mirror of
https://github.com/Alexander-D-Karpov/akarpov
synced 2024-11-27 15:53:44 +03:00
Compare commits
3 Commits
0189377aeb
...
a309d5653d
Author | SHA1 | Date | |
---|---|---|---|
a309d5653d | |||
c81b387689 | |||
db72084d64 |
|
@ -63,7 +63,7 @@ def filter(self, queryset):
|
||||||
|
|
||||||
if search_type in search_classes:
|
if search_type in search_classes:
|
||||||
search_instance = search_classes[search_type](
|
search_instance = search_classes[search_type](
|
||||||
queryset=File.objects.filter(user=self.request.user)
|
queryset=File.objects.filter(user=self.request.user).nocache()
|
||||||
)
|
)
|
||||||
queryset = search_instance.search(query)
|
queryset = search_instance.search(query)
|
||||||
return queryset
|
return queryset
|
||||||
|
|
|
@ -1,9 +1,11 @@
|
||||||
import os
|
import os
|
||||||
import re
|
import re
|
||||||
|
|
||||||
|
import requests
|
||||||
from deep_translator import GoogleTranslator
|
from deep_translator import GoogleTranslator
|
||||||
from django.core.files import File
|
from django.core.files import File
|
||||||
from django.db import transaction
|
from django.db import transaction
|
||||||
|
from django.db.models import Min
|
||||||
from django.utils.text import slugify
|
from django.utils.text import slugify
|
||||||
from mutagen import File as MutagenFile
|
from mutagen import File as MutagenFile
|
||||||
from mutagen.id3 import APIC, ID3, TCON, TORY, TextFrame
|
from mutagen.id3 import APIC, ID3, TCON, TORY, TextFrame
|
||||||
|
@ -16,6 +18,32 @@
|
||||||
from akarpov.users.models import User
|
from akarpov.users.models import User
|
||||||
|
|
||||||
|
|
||||||
|
def get_or_create_author(author_name):
|
||||||
|
retry = True
|
||||||
|
while retry:
|
||||||
|
retry = False
|
||||||
|
try:
|
||||||
|
with transaction.atomic():
|
||||||
|
author, created = Author.objects.get_or_create(
|
||||||
|
name__iexact=author_name, defaults={"name": author_name}
|
||||||
|
)
|
||||||
|
return author
|
||||||
|
except Author.MultipleObjectsReturned:
|
||||||
|
with transaction.atomic():
|
||||||
|
# If multiple authors are found, get the first one and delete the rest
|
||||||
|
min_id = Author.objects.filter(name__iexact=author_name).aggregate(
|
||||||
|
Min("id")
|
||||||
|
)["id__min"]
|
||||||
|
author = Author.objects.get(id=min_id)
|
||||||
|
Author.objects.filter(name__iexact=author_name).exclude(
|
||||||
|
id=min_id
|
||||||
|
).delete()
|
||||||
|
return author
|
||||||
|
except Exception as e:
|
||||||
|
if "could not serialize access due to concurrent update" in str(e):
|
||||||
|
retry = True
|
||||||
|
|
||||||
|
|
||||||
def process_track_name(track_name: str) -> str:
|
def process_track_name(track_name: str) -> str:
|
||||||
# Split the track name by dash and parentheses
|
# Split the track name by dash and parentheses
|
||||||
parts = track_name.split(" - ")
|
parts = track_name.split(" - ")
|
||||||
|
@ -78,23 +106,12 @@ def load_track(
|
||||||
if album and type(album) is str and album.startswith("['"):
|
if album and type(album) is str and album.startswith("['"):
|
||||||
album = album.replace("['", "").replace("']", "")
|
album = album.replace("['", "").replace("']", "")
|
||||||
|
|
||||||
re_authors = []
|
processed_authors = []
|
||||||
if authors:
|
if authors:
|
||||||
for x in authors:
|
for author_name in authors:
|
||||||
while True:
|
author = get_or_create_author(author_name)
|
||||||
try:
|
processed_authors.append(author)
|
||||||
with transaction.atomic():
|
authors = processed_authors
|
||||||
author, created = Author.objects.get_or_create(
|
|
||||||
name__iexact=x, defaults={"name": x}
|
|
||||||
)
|
|
||||||
re_authors.append(author)
|
|
||||||
break
|
|
||||||
except Author.MultipleObjectsReturned:
|
|
||||||
# If multiple authors are found, delete all but one
|
|
||||||
Author.objects.filter(name__iexact=x).exclude(
|
|
||||||
id=Author.objects.filter(name__iexact=x).first().id
|
|
||||||
).delete()
|
|
||||||
authors = re_authors
|
|
||||||
|
|
||||||
if album:
|
if album:
|
||||||
if type(album) is str:
|
if type(album) is str:
|
||||||
|
@ -122,6 +139,14 @@ def load_track(
|
||||||
path = mp3_path
|
path = mp3_path
|
||||||
|
|
||||||
tag = MP3(path, ID3=ID3)
|
tag = MP3(path, ID3=ID3)
|
||||||
|
|
||||||
|
if image_path and image_path.startswith("http"):
|
||||||
|
response = requests.get(image_path)
|
||||||
|
se = image_path.split("/")[-1]
|
||||||
|
