)?rtve\.es/infantil/serie/[^/]+/video/[^/]+/(?P<id>[0-9]+)/'
_TESTS = [{
'url': 'https://www.rtve.es/infantil/serie/dino-ranch/video/pequeno-gran-ayudante/6693248/',
'md5': '06d3f57eec593ad93fe9dcf079fbd940',
'info_dict': {
'id': '6693248',
'ext': 'mp4',
'title': 'Un pequeño gran ayudante',
'description': 'md5:144ca351e31f9ee99a637ab9fc2787d5',
'thumbnail': r're:https?://.+/1663318364501\.jpg',
'duration': 691.44,
},
'expected_warnings': ['Failed to download MPD manifest', 'Failed to download m3u8 information'],
}]
class RTVELiveIE(RTVEPlayIE):
IE_NAME = 'rtve.es:live'
IE_DESC = 'RTVE.es live streams'
_VALID_URL = r'https?://(?:www\.)?rtve\.es/play/videos/directo/(?P<id>.+)'
_TESTS = [{
'url': 'https://www.rtve.es/play/videos/directo/la-1/',
'info_dict': {
'id': '1688877',
'ext': 'mp4',
'title': 're:^La 1 [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$',
'description': 'La 1',
},
'params': {
'skip_download': 'live stream',
}
}, {
'url': 'https://www.rtve.es/play/videos/directo/canales-lineales/la-1/',
'info_dict': {
'id': '1688877',
'ext': 'mp4',
'title': 're:^La 1 [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$',
'description': 'La 1',
},
'params': {
'skip_download': 'live stream',
}
}, {
'url': 'https://www.rtve.es/play/videos/directo/canales-lineales/capilla-ardiente-isabel-westminster/10886/',
'info_dict': {
'id': '1938028',
'ext': 'mp4',
'title': 're:^Mas24 - 1 [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$',
'description': 'Mas24 - 1',
},
'params': {
'skip_download': 'live stream',
}
}]
def _real_extract(self, url):
webpage = self._download_webpage(url, self._match_id(url))
asset_id = self._search_regex(
r'class=["\'].*?\bvideoPlayer\b.*?["\'][^>]+data-setup=[^>]+?(?:"|&quot;)idAsset(?:"|&quot;)\s*:\s*(?:"|&quot;)(\d+)(?:"|&quot;)',
webpage, 'internal video ID')
return self._real_extract_from_id(asset_id)
class RTVETelevisionIE(InfoExtractor):
IE_NAME = 'rtve.es:television'
# https://www.rtve.es/SECTION/YYYYMMDD/CONTENT_SLUG/CONTENT_ID.shtml
_VALID_URL = r'https?://(?:www\.)?rtve\.es/[^/]+/\d{8}/[^/]+/(?P<id>\d+)\.shtml'
_TESTS = [{
'url': 'https://www.rtve.es/television/20220916/destacados-festival-san-sebastian-rtve-play/2395620.shtml',
'info_dict': {
'id': '6668919',
'ext': 'mp4',
'title': 'Las películas del Festival de San Sebastián en RTVE Play',
'description': 'El\xa0Festival de San Sebastián vuelve a llenarse de artistas. Y en su honor,\xa0RTVE Play\xa0destacará cada viernes una\xa0película galardonada\xa0con la\xa0Concha de Oro\xa0en su catálogo.',
'duration': 20.048,
},
'params': {
'skip_download': True,
},
}, {
'url': 'https://www.rtve.es/noticias/20220917/penelope-cruz-san-sebastian-premio-nacional/2402565.shtml',
'info_dict': {
'id': '6694087',
'ext': 'mp4',
'title': 'Penélope Cruz recoge el Premio Nacional de Cinematografía: "No dejen nunca de proteger nuestro cine"',
'description': 'md5:eda9e6baa78dbbbcc7708c0cc8150a91',
'duration': 388.2,
},
'params': {
'skip_download': True,
},
}, {
'url': 'https://www.rtve.es/deportes/20220917/motogp-bagnaia-pole-marquez-decimotercero-motorland-aragon/2402566.shtml',
'info_dict': {
'id': '6694142',
'ext': 'mp4',
'title': "Bagnaia logra su quinta 'pole' del año y Márquez partirá decimotercero",
'description': 'md5:07e2ccb983a046cb42f896cce225f0a7',
'duration': 153.44,
},
'params': {
'skip_download': True,
},
}, {
'url': 'https://www.rtve.es/playz/20220807/covaleda-fest-final/2394809.shtml',
'info_dict': {
'id': '6665408',
'ext': 'mp4',
'title': 'Covaleda Fest (Soria) - Día 3 con Marc Seguí y Paranoid 1966',
'description': 'Festivales Playz viaja a Covaleda, Soria, para contarte todo lo que sucede en el Covaleda Fest. Entrevistas, challenges a los artistas, juegos... Khan, Adriana Jiménez y María García no dejarán pasar ni una. ¡No te lo pierdas!',
'duration': 12009.92,
},
'params': {
'skip_download': True,
},
}]
def _real_extract(self, url):
page_id = self._match_id(url)
webpage = self._download_webpage(url, page_id)
alacarta_url = self._search_regex(
r'data-location="alacarta_videos"[^<]+url&quot;:&quot;(https?://www\.rtve\.es/play.+?)&',
webpage, 'alacarta url', default=None)
if alacarta_url is None:
raise ExtractorError(
'The webpage doesn\'t contain any video', expected=True)
return self.url_result(alacarta_url, ie=RTVEPlayIE.ie_key())