mirror of
https://github.com/nathom/streamrip.git
synced 2025-05-13 06:34:45 -04:00
Merge pull request #158 from nathom/exclude_metadata
Add ability to exclude certain tags from embedded metadata
This commit is contained in:
commit
a5ff4e4cbb
10 changed files with 425 additions and 137 deletions
115
rip/cli.py
115
rip/cli.py
|
@ -2,6 +2,7 @@ import concurrent.futures
|
||||||
import logging
|
import logging
|
||||||
import os
|
import os
|
||||||
import threading
|
import threading
|
||||||
|
from typing import Optional
|
||||||
|
|
||||||
import requests
|
import requests
|
||||||
from cleo.application import Application as BaseApplication
|
from cleo.application import Application as BaseApplication
|
||||||
|
@ -85,7 +86,9 @@ class DownloadCommand(Command):
|
||||||
if len(core) > 0:
|
if len(core) > 0:
|
||||||
core.download()
|
core.download()
|
||||||
elif not urls and path is None:
|
elif not urls and path is None:
|
||||||
self.line("<error>Must pass arguments. See </><cmd>rip url -h</cmd>.")
|
self.line(
|
||||||
|
"<error>Must pass arguments. See </><cmd>rip url -h</cmd>."
|
||||||
|
)
|
||||||
|
|
||||||
update_check.join()
|
update_check.join()
|
||||||
if outdated:
|
if outdated:
|
||||||
|
@ -112,10 +115,16 @@ class DownloadCommand(Command):
|
||||||
"https://api.github.com/repos/nathom/streamrip/releases/latest"
|
"https://api.github.com/repos/nathom/streamrip/releases/latest"
|
||||||
).json()["body"]
|
).json()["body"]
|
||||||
|
|
||||||
release_notes = md_header.sub(r"<header>\1</header>", release_notes)
|
release_notes = md_header.sub(
|
||||||
release_notes = bullet_point.sub(r"<options=bold>•</> \1", release_notes)
|
r"<header>\1</header>", release_notes
|
||||||
|
)
|
||||||
|
release_notes = bullet_point.sub(
|
||||||
|
r"<options=bold>•</> \1", release_notes
|
||||||
|
)
|
||||||
release_notes = code.sub(r"<cmd>\1</cmd>", release_notes)
|
release_notes = code.sub(r"<cmd>\1</cmd>", release_notes)
|
||||||
release_notes = issue_reference.sub(r"<options=bold>\1</>", release_notes)
|
release_notes = issue_reference.sub(
|
||||||
|
r"<options=bold>\1</>", release_notes
|
||||||
|
)
|
||||||
|
|
||||||
self.line(release_notes)
|
self.line(release_notes)
|
||||||
|
|
||||||
|
@ -145,7 +154,9 @@ class SearchCommand(Command):
|
||||||
|
|
||||||
def handle(self):
|
def handle(self):
|
||||||
query = self.argument("query")
|
query = self.argument("query")
|
||||||
source, type = clean_options(self.option("source"), self.option("type"))
|
source, type = clean_options(
|
||||||
|
self.option("source"), self.option("type")
|
||||||
|
)
|
||||||
|
|
||||||
config = Config()
|
config = Config()
|
||||||
core = RipCore(config)
|
core = RipCore(config)
|
||||||
|
@ -197,7 +208,9 @@ class DiscoverCommand(Command):
|
||||||
max_items = self.option("max-items")
|
max_items = self.option("max-items")
|
||||||
|
|
||||||
if chosen_list not in QOBUZ_FEATURED_KEYS:
|
if chosen_list not in QOBUZ_FEATURED_KEYS:
|
||||||
self.line(f'<error>Error: list "{chosen_list}" not available</error>')
|
self.line(
|
||||||
|
f'<error>Error: list "{chosen_list}" not available</error>'
|
||||||
|
)
|
||||||
self.line(self.help)
|
self.line(self.help)
|
||||||
return 1
|
return 1
|
||||||
|
|
||||||
|
@ -259,29 +272,34 @@ class ConfigCommand(Command):
|
||||||
{--qobuz : Set the credentials for Qobuz}
|
{--qobuz : Set the credentials for Qobuz}
|
||||||
{--tidal : Log into Tidal}
|
{--tidal : Log into Tidal}
|
||||||
{--deezer : Set the Deezer ARL}
|
{--deezer : Set the Deezer ARL}
|
||||||
|
{--music-app : Configure the config file for usage with the macOS Music App}
|
||||||
{--reset : Reset the config file}
|
{--reset : Reset the config file}
|
||||||
{--update : Reset the config file, keeping the credentials}
|
{--update : Reset the config file, keeping the credentials}
|
||||||
"""
|
"""
|
||||||
|
|
||||||
|
_config: Optional[Config]
|
||||||
|
|
||||||
def handle(self):
|
def handle(self):
|
||||||
import shutil
|
import shutil
|
||||||
|
|
||||||
from .constants import CONFIG_DIR, CONFIG_PATH
|
from .constants import CONFIG_DIR, CONFIG_PATH
|
||||||
|
|
||||||
config = Config()
|
self._config = Config()
|
||||||
|
|
||||||
if self.option("path"):
|
if self.option("path"):
|
||||||
self.line(f"<info>{CONFIG_PATH}</info>")
|
self.line(f"<info>{CONFIG_PATH}</info>")
|
||||||
|
|
||||||
if self.option("open"):
|
if self.option("open"):
|
||||||
self.line(f"Opening <url>{CONFIG_PATH}</url> in default application")
|
self.line(
|
||||||
|
f"Opening <url>{CONFIG_PATH}</url> in default application"
|
||||||
|
)
|
||||||
launch(CONFIG_PATH)
|
launch(CONFIG_PATH)
|
||||||
|
|
||||||
if self.option("reset"):
|
if self.option("reset"):
|
||||||
config.reset()
|
self._config.reset()
|
||||||
|
|
||||||
if self.option("update"):
|
if self.option("update"):
|
||||||
config.update()
|
self._config.update()
|
||||||
|
|
||||||
if self.option("open-vim"):
|
if self.option("open-vim"):
|
||||||
if shutil.which("nvim") is not None:
|
if shutil.which("nvim") is not None:
|
||||||
|
@ -298,8 +316,8 @@ class ConfigCommand(Command):
|
||||||
|
|
||||||
client = TidalClient()
|
client = TidalClient()
|
||||||
client.login()
|
client.login()
|
||||||
config.file["tidal"].update(client.get_tokens())
|
self._config.file["tidal"].update(client.get_tokens())
|
||||||
config.save()
|
self._config.save()
|
||||||
self.line("<info>Credentials saved to config.</info>")
|
self.line("<info>Credentials saved to config.</info>")
|
||||||
|
|
||||||
if self.option("deezer"):
|
if self.option("deezer"):
|
||||||
|
@ -316,22 +334,71 @@ class ConfigCommand(Command):
|
||||||
|
|
||||||
try:
|
try:
|
||||||
DeezerClient().login(arl=given_arl)
|
DeezerClient().login(arl=given_arl)
|
||||||
config.file["deezer"]["arl"] = given_arl
|
self._config.file["deezer"]["arl"] = given_arl
|
||||||
config.save()
|
self._config.save()
|
||||||
self.line("<b>Sucessfully logged in!</b>")
|
self.line("<b>Sucessfully logged in!</b>")
|
||||||
|
|
||||||
except AuthenticationError:
|
except AuthenticationError:
|
||||||
self.line("<error>Could not log in. Double check your ARL</error>")
|
self.line(
|
||||||
|
"<error>Could not log in. Double check your ARL</error>"
|
||||||
|
)
|
||||||
|
|
||||||
if self.option("qobuz"):
|
if self.option("qobuz"):
|
||||||
import getpass
|
import getpass
|
||||||
import hashlib
|
import hashlib
|
||||||
|
|
||||||
config.file["qobuz"]["email"] = self.ask("Qobuz email:")
|
self._config.file["qobuz"]["email"] = self.ask("Qobuz email:")
|
||||||
config.file["qobuz"]["password"] = hashlib.md5(
|
self._config.file["qobuz"]["password"] = hashlib.md5(
|
||||||
getpass.getpass("Qobuz password (won't show on screen): ").encode()
|
getpass.getpass(
|
||||||
|
"Qobuz password (won't show on screen): "
|
||||||
|
).encode()
|
||||||
).hexdigest()
|
).hexdigest()
|
||||||
config.save()
|
self._config.save()
|
||||||
|
|
||||||
|
if self.option("music-app"):
|
||||||
|
self._conf_music_app()
|
||||||
|
|
||||||
|
def _conf_music_app(self):
|
||||||
|
import subprocess
|
||||||
|
import xml.etree.ElementTree as ET
|
||||||
|
from pathlib import Path
|
||||||
|
from tempfile import mktemp
|
||||||
|
|
||||||
|
# Find the Music library folder
|
||||||
|
temp_file = mktemp()
|
||||||
|
music_pref_plist = Path(Path.home()) / Path(
|
||||||
|
"Library/Preferences/com.apple.Music.plist"
|
||||||
|
)
|
||||||
|
# copy preferences to tempdir
|
||||||
|
subprocess.run(["cp", music_pref_plist, temp_file])
|
||||||
|
# convert binary to xml for parsing
|
||||||
|
subprocess.run(["plutil", "-convert", "xml1", temp_file])
|
||||||
|
items = iter(ET.parse(temp_file).getroot()[0])
|
||||||
|
|
||||||
|
for item in items:
|
||||||
|
if item.text == "NSNavLastRootDirectory":
|
||||||
|
break
|
||||||
|
|
||||||
|
library_folder = Path(next(items).text)
|
||||||
|
os.remove(temp_file)
|
||||||
|
|
||||||
|
# cp ~/library/preferences/com.apple.music.plist music.plist
|
||||||
|
# plutil -convert xml1 music.plist
|
||||||
|
# cat music.plist | pbcopy
|
||||||
|
|
||||||
|
self._config.file["downloads"]["folder"] = os.path.join(
|
||||||
|
library_folder, "Automatically Add to Music.localized"
|
||||||
|
)
|
||||||
|
|
||||||
|
conversion_config = self._config.file["conversion"]
|
||||||
|
conversion_config["enabled"] = True
|
||||||
|
conversion_config["codec"] = "ALAC"
|
||||||
|
conversion_config["sampling_rate"] = 48000
|
||||||
|
conversion_config["bit_depth"] = 24
|
||||||
|
|
||||||
|
self._config.file["filepaths"]["folder_format"] = ""
|
||||||
|
self._config.file["artwork"]["keep_hires_cover"] = False
|
||||||
|
self._config.save()
|
||||||
|
|
||||||
|
|
||||||
class ConvertCommand(Command):
|
class ConvertCommand(Command):
|
||||||
|
@ -409,7 +476,8 @@ class ConvertCommand(Command):
|
||||||
futures.append(
|
futures.append(
|
||||||
executor.submit(
|
executor.submit(
|
||||||
ConverterCls(
|
ConverterCls(
|
||||||
filename=os.path.join(dirname, file), **converter_args
|
filename=os.path.join(dirname, file),
|
||||||
|
**converter_args,
|
||||||
).convert
|
).convert
|
||||||
)
|
)
|
||||||
)
|
)
|
||||||
|
@ -428,7 +496,8 @@ class ConvertCommand(Command):
|
||||||
ConverterCls(filename=path, **converter_args).convert()
|
ConverterCls(filename=path, **converter_args).convert()
|
||||||
else:
|
else:
|
||||||
self.line(
|
self.line(
|
||||||
f'<error>Path <path>"{path}"</path> does not exist.</error>', fg="red"
|
f'<error>Path <path>"{path}"</path> does not exist.</error>',
|
||||||
|
fg="red",
|
||||||
)
|
)
|
||||||
|
|
||||||
|
|
||||||
|
@ -534,7 +603,9 @@ class Application(BaseApplication):
|
||||||
formatter.set_style("path", Style("green", options=["bold"]))
|
formatter.set_style("path", Style("green", options=["bold"]))
|
||||||
formatter.set_style("cmd", Style("magenta"))
|
formatter.set_style("cmd", Style("magenta"))
|
||||||
formatter.set_style("title", Style("yellow", options=["bold"]))
|
formatter.set_style("title", Style("yellow", options=["bold"]))
|
||||||
formatter.set_style("header", Style("yellow", options=["bold", "underline"]))
|
formatter.set_style(
|
||||||
|
"header", Style("yellow", options=["bold", "underline"])
|
||||||
|
)
|
||||||
io.output.set_formatter(formatter)
|
io.output.set_formatter(formatter)
|
||||||
io.error_output.set_formatter(formatter)
|
io.error_output.set_formatter(formatter)
|
||||||
|
|
||||||
|
|
|
@ -130,6 +130,9 @@ keep_hires_cover = true
|
||||||
set_playlist_to_album = true
|
set_playlist_to_album = true
|
||||||
# Replaces the original track's tracknumber with it's position in the playlist
|
# Replaces the original track's tracknumber with it's position in the playlist
|
||||||
new_playlist_tracknumbers = true
|
new_playlist_tracknumbers = true
|
||||||
|
# The following metadata tags won't be applied
|
||||||
|
# See https://github.com/nathom/streamrip/wiki/Metadata-Tag-Names for more info
|
||||||
|
exclude = []
