[yandexmusic:playlist][yandexmusic:privatetrack] add download private track
This commit is contained in:
parent
a803582717
commit
35c656d3b9
@ -1574,6 +1574,7 @@ from .yahoo import (
|
||||
from .yandexdisk import YandexDiskIE
|
||||
from .yandexmusic import (
|
||||
YandexMusicTrackIE,
|
||||
YandexMusicPrivateTrackIE,
|
||||
YandexMusicAlbumIE,
|
||||
YandexMusicPlaylistIE,
|
||||
YandexMusicArtistTracksIE,
|
||||
|
@ -196,42 +196,127 @@ class YandexMusicTrackIE(YandexMusicBaseIE):
|
||||
return track_info
|
||||
|
||||
|
||||
class YandexMusicPrivateTrackIE(YandexMusicBaseIE):
|
||||
IE_NAME = 'yandexmusic:privatetrack'
|
||||
IE_DESC = 'Яндекс.Музыка - Трек'
|
||||
_VALID_URL = r'%s/track/(?P<id>[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12})' % YandexMusicBaseIE._VALID_URL_BASE
|
||||
|
||||
def _real_extract(self, url):
|
||||
mobj = re.match(self._VALID_URL, url)
|
||||
tld, track_id = mobj.group('tld'), mobj.group('id')
|
||||
|
||||
track = self._call_api(
|
||||
'track', tld, url, track_id, 'Downloading track JSON',
|
||||
{'track': '%s' % (track_id)})['track']
|
||||
track_title = track['title']
|
||||
|
||||
download_data = self._download_json(
|
||||
'https://music.yandex.ru/api/v2.1/handlers/track/%s/web-album_track-track-track-main/download/m' % (track_id),
|
||||
track_id, 'Downloading track location url JSON',
|
||||
headers={'X-Retpath-Y': url})
|
||||
|
||||
fd_data = self._download_json(
|
||||
download_data['src'], track_id,
|
||||
'Downloading track location JSON',
|
||||
query={'format': 'json'})
|
||||
key = hashlib.md5(('XGRlBW9FXlekgbPrRHuSiA' + fd_data['path'][1:] + fd_data['s']).encode('utf-8')).hexdigest()
|
||||
f_url = 'http://%s/get-mp3/%s/%s?track-id=%s ' % (fd_data['host'], key, fd_data['ts'] + fd_data['path'], track['id'])
|
||||
|
||||
thumbnail = None
|
||||
cover_uri = track.get('coverUri')
|
||||
if cover_uri:
|
||||
thumbnail = cover_uri.replace('%%', 'orig')
|
||||
if not thumbnail.startswith('http'):
|
||||
thumbnail = 'http://' + thumbnail
|
||||
|
||||
track_info = {
|
||||
'id': track_id,
|
||||
'ext': 'mp3',
|
||||
'url': f_url,
|
||||
'filesize': int_or_none(track.get('fileSize')),
|
||||
'duration': float_or_none(track.get('durationMs'), 1000),
|
||||
'thumbnail': thumbnail,
|
||||
'track': track_title,
|
||||
'acodec': download_data.get('codec'),
|
||||
'abr': int_or_none(download_data.get('bitrate')),
|
||||
}
|
||||
|
||||
def extract_artist_name(artist):
|
||||
decomposed = artist.get('decomposed')
|
||||
if not isinstance(decomposed, list):
|
||||
return artist['name']
|
||||
parts = [artist['name']]
|
||||
for element in decomposed:
|
||||
if isinstance(element, dict) and element.get('name'):
|
||||
parts.append(element['name'])
|
||||
elif isinstance(element, compat_str):
|
||||
parts.append(element)
|
||||
return ''.join(parts)
|
||||
|
||||
def extract_artist(artist_list):
|
||||
if artist_list and isinstance(artist_list, list):
|
||||
artists_names = [extract_artist_name(a) for a in artist_list if a.get('name')]
|
||||
if artists_names:
|
||||
return ', '.join(artists_names)
|
||||
|
||||
albums = track.get('albums')
|
||||
if albums and isinstance(albums, list):
|
||||
album = albums[0]
|
||||
if isinstance(album, dict):
|
||||
year = album.get('year')
|
||||
disc_number = int_or_none(try_get(
|
||||
album, lambda x: x['trackPosition']['volume']))
|
||||
track_number = int_or_none(try_get(
|
||||
album, lambda x: x['trackPosition']['index']))
|
||||
track_info.update({
|
||||
'album': album.get('title'),
|
||||
'album_artist': extract_artist(album.get('artists')),
|
||||
'release_year': int_or_none(year),
|
||||
'genre': album.get('genre'),
|
||||
'disc_number': disc_number,
|
||||
'track_number': track_number,
|
||||
})
|
||||
|
||||
track_artist = extract_artist(track.get('artists'))
|
||||
if track_artist:
|
||||
track_info.update({
|
||||
'artist': track_artist,
|
||||
'title': '%s - %s' % (track_artist, track_title),
|
||||
})
|
||||
else:
|
||||
track_info['title'] = track_title
|
||||
|
||||
return track_info
|
||||
|
||||
|
||||
class YandexMusicPlaylistBaseIE(YandexMusicBaseIE):
|
||||
def _extract_tracks(self, source, item_id, url, tld):
|
||||
tracks = source['tracks']
|
||||
tracks = []
|
||||
track_ids = [compat_str(track_id) for track_id in source['trackIds']]
|
||||
|
||||
def create_chunks(lst, n):
|
||||
"""Yield successive n-sized chunks from lst."""
