mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-30 04:10:23 +00:00
4b5eec0aaa
Some checks failed
CodeQL / Analyze (python) (push) Has been cancelled
Download Tests / Quick Download Tests (push) Has been cancelled
Download Tests / Full Download Tests (ubuntu-latest, 3.10) (push) Has been cancelled
Download Tests / Full Download Tests (ubuntu-latest, 3.11) (push) Has been cancelled
Download Tests / Full Download Tests (ubuntu-latest, 3.12) (push) Has been cancelled
Download Tests / Full Download Tests (ubuntu-latest, 3.13) (push) Has been cancelled
Download Tests / Full Download Tests (ubuntu-latest, pypy-3.10) (push) Has been cancelled
Download Tests / Full Download Tests (windows-latest, 3.9) (push) Has been cancelled
Download Tests / Full Download Tests (windows-latest, pypy-3.10) (push) Has been cancelled
Quick Test / Core Test (push) Has been cancelled
Quick Test / Code check (push) Has been cancelled
Release (master) / release (push) Has been cancelled
Release (master) / publish_pypi (push) Has been cancelled
Fix bug in 720b3dc453
Closes #11623
Authored by: jkruse
154 lines
5.7 KiB
Python
154 lines
5.7 KiB
Python
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
ExtractorError,
|
|
lowercase_escape,
|
|
url_or_none,
|
|
urlencode_postdata,
|
|
)
|
|
|
|
|
|
class ChaturbateIE(InfoExtractor):
|
|
_VALID_URL = r'https?://(?:[^/]+\.)?chaturbate\.(?P<tld>com|eu|global)/(?:fullvideo/?\?.*?\bb=)?(?P<id>[^/?&#]+)'
|
|
_TESTS = [{
|
|
'url': 'https://www.chaturbate.com/siswet19/',
|
|
'info_dict': {
|
|
'id': 'siswet19',
|
|
'ext': 'mp4',
|
|
'title': 're:^siswet19 [0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}$',
|
|
'age_limit': 18,
|
|
'is_live': True,
|
|
},
|
|
'params': {
|
|
'skip_download': True,
|
|
},
|
|
'skip': 'Room is offline',
|
|
}, {
|
|
'url': 'https://chaturbate.com/fullvideo/?b=caylin',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://en.chaturbate.com/siswet19/',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://chaturbate.eu/siswet19/',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://chaturbate.eu/fullvideo/?b=caylin',
|
|
'only_matching': True,
|
|
}, {
|
|
'url': 'https://chaturbate.global/siswet19/',
|
|
'only_matching': True,
|
|
}]
|
|
|
|
_ERROR_MAP = {
|
|
'offline': 'Room is currently offline',
|
|
'private': 'Room is currently in a private show',
|
|
'away': 'Performer is currently away',
|
|
'password protected': 'Room is password protected',
|
|
'hidden': 'Hidden session in progress',
|
|
}
|
|
|
|
def _extract_from_api(self, video_id, tld):
|
|
response = self._download_json(
|
|
f'https://chaturbate.{tld}/get_edge_hls_url_ajax/', video_id,
|
|
data=urlencode_postdata({'room_slug': video_id}),
|
|
headers={
|
|
**self.geo_verification_headers(),
|
|
'X-Requested-With': 'XMLHttpRequest',
|
|
'Accept': 'application/json',
|
|
}, fatal=False, impersonate=True) or {}
|
|
|
|
m3u8_url = response.get('url')
|
|
if not m3u8_url:
|
|
status = response.get('room_status')
|
|
if error := self._ERROR_MAP.get(status):
|
|