|
||||||
|
|
||||||
# Changes the folder and file names generated by streamrip.
|
# Changes the folder and file names generated by streamrip.
|
||||||
[filepaths]
|
[filepaths]
|
||||||
|
@ -157,4 +160,4 @@ progress_bar = "dainty"
|
||||||
|
|
||||||
[misc]
|
[misc]
|
||||||
# Metadata to identify this config file. Do not change.
|
# Metadata to identify this config file. Do not change.
|
||||||
version = "1.3"
|
version = "1.4"
|
||||||
|
|
91
rip/core.py
91
rip/core.py
|
@ -111,14 +111,18 @@ class RipCore(list):
|
||||||
else:
|
else:
|
||||||
self.config = config
|
self.config = config
|
||||||
|
|
||||||
if (theme := self.config.file["theme"]["progress_bar"]) != TQDM_DEFAULT_THEME:
|
if (
|
||||||
|
theme := self.config.file["theme"]["progress_bar"]
|
||||||
|
) != TQDM_DEFAULT_THEME:
|
||||||
set_progress_bar_theme(theme.lower())
|
set_progress_bar_theme(theme.lower())
|
||||||
|
|
||||||
def get_db(db_type: str) -> db.Database:
|
def get_db(db_type: str) -> db.Database:
|
||||||
db_settings = self.config.session["database"]
|
db_settings = self.config.session["database"]
|
||||||
db_class = db.CLASS_MAP[db_type]
|
db_class = db.CLASS_MAP[db_type]
|
||||||
|
|
||||||
if db_settings[db_type]["enabled"] and db_settings.get("enabled", True):
|
if db_settings[db_type]["enabled"] and db_settings.get(
|
||||||
|
"enabled", True
|
||||||
|
):
|
||||||
default_db_path = DB_PATH_MAP[db_type]
|
default_db_path = DB_PATH_MAP[db_type]
|
||||||
path = db_settings[db_type]["path"]
|
path = db_settings[db_type]["path"]
|
||||||
|
|
||||||
|
@ -212,8 +216,9 @@ class RipCore(list):
|
||||||
session = self.config.session
|
session = self.config.session
|
||||||
logger.debug(session)
|
logger.debug(session)
|
||||||
# So that the dictionary isn't searched for the same keys multiple times
|
# So that the dictionary isn't searched for the same keys multiple times
|
||||||
artwork, conversion, filepaths = tuple(
|
artwork, conversion, filepaths, metadata = (
|
||||||
session[key] for key in ("artwork", "conversion", "filepaths")
|
session[key]
|
||||||
|
for key in ("artwork", "conversion", "filepaths", "metadata")
|
||||||
)
|
)
|
||||||
concurrency = session["downloads"]["concurrency"]
|
concurrency = session["downloads"]["concurrency"]
|
||||||
return {
|
return {
|
||||||
|
@ -223,12 +228,12 @@ class RipCore(list):
|
||||||
"embed_cover": artwork["embed"],
|
"embed_cover": artwork["embed"],
|
||||||
"embed_cover_size": artwork["size"],
|
"embed_cover_size": artwork["size"],
|
||||||
"keep_hires_cover": artwork["keep_hires_cover"],
|
"keep_hires_cover": artwork["keep_hires_cover"],
|
||||||
"set_playlist_to_album": session["metadata"]["set_playlist_to_album"],
|
"set_playlist_to_album": metadata["set_playlist_to_album"],
|
||||||
"stay_temp": conversion["enabled"],
|
"stay_temp": conversion["enabled"],
|
||||||
"conversion": conversion,
|
"conversion": conversion,
|
||||||
"concurrent_downloads": concurrency["enabled"],
|
"concurrent_downloads": concurrency["enabled"],
|
||||||
"max_connections": concurrency["max_connections"],
|
"max_connections": concurrency["max_connections"],
|
||||||
"new_tracknumbers": session["metadata"]["new_playlist_tracknumbers"],
|
"new_tracknumbers": metadata["new_playlist_tracknumbers"],
|
||||||
"download_videos": session["tidal"]["download_videos"],
|
"download_videos": session["tidal"]["download_videos"],
|
||||||
"download_booklets": session["qobuz"]["download_booklets"],
|
"download_booklets": session["qobuz"]["download_booklets"],
|
||||||
"download_youtube_videos": session["youtube"]["download_videos"],
|
"download_youtube_videos": session["youtube"]["download_videos"],
|
||||||
|
@ -238,6 +243,7 @@ class RipCore(list):
|
||||||
"add_singles_to_folder": filepaths["add_singles_to_folder"],
|
"add_singles_to_folder": filepaths["add_singles_to_folder"],
|
||||||
"max_artwork_width": int(artwork["max_width"]),
|
"max_artwork_width": int(artwork["max_width"]),
|
||||||
"max_artwork_height": int(artwork["max_height"]),
|
"max_artwork_height": int(artwork["max_height"]),
|
||||||
|
"exclude_tags": metadata["exclude"],
|
||||||
}
|
}
|
||||||
|
|
||||||
def repair(self, max_items=None):
|
def repair(self, max_items=None):
|
||||||
|
@ -257,7 +263,9 @@ class RipCore(list):
|
||||||
)
|
)
|
||||||
exit()
|
exit()
|
||||||
|
|
||||||
for counter, (source, media_type, item_id) in enumerate(self.failed_db):
|
for counter, (source, media_type, item_id) in enumerate(
|
||||||
|
self.failed_db
|
||||||
|
):
|
||||||
if counter >= max_items:
|
if counter >= max_items:
|
||||||
break
|
break
|
||||||
|
|
||||||
|
@ -280,7 +288,9 @@ class RipCore(list):
|
||||||
|
|
||||||
logger.debug("Arguments from config: %s", arguments)
|
logger.debug("Arguments from config: %s", arguments)
|
||||||
|
|
||||||
source_subdirs = self.config.session["downloads"]["source_subdirectories"]
|
source_subdirs = self.config.session["downloads"][
|
||||||
|
"source_subdirectories"
|
||||||
|
]
|
||||||
for item in self:
|
for item in self:
|
||||||
# Item already checked in database in handle_urls
|
# Item already checked in database in handle_urls
|
||||||
if source_subdirs:
|
if source_subdirs:
|
||||||
|
@ -292,20 +302,26 @@ class RipCore(list):
|
||||||
item.download(**arguments)
|
item.download(**arguments)
|
||||||
continue
|
continue
|
||||||
|
|
||||||
arguments["quality"] = self.config.session[item.client.source]["quality"]
|
arguments["quality"] = self.config.session[item.client.source][
|
||||||
|
"quality"
|
||||||
|
]
|
||||||
if isinstance(item, Artist):
|
if isinstance(item, Artist):
|
||||||
filters_ = tuple(
|
filters_ = tuple(
|
||||||
k for k, v in self.config.session["filters"].items() if v
|
k for k, v in self.config.session["filters"].items() if v
|
||||||
)
|
)
|
||||||
arguments["filters"] = filters_
|
arguments["filters"] = filters_
|
||||||
logger.debug("Added filter argument for artist/label: %s", filters_)
|
logger.debug(
|
||||||
|
"Added filter argument for artist/label: %s", filters_
|
||||||
|
)
|
||||||
|
|
||||||
if not isinstance(item, Tracklist) or not item.loaded:
|
if not isinstance(item, Tracklist) or not item.loaded:
|
||||||
logger.debug("Loading metadata")
|
logger.debug("Loading metadata")
|
||||||
try:
|
try:
|
||||||
item.load_meta(**arguments)
|
item.load_meta(**arguments)
|
||||||
except NonStreamable:
|
except NonStreamable:
|
||||||
self.failed_db.add((item.client.source, item.type, item.id))
|
self.failed_db.add(
|
||||||
|
(item.client.source, item.type, item.id)
|
||||||
|
)
|
||||||
secho(f"{item!s} is not available, skipping.", fg="red")
|
secho(f"{item!s} is not available, skipping.", fg="red")
|
||||||
continue
|
continue
|
||||||
|
|
||||||
|
@ -332,7 +348,7 @@ class RipCore(list):
|
||||||
self.db.add(str(item_id))
|
self.db.add(str(item_id))
|
||||||
|
|
||||||
if isinstance(item, Track):
|
if isinstance(item, Track):
|
||||||
item.tag()
|
item.tag(exclude_tags=arguments["exclude_tags"])
|
||||||
if arguments["conversion"]["enabled"]:
|
if arguments["conversion"]["enabled"]:
|
||||||
item.convert(**arguments["conversion"])
|
item.convert(**arguments["conversion"])
|
||||||
|
|
||||||
|
@ -342,7 +358,9 @@ class RipCore(list):
|
||||||
:param featured_list: The name of the list. See `rip discover --help`.
|
:param featured_list: The name of the list. See `rip discover --help`.
|
||||||
:type featured_list: str
|
:type featured_list: str
|
||||||
"""
|
"""
|
||||||
self.extend(self.search("qobuz", featured_list, "featured", limit=max_items))
|
self.extend(
|
||||||
|
self.search("qobuz", featured_list, "featured", limit=max_items)
|
||||||
|
)
|
||||||
|
|
||||||
def get_client(self, source: str) -> Client:
|
def get_client(self, source: str) -> Client:
|
||||||
"""Get a client given the source and log in.