|
||||
for i in range(0, len(lst), n):
|
||||
yield lst[i:i + n]
|
||||
# tracks dictionary shipped with playlist.jsx API is limited to 150 tracks,
|
||||
# missing tracks should be retrieved manually.
|
||||
if len(tracks) < len(track_ids):
|
||||
present_track_ids = set([
|
||||
compat_str(track['id'])
|
||||
for track in tracks if track.get('id')])
|
||||
missing_track_ids = [
|
||||
track_id for track_id in track_ids
|
||||
if track_id not in present_track_ids]
|
||||
# Request missing tracks in chunks to avoid exceeding max HTTP header size,
|
||||
# see https://github.com/ytdl-org/youtube-dl/issues/27355
|
||||
_TRACKS_PER_CHUNK = 250
|
||||
for chunk_num in itertools.count(0):
|
||||
start = chunk_num * _TRACKS_PER_CHUNK
|
||||
end = start + _TRACKS_PER_CHUNK
|
||||
missing_track_ids_req = missing_track_ids[start:end]
|
||||
assert missing_track_ids_req
|
||||
missing_tracks = self._call_api(
|
||||
'track-entries', tld, url, item_id,
|
||||
'Downloading missing tracks JSON chunk %d' % (chunk_num + 1), {
|
||||
'entries': ','.join(missing_track_ids_req),
|
||||
'lang': tld,
|
||||
'external-domain': 'music.yandex.%s' % tld,
|
||||
'overembed': 'false',
|
||||
'strict': 'true',
|
||||
})
|
||||
if missing_tracks:
|
||||
tracks.extend(missing_tracks)
|
||||
if end >= len(missing_track_ids):
|
||||
break
|
||||
|
||||
_TRACKS_PER_CHUNK = 100
|
||||
# Request missing tracks in chunks to avoid exceeding max HTTP header size,
|
||||
# see https://github.com/ytdl-org/youtube-dl/issues/27355
|
||||
chunks = create_chunks(track_ids, _TRACKS_PER_CHUNK)
|
||||
for chunk_num, chunk in enumerate(chunks):
|
||||
missing_tracks = self._call_api(
|
||||
'track-entries', tld, url, item_id,
|
||||
'Downloading tracks JSON chunk %d and size %d' % (chunk_num + 1, len(chunk)), {
|
||||
'entries': ','.join(chunk),
|
||||
'lang': tld,
|
||||
'external-domain': 'music.yandex.%s' % tld,
|
||||
'experiments':'{"userFeed":"old","similarities":"default","genreRadio":"new-ichwill-matrixnet6","recommendedArtists":"ichwill_similar_artists","recommendedTracks":"recommended_tracks_by_artist_from_history","recommendedAlbumsOfFavoriteGenre":"recent","recommendedSimilarArtists":"default","recommendedArtistsWithArtistsFromHistory":"force_recent","adv":"a","loserArtistsWithArtists":"off","ny2015":"no"}',
|
||||
'overembed': 'false',
|
||||
'strict': 'true',
|
||||
})
|
||||
if missing_tracks:
|
||||
tracks.extend(missing_tracks)
|
||||
|
||||
return tracks
|
||||
|
||||
def _build_playlist(self, tracks):
|
||||
@ -240,6 +325,11 @@ class YandexMusicPlaylistBaseIE(YandexMusicBaseIE):
|
||||
track_id = track.get('id') or track.get('realId')
|
||||
if not track_id:
|
||||
continue
|
||||
if re.match(r"[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}", track_id):
|
||||
entries.append(self.url_result(
|
||||
'http://music.yandex.ru/track/%s' % (track_id),
|
||||
ie=YandexMusicTrackIE.ie_key(), video_id=track_id))
|
||||
continue
|
||||
albums = track.get('albums')
|
||||
if not albums or not isinstance(albums, list):
|
||||
continue
|
||||
|
Loading…
Reference in New Issue
Block a user