1077 lines
40 KiB
Python
1077 lines
40 KiB
Python
from collections import defaultdict
|
||
from dataclasses import dataclass
|
||
from enum import Enum
|
||
from typing import List, Optional, Type, Union
|
||
from urllib.parse import urlparse
|
||
|
||
import pycountry
|
||
import requests
|
||
from bs4 import BeautifulSoup
|
||
|
||
from ..connection import Connection
|
||
from .abstract import Page
|
||
from ..utils.enums.source import SourcePages
|
||
from ..utils.enums.album import AlbumType, AlbumStatus
|
||
from ..objects import (
|
||
Artist,
|
||
Source,
|
||
Song,
|
||
Album,
|
||
ID3Timestamp,
|
||
FormattedText,
|
||
Label,
|
||
Options,
|
||
Target,
|
||
DatabaseObject
|
||
)
|
||
from ..utils.shared import MUSIFY_LOGGER
|
||
from ..utils import string_processing, shared
|
||
from ..utils.support_classes import DownloadResult, Query
|
||
|
||
"""
|
||
https://musify.club/artist/ghost-bath-280348?_pjax=#bodyContent
|
||
https://musify.club/artist/ghost-bath-280348/releases?_pjax=#bodyContent
|
||
https://musify.club/artist/ghost-bath-280348/clips?_pjax=#bodyContent
|
||
https://musify.club/artist/ghost-bath-280348/photos?_pjax=#bodyContent
|
||
|
||
POST https://musify.club/artist/filtersongs
|
||
ID: 280348
|
||
NameForUrl: ghost-bath
|
||
Page: 1
|
||
IsAllowed: True
|
||
SortOrder.Property: dateCreated
|
||
SortOrder.IsAscending: false
|
||
X-Requested-With: XMLHttpRequest
|
||
|
||
POST https://musify.club/artist/filteralbums
|
||
ArtistID: 280348
|
||
SortOrder.Property: dateCreated
|
||
SortOrder.IsAscending: false
|
||
X-Requested-With: XMLHttpRequest
|
||
"""
|
||
|
||
|
||
class MusifyTypes(Enum):
|
||
ARTIST = "artist"
|
||
RELEASE = "release"
|
||
SONG = "track"
|
||
|
||
|
||
@dataclass
|
||
class MusifyUrl:
|
||
source_type: MusifyTypes
|
||
name_without_id: str
|
||
name_with_id: str
|
||
musify_id: str
|
||
url: str
|
||
|
||
|
||
ALBUM_TYPE_MAP = defaultdict(lambda: AlbumType.OTHER, {
|
||
1: AlbumType.OTHER, # literally other xD
|
||
2: AlbumType.STUDIO_ALBUM,
|
||
3: AlbumType.EP,
|
||
4: AlbumType.SINGLE,
|
||
5: AlbumType.OTHER, # BOOTLEG
|
||
6: AlbumType.LIVE_ALBUM,
|
||
7: AlbumType.COMPILATION_ALBUM, # compilation of different artists
|
||
8: AlbumType.MIXTAPE,
|
||
9: AlbumType.DEMO,
|
||
10: AlbumType.MIXTAPE, # DJ Mixes
|
||
11: AlbumType.COMPILATION_ALBUM, # compilation of only this artist
|
||
12: AlbumType.STUDIO_ALBUM, # split
|
||
13: AlbumType.COMPILATION_ALBUM, # unofficial
|
||
14: AlbumType.MIXTAPE # "Soundtracks"
|
||
})
|
||
|
||
|
||
def parse_url(url: str) -> MusifyUrl:
|
||
parsed = urlparse(url)
|
||
|
||
path = parsed.path.split("/")
|
||
|
||
split_name = path[2].split("-")
|
||
url_id = split_name[-1]
|
||
name_for_url = "-".join(split_name[:-1])
|
||
|
||
try:
|
||
type_enum = MusifyTypes(path[1])
|
||
except ValueError as e:
|
||
MUSIFY_LOGGER.warning(f"{path[1]} is not yet implemented, add it to MusifyTypes")
|
||
raise e
|
||
|
||
return MusifyUrl(
|
||
source_type=type_enum,
|
||
name_without_id=name_for_url,
|
||
name_with_id=path[2],
|
||
musify_id=url_id,
|
||
url=url
|
||
)
|
||
|
||
|
||
class Musify(Page):
|
||
# CHANGE
|
||
SOURCE_TYPE = SourcePages.MUSIFY
|
||
LOGGER = MUSIFY_LOGGER
|
||
|
||
HOST = "https://musify.club"
|
||
|
||
def __init__(self):
|