|
"""Get a client given the source and log in.
|
||||||
|
@ -448,12 +466,15 @@ class RipCore(list):
|
||||||
fg="yellow",
|
fg="yellow",
|
||||||
)
|
)
|
||||||
parsed.extend(
|
parsed.extend(
|
||||||
("deezer", *extract_deezer_dynamic_link(url)) for url in dynamic_urls
|
("deezer", *extract_deezer_dynamic_link(url))
|
||||||
|
for url in dynamic_urls
|
||||||
)
|
)
|
||||||
|
|
||||||
parsed.extend(URL_REGEX.findall(url)) # Qobuz, Tidal, Dezer
|
parsed.extend(URL_REGEX.findall(url)) # Qobuz, Tidal, Dezer
|
||||||
soundcloud_urls = SOUNDCLOUD_URL_REGEX.findall(url)
|
soundcloud_urls = SOUNDCLOUD_URL_REGEX.findall(url)
|
||||||
soundcloud_items = [self.clients["soundcloud"].get(u) for u in soundcloud_urls]
|
soundcloud_items = [
|
||||||
|
self.clients["soundcloud"].get(u) for u in soundcloud_urls
|
||||||
|
]
|
||||||
|
|
||||||
parsed.extend(
|
parsed.extend(
|
||||||
("soundcloud", item["kind"], url)
|
("soundcloud", item["kind"], url)
|
||||||
|
@ -484,11 +505,15 @@ class RipCore(list):
|
||||||
|
|
||||||
# For testing:
|
# For testing:
|
||||||
# https://www.last.fm/user/nathan3895/playlists/12058911
|
# https://www.last.fm/user/nathan3895/playlists/12058911
|
||||||
user_regex = re.compile(r"https://www\.last\.fm/user/([^/]+)/playlists/\d+")
|
user_regex = re.compile(
|
||||||
|
r"https://www\.last\.fm/user/([^/]+)/playlists/\d+"
|
||||||
|
)
|
||||||
lastfm_urls = LASTFM_URL_REGEX.findall(urls)
|
lastfm_urls = LASTFM_URL_REGEX.findall(urls)
|
||||||
try:
|
try:
|
||||||
lastfm_source = self.config.session["lastfm"]["source"]
|
lastfm_source = self.config.session["lastfm"]["source"]
|
||||||
lastfm_fallback_source = self.config.session["lastfm"]["fallback_source"]
|
lastfm_fallback_source = self.config.session["lastfm"][
|
||||||
|
"fallback_source"
|
||||||
|
]
|
||||||
except KeyError:
|
except KeyError:
|
||||||
self._config_updating_message()
|
self._config_updating_message()
|
||||||
self.config.update()
|
self.config.update()
|
||||||
|
@ -522,12 +547,16 @@ class RipCore(list):
|
||||||
)
|
)
|
||||||
query_is_clean = banned_words_plain.search(query) is None
|
query_is_clean = banned_words_plain.search(query) is None
|
||||||
|
|
||||||
search_results = self.search(source, query, media_type="track")
|
search_results = self.search(
|
||||||
|
source, query, media_type="track"
|
||||||
|
)
|
||||||
track = next(search_results)
|
track = next(search_results)
|
||||||
|
|
||||||
if query_is_clean:
|
if query_is_clean:
|
||||||
while banned_words.search(track["title"]) is not None:
|
while banned_words.search(track["title"]) is not None:
|
||||||
logger.debug("Track title banned for query=%s", query)
|
logger.debug(
|
||||||
|
"Track title banned for query=%s", query
|
||||||
|
)
|
||||||
track = next(search_results)
|
track = next(search_results)
|
||||||
|
|
||||||
# Because the track is searched as a single we need to set
|
# Because the track is searched as a single we need to set
|
||||||
|
@ -537,7 +566,9 @@ class RipCore(list):
|
||||||
except (NoResultsFound, StopIteration):
|
except (NoResultsFound, StopIteration):
|
||||||
return None
|
return None
|
||||||
|
|
||||||
track = try_search(lastfm_source) or try_search(lastfm_fallback_source)
|
track = try_search(lastfm_source) or try_search(
|
||||||
|
lastfm_fallback_source
|
||||||
|
)
|
||||||
if track is None:
|
if track is None:
|
||||||
return False
|
return False
|
||||||
|
|
||||||
|
@ -561,7 +592,9 @@ class RipCore(list):
|
||||||
pl.creator = creator_match.group(1)
|
pl.creator = creator_match.group(1)
|
||||||
|
|
||||||
tracks_not_found = 0
|
tracks_not_found = 0
|
||||||
with concurrent.futures.ThreadPoolExecutor(max_workers=15) as executor:
|
with concurrent.futures.ThreadPoolExecutor(
|
||||||
|
max_workers=15
|
||||||
|
) as executor:
|
||||||
futures = [
|
futures = [
|
||||||
executor.submit(search_query, title, artist, pl)
|
executor.submit(search_query, title, artist, pl)
|
||||||
for title, artist in queries
|
for title, artist in queries
|
||||||
|
@ -639,7 +672,9 @@ class RipCore(list):
|
||||||
else:
|
else:
|
||||||
logger.debug("Not generator")
|
logger.debug("Not generator")
|
||||||
items = (
|
items = (
|
||||||
results.get("data") or results.get("items") or results.get("collection")
|
results.get("data")
|
||||||
|
or results.get("items")
|
||||||
|
or results.get("collection")
|
||||||
)
|
)
|
||||||
if items is None:
|
if items is None:
|
||||||
raise NoResultsFound(query)
|
raise NoResultsFound(query)
|
||||||
|
@ -679,7 +714,9 @@ class RipCore(list):
|
||||||
raise NotImplementedError
|
raise NotImplementedError
|
||||||
|
|
||||||
fields = (fname for _, fname, _, _ in Formatter().parse(fmt) if fname)
|
fields = (fname for _, fname, _, _ in Formatter().parse(fmt) if fname)
|
||||||
ret = fmt.format(**{k: media.get(k, default="Unknown") for k in fields})
|
ret = fmt.format(
|
||||||
|
**{k: media.get(k, default="Unknown") for k in fields}
|
||||||
|
)
|
||||||
return ret
|
return ret
|
||||||
|
|
||||||
def interactive_search(
|
def interactive_search(
|
||||||
|
@ -817,7 +854,9 @@ class RipCore(list):
|
||||||
playlist_title = html.unescape(playlist_title_match.group(1))
|
playlist_title = html.unescape(playlist_title_match.group(1))
|
||||||
|
|
||||||
if remaining_tracks > 0:
|
if remaining_tracks > 0:
|
||||||
with concurrent.futures.ThreadPoolExecutor(max_workers=15) as executor:
|
with concurrent.futures.ThreadPoolExecutor(
|
||||||
|
max_workers=15
|
||||||
|
) as executor:
|
||||||
last_page = int(remaining_tracks // 50) + int(
|
last_page = int(remaining_tracks // 50) + int(
|
||||||
remaining_tracks % 50 != 0
|
remaining_tracks % 50 != 0
|
||||||
)
|
)
|
||||||
|
@ -872,7 +911,9 @@ class RipCore(list):
|
||||||
fg="blue",
|
fg="blue",
|
||||||
)
|
)
|
||||||
|
|
||||||
self.config.file["deezer"]["arl"] = input(style("ARL: ", fg="green"))
|
self.config.file["deezer"]["arl"] = input(
|
||||||
|
style("ARL: ", fg="green")
|
||||||
|
)
|
||||||
self.config.save()
|
self.config.save()
|
||||||
secho(
|
secho(
|
||||||
f'Credentials saved to config file at "{self.config._path}"',
|
f'Credentials saved to config file at "{self.config._path}"',
|
||||||
|
|
12
rip/db.py
12
rip/db.py
|
@ -71,11 +71,15 @@ class Database:
|
||||||
|
|
||||||
with sqlite3.connect(self.path) as conn:
|
with sqlite3.connect(self.path) as conn:
|
||||||
conditions = " AND ".join(f"{key}=?" for key in items.keys())
|
conditions = " AND ".join(f"{key}=?" for key in items.keys())
|
||||||
command = f"SELECT EXISTS(SELECT 1 FROM {self.name} WHERE {conditions})"
|
command = (
|
||||||
|
f"SELECT EXISTS(SELECT 1 FROM {self.name} WHERE {conditions})"
|
||||||
|
)
|
||||||
|
|
||||||
logger.debug("Executing %s", command)
|
logger.debug("Executing %s", command)
|
||||||
|
|
||||||
return bool(conn.execute(command, tuple(items.values())).fetchone()[0])
|
return bool(
|
||||||
|
conn.execute(command, tuple(items.values())).fetchone()[0]
|
||||||
|
)
|
||||||
|
|
||||||
def __contains__(self, keys: Union[str, dict]) -> bool:
|
def __contains__(self, keys: Union[str, dict]) -> bool:
|
||||||
"""Check whether a key-value pair exists in the database.
|
"""Check whether a key-value pair exists in the database.
|
||||||
|
@ -119,7 +123,9 @@ class Database:
|
||||||
|
|
||||||
params = ", ".join(self.structure.keys())
|
params = ", ".join(self.structure.keys())
|
||||||
question_marks = ", ".join("?" for _ in items)
|
question_marks = ", ".join("?" for _ in items)
|
||||||
command = f"INSERT INTO {self.name} ({params}) VALUES ({question_marks})"
|
command = (
|
||||||
|
f"INSERT INTO {self.name} ({params}) VALUES ({question_marks})"
|
||||||
|
)
|
||||||
|
|
||||||
logger.debug("Executing %s", command)
|
logger.debug("Executing %s", command)
|
||||||
logger.debug("Items to add: %s", items)
|
logger.debug("Items to add: %s", items)
|
||||||
|
|
|
@ -132,7 +132,10 @@ class QobuzClient(Client):
|
||||||
if not kwargs.get("app_id") or not kwargs.get("secrets"):
|
if not kwargs.get("app_id") or not kwargs.get("secrets"):
|
||||||
self._get_app_id_and_secrets() # can be async
|
self._get_app_id_and_secrets() # can be async
|
||||||
else:
|
else:
|
||||||
self.app_id, self.secrets = str(kwargs["app_id"]), kwargs["secrets"]
|
self.app_id, self.secrets = (
|
||||||
|
str(kwargs["app_id"]),
|
||||||
|
kwargs["secrets"],
|
||||||
|
)
|
||||||
self.session = gen_threadsafe_session(
|
self.session = gen_threadsafe_session(
|
||||||
headers={"User-Agent": AGENT, "X-App-Id": self.app_id}
|
headers={"User-Agent": AGENT, "X-App-Id": self.app_id}
|
||||||
)
|
)
|
||||||
|
@ -215,7 +218,10 @@ class QobuzClient(Client):
|
||||||
def _get_app_id_and_secrets(self):
|
def _get_app_id_and_secrets(self):
|
||||||
if not hasattr(self, "app_id") or not hasattr(self, "secrets"):
|
if not hasattr(self, "app_id") or not hasattr(self, "secrets"):
|
||||||
spoofer = Spoofer()
|
spoofer = Spoofer()
|
||||||
self.app_id, self.secrets = str(spoofer.get_app_id()), spoofer.get_secrets()
|
self.app_id, self.secrets = (
|
||||||
|
str(spoofer.get_app_id()),
|
||||||
|
spoofer.get_secrets(),
|
||||||
|
)
|
||||||
|
|
||||||
if not hasattr(self, "sec"):
|
if not hasattr(self, "sec"):
|
||||||
if not hasattr(self, "session"):
|
if not hasattr(self, "session"):
|
||||||
|
@ -234,7 +240,9 @@ class QobuzClient(Client):
|
||||||
:rtype: dict
|
:rtype: dict
|
||||||
"""
|
"""
|
||||||
page, status_code = self._api_request(epoint, params)
|
page, status_code = self._api_request(epoint, params)
|
||||||
logger.debug("Keys returned from _gen_pages: %s", ", ".join(page.keys()))
|
logger.debug(
|
||||||
|
"Keys returned from _gen_pages: %s", ", ".join(page.keys())
|
||||||
|
)
|
||||||
key = epoint.split("/")[0] + "s"
|
key = epoint.split("/")[0] + "s"
|
||||||
total = page.get(key, {})
|
total = page.get(key, {})
|
||||||
total = total.get("total") or total.get("items")
|
total = total.get("total") or total.get("items")
|
||||||
|
@ -257,7 +265,8 @@ class QobuzClient(Client):
|
||||||
"""Check if the secrets are usable."""
