[extractor/rcs] Fix extractors (#5700)

Authored by: nixxo, pukkandan
Closes #5683
This commit is contained in:
nixxo 2023-02-12 15:43:20 +01:00 committed by GitHub
parent a4f1683221
commit c6b657867a
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

View File

@ -3,9 +3,18 @@
from .common import InfoExtractor
from ..utils import (
ExtractorError,
HEADRequest,
base_url,
clean_html,
extract_attributes,
get_element_html_by_class,
get_element_html_by_id,
int_or_none,
js_to_json,
mimetype2ext,
sanitize_url,
traverse_obj,
try_call,
url_basename,
urljoin,
)
@ -15,41 +24,8 @@ class RCSBaseIE(InfoExtractor):
# based on VideoPlayerLoader.prototype.getVideoSrc
# and VideoPlayerLoader.prototype.transformSrc from
# https://js2.corriereobjects.it/includes2013/LIBS/js/corriere_video.sjs
_ALL_REPLACE = {
'media2vam.corriere.it.edgesuite.net':
'media2vam-corriere-it.akamaized.net',
'media.youreporter.it.edgesuite.net':
'media-youreporter-it.akamaized.net',
'corrierepmd.corriere.it.edgesuite.net':
'corrierepmd-corriere-it.akamaized.net',
'media2vam-corriere-it.akamaized.net/fcs.quotidiani/vr/videos/':
'video.corriere.it/vr360/videos/',
'.net//': '.net/',
}
_MP4_REPLACE = {
'media2vam.corbologna.corriere.it.edgesuite.net':
'media2vam-bologna-corriere-it.akamaized.net',
'media2vam.corfiorentino.corriere.it.edgesuite.net':
'media2vam-fiorentino-corriere-it.akamaized.net',
'media2vam.cormezzogiorno.corriere.it.edgesuite.net':
'media2vam-mezzogiorno-corriere-it.akamaized.net',
'media2vam.corveneto.corriere.it.edgesuite.net':
'media2vam-veneto-corriere-it.akamaized.net',
'media2.oggi.it.edgesuite.net':
'media2-oggi-it.akamaized.net',
'media2.quimamme.it.edgesuite.net':
'media2-quimamme-it.akamaized.net',
'media2.amica.it.edgesuite.net':
'media2-amica-it.akamaized.net',
'media2.living.corriere.it.edgesuite.net':
'media2-living-corriere-it.akamaized.net',
'media2.style.corriere.it.edgesuite.net':
'media2-style-corriere-it.akamaized.net',
'media2.iodonna.it.edgesuite.net':
'media2-iodonna-it.akamaized.net',
'media2.leitv.it.edgesuite.net':
'media2-leitv-it.akamaized.net',
}
_UUID_RE = r'[\da-f]{8}-[\da-f]{4}-[\da-f]{4}-[\da-f]{4}-[\da-f]{12}'
_RCS_ID_RE = r'[\w-]+-\d{10}'
_MIGRATION_MAP = {
'videoamica-vh.akamaihd': 'amica',
'media2-amica-it.akamaized': 'amica',
@ -90,183 +66,140 @@ class RCSBaseIE(InfoExtractor):
'vivimilano-vh.akamaihd': 'vivimilano',
'media2-youreporter-it.akamaized': 'youreporter'
}
_MIGRATION_MEDIA = {
'advrcs-vh.akamaihd': '',
'corriere-f.akamaihd': '',
'corrierepmd-corriere-it.akamaized': '',
'corrprotetto-vh.akamaihd': '',
'gazzetta-f.akamaihd': '',
'gazzettapmd-gazzetta-it.akamaized': '',
'gazzprotetto-vh.akamaihd': '',
'periodici-f.akamaihd': '',
'periodicisecure-vh.akamaihd': '',
'videocoracademy-vh.akamaihd': ''
}
def _get_video_src(self, video):
mediaFiles = video.get('mediaProfile').get('mediaFile')
src = {}
# audio
if video.get('mediaType') == 'AUDIO':
for aud in mediaFiles:
# todo: check
src['mp3'] = aud.get('value')
# video
else:
for vid in mediaFiles:
if vid.get('mimeType') == 'application/vnd.apple.mpegurl':