||
self.connection: Connection = Connection(
|
||
host="https://musify.club/",
|
||
logger=self.LOGGER
|
||
)
|
||
|
||
super().__init__()
|
||
|
||
def get_source_type(self, source: Source) -> Optional[Type[DatabaseObject]]:
|
||
if source.url is None:
|
||
return None
|
||
|
||
musify_url = parse_url(source.url)
|
||
|
||
# Has no labels, because afaik musify has no Labels
|
||
musify_type_to_database_type = {
|
||
MusifyTypes.SONG: Song,
|
||
MusifyTypes.RELEASE: Album,
|
||
MusifyTypes.ARTIST: Artist
|
||
}
|
||
|
||
return musify_type_to_database_type.get(musify_url.source_type)
|
||
|
||
def _parse_artist_contact(self, contact: BeautifulSoup) -> Artist:
|
||
source_list: List[Source] = []
|
||
name = None
|
||
_id = None
|
||
|
||
# source
|
||
anchor = contact.find("a")
|
||
if anchor is not None:
|
||
href = anchor.get("href")
|
||
name = anchor.get("title")
|
||
|
||
if "-" in href:
|
||
_id = href.split("-")[-1]
|
||
|
||
source_list.append(Source(self.SOURCE_TYPE, self.HOST + href))
|
||
|
||
# artist image
|
||
image_soup = contact.find("img")
|
||
if image_soup is not None:
|
||
alt = image_soup.get("alt")
|
||
if alt is not None:
|
||
name = alt
|
||
|
||
artist_thumbnail = image_soup.get("src")
|
||
|
||
return Artist(
|
||
_id=_id,
|
||
name=name,
|
||
source_list=source_list
|
||
)
|
||
|
||
def _parse_album_contact(self, contact: BeautifulSoup) -> Album:
|
||
"""
|
||
<div class="contacts__item">
|
||
<a href="/release/ghost-bath-ghost-bath-2013-602489" title="Ghost Bath - 2013">
|
||
|
||
<div class="contacts__img release">
|
||
<img alt="Ghost Bath" class="lozad" data-src="https://37s.musify.club/img/69/9060265/24178833.jpg"/>
|
||
<noscript><img alt="Ghost Bath" src="https://37s.musify.club/img/69/9060265/24178833.jpg"/></noscript>
|
||
</div>
|
||
|
||
<div class="contacts__info">
|
||
<strong>Ghost Bath - 2013</strong>
|
||
<small>Ghost Bath</small>
|
||
<small>Треков: 4</small> <!--tracks-->
|
||
<small><i class="zmdi zmdi-star zmdi-hc-fw"></i> 9,04</small>
|
||
</div>
|
||
</a>
|
||
</div>
|
||
"""
|
||
|
||
source_list: List[Source] = []
|
||
title = None
|
||
_id = None
|
||
year = None
|
||
artist_list: List[Artist] = []
|
||
|
||
def parse_title_date(title_date: Optional[str], delimiter: str = " - "):
|
||
nonlocal year
|
||
nonlocal title
|
||
|
||
if title_date is None:
|
||
return
|
||
|
||
title_date = title_date.strip()
|
||
split_attr = title_date.split(delimiter)
|
||
|
||
if len(split_attr) < 2:
|
||
return
|
||
if not split_attr[-1].isdigit():
|
||
return
|
||
|
||
year = int(split_attr[-1])
|
||
title = delimiter.join(split_attr[:-1])
|
||
|
||
# source
|
||
anchor = contact.find("a")
|
||
if anchor is not None:
|
||
href = anchor.get("href")
|
||
|
||
# get the title and year
|
||
parse_title_date(anchor.get("title"))
|
||
|
||
if "-" in href:
|
||
_id = href.split("-")[-1]
|
||
|
||
source_list.append(Source(self.SOURCE_TYPE, self.HOST + href))
|
||
|
||
# cover art
|
||
image_soup = contact.find("img")
|
||
if image_soup is not None:
|
||
alt = image_soup.get("alt")
|
||
if alt is not None:
|
||
title = alt
|
||
|
||
cover_art = image_soup.get("src")
|
||
|
||
contact_info_soup = contact.find("div", {"class": "contacts__info"})
|
||
if contact_info_soup is not None:
|
||
"""
|
||