|
"""Check if the secrets are usable."""
|
||||||
with concurrent.futures.ThreadPoolExecutor() as executor:
|
with concurrent.futures.ThreadPoolExecutor() as executor:
|
||||||
futures = [
|
futures = [
|
||||||
executor.submit(self._test_secret, secret) for secret in self.secrets
|
executor.submit(self._test_secret, secret)
|
||||||
|
for secret in self.secrets
|
||||||
]
|
]
|
||||||
|
|
||||||
for future in concurrent.futures.as_completed(futures):
|
for future in concurrent.futures.as_completed(futures):
|
||||||
|
@ -300,11 +309,15 @@ class QobuzClient(Client):
|
||||||
|
|
||||||
response, status_code = self._api_request(epoint, params)
|
response, status_code = self._api_request(epoint, params)
|
||||||
if status_code != 200:
|
if status_code != 200:
|
||||||
raise Exception(f'Error fetching metadata. "{response["message"]}"')
|
raise Exception(
|
||||||
|
f'Error fetching metadata. "{response["message"]}"'
|
||||||
|
)
|
||||||
|
|
||||||
return response
|
return response
|
||||||
|
|
||||||
def _api_search(self, query: str, media_type: str, limit: int = 500) -> Generator:
|
def _api_search(
|
||||||
|
self, query: str, media_type: str, limit: int = 500
|
||||||
|
) -> Generator:
|
||||||
"""Send a search request to the API.
|
"""Send a search request to the API.
|
||||||
|
|
||||||
:param query:
|
:param query:
|
||||||
|
@ -356,7 +369,9 @@ class QobuzClient(Client):
|
||||||
resp, status_code = self._api_request(epoint, params)
|
resp, status_code = self._api_request(epoint, params)
|
||||||
|
|
||||||
if status_code == 401:
|
if status_code == 401:
|
||||||
raise AuthenticationError(f"Invalid credentials from params {params}")
|
raise AuthenticationError(
|
||||||
|
f"Invalid credentials from params {params}"
|
||||||
|
)
|
||||||
elif status_code == 400:
|
elif status_code == 400:
|
||||||
logger.debug(resp)
|
logger.debug(resp)
|
||||||
raise InvalidAppIdError(f"Invalid app id from params {params}")
|
raise InvalidAppIdError(f"Invalid app id from params {params}")
|
||||||
|
@ -364,7 +379,9 @@ class QobuzClient(Client):
|
||||||
logger.info("Logged in to Qobuz")
|
logger.info("Logged in to Qobuz")
|
||||||
|
|
||||||
if not resp["user"]["credential"]["parameters"]:
|
if not resp["user"]["credential"]["parameters"]:
|
||||||
raise IneligibleError("Free accounts are not eligible to download tracks.")
|
raise IneligibleError(
|
||||||
|
"Free accounts are not eligible to download tracks."
|
||||||
|
)
|
||||||
|
|
||||||
self.uat = resp["user_auth_token"]
|
self.uat = resp["user_auth_token"]
|
||||||
self.session.headers.update({"X-User-Auth-Token": self.uat})
|
self.session.headers.update({"X-User-Auth-Token": self.uat})
|
||||||
|
@ -413,7 +430,9 @@ class QobuzClient(Client):
|
||||||
}
|
}
|
||||||
response, status_code = self._api_request("track/getFileUrl", params)
|
response, status_code = self._api_request("track/getFileUrl", params)
|
||||||
if status_code == 400:
|
if status_code == 400:
|
||||||
raise InvalidAppSecretError("Invalid app secret from params %s" % params)
|
raise InvalidAppSecretError(
|
||||||
|
"Invalid app secret from params %s" % params
|
||||||
|
)
|
||||||
|
|
||||||
return response
|
return response
|
||||||
|
|
||||||
|
@ -432,7 +451,9 @@ class QobuzClient(Client):
|
||||||
logger.debug(r.text)
|
logger.debug(r.text)
|
||||||
return r.json(), r.status_code
|
return r.json(), r.status_code
|
||||||
except Exception:
|
except Exception:
|
||||||
logger.error("Problem getting JSON. Status code: %s", r.status_code)
|
logger.error(
|
||||||
|
"Problem getting JSON. Status code: %s", r.status_code
|
||||||
|
)
|
||||||
raise
|
raise
|
||||||
|
|
||||||
def _test_secret(self, secret: str) -> Optional[str]:
|
def _test_secret(self, secret: str) -> Optional[str]:
|
||||||
|
@ -464,7 +485,9 @@ class DeezerClient(Client):
|
||||||
# no login required
|
# no login required
|
||||||
self.logged_in = False
|
self.logged_in = False
|
||||||
|
|
||||||
def search(self, query: str, media_type: str = "album", limit: int = 200) -> dict:
|
def search(
|
||||||
|
self, query: str, media_type: str = "album", limit: int = 200
|
||||||
|
) -> dict:
|
||||||
"""Search API for query.
|
"""Search API for query.
|
||||||
|
|
||||||
:param query:
|
:param query:
|
||||||
|
@ -550,9 +573,9 @@ class DeezerClient(Client):
|
||||||
format_no, format_str = format_info
|
format_no, format_str = format_info
|
||||||
|
|
||||||
dl_info["size_to_quality"] = {
|
dl_info["size_to_quality"] = {
|
||||||
int(track_info.get(f"FILESIZE_{format}")): self._quality_id_from_filetype(
|
int(
|
||||||
format
|
track_info.get(f"FILESIZE_{format}")
|
||||||
)
|
): self._quality_id_from_filetype(format)
|
||||||
for format in DEEZER_FORMATS
|
for format in DEEZER_FORMATS
|
||||||
}
|
}
|
||||||
|
|
||||||
|
@ -593,7 +616,9 @@ class DeezerClient(Client):
|
||||||
logger.debug("Info bytes: %s", info_bytes)
|
logger.debug("Info bytes: %s", info_bytes)
|
||||||
path = self._gen_url_path(info_bytes)
|
path = self._gen_url_path(info_bytes)
|
||||||
logger.debug(path)
|
logger.debug(path)
|
||||||
return f"https://e-cdns-proxy-{track_hash[0]}.dzcdn.net/mobile/1/{path}"
|
return (
|
||||||
|
f"https://e-cdns-proxy-{track_hash[0]}.dzcdn.net/mobile/1/{path}"
|
||||||
|
)
|
||||||
|
|
||||||
def _gen_url_path(self, data):
|
def _gen_url_path(self, data):
|
||||||
return binascii.hexlify(
|
return binascii.hexlify(
|
||||||
|
@ -623,7 +648,9 @@ class DeezloaderClient(Client):
|
||||||
# no login required
|
# no login required
|
||||||
self.logged_in = True
|
self.logged_in = True
|
||||||
|
|
||||||
def search(self, query: str, media_type: str = "album", limit: int = 200) -> dict:
|
def search(
|
||||||
|
self, query: str, media_type: str = "album", limit: int = 200
|
||||||
|
) -> dict:
|
||||||
"""Search API for query.
|
"""Search API for query.
|
||||||
|
|
||||||
:param query:
|
:param query:
|
||||||
|
@ -660,7 +687,9 @@ class DeezloaderClient(Client):
|
||||||
url = f"{DEEZER_BASE}/{media_type}/{meta_id}"
|
url = f"{DEEZER_BASE}/{media_type}/{meta_id}"
|
||||||
item = self.session.get(url).json()
|
item = self.session.get(url).json()
|
||||||
if media_type in ("album", "playlist"):
|
if media_type in ("album", "playlist"):
|
||||||
tracks = self.session.get(f"{url}/tracks", params={"limit": 1000}).json()
|
tracks = self.session.get(
|
||||||
|
f"{url}/tracks", params={"limit": 1000}
|
||||||
|
).json()
|
||||||
item["tracks"] = tracks["data"]
|
item["tracks"] = tracks["data"]
|
||||||
item["track_total"] = len(tracks["data"])
|
item["track_total"] = len(tracks["data"])
|
||||||
elif media_type == "artist":
|
elif media_type == "artist":
|
||||||
|
@ -756,7 +785,9 @@ class TidalClient(Client):
|
||||||
logger.debug(resp)
|
logger.debug(resp)
|
||||||
return resp
|
return resp
|
||||||
|
|
||||||
def search(self, query: str, media_type: str = "album", limit: int = 100) -> dict:
|
def search(
|
||||||
|
self, query: str, media_type: str = "album", limit: int = 100
|
||||||
|
) -> dict:
|
||||||
"""Search for a query.
|
"""Search for a query.
