mirror of
https://github.com/yt-dlp/yt-dlp.git
synced 2024-11-04 23:43:20 +00:00
89dbf08483
Authored by: C0D3D3V
184 lines
7.8 KiB
Python
184 lines
7.8 KiB
Python
import re
|
|
|
|
from .common import InfoExtractor
|
|
from ..utils import (
|
|
determine_ext,
|
|
ExtractorError,
|
|
int_or_none,
|
|
parse_iso8601,
|
|
traverse_obj,
|
|
variadic,
|
|
)
|
|
|
|
|
|
class OpencastBaseIE(InfoExtractor):
|
|
_INSTANCES_RE = r'''(?:
|
|
opencast\.informatik\.kit\.edu|
|
|
electures\.uni-muenster\.de|
|
|
oc-presentation\.ltcc\.tuwien\.ac\.at|
|
|
medien\.ph-noe\.ac\.at|
|
|
oc-video\.ruhr-uni-bochum\.de|
|
|
oc-video1\.ruhr-uni-bochum\.de|
|
|
opencast\.informatik\.uni-goettingen\.de|
|
|
heicast\.uni-heidelberg\.de|
|
|
opencast\.hawk\.de:8080|
|
|
opencast\.hs-osnabrueck\.de|
|
|
video[0-9]+\.virtuos\.uni-osnabrueck\.de|
|
|
opencast\.uni-koeln\.de|
|
|
media\.opencast\.hochschule-rhein-waal\.de|
|
|
matterhorn\.dce\.harvard\.edu|
|
|
hs-harz\.opencast\.uni-halle\.de|
|
|
videocampus\.urz\.uni-leipzig\.de|
|
|
media\.uct\.ac\.za|
|
|
vid\.igb\.illinois\.edu|
|
|
cursosabertos\.c3sl\.ufpr\.br|
|
|
mcmedia\.missioncollege\.org|
|
|
clases\.odon\.edu\.uy
|
|
)'''
|
|
_UUID_RE = r'[\da-fA-F]{8}-[\da-fA-F]{4}-[\da-fA-F]{4}-[\da-fA-F]{4}-[\da-fA-F]{12}'
|
|
|
|
def _call_api(self, host, video_id, **kwargs):
|
|
return self._download_json(self._API_BASE % (host, video_id), video_id, **kwargs)
|
|
|
|
def _parse_mediapackage(self, video):
|
|
video_id = video.get('id')
|
|
if video_id is None:
|
|
raise ExtractorError('Video id was not found')
|
|
|
|
formats = []
|
|
for track in variadic(traverse_obj(video, ('media', 'track')) or []):
|
|
href = track.get('url')
|
|
if href is None:
|
|
continue
|
|
ext = determine_ext(href, None)
|
|
|
|
transport = track.get('transport')
|
|
|
|
if transport == 'DASH' or ext == 'mpd':
|
|
formats.extend(self._extract_mpd_formats(href, video_id, mpd_id='dash', fatal=False))
|
|
elif transport == 'HLS' or ext == 'm3u8':
|
|
formats.extend(self._extract_m3u8_formats(
|
|
href, video_id, m3u8_id='hls', entry_protocol='m3u8_native', fatal=False))
|
|
elif transport == 'HDS' or ext == 'f4m':
|
|
formats.extend(self._extract_f4m_formats(href, video_id, f4m_id='hds', fatal=False))
|
|
elif transport == 'SMOOTH':
|
|
formats.extend(self._extract_ism_formats(href, video_id, ism_id='smooth', fatal=False))
|
|
elif ext == 'smil':
|
|
formats.extend(self._extract_smil_formats(href, video_id, fatal=False))
|
|
else:
|
|
track_obj = {
|
|
'url': href,
|
|
'ext': ext,
|
|
'format_note': track.get('transport'),
|
|
'resolution': traverse_obj(track, ('video', 'resolution')),
|
|
'fps': int_or_none(traverse_obj(track, ('video', 'framerate'))),
|
|
'vbr': int_or_none(traverse_obj(track, ('video', 'bitrate')), scale=1000),
|
|
'vcodec': traverse_obj(track, ('video', 'encoder', 'type')) if track.get('video') else 'none',
|
|
'abr': int_or_none(traverse_obj(track, ('audio', 'bitrate')), scale=1000),
|
|
'asr': int_or_none(traverse_obj(track, ('audio', 'samplingrate'))),
|
|
'acodec': traverse_obj(track, ('audio', 'encoder', 'type')) if track.get('audio') else 'none',
|
|
}
|
|
|
|
if transport == 'RTMP':
|
|
m_obj = re.search(r'(?:rtmp://[^/]+/(?P<app>[^/]+))/(?P<ext>.+):(?P<playpath>.+)', href)
|
|
if not m_obj:
|
|
continue
|
|
track_obj.update({
|
|
'app': m_obj.group('app'),
|
|
'ext': m_obj.group('ext'),
|
|
'play_path': m_obj.group('ext') + ':' + m_obj.group('playpath'),
|
|
'rtmp_live': True,
|
|
'preference': -2,
|
|
})
|
|
formats.append(track_obj)
|
|
|
|
return {
|
|
'id': video_id,
|
|
'formats': formats,
|
|
'title': video.get('title'),
|
|
'series': video.get('seriestitle'),
|
|
'season_id': video.get('series'),
|
|
'creator': traverse_obj(video, ('creators', 'creator')),
|
|
'timestamp': parse_iso8601(video.get('start')),
|
|
'thumbnail': traverse_obj(video, ('attachments', 'attachment', ..., 'url'), get_all=False),
|
|
}
|
|
|
|
|
|
class OpencastIE(OpencastBaseIE):
|
|
_VALID_URL = rf'''(?x)
|
|
https?://(?P<host>{OpencastBaseIE._INSTANCES_RE})/paella/ui/watch\.html\?