<strong>Ghost Bath - 2013</strong>
|
||
<small>Ghost Bath</small>
|
||
<small>Треков: 4</small> <!--tracks-->
|
||
<small><i class="zmdi zmdi-star zmdi-hc-fw"></i> 9,04</small>
|
||
"""
|
||
|
||
title_soup = contact_info_soup.find("strong")
|
||
if title_soup is None:
|
||
parse_title_date(title_soup)
|
||
|
||
small_list = contact_info_soup.find_all("small")
|
||
if len(small_list) == 3:
|
||
# artist
|
||
artist_soup: BeautifulSoup = small_list[0]
|
||
raw_artist_str = artist_soup.text
|
||
|
||
for artist_str in raw_artist_str.split("&\r\n"):
|
||
artist_str = artist_str.rstrip("& ...\r\n")
|
||
artist_str = artist_str.strip()
|
||
|
||
if artist_str.endswith("]") and "[" in artist_str:
|
||
artist_str = artist_str.rsplit("[", maxsplit=1)[0]
|
||
|
||
artist_list.append(Artist(name=artist_str))
|
||
|
||
track_count_soup: BeautifulSoup = small_list[1]
|
||
rating_soup: BeautifulSoup = small_list[2]
|
||
else:
|
||
self.LOGGER.warning("got an unequal ammount than 3 small elements")
|
||
|
||
return Album(
|
||
_id=_id,
|
||
title=title,
|
||
source_list=source_list,
|
||
date=ID3Timestamp(year=year),
|
||
artist_list=artist_list
|
||
)
|
||
|
||
def _parse_contact_container(self, contact_container_soup: BeautifulSoup) -> List[Union[Artist, Album]]:
|
||
contacts = []
|
||
|
||
contact: BeautifulSoup
|
||
for contact in contact_container_soup.find_all("div", {"class": "contacts__item"}):
|
||
|
||
anchor_soup = contact.find("a")
|
||
|
||
if anchor_soup is not None:
|
||
url = anchor_soup.get("href")
|
||
|
||
if url is not None:
|
||
if "artist" in url:
|
||
contacts.append(self._parse_artist_contact(contact))
|
||
elif "release" in url:
|
||
contacts.append(self._parse_album_contact(contact))
|
||
return contacts
|
||
|
||
def _parse_playlist_item(self, playlist_item_soup: BeautifulSoup) -> Song:
|
||
_id = None
|
||
song_title = playlist_item_soup.get("data-name")
|
||
artist_list: List[Artist] = []
|
||
source_list: List[Source] = []
|
||
|
||
# details
|
||
playlist_details: BeautifulSoup = playlist_item_soup.find("div", {"class", "playlist__heading"})
|
||
if playlist_details is not None:
|
||
anchor_list = playlist_details.find_all("a")
|
||
|
||
if len(anchor_list) >= 2:
|
||
# artists
|
||
artist_anchor: BeautifulSoup
|
||
for artist_anchor in anchor_list[:-1]:
|
||
_id = None
|
||
href = artist_anchor.get("href")
|
||
artist_source: Source = Source(self.SOURCE_TYPE, self.HOST + href)
|
||
if "-" in href:
|
||
_id = href.split("-")[-1]
|
||
|
||
artist_list.append(Artist(
|
||
_id=_id,
|
||
name=artist_anchor.get_text(strip=True),
|
||
source_list=[artist_source]
|
||
))
|
||
|
||
# track
|
||
track_soup: BeautifulSoup = anchor_list[-1]
|
||
"""
|
||
TODO
|
||
this anchor text may have something like (feat. some artist)
|
||
which is not acceptable
|
||
"""
|
||
href = track_soup.get("href")
|
||
if href is not None:
|
||
if "-" in href:
|
||
raw_id: str = href.split("-")[-1]
|
||
if raw_id.isdigit():
|
||
_id = raw_id
|
||
source_list.append(Source(self.SOURCE_TYPE, self.HOST + href))
|
||
|
||
else:
|
||
self.LOGGER.debug("there are not enough anchors (2) for artist and track")
|
||
self.LOGGER.debug(str(artist_list))
|
||
|
||
"""
|
||
artist_name = playlist_item_soup.get("data-artist")
|
||
if artist_name is not None:
