mirror of
https://github.com/ytdl-org/youtube-dl
synced 2024-12-22 20:20:09 +09:00
Add Ant1NewsGr IEs
* Add new IEs * Ant1NewsGrBaseIE: Base IE class * Ant1NewsGrWatchIE: Extract videos from TV VOD pages * Ant1NewsGrArticleIE: Extract videos from news articles * Ant1NewsGrEmbedIE: Extract iframe-embeddable ant1news.gr videos * NB: There is a different platform at vod.antenna.gr, which is not covered here * The Generic extractor can also be used to extract videos from news article pages (through Ant1NewsGrEmbed._extract_urls), however a specific IE is used to extract JSON-LD @type='NewsArticle' metadata * However that does not work currently, as _json_ld does not support @graph nesting * Ant1NewsGrArticleIE defers to Ant1NewsGrEmbedIE, either as a playlist or a single video * Ant1NewsGrWatchIE and Ant1NewsGrEmbedIE query an API endpoint to extract metadata, get the respective stream/source URLs and detect video formats * The endpoint HTTP path varies per IE * Ant1NewsGrEmbedIE first resolves any redirects for its' own URL, to derive the properl base URL for the API query
This commit is contained in:
parent
a803582717
commit
9af0f299bf
190
youtube_dl/extractor/ant1newsgr.py
Normal file
190
youtube_dl/extractor/ant1newsgr.py
Normal file
@ -0,0 +1,190 @@
|
|||||||
|
# coding: utf-8
|
||||||
|
from __future__ import unicode_literals
|
||||||
|
|
||||||
|
import re
|
||||||
|
|
||||||
|
from .common import InfoExtractor
|
||||||
|
from ..compat import (
|
||||||
|
compat_parse_qs,
|
||||||
|
compat_urllib_parse_urlparse,
|
||||||
|
compat_urlparse,
|
||||||
|
)
|
||||||
|
from ..utils import (
|
||||||
|
HEADRequest,
|
||||||
|
ExtractorError,
|
||||||
|
determine_ext,
|
||||||
|
smuggle_url,
|
||||||
|
unsmuggle_url,
|
||||||
|
unescapeHTML,
|
||||||
|
)
|
||||||
|
|
||||||
|
|
||||||
|
class Ant1NewsGrBaseIE(InfoExtractor):
|
||||||
|
@staticmethod
|
||||||
|
def _smuggle_parent_info(url, **info_dict):
|
||||||
|
return smuggle_url(url, {'parent_info': info_dict})
|
||||||
|
|
||||||
|
@staticmethod
|
||||||
|
def _unsmuggle_parent_info(url):
|
||||||
|
unsmuggled_url, data = unsmuggle_url(url, default={'parent_info': {}})
|
||||||
|
return unsmuggled_url, data['parent_info']
|
||||||
|
|
||||||
|
def _download_api_data(self, netloc, cid, scheme='https'):
|
||||||
|
url_parts = (scheme, netloc, self._API_PATH, None, None, None)
|
||||||
|
url = compat_urlparse.urlunparse(url_parts)
|
||||||
|
query = {'cid': cid}
|
||||||
|
return self._download_json(
|
||||||
|
url, cid,
|
||||||
|
'Downloading JSON',
|
||||||
|
'Unable to download JSON',
|
||||||
|
query=query)
|
||||||
|
|
||||||
|
def _download_and_extract_api_data(self, video_id, *args, **kwargs):
|
||||||
|
info = self._download_api_data(*args, **kwargs)
|
||||||
|
try:
|
||||||
|
source = info['url']
|
||||||
|
except KeyError:
|
||||||
|
raise ExtractorError('no source found for %s' % video_id)
|
||||||
|
formats = self._extract_m3u8_formats(source, video_id, 'mp4') \
|
||||||
|
if determine_ext(source) == 'm3u8' else [source]
|
||||||
|
self._sort_formats(formats)
|
||||||
|
return {
|
||||||
|
'id': video_id,
|
||||||
|
'title': info['title'],
|
||||||
|
'thumbnail': info['thumb'],
|
||||||
|
'formats': formats,
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
|
class Ant1NewsGrWatchIE(Ant1NewsGrBaseIE):
|
||||||
|