|
|
(?:[^#]+&)?id=(?P<id>{OpencastBaseIE._UUID_RE})'''
|
|
|
|
_API_BASE = 'https://%s/search/episode.json?id=%s'
|
|
|
|
_TESTS = [
|
|
{
|
|
'url': 'https://oc-video1.ruhr-uni-bochum.de/paella/ui/watch.html?id=ed063cd5-72c8-46b5-a60a-569243edcea8',
|
|
'md5': '554c8e99a90f7be7e874619fcf2a3bc9',
|
|
'info_dict': {
|
|
'id': 'ed063cd5-72c8-46b5-a60a-569243edcea8',
|
|
'ext': 'mp4',
|
|
'title': '11 - Kryptographie - 24.11.2015',
|
|
'thumbnail': r're:^https?://.*\.jpg$',
|
|
'timestamp': 1606208400,
|
|
'upload_date': '20201124',
|
|
'season_id': 'cf68a4a1-36b1-4a53-a6ba-61af5705a0d0',
|
|
'series': 'Kryptographie - WiSe 15/16',
|
|
'creator': 'Alexander May',
|
|
},
|
|
}
|
|
]
|
|
|
|
def _real_extract(self, url):
|
|
host, video_id = self._match_valid_url(url).group('host', 'id')
|
|
return self._parse_mediapackage(
|
|
self._call_api(host, video_id)['search-results']['result']['mediapackage'])
|
|
|
|
|
|
class OpencastPlaylistIE(OpencastBaseIE):
|
|
_VALID_URL = rf'''(?x)
|
|
https?://(?P<host>{OpencastBaseIE._INSTANCES_RE})(?:
|
|
/engage/ui/index\.html\?(?:[^#]+&)?epFrom=|
|
|
/ltitools/index\.html\?(?:[^#]+&)?series=
|
|
)(?P<id>{OpencastBaseIE._UUID_RE})'''
|
|
|
|
_API_BASE = 'https://%s/search/episode.json?sid=%s'
|
|
|
|
_TESTS = [
|
|
{
|
|
'url': 'https://oc-video1.ruhr-uni-bochum.de/engage/ui/index.html?epFrom=cf68a4a1-36b1-4a53-a6ba-61af5705a0d0',
|
|
'info_dict': {
|
|
'id': 'cf68a4a1-36b1-4a53-a6ba-61af5705a0d0',
|
|
'title': 'Kryptographie - WiSe 15/16',
|
|
},
|
|
'playlist_mincount': 29,
|
|
},
|
|
{
|
|
'url': 'https://oc-video1.ruhr-uni-bochum.de/ltitools/index.html?subtool=series&series=cf68a4a1-36b1-4a53-a6ba-61af5705a0d0&lng=de',
|
|
'info_dict': {
|
|
'id': 'cf68a4a1-36b1-4a53-a6ba-61af5705a0d0',
|
|
'title': 'Kryptographie - WiSe 15/16',
|
|
},
|
|
'playlist_mincount': 29,
|
|
},
|
|
{
|
|
'url': 'https://electures.uni-muenster.de/engage/ui/index.html?e=1&p=1&epFrom=39391d10-a711-4d23-b21d-afd2ed7d758c',
|
|
'info_dict': {
|
|
'id': '39391d10-a711-4d23-b21d-afd2ed7d758c',
|
|
'title': '021670 Theologische Themen bei Hans Blumenberg WiSe 2017/18',
|
|
},
|
|
'playlist_mincount': 13,
|
|
},
|
|
]
|
|
|
|
def _real_extract(self, url):
|
|
host, video_id = self._match_valid_url(url).group('host', 'id')
|
|
|
|
entries = [
|
|
self._parse_mediapackage(episode['mediapackage'])
|
|
for episode in variadic(self._call_api(host, video_id)['search-results']['result'])
|
|
if episode.get('mediapackage')
|
|
]
|
|
|
|
return self.playlist_result(entries, video_id, traverse_obj(entries, (0, 'series')))
|