src['m3u8'] = vid.get('value')
if vid.get('mimeType') == 'video/mp4':
src['mp4'] = vid.get('value')
for source in traverse_obj(video, (
'mediaProfile', 'mediaFile', lambda _, v: v.get('mimeType'))):
url = source['value']
for s, r in (
('media2vam.corriere.it.edgesuite.net', 'media2vam-corriere-it.akamaized.net'),
('media.youreporter.it.edgesuite.net', 'media-youreporter-it.akamaized.net'),
('corrierepmd.corriere.it.edgesuite.net', 'corrierepmd-corriere-it.akamaized.net'),
('media2vam-corriere-it.akamaized.net/fcs.quotidiani/vr/videos/', 'video.corriere.it/vr360/videos/'),
('http://', 'https://'),
):
url = url.replace(s, r)
# replace host
for t in src:
for s, r in self._ALL_REPLACE.items():
src[t] = src[t].replace(s, r)
for s, r in self._MP4_REPLACE.items():
src[t] = src[t].replace(s, r)
type_ = mimetype2ext(source['mimeType'])
if type_ == 'm3u8' and '-vh.akamaihd' in url:
# still needed for some old content: see _TESTS #3
matches = re.search(r'(?:https?:)?//(?P<host>[\w\.\-]+)\.net/i(?P<path>.+)$', url)
if matches:
url = f'https://vod.rcsobjects.it/hls/{self._MIGRATION_MAP[matches.group("host")]}{matches.group("path")}'
if traverse_obj(video, ('mediaProfile', 'geoblocking')) or (
type_ == 'm3u8' and 'fcs.quotidiani_!' in url):
url = url.replace('vod.rcsobjects', 'vod-it.rcsobjects')
if type_ == 'm3u8' and 'vod' in url:
url = url.replace('.csmil', '.urlset')
if type_ == 'mp3':
url = url.replace('media2vam-corriere-it.akamaized.net', 'vod.rcsobjects.it/corriere')
# switch cdn
if 'mp4' in src and 'm3u8' in src:
if ('-lh.akamaihd' not in src.get('m3u8')
and 'akamai' in src.get('mp4')):
if 'm3u8' in src:
matches = re.search(r'(?:https*:)?\/\/(?P<host>.*)\.net\/i(?P<path>.*)$', src.get('m3u8'))
src['m3u8'] = 'https://vod.rcsobjects.it/hls/%s%s' % (
self._MIGRATION_MAP[matches.group('host')],
matches.group('path').replace(
'///', '/').replace(
'//', '/').replace(
'.csmil', '.urlset'
)
)
if 'mp4' in src:
matches = re.search(r'(?:https*:)?\/\/(?P<host>.*)\.net\/i(?P<path>.*)$', src.get('mp4'))
if matches:
if matches.group('host') in self._MIGRATION_MEDIA:
vh_stream = 'https://media2.corriereobjects.it'
if src.get('mp4').find('fcs.quotidiani_!'):
vh_stream = 'https://media2-it.corriereobjects.it'
src['mp4'] = '%s%s' % (
vh_stream,
matches.group('path').replace(
'///', '/').replace(
'//', '/').replace(
'/fcs.quotidiani/mediacenter', '').replace(
'/fcs.quotidiani_!/mediacenter', '').replace(
'corriere/content/mediacenter/', '').replace(
'gazzetta/content/mediacenter/', '')
)
else:
src['mp4'] = 'https://vod.rcsobjects.it/%s%s' % (
self._MIGRATION_MAP[matches.group('host')],
matches.group('path').replace('///', '/').replace('//', '/')
)
yield {
'type': type_,
'url': url,
'bitrate': source.get('bitrate')
}
if 'mp3' in src:
src['mp3'] = src.get('mp3').replace(
'media2vam-corriere-it.akamaized.net',
'vod.rcsobjects.it/corriere')
if 'mp4' in src:
if src.get('mp4').find('fcs.quotidiani_!'):
src['mp4'] = src.get('mp4').replace('vod.rcsobjects', 'vod-it.rcsobjects')
if 'm3u8' in src:
if src.get('m3u8').find('fcs.quotidiani_!'):