IE_NAME = 'ant1newsgr:watch'
|
||||||
|
IE_DESC = 'ant1news.gr videos'
|
||||||
|
_VALID_URL = r'https?://(?:www\.)?ant1news\.gr/watch/(?P<id>\d+)/'
|
||||||
|
_API_PATH = '/templates/data/player'
|
||||||
|
|
||||||
|
_TEST = {
|
||||||
|
'url': 'https://www.ant1news.gr/watch/1506168/ant1-news-09112021-stis-18-45',
|
||||||
|
'md5': '60a984da5ffc98c9924e6d9dd46c6f04',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '1506168',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'md5:0ad00fa66ecf8aa233d26ab0dba7514a',
|
||||||
|
'description': 'md5:18665af715a6dcfeac1d6153a44f16b0',
|
||||||
|
},
|
||||||
|
}
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
video_id = self._match_id(url)
|
||||||
|
webpage = self._download_webpage(url, video_id)
|
||||||
|
scheme, netloc, _, _, _, _ = compat_urllib_parse_urlparse(url)
|
||||||
|
info = self._download_and_extract_api_data(
|
||||||
|
video_id, netloc, video_id, scheme=scheme)
|
||||||
|
info['description'] = self._og_search_description(webpage)
|
||||||
|
return info
|
||||||
|
|
||||||
|
|
||||||
|
class Ant1NewsGrArticleIE(Ant1NewsGrBaseIE):
|
||||||
|
IE_NAME = 'ant1newsgr:article'
|
||||||
|
IE_DESC = 'ant1news.gr articles'
|
||||||
|
_VALID_URL = r'https?://(?:www\.)?ant1news\.gr/[^/]+/article/(?P<id>\d+)/'
|
||||||
|
|
||||||
|
_TESTS = [{
|
||||||
|
'url': 'https://www.ant1news.gr/afieromata/article/549468/o-tzeims-mpont-sta-meteora-oi-apeiles-kai-o-xesikomos-ton-kalogeron',
|
||||||
|
'md5': 'eb635a194c15272c2611a751766b0200',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '_xvg/m_cmbatw=',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'md5:a93e8ecf2e4073bfdffcb38f59945411',
|
||||||
|
},
|
||||||
|
'expected_warnings': [r'^[Uu]nable to extract JSON-LD'],
|
||||||
|
}, {
|
||||||
|
'url': 'https://ant1news.gr/Society/article/620286/symmoria-anilikon-dikigoros-thymaton-ithelan-na-toys-apoteleiosoyn',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '620286',
|
||||||
|
'title': 'md5:91fe569e952e4d146485740ae927662b',
|
||||||
|
},
|
||||||
|
'expected_warnings': [r'^[Uu]nable to extract JSON-LD'],
|
||||||
|
'playlist_mincount': 2,
|
||||||
|
'params': {
|
||||||
|
'skip_download': True,
|
||||||
|
},
|
||||||
|
}]
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
video_id = self._match_id(url)
|
||||||
|
webpage = self._download_webpage(url, video_id)
|
||||||
|
info = self._search_json_ld(webpage, video_id,
|
||||||
|
expected_type='NewsArticle',
|
||||||
|
fatal=False)
|
||||||
|
# workaround as _json_ld does not recognize @graph nesting
|
||||||
|
if not info:
|
||||||
|
info['title'] = self._og_search_title(webpage)
|
||||||
|
embed_urls = list(Ant1NewsGrEmbedIE._extract_urls(webpage, url, **info))
|
||||||
|
if not embed_urls:
|
||||||
|
raise ExtractorError('no videos found for %s' % video_id)
|
||||||
|
if len(embed_urls) == 1:
|
||||||
|
return self.url_result(embed_urls[0], ie=Ant1NewsGrEmbedIE.ie_key(),
|
||||||
|
video_title=info['title'])
|
||||||
|
return self.playlist_from_matches(
|
||||||
|
embed_urls, video_id, info['title'], ie=Ant1NewsGrEmbedIE.ie_key())
|
||||||
|
|
||||||
|
|
||||||
|
class Ant1NewsGrEmbedIE(Ant1NewsGrBaseIE):
|
||||||
|
IE_NAME = 'ant1newsgr:embed'
|
||||||
|
IE_DESC = 'ant1news.gr embedded videos'
|
||||||
|
_VALID_URL = r'''(?x)https?://(?:[a-zA-Z0-9\-]+\.)?