|
||
artist_list.append(Artist(name=artist_name))
|
||
"""
|
||
id_attribute = playlist_item_soup.get("id")
|
||
if id_attribute is not None:
|
||
raw_id = id_attribute.replace("playerDiv", "")
|
||
if raw_id.isdigit():
|
||
_id = raw_id
|
||
|
||
return Song(
|
||
_id=_id,
|
||
title=song_title,
|
||
main_artist_list=artist_list,
|
||
source_list=source_list
|
||
)
|
||
|
||
def _parse_playlist_soup(self, playlist_soup: BeautifulSoup) -> List[Song]:
|
||
song_list = []
|
||
|
||
for playlist_item_soup in playlist_soup.find_all("div", {"class": "playlist__item"}):
|
||
song_list.append(self._parse_playlist_item(playlist_item_soup))
|
||
|
||
return song_list
|
||
|
||
def general_search(self, search_query: str) -> List[DatabaseObject]:
|
||
search_results = []
|
||
|
||
r = self.connection.get(f"https://musify.club/search?searchText={search_query}")
|
||
if r is None:
|
||
return []
|
||
search_soup: BeautifulSoup = self.get_soup_from_response(r)
|
||
|
||
# album and songs
|
||
# child of div class: contacts row
|
||
for contact_container_soup in search_soup.find_all("div", {"class": "contacts"}):
|
||
search_results.extend(self._parse_contact_container(contact_container_soup))
|
||
|
||
# song
|
||
# div class: playlist__item
|
||
for playlist_soup in search_soup.find_all("div", {"class": "playlist"}):
|
||
search_results.extend(self._parse_playlist_soup(playlist_soup))
|
||
|
||
return search_results
|
||
|
||
def fetch_song(self, source: Source, stop_at_level: int = 1) -> Song:
|
||
return Song()
|
||
|
||
def _parse_song_card(self, song_card: BeautifulSoup) -> Song:
|
||
"""
|
||
<div id="playerDiv3051" class="playlist__item" itemprop="track" itemscope="itemscope" itemtype="http://schema.org/MusicRecording" data-artist="Linkin Park" data-name="Papercut">
|
||
<div id="play_3051" class="playlist__control play" data-url="/track/play/3051/linkin-park-papercut.mp3" data-position="1" data-title="Linkin Park - Papercut" title="Слушать Linkin Park - Papercut">
|
||
<span class="ico-play"><i class="zmdi zmdi-play-circle-outline zmdi-hc-2-5x"></i></span>
|
||
<span class="ico-pause"><i class="zmdi zmdi-pause-circle-outline zmdi-hc-2-5x"></i></span>
|
||
</div>
|
||
<div class="playlist__position">
|
||
1
|
||
</div>
|
||
<div class="playlist__details">
|
||
<div class="playlist__heading">
|
||
<a href="/artist/linkin-park-5" rel="nofollow">Linkin Park</a> - <a class="strong" href="/track/linkin-park-papercut-3051">Papercut</a>
|
||
<span itemprop="byArtist" itemscope="itemscope" itemtype="http://schema.org/MusicGroup">
|
||
<meta content="/artist/linkin-park-5" itemprop="url" />
|
||
<meta content="Linkin Park" itemprop="name" />
|
||
</span>
|
||
</div>
|
||
</div>
|
||
<div>
|
||
<div class="track__details track__rating hidden-xs-down">
|
||
<span class="text-muted">
|
||
<i class="zmdi zmdi-star-circle zmdi-hc-1-3x" title="Рейтинг"></i>
|
||
326,3K
|
||
</span>
|
||
</div>
|
||
</div>
|
||
<div class="track__details hidden-xs-down">
|
||
<span class="text-muted">03:05</span>
|
||
<span class="text-muted">320 Кб/с</span>
|
||
</div>
|
||
<div class="track__details hidden-xs-down">
|
||
<span title='Есть видео Linkin Park - Papercut'><i class='zmdi zmdi-videocam zmdi-hc-1-3x'></i></span>
|
||