src['m3u8'] = src.get('m3u8').replace('vod.rcsobjects', 'vod-it.rcsobjects')
def _create_http_formats(self, m3u8_formats, video_id):
for f in m3u8_formats:
if f['vcodec'] == 'none':
continue
http_url = re.sub(r'(https?://[^/]+)/hls/([^?#]+?\.mp4).+', r'\g<1>/\g<2>', f['url'])
if http_url == f['url']:
continue
if 'geoblocking' in video.get('mediaProfile'):
if 'm3u8' in src:
src['m3u8'] = src.get('m3u8').replace('vod.rcsobjects', 'vod-it.rcsobjects')
if 'mp4' in src:
src['mp4'] = src.get('mp4').replace('vod.rcsobjects', 'vod-it.rcsobjects')
if 'm3u8' in src:
if src.get('m3u8').find('csmil') and src.get('m3u8').find('vod'):
src['m3u8'] = src.get('m3u8').replace('.csmil', '.urlset')
http_f = f.copy()
del http_f['manifest_url']
format_id = try_call(lambda: http_f['format_id'].replace('hls-', 'https-'))
urlh = self._request_webpage(HEADRequest(http_url), video_id, fatal=False,
note=f'Check filesize for {format_id}')
if not urlh:
continue
return src
def _create_formats(self, urls, video_id):
formats = []
formats = self._extract_m3u8_formats(
urls.get('m3u8'), video_id, 'mp4', entry_protocol='m3u8_native',
m3u8_id='hls', fatal=False)
if urls.get('mp4'):
formats.append({
'format_id': 'http-mp4',
'url': urls['mp4']
http_f.update({
'format_id': format_id,
'url': http_url,
'protocol': 'https',
'filesize_approx': int_or_none(urlh.headers.get('Content-Length', None)),
})
return formats
yield http_f
def _create_formats(self, sources, video_id):
for source in sources:
if source['type'] == 'm3u8':
m3u8_formats = self._extract_m3u8_formats(
source['url'], video_id, 'mp4', m3u8_id='hls', fatal=False)
yield from m3u8_formats
yield from self._create_http_formats(m3u8_formats, video_id)
elif source['type'] == 'mp3':
yield {
'format_id': 'https-mp3',
'ext': 'mp3',
'acodec': 'mp3',
'vcodec': 'none',
'abr': source.get('bitrate'),
'url': source['url'],
}
def _real_extract(self, url):
mobj = self._match_valid_url(url)
video_id = mobj.group('id')
cdn, video_id = self._match_valid_url(url).group('cdn', 'id')
display_id, video_data = None, None
if 'cdn' not in mobj.groupdict():
raise ExtractorError('CDN not found in url: %s' % url)
# for leitv/youreporter/viaggi don't use the embed page
if ((mobj.group('cdn') not in ['leitv.it', 'youreporter.it'])
and (mobj.group('vid') == 'video')):
url = 'https://video.%s/video-embed/%s' % (mobj.group('cdn'), video_id)
page = self._download_webpage(url, video_id)
video_data = None
# look for json video data url
json = self._search_regex(
r'''(?x)url\s*=\s*(["'])
(?P<url>
(?:https?:)?//video\.rcs\.it
/fragment-includes/video-includes/.+?\.json
)\1;''',
page, video_id, group='url', default=None)
if json:
if json.startswith('//'):
json = 'https:%s' % json
video_data = self._download_json(json, video_id)
# if json url not found, look for json video data directly in the page
if re.match(self._UUID_RE, video_id) or re.match(self._RCS_ID_RE, video_id):
url = f'https://video.{cdn}/video-json/{video_id}'
else:
# RCS normal pages and most of the embeds
json = self._search_regex(