|
||||||
|
(?:antenna|ant1news)\.gr/templates/pages/player
|
||||||
|
\?(?:(?:cid=(?P<id>[^&#]+)|[^&=#]+=[^&#]+)&?)+'''
|
||||||
|
_API_PATH = '/news/templates/data/jsonPlayer'
|
||||||
|
|
||||||
|
_TEST = {
|
||||||
|
'url': 'https://www.antenna.gr/templates/pages/player?cid=3f_li_c_az_jw_y_u=&w=670&h=377',
|
||||||
|
'md5': '12872b12af18b5dbf76528786728de8c',
|
||||||
|
'info_dict': {
|
||||||
|
'id': '3f_li_c_az_jw_y_u=',
|
||||||
|
'ext': 'mp4',
|
||||||
|
'title': 'md5:a30c93332455f53e1e84ae0724f0adf7',
|
||||||
|
},
|
||||||
|
}
|
||||||
|
|
||||||
|
@classmethod
|
||||||
|
def _extract_urls(cls, webpage, origin_url=None, **parent_info):
|
||||||
|
# make the scheme in _VALID_URL optional
|
||||||
|
_URL_RE = r'(?:https?:)?//' + cls._VALID_URL.split('://', 1)[1]
|
||||||
|
# simplify the query string part of _VALID_URL; after extracting iframe
|
||||||
|
# src, the URL will be matched again
|
||||||
|
_URL_RE = _URL_RE.split(r'\?', 1)[0] + r'\?(?:(?!(?P=_q1)).)+'
|
||||||
|
EMBED_RE = r'''(?x)
|
||||||
|
<iframe[^>]+?src=(?P<_q1>%(quot_re)s)(?P<url>%(url_re)s)(?P=_q1)
|
||||||
|
''' % {'quot_re': r'[\"\']', 'url_re': _URL_RE}
|
||||||
|
for mobj in re.finditer(EMBED_RE, webpage):
|
||||||
|
url = unescapeHTML(mobj.group('url'))
|
||||||
|
if url.startswith('//'):
|
||||||
|
scheme = compat_urllib_parse_urlparse(origin_url).scheme \
|
||||||
|
if origin_url else 'https'
|
||||||
|
url = '%s:%s' % (scheme, url)
|
||||||
|
if not cls.suitable(url):
|
||||||
|
continue
|
||||||
|
if isinstance(parent_info, dict):
|
||||||
|
url = cls._smuggle_parent_info(url, **parent_info)
|
||||||
|
yield url
|
||||||
|
|
||||||
|
def _real_extract(self, url):
|
||||||
|
url, parent_info = type(self)._unsmuggle_parent_info(url)
|
||||||
|
video_id = self._match_id(url)
|
||||||
|
|
||||||
|
# resolve any redirects, to derive the proper base URL for the API query
|
||||||
|
canonical_url = self._request_webpage(
|
||||||
|
HEADRequest(url), video_id,
|
||||||
|
note='Resolve canonical player URL',
|
||||||
|
errnote='Could not resolve canonical player URL').geturl()
|
||||||
|
scheme, netloc, _, _, query, _ = compat_urllib_parse_urlparse(
|
||||||
|
canonical_url)
|
||||||
|
query = compat_parse_qs(query)
|
||||||
|
cid = query['cid'][0]
|
||||||
|
|
||||||
|
info = self._download_and_extract_api_data(
|
||||||
|
video_id, netloc, cid, scheme=scheme)
|
||||||
|
if 'timestamp' not in info and 'timestamp' in parent_info:
|
||||||
|
info['timestamp'] = parent_info['timestamp']
|
||||||
|
return info
|
@ -1038,6 +1038,11 @@ from .rutube import (
|
|||||||
RutubePersonIE,
|
RutubePersonIE,
|
||||||
RutubePlaylistIE,
|
RutubePlaylistIE,
|
||||||
)
|
)
|
||||||
|
from .ant1newsgr import (
|
||||||
|
Ant1NewsGrWatchIE,
|
||||||
|
Ant1NewsGrArticleIE,
|
||||||
|
Ant1NewsGrEmbedIE,
|
||||||
|
)
|
||||||
from .rutv import RUTVIE
|
from .rutv import RUTVIE
|
||||||
from .ruutu import RuutuIE
|
from .ruutu import RuutuIE
|
||||||
from .ruv import RuvIE
|
from .ruv import RuvIE
|
||||||
|
@ -100,6 +100,7 @@ from .ustream import UstreamIE
|
|||||||
from .arte import ArteTVEmbedIE
|
from .arte import ArteTVEmbedIE
|
||||||
from .videopress import VideoPressIE
|
from .videopress import VideoPressIE
|
||||||
from .rutube import RutubeIE
|
from .rutube import RutubeIE
|
||||||
|
from .ant1newsgr import Ant1NewsGrEmbedIE
|
||||||
from .limelight import LimelightBaseIE
|
from .limelight import LimelightBaseIE
|
||||||
from .anvato import AnvatoIE
|
from .anvato import AnvatoIE
|
||||||
from .washingtonpost import WashingtonPostIE
|
from .washingtonpost import WashingtonPostIE
|
||||||
@ -3199,6 +3200,13 @@ class GenericIE(InfoExtractor):
|
|||||||
return self.playlist_from_matches(
|
return self.playlist_from_matches(
|
||||||
rutube_urls, video_id, video_title, ie=RutubeIE.ie_key())
|
rutube_urls, video_id, video_title, ie=RutubeIE.ie_key())
|
||||||
|
|
||||||
|
# Look for ant1news.gr embeds
|
||||||
|
ant1newsgr_urls = list(Ant1NewsGrEmbedIE._extract_urls(webpage, url,
|
||||||
|
title=video_title))
|
||||||
|
if ant1newsgr_urls:
|
||||||
|
return self.playlist_from_matches(
|
||||||
|
ant1newsgr_urls, video_id, video_title, ie=Ant1NewsGrEmbedIE.ie_key())
|
||||||
|
|
||||||
# Look for WashingtonPost embeds
|
# Look for WashingtonPost embeds
|
||||||
wapo_urls = WashingtonPostIE._extract_urls(webpage)
|
wapo_urls = WashingtonPostIE._extract_urls(webpage)
|
||||||
if wapo_urls:
|
if wapo_urls:
|
||||||
|
Loading…
Reference in New Issue
Block a user