<span title='Есть текст Linkin Park - Papercut'><i class='zmdi zmdi-file-text zmdi-hc-1-3x'></i></span>
|
||
</div>
|
||
<div class="playlist__actions">
|
||
<span class="pl-btn save-to-pl" id="add_3051" title="Сохранить в плейлист"><i class="zmdi zmdi-plus zmdi-hc-1-5x"></i></span>
|
||
<a target="_blank" itemprop="audio" download="Linkin Park - Papercut.mp3" href="/track/dl/3051/linkin-park-papercut.mp3" class="no-ajaxy yaBrowser" id="dl_3051" title='Скачать Linkin Park - Papercut'>
|
||
<span><i class="zmdi zmdi-download zmdi-hc-2-5x"></i></span>
|
||
</a>
|
||
</div>
|
||
</div>
|
||
"""
|
||
song_name = song_card.get("data-name")
|
||
artist_list: List[Artist] = []
|
||
source_list: List[Source] = []
|
||
tracksort = None
|
||
|
||
current_url = None
|
||
|
||
def parse_title(_title: str) -> str:
|
||
return _title
|
||
|
||
"""
|
||
# get from parent div
|
||
_artist_name = song_card.get("data-artist")
|
||
if _artist_name is not None:
|
||
artist_list.append(Artist(name=_artist_name))
|
||
"""
|
||
|
||
# get tracksort
|
||
tracksort_soup: BeautifulSoup = song_card.find("div", {"class": "playlist__position"})
|
||
if tracksort_soup is not None:
|
||
raw_tracksort: str = tracksort_soup.get_text(strip=True)
|
||
if raw_tracksort.isdigit():
|
||
tracksort = int(raw_tracksort)
|
||
|
||
# playlist details
|
||
playlist_details: BeautifulSoup = song_card.find("div", {"class": "playlist__details"})
|
||
if playlist_details is not None:
|
||
"""
|
||
<div class="playlist__heading">
|
||
<a href="/artist/tamas-141317" rel="nofollow">Tamas</a> ft.<a href="/artist/zombiez-630767" rel="nofollow">Zombiez</a> - <a class="strong" href="/track/tamas-zombiez-voodoo-feat-zombiez-16185276">Voodoo (Feat. Zombiez)</a>
|
||
<span itemprop="byArtist" itemscope="itemscope" itemtype="http://schema.org/MusicGroup">
|
||
<meta content="/artist/tamas-141317" itemprop="url" />
|
||
<meta content="Tamas" itemprop="name" />
|
||
</span>
|
||
<span itemprop="byArtist" itemscope="itemscope" itemtype="http://schema.org/MusicGroup">
|
||
<meta content="/artist/zombiez-630767" itemprop="url" />
|
||
<meta content="Zombiez" itemprop="name" />
|
||
</span>
|
||
</div>
|
||
"""
|
||
# track
|
||
anchor_list: List[BeautifulSoup] = playlist_details.find_all("a")
|
||
if len(anchor_list) > 1:
|
||
track_anchor: BeautifulSoup = anchor_list[-1]
|
||
href: str = track_anchor.get("href")
|
||
if href is not None:
|
||
current_url = self.HOST + href
|
||
source_list.append(Source(self.SOURCE_TYPE, self.HOST + href))
|
||
song_name = parse_title(track_anchor.get_text(strip=True))
|
||
|
||
# artist
|
||
artist_span: BeautifulSoup
|
||
for artist_span in playlist_details.find_all("span", {"itemprop": "byArtist"}):
|
||
_artist_src = None
|
||
_artist_name = None
|
||
meta_artist_src = artist_span.find("meta", {"itemprop": "url"})
|
||
if meta_artist_src is not None:
|
||
meta_artist_url = meta_artist_src.get("content")
|
||
if meta_artist_url is not None:
|
||
_artist_src = [Source(self.SOURCE_TYPE, self.HOST + meta_artist_url)]
|
||
|
||
meta_artist_name = artist_span.find("meta", {"itemprop": "name"})
|
||
if meta_artist_name is not None:
|
||
meta_artist_name_text = meta_artist_name.get("content")
|
||
_artist_name = meta_artist_name_text
|
||
|
||