r'[\s;]video\s*=\s*({[\s\S]+?})(?:;|,playlist=)',
page, video_id, default=None)
if not json and 'video-embed' in url:
page = self._download_webpage(url.replace('video-embed', 'video-json'), video_id)
json = self._search_regex(
r'##start-video##({[\s\S]+?})##end-video##',
page, video_id, default=None)
if not json:
# if no video data found try search for iframes
emb = RCSEmbedsIE._extract_url(page)
webpage = self._download_webpage(url, video_id)
data_config = get_element_html_by_id('divVideoPlayer', webpage) or get_element_html_by_class('divVideoPlayer', webpage)
if data_config:
data_config = self._parse_json(
extract_attributes(data_config).get('data-config'),
video_id, fatal=False) or {}
if data_config.get('newspaper'):
cdn = f'{data_config["newspaper"]}.it'
display_id, video_id = video_id, data_config.get('uuid') or video_id
url = f'https://video.{cdn}/video-json/{video_id}'
else:
json_url = self._search_regex(
r'''(?x)url\s*=\s*(["'])
(?P<url>
(?:https?:)?//video\.rcs\.it
/fragment-includes/video-includes/[^"']+?\.json
)\1;''',
webpage, video_id, group='url', default=None)
if json_url:
video_data = self._download_json(sanitize_url(json_url, scheme='https'), video_id)
display_id, video_id = video_id, video_data.get('id') or video_id
if not video_data:
webpage = self._download_webpage(url, video_id)
video_data = self._search_json(
'##start-video##', webpage, 'video data', video_id, default=None,
end_pattern='##end-video##', transform_source=js_to_json)
if not video_data:
# try search for iframes
emb = RCSEmbedsIE._extract_url(webpage)
if emb:
return {
'_type': 'url_transparent',
'url': emb,
'ie_key': RCSEmbedsIE.ie_key()
}
if json:
video_data = self._parse_json(
json, video_id, transform_source=js_to_json)
if not video_data:
raise ExtractorError('Video data not found in the page')
formats = self._create_formats(
self._get_video_src(video_data), video_id)
description = (video_data.get('description')
or clean_html(video_data.get('htmlDescription'))
or self._html_search_meta('description', page))
uploader = video_data.get('provider') or mobj.group('cdn')
return {
'id': video_id,
'display_id': display_id,
'title': video_data.get('title'),
'description': description,
'uploader': uploader,
'formats': formats
'description': (clean_html(video_data.get('description'))
or clean_html(video_data.get('htmlDescription'))
or self._html_search_meta('description', webpage)),
'uploader': video_data.get('provider') or cdn,
'formats': list(self._create_formats(self._get_video_src(video_data), video_id)),
}
@ -296,7 +229,7 @@ class RCSEmbedsIE(RCSBaseIE):
\1''']
_TESTS = [{
'url': 'https://video.rcs.it/video-embed/iodonna-0001585037',
'md5': '623ecc8ffe7299b2d0c1046d8331a9df',
'md5': '0faca97df525032bb9847f690bc3720c',
'info_dict': {
'id': 'iodonna-0001585037',
'ext': 'mp4',
@ -305,38 +238,31 @@ class RCSEmbedsIE(RCSBaseIE):
'uploader': 'rcs.it',
}
}, {
# redownload the page changing 'video-embed' in 'video-json'
'url': 'https://video.gazzanet.gazzetta.it/video-embed/gazzanet-mo05-0000260789',
'md5': 'a043e3fecbe4d9ed7fc5d888652a5440',
