mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-14 21:23:05 +00:00
[rudovideo] improvements to rudovideo extractor:
* implementing PR suggestions * removing c13cl extractor in favor of RudoVideo with _EMBED_REGEX * supporting VODs and Podcasts from rudo.video * supporting embeded youtube * adding tests
This commit is contained in:
parent
9df466edfc
commit
cfce954490
@ -278,7 +278,6 @@ from .businessinsider import BusinessInsiderIE
|
||||
from .bundesliga import BundesligaIE
|
||||
from .buzzfeed import BuzzFeedIE
|
||||
from .byutv import BYUtvIE
|
||||
from .c13cl import C13ClIE
|
||||
from .c56 import C56IE
|
||||
from .cableav import CableAVIE
|
||||
from .callin import CallinIE
|
||||
@ -1644,7 +1643,7 @@ from .rumble import (
|
||||
RumbleIE,
|
||||
RumbleChannelIE,
|
||||
)
|
||||
from .rudovideo import RudoVideoIE
|
||||
from .rudovideo import RudoVideoLiveIE
|
||||
from .rutube import (
|
||||
RutubeIE,
|
||||
RutubeChannelIE,
|
||||
|
@ -1,13 +0,0 @@
|
||||
from .common import InfoExtractor
|
||||
|
||||
|
||||
class C13ClIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://(?:www\.)?13\.cl/en-vivo'
|
||||
|
||||
def _real_extract(self, url):
|
||||
display_id = 'C13'
|
||||
webpage = self._download_webpage(url, display_id)
|
||||
|
||||
stream_url = self._search_regex(r'<div\s+.*id=(?:"|\')player(?:"|\')[^>]+><iframe\s+.*src=(?:"|\')([^?]+)', webpage, 'stream_url')
|
||||
|
||||
return self.url_result(stream_url, display_id=display_id, url_transparent=True)
|
@ -1,31 +1,116 @@
|
||||
from .common import InfoExtractor
|
||||
from ..utils import ExtractorError, traverse_obj
|
||||
from ..utils import ExtractorError, traverse_obj, js_to_json, update_url_query
|
||||
|
||||
|
||||
class RudoVideoIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://rudo\.video/live/(?P<id>[^/]+)'
|
||||
class RudoVideoLiveIE(InfoExtractor):
|
||||
_VALID_URL = r'https?://rudo\.video/(?P<type>live|vod|podcast)/(?P<id>[^/?]+)'
|
||||
_EMBED_REGEX = [r'<iframe[^>]+src=[\'"](?P<url>(?:https?:)//rudo\.video/(?:live|vod|podcast)/[^\'"]+)']
|
||||
_TESTS = [{
|
||||
'url': 'https://rudo.video/podcast/cz2wrUy8l0o',
|
||||
'md5': '28ed82b477708dc5e12e072da2449221',
|
||||
'info_dict': {
|
||||
'id': 'cz2wrUy8l0o',
|
||||
'title': 'Diego Cabot',
|
||||
'ext': 'mp4',
|
||||
'thumbnail': r're:^(?:https?:)?//.*\.(png|jpg)$',
|
||||
},
|
||||
}, {
|
||||
'url': 'https://rudo.video/podcast/bQkt07',
|
||||
'md5': '36b22a9863de0f47f00fc7532a32a898',
|
||||
'info_dict': {
|
||||
'id': 'bQkt07',
|
||||
'title': 'Tubular Bells',
|
||||
'ext': 'mp4',
|
||||
'thumbnail': r're:^(?:https?:)?//.*\.(png|jpg)$',
|
||||
},
|
||||
}, {
|
||||
'url': 'https://rudo.video/vod/bN5AaJ',
|
||||
'md5': '01324a329227e2591530ecb4f555c881',
|
||||
'info_dict': {
|
||||
'id': 'bN5AaJ',
|
||||
'title': 'Ucrania 19.03',
|
||||
'creator': 'La Tercera',
|
||||
'ext': 'mp4',
|
||||
'thumbnail': r're:^(?:https?:)?//.*\.(png|jpg)$',
|
||||
},
|
||||
}, {
|
||||
'url': 'https://rudo.video/live/bbtv',
|
||||
'info_dict': {
|
||||
'id': 'bbtv',
|
||||
'ext': 'mp4',
|
||||
'creator': 'BioBioTV',
|
||||
'live_status': 'is_live',
|
||||
'title': r're:^LIVE BBTV\s\d{4}-\d{2}-\d{2}\s\d{2}:\d{2}$',
|
||||
'thumbnail': r're:^(?:https?:)?//.*\.(png|jpg)$',
|
||||
},
|
||||
}, {
|
||||
'url': 'https://rudo.video/live/c13',
|
||||
'info_dict': {
|
||||
'id': 'c13',
|
||||
'title': 'CANAL13',
|
||||
'ext': 'mp4',
|
||||
},
|
||||
'skip': 'Geo-restricted to Chile',
|
||||
}, {
|
||||
'url': 'https://rudo.video/live/t13-13cl',
|
||||
'info_dict': {
|
||||
'id': 't13-13cl',
|
||||
'title': 'T13',
|
||||
'ext': 'mp4',
|
||||
},
|
||||
'skip': 'Geo-restricted to Chile',
|
||||
}]
|
||||
|
||||
def get_title(self, webpage):