if _artist_name is not None or _artist_src is not None:
|
||
artist_list.append(Artist(name=_artist_name, source_list=_artist_src))
|
||
|
||
# playlist actions
|
||
playlist_actions: BeautifulSoup = song_card.find("div", {"class": "playlist__actions"})
|
||
if playlist_actions is not None:
|
||
"""
|
||
<div class="playlist__actions">
|
||
<span class="pl-btn save-to-pl" id="add_3051" title="Сохранить в плейлист"><i class="zmdi zmdi-plus zmdi-hc-1-5x"></i></span>
|
||
<a target="_blank" itemprop="audio" download="Linkin Park - Papercut.mp3" href="/track/dl/3051/linkin-park-papercut.mp3" class="no-ajaxy yaBrowser" id="dl_3051" title='Скачать Linkin Park - Papercut'>
|
||
<span><i class="zmdi zmdi-download zmdi-hc-2-5x"></i></span>
|
||
</a>
|
||
</div>
|
||
"""
|
||
# getting the actual download link:
|
||
download_anchor = playlist_actions.find("a", {"itemprop": "audio"})
|
||
if download_anchor is not None:
|
||
download_href = download_anchor.get("href")
|
||
if download_href is not None and current_url is not None:
|
||
source_list.append(Source(
|
||
self.SOURCE_TYPE,
|
||
url=current_url,
|
||
adio_url=self.HOST + download_href
|
||
))
|
||
|
||
return Song(
|
||
title=song_name,
|
||
tracksort=tracksort,
|
||
main_artist_list=artist_list,
|
||
source_list=source_list
|
||
)
|
||
|
||
def fetch_album(self, source: Source, stop_at_level: int = 1) -> Album:
|
||
"""
|
||
fetches album from source:
|
||
eg. 'https://musify.club/release/linkin-park-hybrid-theory-2000-188'
|
||
|
||
/html/musify/album_overview.html
|
||
- [x] tracklist
|
||
- [x] attributes
|
||
- [ ] ratings
|
||
|
||
:param stop_at_level:
|
||
:param source:
|
||
:return:
|
||
"""
|
||
|
||
url = parse_url(source.url)
|
||
|
||
endpoint = self.HOST + "/release/" + url.name_with_id
|
||
r = self.connection.get(endpoint)
|
||
if r is None:
|
||
return Album()
|
||
|
||
soup = BeautifulSoup(r.content, "html.parser")
|
||
|
||
album = self._parse_album(soup)
|
||
|
||
# <div class="card"><div class="card-body">...</div></div>
|
||
cards_soup: BeautifulSoup = soup.find("div", {"class": "card-body"})
|
||
if cards_soup is not None:
|
||
card_soup: BeautifulSoup
|
||
for card_soup in cards_soup.find_all("div", {"class": "playlist__item"}):
|
||
new_song = self._parse_song_card(card_soup)
|
||
album.song_collection.append(new_song)
|
||
|
||
if stop_at_level > 1:
|
||
song: Song
|
||
for song in album.song_collection:
|
||
sources = song.source_collection.get_sources_from_page(cls.SOURCE_TYPE)
|
||
for source in sources:
|
||
song.merge(self.fetch_song(source=source))
|
||
|
||
album.update_tracksort()
|
||
|
||
return album
|
||
|
||
def _get_artist_attributes(self, url: MusifyUrl) -> Artist:
|
||
"""
|
||
fetches the main Artist attributes from this endpoint
|
||
https://musify.club/artist/ghost-bath-280348?_pjax=#bodyContent
|
||
it needs to parse html
|
||
|
||
:param url:
|
||
:return:
|
||
"""
|
||
|
||
r = self.connection.get(f"https://musify.club/{url.source_type.value}/{url.name_with_id}?_pjax=#bodyContent")
|
||
if r is None:
|
||
return Artist()
|
||
|
||
soup = self.get_soup_from_response(r)
|
||
|
||
"""
|
||
<ol class="breadcrumb" itemscope="" itemtype="http://schema.org/BreadcrumbList">
|
||