image_path = f'/tmp/{generate_readable_slug(name, Song)}.{"png" if "." not in se else se.split(".")[-1]}'
|
||||||
|
with open(image_path, "wb") as f:
|
||||||
|
f.write(response.content)
|
||||||
|
|
||||||
if image_path:
|
if image_path:
|
||||||
if not image_path.endswith(".png"):
|
if not image_path.endswith(".png"):
|
||||||
nm = image_path
|
nm = image_path
|
||||||
|
|
|
@ -1,7 +1,10 @@
|
||||||
import spotipy
|
import spotipy
|
||||||
from django.conf import settings
|
from django.conf import settings
|
||||||
|
from spotdl import Song, Spotdl
|
||||||
from spotipy.oauth2 import SpotifyClientCredentials
|
from spotipy.oauth2 import SpotifyClientCredentials
|
||||||
|
|
||||||
|
from akarpov.music.services.db import load_track
|
||||||
|
|
||||||
|
|
||||||
def create_session() -> spotipy.Spotify:
|
def create_session() -> spotipy.Spotify:
|
||||||
if not settings.MUSIC_SPOTIFY_ID or not settings.MUSIC_SPOTIFY_SECRET:
|
if not settings.MUSIC_SPOTIFY_ID or not settings.MUSIC_SPOTIFY_SECRET:
|
||||||
|
@ -18,3 +21,61 @@ def create_session() -> spotipy.Spotify:
|
||||||
def search(name: str, session: spotipy.Spotify, search_type="track"):
|
def search(name: str, session: spotipy.Spotify, search_type="track"):
|
||||||
res = session.search(name, type=search_type)
|
res = session.search(name, type=search_type)
|
||||||
return res
|
return res
|
||||||
|
|
||||||
|
|
||||||
|
def download_url(url, user_id=None):
|
||||||
|
spot_settings = {
|
||||||
|
"simple_tui": True,
|
||||||
|
"log_level": "ERROR",
|
||||||
|
"lyrics_providers": ["genius", "musixmatch"],
|
||||||
|
}
|
||||||
|
|
||||||
|
spotdl_client = Spotdl(
|
||||||
|
client_id=settings.MUSIC_SPOTIFY_ID,
|
||||||
|
client_secret=settings.MUSIC_SPOTIFY_SECRET,
|
||||||
|
user_auth=False,
|
||||||
|
headless=False,
|
||||||
|
downloader_settings=spot_settings,
|
||||||
|
)
|
||||||
|
|
||||||
|
session = create_session()
|
||||||
|
|
||||||
|
if "track" in url:
|
||||||
|
songs = [Song.from_url(url)]
|
||||||
|
elif "album" in url:
|
||||||
|
album_tracks = session.album(url)["tracks"]["items"]
|
||||||
|
songs = [
|
||||||
|
Song.from_url(track["external_urls"]["spotify"]) for track in album_tracks
|
||||||
|
]
|
||||||
|
elif "artist" in url:
|
||||||
|
artist_top_tracks = session.artist_top_tracks(url)["tracks"]
|
||||||
|
songs = [
|
||||||
|
Song.from_url(track["external_urls"]["spotify"])
|
||||||
|
for track in artist_top_tracks
|
||||||
|
]
|
||||||
|
elif "playlist" in url:
|
||||||
|
playlist_tracks = session.playlist_items(url)["items"]
|
||||||
|
songs = [
|
||||||
|
Song.from_url(track["track"]["external_urls"]["spotify"])
|
||||||
|
for track in playlist_tracks
|
||||||
|
]
|
||||||
|
else:
|
||||||
|
return None
|
||||||
|
|
||||||
|
for song in songs:
|
||||||
|
res = spotdl_client.download(song)
|
||||||
|
if res:
|
||||||
|
song, path = res
|
||||||
|
else:
|
||||||
|
return None
|
||||||
|
load_track(
|
||||||
|
path=str(path),
|
||||||
|
image_path=song.cover_url,
|
||||||
|
user_id=user_id,
|
||||||
|
authors=song.artists,
|
||||||
|
album=song.album_name,
|
||||||
|
name=song.name,
|
||||||
|
link=song.url,
|
||||||
|
genre=song.genres[0] if song.genres else None,
|
||||||
|
release=song.date,
|
||||||
|
)
|
||||||
|
|
|
@ -19,7 +19,7 @@
|
||||||
UserListenHistory,
|
UserListenHistory,
|
||||||
UserMusicProfile,
|
UserMusicProfile,
|
||||||
)
|
)
|
||||||
from akarpov.music.services import yandex, youtube
|
from akarpov.music.services import spotify, yandex, youtube
|
||||||
from akarpov.music.services.file import load_dir, load_file
|
from akarpov.music.services.file import load_dir, load_file
|
||||||
from akarpov.utils.celery import get_scheduled_tasks_name
|
from akarpov.utils.celery import get_scheduled_tasks_name
|
||||||
|
|
||||||
|
@ -28,7 +28,11 @@
|
||||||
|
|
||||||
@shared_task
|
@shared_task
|
||||||
def list_tracks(url, user_id):
|
def list_tracks(url, user_id):
|
||||||
if "music.yandex.ru" in url:
|
if "music.youtube.com" in url:
|
||||||
|
url = url.replace("music.youtube.com", "youtube.com")
|
||||||
|
if "spotify.com" in url:
|
||||||
|
spotify.download_url(url, user_id)
|
||||||
|
elif "music.yandex.ru" in url:
|
||||||
yandex.load_playlist(url, user_id)
|
yandex.load_playlist(url, user_id)
|
||||||
elif "channel" in url or "/c/" in url:
|
elif "channel" in url or "/c/" in url:
|
||||||
p = Channel(url)
|
p = Channel(url)
|
||||||
|
|
Loading…
Reference in New Issue
Block a user