|
||||||
|
|
||||||
:param query:
|
:param query:
|
||||||
|
@ -785,13 +816,19 @@ class TidalClient(Client):
|
||||||
return self._get_video_stream_url(track_id)
|
return self._get_video_stream_url(track_id)
|
||||||
|
|
||||||
params = {
|
params = {
|
||||||
"audioquality": get_quality(min(quality, TIDAL_MAX_Q), self.source),
|
"audioquality": get_quality(
|
||||||
|
min(quality, TIDAL_MAX_Q), self.source
|
||||||
|
),
|
||||||
"playbackmode": "STREAM",
|
"playbackmode": "STREAM",
|
||||||
"assetpresentation": "FULL",
|
"assetpresentation": "FULL",
|
||||||
}
|
}
|
||||||
resp = self._api_request(f"tracks/{track_id}/playbackinfopostpaywall", params)
|
resp = self._api_request(
|
||||||
|
f"tracks/{track_id}/playbackinfopostpaywall", params
|
||||||
|
)
|
||||||
try:
|
try:
|
||||||
manifest = json.loads(base64.b64decode(resp["manifest"]).decode("utf-8"))
|
manifest = json.loads(
|
||||||
|
base64.b64decode(resp["manifest"]).decode("utf-8")
|
||||||
|
)
|
||||||
except KeyError:
|
except KeyError:
|
||||||
raise Exception(resp["userMessage"])
|
raise Exception(resp["userMessage"])
|
||||||
|
|
||||||
|
@ -996,7 +1033,9 @@ class TidalClient(Client):
|
||||||
offset += 100
|
offset += 100
|
||||||
tracks_left -= 100
|
tracks_left -= 100
|
||||||
resp["items"].extend(
|
resp["items"].extend(
|
||||||
self._api_request(f"{url}/items", {"offset": offset})["items"]
|
self._api_request(f"{url}/items", {"offset": offset})[
|
||||||
|
"items"
|
||||||
|
]
|
||||||
)
|
)
|
||||||
|
|
||||||
item["tracks"] = [item["item"] for item in resp["items"]]
|
item["tracks"] = [item["item"] for item in resp["items"]]
|
||||||
|
@ -1048,7 +1087,9 @@ class TidalClient(Client):
|
||||||
r'#EXT-X-STREAM-INF:BANDWIDTH=\d+,AVERAGE-BANDWIDTH=\d+,CODECS="[^"]+"'
|
r'#EXT-X-STREAM-INF:BANDWIDTH=\d+,AVERAGE-BANDWIDTH=\d+,CODECS="[^"]+"'
|
||||||
r",RESOLUTION=\d+x\d+\n(.+)"
|
r",RESOLUTION=\d+x\d+\n(.+)"
|
||||||
)
|
)
|
||||||
manifest = json.loads(base64.b64decode(resp["manifest"]).decode("utf-8"))
|
manifest = json.loads(
|
||||||
|
base64.b64decode(resp["manifest"]).decode("utf-8")
|
||||||
|
)
|
||||||
available_urls = self.session.get(manifest["urls"][0])
|
available_urls = self.session.get(manifest["urls"][0])
|
||||||
url_info = re.findall(stream_url_regex, available_urls.text)
|
url_info = re.findall(stream_url_regex, available_urls.text)
|
||||||
|
|
||||||
|
@ -1138,7 +1179,10 @@ class SoundCloudClient(Client):
|
||||||
url = None
|
url = None
|
||||||
for tc in track["media"]["transcodings"]:
|
for tc in track["media"]["transcodings"]:
|
||||||
fmt = tc["format"]
|
fmt = tc["format"]
|
||||||
if fmt["protocol"] == "hls" and fmt["mime_type"] == "audio/mpeg":
|
if (
|
||||||
|
fmt["protocol"] == "hls"
|
||||||
|
and fmt["mime_type"] == "audio/mpeg"
|
||||||
|
):
|
||||||
url = tc["url"]
|
url = tc["url"]
|
||||||
break
|
break
|
||||||
|
|
||||||
|
|
|
@ -4,7 +4,9 @@ import mutagen.id3 as id3
|
||||||
|
|
||||||
AGENT = "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:83.0) Gecko/20100101 Firefox/83.0"
|
AGENT = "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:83.0) Gecko/20100101 Firefox/83.0"
|
||||||
|
|
||||||
TIDAL_COVER_URL = "https://resources.tidal.com/images/{uuid}/{width}x{height}.jpg"
|
TIDAL_COVER_URL = (
|
||||||
|
"https://resources.tidal.com/images/{uuid}/{width}x{height}.jpg"
|
||||||
|
)
|
||||||
SOUNDCLOUD_CLIENT_ID = "QFciLWLC1GS4P3EZvXIjA3jKhKO5pKB3"
|
SOUNDCLOUD_CLIENT_ID = "QFciLWLC1GS4P3EZvXIjA3jKhKO5pKB3"
|
||||||
SOUNDCLOUD_USER_ID = "672320-86895-162383-801513"
|
SOUNDCLOUD_USER_ID = "672320-86895-162383-801513"
|
||||||
SOUNDCLOUD_APP_VERSION = "1626941202"
|
SOUNDCLOUD_APP_VERSION = "1626941202"
|
||||||
|
@ -136,9 +138,7 @@ ALBUM_KEYS = (
|
||||||
"albumcomposer",
|
"albumcomposer",
|
||||||
)
|
)
|
||||||
# TODO: rename these to DEFAULT_FOLDER_FORMAT etc
|
# TODO: rename these to DEFAULT_FOLDER_FORMAT etc
|
||||||
FOLDER_FORMAT = (
|
FOLDER_FORMAT = "{albumartist} - {title} ({year}) [{container}] [{bit_depth}B-{sampling_rate}kHz]"
|
||||||
"{albumartist} - {title} ({year}) [{container}] [{bit_depth}B-{sampling_rate}kHz]"
|
|
||||||
)
|
|
||||||
TRACK_FORMAT = "{tracknumber}. {artist} - {title}"
|
TRACK_FORMAT = "{tracknumber}. {artist} - {title}"
|
||||||
|
|
||||||
|
|
||||||
|
|
|
@ -52,7 +52,9 @@ class Converter:
|
||||||
|
|
||||||
self.filename = filename
|
self.filename = filename
|
||||||
self.final_fn = f"{os.path.splitext(filename)[0]}.{self.container}"
|
self.final_fn = f"{os.path.splitext(filename)[0]}.{self.container}"
|
||||||
self.tempfile = os.path.join(gettempdir(), os.path.basename(self.final_fn))
|
self.tempfile = os.path.join(
|
||||||
|
gettempdir(), os.path.basename(self.final_fn)
|
||||||
|
)
|
||||||
self.remove_source = remove_source
|
self.remove_source = remove_source
|
||||||
self.sampling_rate = sampling_rate
|
self.sampling_rate = sampling_rate
|
||||||
self.bit_depth = bit_depth
|
self.bit_depth = bit_depth
|
||||||
|
@ -117,9 +119,13 @@ class Converter:
|
||||||
if self.lossless:
|
if self.lossless:
|
||||||
if isinstance(self.sampling_rate, int):
|
if isinstance(self.sampling_rate, int):
|
||||||
sampling_rates = "|".join(
|
sampling_rates = "|".join(
|
||||||
str(rate) for rate in SAMPLING_RATES if rate <= self.sampling_rate
|
str(rate)
|
||||||
|
for rate in SAMPLING_RATES
|
||||||
|
if rate <= self.sampling_rate
|
||||||
|
)
|
||||||
|
command.extend(
|
||||||
|
["-af", f"aformat=sample_rates={sampling_rates}"]
|
||||||
)
|
)
|
||||||
command.extend(["-af", f"aformat=sample_rates={sampling_rates}"])
|
|
||||||
|
|
||||||
elif self.sampling_rate is not None:
|
elif self.sampling_rate is not None:
|
||||||
raise TypeError(
|
raise TypeError(
|
||||||
|
@ -134,7 +140,9 @@ class Converter:
|
||||||
else:
|
else:
|
||||||
raise ValueError("Bit depth must be 16, 24, or 32")
|
raise ValueError("Bit depth must be 16, 24, or 32")
|
||||||
elif self.bit_depth is not None:
|
elif self.bit_depth is not None:
|
||||||
raise TypeError(f"Bit depth must be int, not {type(self.bit_depth)}")
|
raise TypeError(
|
||||||
|
f"Bit depth must be int, not {type(self.bit_depth)}"
|
||||||
|
)
|
||||||
|
|
||||||
# automatically overwrite
|
# automatically overwrite
|
||||||
command.extend(["-y", self.tempfile])
|
command.extend(["-y", self.tempfile])
|
||||||
|
@ -199,7 +207,9 @@ class Vorbis(Converter):
|
||||||
codec_name = "vorbis"
|
codec_name = "vorbis"
|
||||||
codec_lib = "libvorbis"
|
codec_lib = "libvorbis"
|
||||||
container = "ogg"
|
container = "ogg"
|
||||||
default_ffmpeg_arg = "-q:a 6" # 160, aka the "high" quality profile from Spotify
|
default_ffmpeg_arg = (
|
||||||
|
"-q:a 6" # 160, aka the "high" quality profile from Spotify
|
||||||
|
)
|
||||||
|
|
||||||
|
|
||||||
class OPUS(Converter):
|
class OPUS(Converter):
|
||||||
|
|
|
@ -13,7 +13,17 @@ import re
|
||||||
import shutil
|
import shutil
|
||||||
import subprocess
|
import subprocess
|
||||||
from tempfile import gettempdir
|
from tempfile import gettempdir
|
||||||
from typing import Any, Dict, Generator, Iterable, List, Optional, Tuple, Union
|
from typing import (
|
||||||
|
Any,
|
||||||
|
Dict,
|
||||||
|
Generator,
|
||||||
|
Iterable,
|
||||||
|
List,
|
||||||
|
Optional,
|
||||||
|
Sequence,
|
||||||
|
Tuple,
|
||||||
|
Union,
|
||||||
|
)
|
||||||
|
|
||||||
from click import echo, secho, style
|
from click import echo, secho, style
|
||||||
from mutagen.flac import FLAC, Picture
|
from mutagen.flac import FLAC, Picture
|
||||||
|
@ -56,7 +66,9 @@ logger = logging.getLogger("streamrip")
|
||||||
|
|
||||||
TYPE_REGEXES = {
|
TYPE_REGEXES = {
|
||||||
"remaster": re.compile(r"(?i)(re)?master(ed)?"),
|
"remaster": re.compile(r"(?i)(re)?master(ed)?"),
|
||||||
"extra": re.compile(r"(?i)(anniversary|deluxe|live|collector|demo|expanded)"),
|
"extra": re.compile(
|
||||||
|
r"(?i)(anniversary|deluxe|live|collector|demo|expanded)"
|
||||||
|
),
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
|
@ -198,12 +210,15 @@ class Track(Media):
|
||||||
if source == "qobuz":
|
if source == "qobuz":
|
||||||
self.cover_url = self.resp["album"]["image"]["large"]
|
self.cover_url = self.resp["album"]["image"]["large"]
|
||||||
elif source == "tidal":
|
elif source == "tidal":
|
||||||
self.cover_url = tidal_cover_url(self.resp["album"]["cover"], 320)
|
self.cover_url = tidal_cover_url(
|
||||||
|
self.resp["album"]["cover"], 320
|
||||||
|
)
|
||||||
elif source == "deezer":
|
elif source == "deezer":
|
||||||
self.cover_url = self.resp["album"]["cover_medium"]
|
self.cover_url = self.resp["album"]["cover_medium"]
|
||||||
elif source == "soundcloud":
|
elif source == "soundcloud":
|
||||||
self.cover_url = (
|
self.cover_url = (
|
||||||
self.resp["artwork_url"] or self.resp["user"].get("avatar_url")
|
self.resp["artwork_url"]
|
||||||
|
or self.resp["user"].get("avatar_url")
|
||||||
).replace("large", "t500x500")
|
).replace("large", "t500x500")
|
||||||
else:
|
else:
|
||||||
raise InvalidSourceError(source)
|
raise InvalidSourceError(source)
|
||||||
|
@ -251,7 +266,9 @@ class Track(Media):
|
||||||
except ItemExists as e:
|
except ItemExists as e:
|
||||||
logger.debug(e)
|
logger.debug(e)
|
||||||
|
|
||||||
self.path = os.path.join(gettempdir(), f"{hash(self.id)}_{self.quality}.tmp")
|
self.path = os.path.join(
|
||||||
|
gettempdir(), f"{hash(self.id)}_{self.quality}.tmp"
|
||||||
|
)
|
||||||
|
|
||||||
def download( # noqa
|
def download( # noqa
|
||||||
self,
|
self,
|
||||||
|
@ -306,9 +323,14 @@ class Track(Media):
|
||||||
except KeyError as e:
|
except KeyError as e:
|
||||||
if restrictions := dl_info["restrictions"]:
|
if restrictions := dl_info["restrictions"]:
|
||||||
# Turn CamelCase code into a readable sentence
|
# Turn CamelCase code into a readable sentence
|
||||||
words = re.findall(r"([A-Z][a-z]+)", restrictions[0]["code"])
|
words = re.findall(
|
||||||
|
r"([A-Z][a-z]+)", restrictions[0]["code"]
|
||||||
|
)
|
||||||
raise NonStreamable(
|
raise NonStreamable(
|
||||||
words[0] + " " + " ".join(map(str.lower, words[1:])) + "."