|
||||
title = self._search_regex(r'var\s+titleVideo\s*=\s*[\'"]([^\'"]+)', webpage, 'title', default=None)
|
||||
if title is None:
|
||||
title = self._search_regex(r'<meta[^>]+property=[\'"]og:title[\'"]\s+content=[\'"]([^\'"]+)', webpage, 'title', fatal=False)
|
||||
return title
|
||||
|
||||
def get_thumbnail(self, webpage):
|
||||
thumbnail = self._search_regex(r'var\s+posterIMG\s*=\s*[\'"]([^?\'"]+)', webpage, 'thumbnail', default=None)
|
||||
if thumbnail is None:
|
||||
thumbnail = self._search_regex(r'<meta[^>]+property=[\'"]og:image[\'"]\s+content=[\'"]([^\'"]+)', webpage, 'thumbnail', default=None)
|
||||
return thumbnail
|
||||
|
||||
def _real_extract(self, url):
|
||||
video_id = self._match_id(url)
|
||||
type = self._match_valid_url(url).group('type')
|
||||
webpage = self._download_webpage(url, video_id)
|
||||
|
||||
if "Streaming is not available in your area." in webpage:
|
||||
if 'Streaming is not available in your area.' in webpage:
|
||||
self.raise_geo_restricted()
|
||||
|
||||
stream_url = self._search_regex(r'var\s+streamURL\s*=\s*\'([^?\']+)', webpage, "streamUrl")
|
||||
token_array_string = self._search_regex(r'<script>var\s+_\$_[a-zA-Z0-9]+\s*=\s*(\[[^]]+\])', webpage, 'token_array', default=None)
|
||||
if token_array_string:
|
||||
token_array_string = token_array_string.replace("x", "u00")
|
||||
token_array = self._parse_json(token_array_string, video_id)
|
||||
stream_url = self._search_regex(r'var\s+streamURL\s*=\s*[\'"]([^?\'"]+)', webpage, 'streamUrl', default=None)
|
||||
source_url = self._search_regex(r'<source[^>]+src=[\'"]([^\'"]+)', webpage, 'sourceUrl', default=None)
|
||||
youtube_url = self._search_regex(r'file:\s*[\'"]((?:https?:)//(?:www\.)?youtube.com[^\'"]+)', webpage, 'youtubeUrl', default=None)
|
||||
if stream_url is None:
|
||||
if source_url is not None:
|
||||
stream_url = source_url
|
||||
elif youtube_url is not None:
|
||||
return self.url_result(youtube_url, display_id=video_id)
|
||||
else:
|
||||
raise ExtractorError('Unable to extract stream url')
|
||||
|
||||
title = self.get_title(webpage)
|
||||
thumbnail = self.get_thumbnail(webpage)
|
||||
is_live = None
|
||||
if type == 'live':
|
||||
is_live = True
|
||||
|
||||
token_array = self._search_json(r'<script>var\s+_\$_[a-zA-Z0-9]+\s*=', webpage, 'access token array', video_id,
|
||||
contains_pattern=r'\[(?s:.+)\]', default=None, transform_source=js_to_json)
|
||||
if token_array:
|
||||
if len(token_array) != 9:
|
||||
raise ExtractorError('Couldnt get access token array', video_id=video_id)
|
||||
access_token_webpage = self._download_webpage(token_array[0], video_id)
|
||||
access_token = self._parse_json(access_token_webpage, video_id)
|
||||
if "data" not in access_token or token_array[3] not in access_token.get("data"):
|
||||
raise ExtractorError('Couldnt get access token', video_id=video_id)
|
||||
query_string = token_array[5] + traverse_obj(access_token, ("data", token_array[3]))
|
||||
stream_url = f'{stream_url}{query_string}'
|
||||
access_token = self._download_json(token_array[0], video_id, note='Downloading access token')
|
||||
stream_url = update_url_query(stream_url, {'auth-token': traverse_obj(access_token, ('data', 'authToken'))})
|
||||
|
||||
return self.url_result(
|
||||
stream_url,
|
||||
display_id=video_id, url_transparent=True)
|
||||
return {
|
||||
'id': video_id,
|
||||
'title': title,
|
||||
'formats': self._extract_m3u8_formats(stream_url, video_id, live=True),
|
||||
'is_live': is_live,
|
||||
'creator': self._search_regex(r'var\s+videoAuthor\s*=\s*[\'"]([^?\'"]+)', webpage, "videoAuthor", default=None),
|
||||
'thumbnail': thumbnail,
|
||||
}
|
||||
|
Loading…
Reference in New Issue
Block a user