raise ExtractorError(error, expected=True)
|
|
if status == 'public':
|
|
self.raise_geo_restricted()
|
|
self.report_warning(f'Got status "{status}" from API; falling back to webpage extraction')
|
|
return None
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': video_id,
|
|
'thumbnail': f'https://roomimg.stream.highwebmedia.com/ri/{video_id}.jpg',
|
|
'is_live': True,
|
|
'age_limit': 18,
|
|
'formats': self._extract_m3u8_formats(m3u8_url, video_id, ext='mp4', live=True),
|
|
}
|
|
|
|
def _extract_from_html(self, video_id, tld):
|
|
webpage = self._download_webpage(
|
|
f'https://chaturbate.{tld}/{video_id}/', video_id,
|
|
headers=self.geo_verification_headers(), impersonate=True)
|
|
|
|
found_m3u8_urls = []
|
|
|
|
data = self._parse_json(
|
|
self._search_regex(
|
|
r'initialRoomDossier\s*=\s*(["\'])(?P<value>(?:(?!\1).)+)\1',
|
|
webpage, 'data', default='{}', group='value'),
|
|
video_id, transform_source=lowercase_escape, fatal=False)
|
|
if data:
|
|
m3u8_url = url_or_none(data.get('hls_source'))
|
|
if m3u8_url:
|
|
found_m3u8_urls.append(m3u8_url)
|
|
|
|
if not found_m3u8_urls:
|
|
for m in re.finditer(
|
|
r'(\\u002[27])(?P<url>http.+?\.m3u8.*?)\1', webpage):
|
|
found_m3u8_urls.append(lowercase_escape(m.group('url')))
|
|
|
|
if not found_m3u8_urls:
|
|
for m in re.finditer(
|
|
r'(["\'])(?P<url>http.+?\.m3u8.*?)\1', webpage):
|
|
found_m3u8_urls.append(m.group('url'))
|
|
|
|
m3u8_urls = []
|
|
for found_m3u8_url in found_m3u8_urls:
|
|
m3u8_fast_url, m3u8_no_fast_url = found_m3u8_url, found_m3u8_url.replace('_fast', '')
|
|
for m3u8_url in (m3u8_fast_url, m3u8_no_fast_url):
|
|
if m3u8_url not in m3u8_urls:
|
|
m3u8_urls.append(m3u8_url)
|
|
|
|
if not m3u8_urls:
|
|
error = self._search_regex(
|
|
[r'<span[^>]+class=(["\'])desc_span\1[^>]*>(?P<error>[^<]+)</span>',
|
|
r'<div[^>]+id=(["\'])defchat\1[^>]*>\s*<p><strong>(?P<error>[^<]+)<'],
|
|
webpage, 'error', group='error', default=None)
|
|
if not error:
|
|
if any(p in webpage for p in (
|
|
self._ERROR_MAP['offline'], 'offline_tipping', 'tip_offline')):
|
|
error = self._ERROR_MAP['offline']
|
|
if error:
|
|
raise ExtractorError(error, expected=True)
|
|
raise ExtractorError('Unable to find stream URL')
|
|
|
|
formats = []
|
|
for m3u8_url in m3u8_urls:
|
|
for known_id in ('fast', 'slow'):
|
|
if f'_{known_id}' in m3u8_url:
|
|
m3u8_id = known_id
|
|
break
|
|
else:
|
|
m3u8_id = None
|
|
formats.extend(self._extract_m3u8_formats(
|
|
m3u8_url, video_id, ext='mp4',
|
|
# ffmpeg skips segments for fast m3u8
|
|
preference=-10 if m3u8_id == 'fast' else None,
|
|
m3u8_id=m3u8_id, fatal=False, live=True))
|
|
|
|
return {
|
|
'id': video_id,
|
|
'title': video_id,
|
|
'thumbnail': f'https://roomimg.stream.highwebmedia.com/ri/{video_id}.jpg',
|
|
'age_limit': self._rta_search(webpage),
|
|
'is_live': True,
|
|
'formats': formats,
|
|
}
|
|
|
|
def _real_extract(self, url):
|
|
video_id, tld = self._match_valid_url(url).group('id', 'tld')
|
|
return self._extract_from_api(video_id, tld) or self._extract_from_html(video_id, tld)
|