<li class="breadcrumb-item" itemprop="itemListElement" itemscope="" itemtype="http://schema.org/ListItem"><a href="/" itemprop="item"><span itemprop="name">Главная</span><meta content="1" itemprop="position"/></a></li>
|
||
<li class="breadcrumb-item" itemprop="itemListElement" itemscope="" itemtype="http://schema.org/ListItem"><a href="/artist" itemprop="item"><span itemprop="name">Исполнители</span><meta content="2" itemprop="position"/></a></li>
|
||
<li class="breadcrumb-item active">Ghost Bath</li>
|
||
</ol>
|
||
|
||
<ul class="nav nav-tabs nav-fill">
|
||
<li class="nav-item"><a class="active nav-link" href="/artist/ghost-bath-280348">песни (41)</a></li>
|
||
<li class="nav-item"><a class="nav-link" href="/artist/ghost-bath-280348/releases">альбомы (12)</a></li>
|
||
<li class="nav-item"><a class="nav-link" href="/artist/ghost-bath-280348/clips">видеоклипы (23)</a></li>
|
||
<li class="nav-item"><a class="nav-link" href="/artist/ghost-bath-280348/photos">фото (38)</a></li>
|
||
</ul>
|
||
|
||
<header class="content__title">
|
||
<h1>Ghost Bath</h1>
|
||
<div class="actions">
|
||
...
|
||
</div>
|
||
</header>
|
||
|
||
<ul class="icon-list">
|
||
<li>
|
||
<i class="zmdi zmdi-globe zmdi-hc-fw" title="Страна"></i>
|
||
<i class="flag-icon US shadow"></i>
|
||
Соединенные Штаты
|
||
</li>
|
||
</ul>
|
||
"""
|
||
name = None
|
||
source_list: List[Source] = []
|
||
country = None
|
||
notes: FormattedText = FormattedText()
|
||
|
||
breadcrumbs: BeautifulSoup = soup.find("ol", {"class": "breadcrumb"})
|
||
if breadcrumbs is not None:
|
||
breadcrumb_list: List[BeautifulSoup] = breadcrumbs.find_all("li", {"class": "breadcrumb-item"}, recursive=False)
|
||
if len(breadcrumb_list) == 3:
|
||
name = breadcrumb_list[-1].get_text(strip=True)
|
||
else:
|
||
self.LOGGER.debug("breadcrumb layout on artist page changed")
|
||
|
||
nav_tabs: BeautifulSoup = soup.find("ul", {"class": "nav-tabs"})
|
||
if nav_tabs is not None:
|
||
list_item: BeautifulSoup
|
||
for list_item in nav_tabs.find_all("li", {"class": "nav-item"}, recursive=False):
|
||
if not list_item.get_text(strip=True).startswith("песни"):
|
||
# "песни" translates to "songs"
|
||
continue
|
||
|
||
anchor: BeautifulSoup = list_item.find("a")
|
||
if anchor is None:
|
||
continue
|
||
href = anchor.get("href")
|
||
if href is None:
|
||
continue
|
||
|
||
source_list.append(Source(
|
||
self.SOURCE_TYPE,
|
||
self.HOST + href
|
||
))
|
||
|
||
content_title: BeautifulSoup = soup.find("header", {"class": "content__title"})
|
||
if content_title is not None:
|
||
h1_name: BeautifulSoup = content_title.find("h1", recursive=False)
|
||
if h1_name is not None:
|
||
name = h1_name.get_text(strip=True)
|
||
|
||
# country and sources
|
||
icon_list: BeautifulSoup = soup.find("ul", {"class": "icon-list"})
|
||
if icon_list is not None:
|
||
country_italic: BeautifulSoup = icon_list.find("i", {"class", "flag-icon"})
|
||
if country_italic is not None:
|
||
style_classes: set = {'flag-icon', 'shadow'}
|
||
classes: set = set(country_italic.get("class"))
|
||
|
||
country_set: set = classes.difference(style_classes)
|
||
if len(country_set) != 1:
|
||
self.LOGGER.debug("the country set contains multiple values")
|
||
if len(country_set) != 0:
|
||
"""
|
||
This is the css file, where all flags that can be used on musify
|
||
are laid out and styled.