|
words[0]
|
||||||
|
+ " "
|
||||||
|
+ " ".join(map(str.lower, words[1:]))
|
||||||
|
+ "."
|
||||||
)
|
)
|
||||||
|
|
||||||
secho(f"Panic: {e} dl_info = {dl_info}", fg="red")
|
secho(f"Panic: {e} dl_info = {dl_info}", fg="red")
|
||||||
|
@ -317,7 +339,9 @@ class Track(Media):
|
||||||
_quick_download(download_url, self.path, desc=self._progress_desc)
|
_quick_download(download_url, self.path, desc=self._progress_desc)
|
||||||
|
|
||||||
elif isinstance(self.client, DeezloaderClient):
|
elif isinstance(self.client, DeezloaderClient):
|
||||||
_quick_download(dl_info["url"], self.path, desc=self._progress_desc)
|
_quick_download(
|
||||||
|
dl_info["url"], self.path, desc=self._progress_desc
|
||||||
|
)
|
||||||
|
|
||||||
elif self.client.source == "deezer":
|
elif self.client.source == "deezer":
|
||||||
# We can only find out if the requested quality is available
|
# We can only find out if the requested quality is available
|
||||||
|
@ -437,7 +461,9 @@ class Track(Media):
|
||||||
]
|
]
|
||||||
)
|
)
|
||||||
elif dl_info["type"] == "original":
|
elif dl_info["type"] == "original":
|
||||||
_quick_download(dl_info["url"], self.path, desc=self._progress_desc)
|
_quick_download(
|
||||||
|
dl_info["url"], self.path, desc=self._progress_desc
|
||||||
|
)
|
||||||
|
|
||||||
# if a wav is returned, convert to flac
|
# if a wav is returned, convert to flac
|
||||||
engine = converter.FLAC(self.path)
|
engine = converter.FLAC(self.path)
|
||||||
|
@ -465,7 +491,9 @@ class Track(Media):
|
||||||
|
|
||||||
def download_cover(self, width=999999, height=999999):
|
def download_cover(self, width=999999, height=999999):
|
||||||
"""Download the cover art, if cover_url is given."""
|
"""Download the cover art, if cover_url is given."""
|
||||||
self.cover_path = os.path.join(gettempdir(), f"cover{hash(self.cover_url)}.jpg")
|
self.cover_path = os.path.join(
|
||||||
|
gettempdir(), f"cover{hash(self.cover_url)}.jpg"
|
||||||
|
)
|
||||||
logger.debug("Downloading cover from %s", self.cover_url)
|
logger.debug("Downloading cover from %s", self.cover_url)
|
||||||
|
|
||||||
if not os.path.exists(self.cover_path):
|
if not os.path.exists(self.cover_path):
|
||||||
|
@ -485,9 +513,9 @@ class Track(Media):
|
||||||
formatter = self.meta.get_formatter(max_quality=self.quality)
|
formatter = self.meta.get_formatter(max_quality=self.quality)
|
||||||
logger.debug("Track meta formatter %s", formatter)
|
logger.debug("Track meta formatter %s", formatter)
|
||||||
filename = clean_format(self.file_format, formatter)
|
filename = clean_format(self.file_format, formatter)
|
||||||
self.final_path = os.path.join(self.folder, filename)[:250].strip() + ext(
|
self.final_path = os.path.join(self.folder, filename)[
|
||||||
self.quality, self.client.source
|
:250
|
||||||
)
|
].strip() + ext(self.quality, self.client.source)
|
||||||
|
|
||||||
logger.debug("Formatted path: %s", self.final_path)
|
logger.debug("Formatted path: %s", self.final_path)
|
||||||
|
|
||||||
|
@ -500,7 +528,9 @@ class Track(Media):
|
||||||
return self.final_path
|
return self.final_path
|
||||||
|
|
||||||
@classmethod
|
@classmethod
|
||||||
def from_album_meta(cls, album: TrackMetadata, track: dict, client: Client):
|
def from_album_meta(
|
||||||
|
cls, album: TrackMetadata, track: dict, client: Client
|
||||||
|
):
|
||||||
"""Return a new Track object initialized with info.
|
"""Return a new Track object initialized with info.
|
||||||
|
|
||||||
:param album: album metadata returned by API
|
:param album: album metadata returned by API
|
||||||
|
@ -510,7 +540,9 @@ class Track(Media):
|
||||||
:raises: IndexError
|
:raises: IndexError
|
||||||
"""
|
"""
|
||||||
meta = TrackMetadata(album=album, track=track, source=client.source)
|
meta = TrackMetadata(album=album, track=track, source=client.source)
|
||||||
return cls(client=client, meta=meta, id=track["id"], part_of_tracklist=True)
|
return cls(
|
||||||
|
client=client, meta=meta, id=track["id"], part_of_tracklist=True
|
||||||
|
)
|
||||||
|
|
||||||
@classmethod
|
@classmethod
|
||||||
def from_api(cls, item: dict, client: Client):
|
def from_api(cls, item: dict, client: Client):
|
||||||
|
@ -554,6 +586,7 @@ class Track(Media):
|
||||||
album_meta: dict = None,
|
album_meta: dict = None,
|
||||||
cover: Union[Picture, APIC, MP4Cover] = None,
|
cover: Union[Picture, APIC, MP4Cover] = None,
|
||||||
embed_cover: bool = True,
|
embed_cover: bool = True,
|
||||||
|
exclude_tags: Optional[Sequence] = None,
|
||||||
**kwargs,
|
**kwargs,
|
||||||
):
|
):
|
||||||
"""Tag the track using the stored metadata.
|
"""Tag the track using the stored metadata.
|
||||||
|
@ -569,7 +602,9 @@ class Track(Media):
|
||||||
:param embed_cover: Embed cover art into file
|
:param embed_cover: Embed cover art into file
|
||||||
:type embed_cover: bool
|
:type embed_cover: bool
|
||||||
"""
|
"""
|
||||||
assert isinstance(self.meta, TrackMetadata), "meta must be TrackMetadata"
|
assert isinstance(
|
||||||
|
self.meta, TrackMetadata
|
||||||
|
), "meta must be TrackMetadata"
|
||||||
if not self.downloaded:
|
if not self.downloaded:
|
||||||
logger.info(
|
logger.info(
|
||||||
"Track %s not tagged because it was not downloaded",
|
"Track %s not tagged because it was not downloaded",
|
||||||
|
@ -620,7 +655,10 @@ class Track(Media):
|
||||||
raise InvalidQuality(f'Invalid quality: "{self.quality}"')
|
raise InvalidQuality(f'Invalid quality: "{self.quality}"')
|
||||||
|
|
||||||
# automatically generate key, value pairs based on container
|
# automatically generate key, value pairs based on container
|
||||||
tags = self.meta.tags(self.container)
|
tags = self.meta.tags(
|
||||||
|
self.container,
|
||||||
|
set(exclude_tags) if exclude_tags is not None else None,
|
||||||
|
)
|
||||||
for k, v in tags:
|
for k, v in tags:
|
||||||
logger.debug("Setting %s tag to %s", k, v)
|
logger.debug("Setting %s tag to %s", k, v)
|
||||||
audio[k] = v
|
audio[k] = v
|
||||||
|
@ -690,7 +728,9 @@ class Track(Media):
|
||||||
self.format_final_path()
|
self.format_final_path()
|
||||||
|
|
||||||
if not os.path.isfile(self.path):
|
if not os.path.isfile(self.path):
|
||||||
logger.info("File %s does not exist. Skipping conversion.", self.path)
|
logger.info(
|
||||||
|
"File %s does not exist. Skipping conversion.", self.path
|
||||||
|
)
|
||||||
secho(f"{self!s} does not exist. Skipping conversion.", fg="red")
|
secho(f"{self!s} does not exist. Skipping conversion.", fg="red")
|
||||||
return
|
return
|
||||||
|
|
||||||
|
@ -1093,7 +1133,8 @@ class Tracklist(list):
|
||||||
kwargs.get("max_connections", 3)
|
kwargs.get("max_connections", 3)
|
||||||
) as executor:
|
) as executor:
|
||||||
future_map = {
|
future_map = {
|
||||||
executor.submit(target, item, **kwargs): item for item in self
|
executor.submit(target, item, **kwargs): item
|
||||||
|
for item in self
|
||||||
}
|
}
|
||||||
try:
|
try:
|
||||||
concurrent.futures.wait(future_map.keys())
|
concurrent.futures.wait(future_map.keys())
|
||||||
|
@ -1124,7 +1165,9 @@ class Tracklist(list):
|
||||||
secho(f"{item!s} exists. Skipping.", fg="yellow")
|
secho(f"{item!s} exists. Skipping.", fg="yellow")
|
||||||
except NonStreamable as e:
|
except NonStreamable as e:
|
||||||
e.print(item)
|
e.print(item)
|
||||||
failed_downloads.append((item.client.source, item.type, item.id))
|
failed_downloads.append(
|
||||||
|
(item.client.source, item.type, item.id)
|
||||||
|
)
|
||||||
|
|
||||||
self.downloaded = True
|
self.downloaded = True
|
||||||
|
|
||||||
|
@ -1453,7 +1496,9 @@ class Album(Tracklist, Media):
|
||||||
_cover_download(embed_cover_url, cover_path)
|
_cover_download(embed_cover_url, cover_path)
|
||||||
|
|
||||||
hires_cov_path = os.path.join(self.folder, "cover.jpg")
|
hires_cov_path = os.path.join(self.folder, "cover.jpg")
|
||||||
if kwargs.get("keep_hires_cover", True) and not os.path.exists(hires_cov_path):
|
if kwargs.get("keep_hires_cover", True) and not os.path.exists(
|
||||||
|
hires_cov_path
|
||||||
|
):
|
||||||
logger.debug("Downloading hires cover")
|
logger.debug("Downloading hires cover")
|
||||||
_cover_download(self.cover_urls["original"], hires_cov_path)
|
_cover_download(self.cover_urls["original"], hires_cov_path)
|
||||||
|
|
||||||
|
@ -1507,7 +1552,9 @@ class Album(Tracklist, Media):
|
||||||
and isinstance(item, Track)
|
and isinstance(item, Track)
|
||||||
and kwargs.get("folder_format")
|
and kwargs.get("folder_format")
|
||||||
):
|
):
|
||||||
disc_folder = os.path.join(self.folder, f"Disc {item.meta.discnumber}")
|
disc_folder = os.path.join(
|
||||||
|
self.folder, f"Disc {item.meta.discnumber}"
|
||||||
|
)
|
||||||
kwargs["parent_folder"] = disc_folder
|
kwargs["parent_folder"] = disc_folder
|
||||||
else:
|
else:
|
||||||
kwargs["parent_folder"] = self.folder
|
kwargs["parent_folder"] = self.folder
|
||||||
|
@ -1522,6 +1569,7 @@ class Album(Tracklist, Media):
|
||||||
item.tag(
|
item.tag(
|
||||||
cover=self.cover_obj,
|
cover=self.cover_obj,
|
||||||
embed_cover=kwargs.get("embed_cover", True),
|
embed_cover=kwargs.get("embed_cover", True),
|
||||||
|
exclude_tags=kwargs.get("exclude_tags"),
|
||||||
)
|