'info_dict': {
'id': 'gazzanet-mo05-0000260789',
'ext': 'mp4',
'title': 'Valentino Rossi e papà Graziano si divertono col drifting',
'description': 'md5:a8bf90d6adafd9815f70fc74c0fc370a',
'uploader': 'rcd',
}
}, {
'url': 'https://video.corriere.it/video-embed/b727632a-f9d0-11ea-91b0-38d50a849abb?player',
'match_only': True
}, {
'url': 'https://video.gazzetta.it/video-embed/49612410-00ca-11eb-bcd8-30d4253e0140',
'match_only': True
}]
_WEBPAGE_TESTS = [{
'url': 'https://www.iodonna.it/video-iodonna/personaggi-video/monica-bellucci-piu-del-lavoro-oggi-per-me-sono-importanti-lamicizia-e-la-famiglia/',
'info_dict': {
'id': 'iodonna-0002033648',
'ext': 'mp4',
'title': 'Monica Bellucci: «Più del lavoro, oggi per me sono importanti l\'amicizia e la famiglia»',
'description': 'md5:daea6d9837351e56b1ab615c06bebac1',
'uploader': 'rcs.it',
}
}]
@staticmethod
def _sanitize_urls(urls):
# add protocol if missing
for i, e in enumerate(urls):
if e.startswith('//'):
urls[i] = 'https:%s' % e
# clean iframes urls
for i, e in enumerate(urls):
urls[i] = urljoin(base_url(e), url_basename(e))
return urls
def _sanitize_url(url):
url = sanitize_url(url, scheme='https')
return urljoin(base_url(url), url_basename(url))
@classmethod
def _extract_embed_urls(cls, url, webpage):
return cls._sanitize_urls(list(super()._extract_embed_urls(url, webpage)))
return map(cls._sanitize_url, super()._extract_embed_urls(url, webpage))
class RCSIE(RCSBaseIE):
@ -349,37 +275,53 @@ class RCSIE(RCSBaseIE):
|corrierefiorentino\.
)?corriere\.it
|(?:gazzanet\.)?gazzetta\.it)
/(?!video-embed/).+?/(?P<id>[^/\?]+)(?=\?|/$|$)'''
/(?!video-embed/)[^?#]+?/(?P<id>[^/\?]+)(?=\?|/$|$)'''
_TESTS = [{
# json iframe directly from id
'url': 'https://video.corriere.it/sport/formula-1/vettel-guida-ferrari-sf90-mugello-suo-fianco-c-elecrerc-bendato-video-esilarante/b727632a-f9d0-11ea-91b0-38d50a849abb',
'md5': '0f4ededc202b0f00b6e509d831e2dcda',
'md5': '14946840dec46ecfddf66ba4eea7d2b2',
'info_dict': {
'id': 'b727632a-f9d0-11ea-91b0-38d50a849abb',
'ext': 'mp4',
'title': 'Vettel guida la Ferrari SF90 al Mugello e al suo fianco c\'è Leclerc (bendato): il video è esilarante',
'description': 'md5:93b51c9161ac8a64fb2f997b054d0152',
'description': 'md5:3915ce5ebb3d2571deb69a5eb85ac9b5',
'uploader': 'Corriere Tv',
}
}, {
# video data inside iframe
# search for video id inside the page
'url': 'https://viaggi.corriere.it/video/norvegia-il-nuovo-ponte-spettacolare-sopra-la-cascata-di-voringsfossen/',
'md5': 'da378e4918d2afbf7d61c35abb948d4c',
'md5': 'f22a92d9e666e80f2fffbf2825359c81',
'info_dict': {
'id': '5b7cd134-e2c1-11ea-89b3-b56dd0df2aa2',
'display_id': 'norvegia-il-nuovo-ponte-spettacolare-sopra-la-cascata-di-voringsfossen',
'ext': 'mp4',
'title': 'La nuova spettacolare attrazione in Norvegia: il ponte sopra Vøringsfossen',
'description': 'md5:18b35a291f6746c0c8dacd16e5f5f4f8',
'uploader': 'DOVE Viaggi',
}
}, {
'url': 'https://video.gazzetta.it/video-motogp-catalogna-cadute-dovizioso-vale-rossi/49612410-00ca-11eb-bcd8-30d4253e0140?vclk=Videobar',
'md5': 'eedc1b5defd18e67383afef51ff7bdf9',