)
|
||||||
|
|
||||||
self.downloaded_ids.add(item.id)
|
self.downloaded_ids.add(item.id)
|
||||||
|
@ -1601,7 +1649,9 @@ class Album(Tracklist, Media):
|
||||||
:rtype: str
|
:rtype: str
|
||||||
"""
|
"""
|
||||||
|
|
||||||
formatted_folder = clean_format(self.folder_format, self._get_formatter())
|
formatted_folder = clean_format(
|
||||||
|
self.folder_format, self._get_formatter()
|
||||||
|
)
|
||||||
|
|
||||||
return os.path.join(parent_folder, formatted_folder)
|
return os.path.join(parent_folder, formatted_folder)
|
||||||
|
|
||||||
|
@ -1719,7 +1769,9 @@ class Playlist(Tracklist, Media):
|
||||||
if self.client.source == "qobuz":
|
if self.client.source == "qobuz":
|
||||||
self.name = self.meta["name"]
|
self.name = self.meta["name"]
|
||||||
self.image = self.meta["images"]
|
self.image = self.meta["images"]
|
||||||
self.creator = safe_get(self.meta, "owner", "name", default="Qobuz")
|
self.creator = safe_get(
|
||||||
|
self.meta, "owner", "name", default="Qobuz"
|
||||||
|
)
|
||||||
|
|
||||||
tracklist = self.meta["tracks"]["items"]
|
tracklist = self.meta["tracks"]["items"]
|
||||||
|
|
||||||
|
@ -1729,7 +1781,9 @@ class Playlist(Tracklist, Media):
|
||||||
elif self.client.source == "tidal":
|
elif self.client.source == "tidal":
|
||||||
self.name = self.meta["title"]
|
self.name = self.meta["title"]
|
||||||
self.image = tidal_cover_url(self.meta["image"], 640)
|
self.image = tidal_cover_url(self.meta["image"], 640)
|
||||||
self.creator = safe_get(self.meta, "creator", "name", default="TIDAL")
|
self.creator = safe_get(
|
||||||
|
self.meta, "creator", "name", default="TIDAL"
|
||||||
|
)
|
||||||
|
|
||||||
tracklist = self.meta["tracks"]
|
tracklist = self.meta["tracks"]
|
||||||
|
|
||||||
|
@ -1742,7 +1796,9 @@ class Playlist(Tracklist, Media):
|
||||||
elif self.client.source == "deezer":
|
elif self.client.source == "deezer":
|
||||||
self.name = self.meta["title"]
|
self.name = self.meta["title"]
|
||||||
self.image = self.meta["picture_big"]
|
self.image = self.meta["picture_big"]
|
||||||
self.creator = safe_get(self.meta, "creator", "name", default="Deezer")
|
self.creator = safe_get(
|
||||||
|
self.meta, "creator", "name", default="Deezer"
|
||||||
|
)
|
||||||
|
|
||||||
tracklist = self.meta["tracks"]
|
tracklist = self.meta["tracks"]
|
||||||
|
|
||||||
|
@ -1783,7 +1839,9 @@ class Playlist(Tracklist, Media):
|
||||||
|
|
||||||
logger.debug("Loaded %d tracks from playlist %s", len(self), self.name)
|
logger.debug("Loaded %d tracks from playlist %s", len(self), self.name)
|
||||||
|
|
||||||
def _prepare_download(self, parent_folder: str = "StreamripDownloads", **kwargs):
|
def _prepare_download(
|
||||||
|
self, parent_folder: str = "StreamripDownloads", **kwargs
|
||||||
|
):
|
||||||
if kwargs.get("folder_format"):
|
if kwargs.get("folder_format"):
|
||||||
fname = sanitize_filename(self.name)
|
fname = sanitize_filename(self.name)
|
||||||
self.folder = os.path.join(parent_folder, fname)
|
self.folder = os.path.join(parent_folder, fname)
|
||||||
|
@ -1813,7 +1871,10 @@ class Playlist(Tracklist, Media):
|
||||||
|
|
||||||
item.download(**kwargs)
|
item.download(**kwargs)
|
||||||
|
|
||||||
item.tag(embed_cover=kwargs.get("embed_cover", True))
|
item.tag(
|
||||||
|
embed_cover=kwargs.get("embed_cover", True),
|
||||||
|
exclude_tags=kwargs.get("exclude_tags"),
|
||||||
|
)
|
||||||
|
|
||||||
# if playlist_to_album and self.client.source == "deezer":
|
# if playlist_to_album and self.client.source == "deezer":
|
||||||
# # Because Deezer tracks come pre-tagged, the `set_playlist_to_album`
|
# # Because Deezer tracks come pre-tagged, the `set_playlist_to_album`
|
||||||
|
@ -1995,7 +2056,9 @@ class Artist(Tracklist, Media):
|
||||||
final = self
|
final = self
|
||||||
|
|
||||||
if isinstance(filters, tuple) and self.client.source == "qobuz":
|
if isinstance(filters, tuple) and self.client.source == "qobuz":
|
||||||
filter_funcs = (getattr(self, f"_{filter_}") for filter_ in filters)
|
filter_funcs = (
|
||||||
|
getattr(self, f"_{filter_}") for filter_ in filters
|
||||||
|
)
|
||||||
for func in filter_funcs:
|
for func in filter_funcs:
|
||||||
final = filter(func, final)
|
final = filter(func, final)
|
||||||
|
|
||||||
|
@ -2108,7 +2171,10 @@ class Artist(Tracklist, Media):
|
||||||
best_bd = bit_depth(a["bit_depth"] for a in group)
|
best_bd = bit_depth(a["bit_depth"] for a in group)
|
||||||
best_sr = sampling_rate(a["sampling_rate"] for a in group)
|
best_sr = sampling_rate(a["sampling_rate"] for a in group)
|
||||||
for album in group:
|
for album in group:
|
||||||
if album["bit_depth"] == best_bd and album["sampling_rate"] == best_sr:
|
if (
|
||||||
|
album["bit_depth"] == best_bd
|
||||||
|
and album["sampling_rate"] == best_sr
|
||||||
|
):
|
||||||
yield album
|
yield album
|
||||||
break
|
break
|
||||||
|
|
||||||
|
|
|
@ -131,7 +131,9 @@ class TrackMetadata:
|
||||||
self.album = resp.get("title", "Unknown Album")
|
self.album = resp.get("title", "Unknown Album")
|
||||||
self.tracktotal = resp.get("tracks_count", 1)
|
self.tracktotal = resp.get("tracks_count", 1)
|
||||||
self.genre = resp.get("genres_list") or resp.get("genre") or []
|
self.genre = resp.get("genres_list") or resp.get("genre") or []
|
||||||
self.date = resp.get("release_date_original") or resp.get("release_date")
|
self.date = resp.get("release_date_original") or resp.get(
|
||||||
|
"release_date"
|
||||||
|
)
|
||||||
self.copyright = resp.get("copyright")
|
self.copyright = resp.get("copyright")
|
||||||
self.albumartist = safe_get(resp, "artist", "name")
|
self.albumartist = safe_get(resp, "artist", "name")
|
||||||
self.albumcomposer = safe_get(resp, "composer", "name")
|
self.albumcomposer = safe_get(resp, "composer", "name")
|
||||||
|
@ -140,7 +142,9 @@ class TrackMetadata:
|
||||||
self.disctotal = (
|
self.disctotal = (
|
||||||
max(
|
max(
|
||||||
track.get("media_number", 1)
|
track.get("media_number", 1)
|
||||||
for track in safe_get(resp, "tracks", "items", default=[{}])
|
for track in safe_get(
|
||||||
|
resp, "tracks", "items", default=[{}]
|
||||||
|
)
|
||||||
)
|
)
|
||||||
or 1
|
or 1
|
||||||
)
|
)
|
||||||
|
@ -179,14 +183,22 @@ class TrackMetadata:
|
||||||
self.cover_urls = get_cover_urls(resp, self.__source)
|
self.cover_urls = get_cover_urls(resp, self.__source)
|
||||||
self.streamable = resp.get("allowStreaming", False)
|
self.streamable = resp.get("allowStreaming", False)
|
||||||
|
|
||||||
if q := resp.get("audioQuality"): # for album entries in single tracks
|
if q := resp.get(
|
||||||
|
"audioQuality"
|
||||||
|
): # for album entries in single tracks
|
||||||
self._get_tidal_quality(q)
|
self._get_tidal_quality(q)
|
||||||
|
|
||||||
elif self.__source == "deezer":
|
elif self.__source == "deezer":
|
||||||
self.album = resp.get("title", "Unknown Album")
|
self.album = resp.get("title", "Unknown Album")
|
||||||
self.tracktotal = resp.get("track_total", 0) or resp.get("nb_tracks", 0)
|
self.tracktotal = resp.get("track_total", 0) or resp.get(
|
||||||
|
"nb_tracks", 0
|
||||||
|
)
|
||||||
self.disctotal = (
|
self.disctotal = (
|
||||||
max(track.get("disk_number") for track in resp.get("tracks", [{}])) or 1
|
max(
|
||||||
|
track.get("disk_number")
|
||||||
|
for track in resp.get("tracks", [{}])
|
||||||
|
)
|
||||||
|
or 1
|
||||||
)
|
)
|
||||||
self.genre = safe_get(resp, "genres", "data")
|
self.genre = safe_get(resp, "genres", "data")
|
||||||
self.date = resp.get("release_date")
|
self.date = resp.get("release_date")
|
||||||
|
@ -343,7 +355,9 @@ class TrackMetadata:
|
||||||
|
|
||||||
if isinstance(self._genres, list):
|
if isinstance(self._genres, list):
|
||||||
if self.__source == "qobuz":
|
if self.__source == "qobuz":
|
||||||
genres: Iterable = re.findall(r"([^\u2192\/]+)", "/".join(self._genres))
|
genres: Iterable = re.findall(
|
||||||
|
r"([^\u2192\/]+)", "/".join(self._genres)
|
||||||
|
)
|
||||||
genres = set(genres)
|
genres = set(genres)
|
||||||
elif self.__source == "deezer":
|
elif self.__source == "deezer":
|
||||||
genres = (g["name"] for g in self._genres)
|
genres = (g["name"] for g in self._genres)
|
||||||
|
@ -377,7 +391,9 @@ class TrackMetadata:
|
||||||
if hasattr(self, "_copyright"):
|
if hasattr(self, "_copyright"):
|
||||||
if self._copyright is None:
|
if self._copyright is None:
|
||||||
return None
|
return None
|
||||||
copyright: str = re.sub(r"(?i)\(P\)", PHON_COPYRIGHT, self._copyright)
|
copyright: str = re.sub(
|
||||||
|
r"(?i)\(P\)", PHON_COPYRIGHT, self._copyright
|
||||||
|
)
|
||||||
copyright = re.sub(r"(?i)\(C\)", COPYRIGHT, copyright)
|
copyright = re.sub(r"(?i)\(C\)", COPYRIGHT, copyright)
|
||||||
return copyright
|
return copyright
|
||||||
|
|
||||||
|
@ -437,7 +453,9 @@ class TrackMetadata:
|
||||||
formatter["sampling_rate"] /= 1000
|
formatter["sampling_rate"] /= 1000
|
||||||
return formatter
|
return formatter
|
||||||
|
|
||||||
def tags(self, container: str = "flac") -> Generator:
|
def tags(
|
||||||
|
self, container: str = "flac", exclude: Optional[set] = None
|
||||||
|
) -> Generator:
|
||||||
"""Create a generator of key, value pairs for use with mutagen.