# only audio format https://github.com/yt-dlp/yt-dlp/issues/5683
'url': 'https://video.corriere.it/cronaca/audio-telefonata-il-papa-becciu-santita-lettera-che-mi-ha-inviato-condanna/b94c0d20-70c2-11ed-9572-e4b947a0ebd2',
'md5': 'aaffb08d02f2ce4292a4654694c78150',
'info_dict': {
'id': '49612410-00ca-11eb-bcd8-30d4253e0140',
'id': 'b94c0d20-70c2-11ed-9572-e4b947a0ebd2',
'ext': 'mp3',
'title': 'L\'audio della telefonata tra il Papa e Becciu: «Santità, la lettera che mi ha inviato è una condanna»',
'description': 'md5:c0ddb61bd94a8d4e0d4bb9cda50a689b',
'uploader': 'Corriere Tv',
'formats': [{'format_id': 'https-mp3', 'ext': 'mp3'}],
}
}, {
# old content still needs cdn migration
'url': 'https://viaggi.corriere.it/video/milano-varallo-sesia-sul-treno-a-vapore/',
'md5': '2dfdce7af249654ad27eeba03fe1e08d',
'info_dict': {
'id': 'd8f6c8d0-f7d7-11e8-bfca-f74cf4634191',
'display_id': 'milano-varallo-sesia-sul-treno-a-vapore',
'ext': 'mp4',
'title': 'Dovizioso, il contatto con Zarco e la caduta. E anche Vale finisce a terra',
'description': 'md5:8c6e905dc3b9413218beca11ebd69778',
'uploader': 'AMorici',
'title': 'Milano-Varallo Sesia sul treno a vapore',
'description': 'md5:6348f47aac230397fe341a74f7678d53',
'uploader': 'DOVE Viaggi',
}
}, {
'url': 'https://video.corriere.it/video-360/metro-copenaghen-tutta-italiana/a248a7f0-e2db-11e9-9830-af2de6b1f945',
@ -391,13 +333,15 @@ class RCSVariousIE(RCSBaseIE):
_VALID_URL = r'''(?x)https?://www\.
(?P<cdn>
leitv\.it|
youreporter\.it
youreporter\.it|
amica\.it
)/(?:[^/]+/)?(?P<id>[^/]+?)(?:$|\?|/)'''
_TESTS = [{
'url': 'https://www.leitv.it/benessere/mal-di-testa-come-combatterlo-ed-evitarne-la-comparsa/',
'md5': '92b4e63667b8f95acb0a04da25ae28a1',
'url': 'https://www.leitv.it/benessere/mal-di-testa/',
'md5': '3b7a683d105a7313ec7513b014443631',
'info_dict': {
'id': 'mal-di-testa-come-combatterlo-ed-evitarne-la-comparsa',
'id': 'leitv-0000125151',
'display_id': 'mal-di-testa',
'ext': 'mp4',
'title': 'Cervicalgia e mal di testa, il video con i suggerimenti dell\'esperto',
'description': 'md5:ae21418f34cee0b8d02a487f55bcabb5',
@ -405,12 +349,24 @@ class RCSVariousIE(RCSBaseIE):
}
}, {
'url': 'https://www.youreporter.it/fiume-sesia-3-ottobre-2020/',
'md5': '8dccd436b47a830bab5b4a88232f391a',
'md5': '3989b6d603482611a2abd2f32b79f739',
'info_dict': {
'id': 'fiume-sesia-3-ottobre-2020',
'id': 'youreporter-0000332574',
'display_id': 'fiume-sesia-3-ottobre-2020',
'ext': 'mp4',
'title': 'Fiume Sesia 3 ottobre 2020',
'description': 'md5:0070eef1cc884d13c970a4125063de55',
'uploader': 'youreporter.it',
}
}, {
'url': 'https://www.amica.it/video-post/saint-omer-al-cinema-il-film-leone-dargento-che-ribalta-gli-stereotipi/',
'md5': '187cce524dfd0343c95646c047375fc4',
'info_dict': {
'id': 'amica-0001225365',
'display_id': 'saint-omer-al-cinema-il-film-leone-dargento-che-ribalta-gli-stereotipi',
'ext': 'mp4',
'title': '"Saint Omer": al cinema il film Leone d\'argento che ribalta gli stereotipi',
'description': 'md5:b1c8869c2dcfd6073a2a311ba0008aa8',
'uploader': 'rcs.it',
}
}]