|
"""Create a generator of key, value pairs for use with mutagen.
|
||||||
|
|
||||||
The *_KEY dicts are organized in the format:
|
The *_KEY dicts are organized in the format:
|
||||||
|
@ -459,41 +477,52 @@ class TrackMetadata:
|
||||||
:type container: str
|
:type container: str
|
||||||
:rtype: Generator
|
:rtype: Generator
|
||||||
"""
|
"""
|
||||||
|
if exclude is None:
|
||||||
|
exclude = set()
|
||||||
|
logger.debug("Excluded tags: %s", exclude)
|
||||||
|
|
||||||
container = container.lower()
|
container = container.lower()
|
||||||
if container in ("flac", "vorbis"):
|
if container in ("flac", "vorbis"):
|
||||||
return self.__gen_flac_tags()
|
return self.__gen_flac_tags(exclude)
|
||||||
if container in ("mp3", "id3"):
|
if container in ("mp3", "id3"):
|
||||||
return self.__gen_mp3_tags()
|
return self.__gen_mp3_tags(exclude)
|
||||||
if container in ("alac", "m4a", "mp4", "aac"):
|
if container in ("alac", "m4a", "mp4", "aac"):
|
||||||
return self.__gen_mp4_tags()
|
return self.__gen_mp4_tags(exclude)
|
||||||
|
|
||||||
raise InvalidContainerError(f"Invalid container {container}")
|
raise InvalidContainerError(f"Invalid container {container}")
|
||||||
|
|
||||||
def __gen_flac_tags(self) -> Generator:
|
def __gen_flac_tags(self, exclude: set) -> Generator:
|
||||||
"""Generate key, value pairs to tag FLAC files.
|
"""Generate key, value pairs to tag FLAC files.
|
||||||
|
|
||||||
:rtype: Tuple[str, str]
|
:rtype: Tuple[str, str]
|
||||||
"""
|
"""
|
||||||
for k, v in FLAC_KEY.items():
|
for k, v in FLAC_KEY.items():
|
||||||
|
logger.debug("attr: %s", k)
|
||||||
|
if k in exclude:
|
||||||
|
continue
|
||||||
|
|
||||||
tag = getattr(self, k)
|
tag = getattr(self, k)
|
||||||
if tag:
|
if tag:
|
||||||
if k in (
|
if k in {
|
||||||
"tracknumber",
|
"tracknumber",
|
||||||
"discnumber",
|
"discnumber",
|
||||||
"tracktotal",
|
"tracktotal",
|
||||||
"disctotal",
|
"disctotal",
|
||||||
):
|
}:
|
||||||
tag = f"{int(tag):02}"
|
tag = f"{int(tag):02}"
|
||||||
|
|
||||||
logger.debug("Adding tag %s: %s", v, tag)
|
logger.debug("Adding tag %s: %s", v, tag)
|
||||||
yield (v, str(tag))
|
yield (v, str(tag))
|
||||||
|
|
||||||
def __gen_mp3_tags(self) -> Generator:
|
def __gen_mp3_tags(self, exclude: set) -> Generator:
|
||||||
"""Generate key, value pairs to tag MP3 files.
|
"""Generate key, value pairs to tag MP3 files.
|
||||||
|
|
||||||
:rtype: Tuple[str, str]
|
:rtype: Tuple[str, str]
|
||||||
"""
|
"""
|
||||||
for k, v in MP3_KEY.items():
|
for k, v in MP3_KEY.items():
|
||||||
|
if k in exclude:
|
||||||
|
continue
|
||||||
|
|
||||||
if k == "tracknumber":
|
if k == "tracknumber":
|
||||||
text = f"{self.tracknumber}/{self.tracktotal}"
|
text = f"{self.tracknumber}/{self.tracktotal}"
|
||||||
elif k == "discnumber":
|
elif k == "discnumber":
|
||||||
|
@ -504,12 +533,15 @@ class TrackMetadata:
|
||||||
if text is not None and v is not None:
|
if text is not None and v is not None:
|
||||||
yield (v.__name__, v(encoding=3, text=text))
|
yield (v.__name__, v(encoding=3, text=text))
|
||||||
|
|
||||||
def __gen_mp4_tags(self) -> Generator:
|
def __gen_mp4_tags(self, exclude: set) -> Generator:
|
||||||
"""Generate key, value pairs to tag ALAC or AAC files.
|
"""Generate key, value pairs to tag ALAC or AAC files.
|
||||||
|
|
||||||
:rtype: Tuple[str, str]
|
:rtype: Tuple[str, str]
|
||||||
"""
|
"""
|
||||||
for k, v in MP4_KEY.items():
|
for k, v in MP4_KEY.items():
|
||||||
|
if k in exclude:
|
||||||
|
continue
|
||||||
|
|
||||||
if k == "tracknumber":
|
if k == "tracknumber":
|
||||||
text = [(self.tracknumber, self.tracktotal)]
|
text = [(self.tracknumber, self.tracktotal)]
|
||||||
elif k == "discnumber":
|
elif k == "discnumber":
|
||||||
|
@ -581,7 +613,9 @@ class TrackMetadata:
|
||||||
|
|
||||||
:rtype: int
|
:rtype: int
|
||||||
"""
|
"""
|
||||||
return sum(hash(v) for v in self.asdict().values() if isinstance(v, Hashable))
|
return sum(
|
||||||
|
hash(v) for v in self.asdict().values() if isinstance(v, Hashable)
|
||||||
|
)
|
||||||
|
|
||||||
def __repr__(self) -> str:
|
def __repr__(self) -> str:
|
||||||
"""Return the string representation of the metadata object.
|
"""Return the string representation of the metadata object.
|
||||||
|
|
|
@ -82,7 +82,9 @@ class DownloadStream:
|
||||||
info = self.request.json()
|
info = self.request.json()
|
||||||
try:
|
try:
|
||||||
# Usually happens with deezloader downloads
|
# Usually happens with deezloader downloads
|
||||||
raise NonStreamable(f"{info['error']} -- {info['message']}")
|
raise NonStreamable(
|
||||||
|
f"{info['error']} -- {info['message']}"
|
||||||
|
)
|
||||||
except KeyError:
|
except KeyError:
|
||||||
raise NonStreamable(info)
|
raise NonStreamable(info)
|
||||||
|
|
||||||
|
@ -94,7 +96,10 @@ class DownloadStream:
|
||||||
|
|
||||||
:rtype: Iterator
|
:rtype: Iterator
|
||||||
"""
|
"""
|
||||||
if self.source == "deezer" and self.is_encrypted.search(self.url) is not None:
|
if (
|
||||||
|
self.source == "deezer"
|
||||||
|
and self.is_encrypted.search(self.url) is not None
|
||||||
|
):
|
||||||
assert isinstance(self.id, str), self.id
|
assert isinstance(self.id, str), self.id
|
||||||
|
|
||||||
blowfish_key = self._generate_blowfish_key(self.id)
|
blowfish_key = self._generate_blowfish_key(self.id)
|
||||||
|
@ -121,7 +126,9 @@ class DownloadStream:
|
||||||
return self.file_size
|
return self.file_size
|
||||||
|
|
||||||
def _create_deezer_decryptor(self, key) -> Blowfish:
|
def _create_deezer_decryptor(self, key) -> Blowfish:
|
||||||
return Blowfish.new(key, Blowfish.MODE_CBC, b"\x00\x01\x02\x03\x04\x05\x06\x07")
|
return Blowfish.new(
|
||||||
|
key, Blowfish.MODE_CBC, b"\x00\x01\x02\x03\x04\x05\x06\x07"
|
||||||
|
)
|
||||||
|
|
||||||
@staticmethod
|
@staticmethod
|
||||||
def _generate_blowfish_key(track_id: str):
|
def _generate_blowfish_key(track_id: str):
|
||||||
|
@ -204,7 +211,9 @@ __QUALITY_MAP: Dict[str, Dict[int, Union[int, str, Tuple[int, str]]]] = {
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
def get_quality(quality_id: int, source: str) -> Union[str, int, Tuple[int, str]]:
|
def get_quality(
|
||||||
|
quality_id: int, source: str
|
||||||
|
) -> Union[str, int, Tuple[int, str]]:
|
||||||
"""Get the source-specific quality id.
|
"""Get the source-specific quality id.
|
||||||
|
|
||||||
:param quality_id: the universal quality id (0, 1, 2, 4)
|
:param quality_id: the universal quality id (0, 1, 2, 4)
|
||||||
|
@ -291,7 +300,9 @@ def tidal_cover_url(uuid, size):
|
||||||
possibles = (80, 160, 320, 640, 1280)
|
possibles = (80, 160, 320, 640, 1280)
|
||||||
assert size in possibles, f"size must be in {possibles}"
|
assert size in possibles, f"size must be in {possibles}"
|
||||||
|
|
||||||
return TIDAL_COVER_URL.format(uuid=uuid.replace("-", "/"), height=size, width=size)
|
return TIDAL_COVER_URL.format(
|
||||||
|
uuid=uuid.replace("-", "/"), height=size, width=size
|
||||||
|
)
|
||||||
|
|
||||||
|
|
||||||
def init_log(path: Optional[str] = None, level: str = "DEBUG"):
|
def init_log(path: Optional[str] = None, level: str = "DEBUG"):
|
||||||
|
@ -393,7 +404,9 @@ def gen_threadsafe_session(
|
||||||
headers = {}
|
headers = {}
|
||||||
|
|
||||||
session = requests.Session()
|
session = requests.Session()
|
||||||
adapter = requests.adapters.HTTPAdapter(pool_connections=100, pool_maxsize=100)
|
adapter = requests.adapters.HTTPAdapter(
|
||||||
|
pool_connections=100, pool_maxsize=100
|
||||||
|
)
|
||||||
session.mount("https://", adapter)
|
session.mount("https://", adapter)
|
||||||
session.headers.update(headers)
|
session.headers.update(headers)
|
||||||
return session
|
return session
|
||||||
|
|
Loading…
Add table
Add a link
Reference in a new issue