|
||
Every flag has two upper case letters, thus I assume they follow the alpha_2
|
||
https://musify.club/content/flags.min.css
|
||
"""
|
||
|
||
country = pycountry.countries.get(alpha_2=list(country_set)[0])
|
||
|
||
# get all additional sources
|
||
additional_source: BeautifulSoup
|
||
for additional_source in icon_list.find_all("a", {"class", "link"}):
|
||
href = additional_source.get("href")
|
||
if href is None:
|
||
continue
|
||
new_src = Source.match_url(href, referer_page=self.SOURCE_TYPE)
|
||
if new_src is None:
|
||
continue
|
||
source_list.append(new_src)
|
||
|
||
note_soup: BeautifulSoup = soup.find(id="text-main")
|
||
if note_soup is not None:
|
||
notes.html = note_soup.decode_contents()
|
||
|
||
return Artist(
|
||
name=name,
|
||
country=country,
|
||
source_list=source_list,
|
||
notes=notes
|
||
)
|
||
|
||
def _parse_album_card(self, album_card: BeautifulSoup, artist_name: str = None) -> Album:
|
||
"""
|
||
<div class="card release-thumbnail" data-type="2">
|
||
<a href="/release/ghost-bath-self-loather-2021-1554266">
|
||
<img alt="Self Loather" class="card-img-top lozad" data-src="https://40s-a.musify.club/img/70/24826582/62624396.jpg"/>
|
||
<noscript><img alt="Self Loather" src="https://40s-a.musify.club/img/70/24826582/62624396.jpg"/></noscript>
|
||
</a>
|
||
<div class="card-body">
|
||
<h4 class="card-subtitle">
|
||
<a href="/release/ghost-bath-self-loather-2021-1554266">Self Loather</a>
|
||
</h4>
|
||
</div>
|
||
<div class="card-footer"><p class="card-text"><a href="/albums/2021">2021</a></p></div>
|
||
<div class="card-footer">
|
||
<p class="card-text genre__labels">
|
||
<a href="/genre/depressive-black-132">Depressive Black</a><a href="/genre/post-black-metal-295">Post-Black Metal</a> </p>
|
||
</div>
|
||
<div class="card-footer">
|
||
<small><i class="zmdi zmdi-calendar" title="Добавлено"></i> 13.11.2021</small>
|
||
<small><i class="zmdi zmdi-star zmdi-hc-fw" title="Рейтинг"></i> 5,88</small>
|
||
</div>
|
||
</div>
|
||
"""
|
||
|
||
_id: Optional[str] = None
|
||
name: str = None
|
||
source_list: List[Source] = []
|
||
timestamp: Optional[ID3Timestamp] = None
|
||
album_status = None
|
||
|
||
def set_name(new_name: str):
|
||
nonlocal name
|
||
nonlocal artist_name
|
||
|
||
# example of just setting not working:
|
||
# https://musify.club/release/unjoy-eurythmie-psychonaut-4-tired-numb-still-alive-2012-324067
|
||
if new_name.count(" - ") != 1:
|
||
name = new_name
|
||
return
|
||
|
||
potential_artist_list, potential_name = new_name.split(" - ")
|
||
unified_artist_list = string_processing.unify(potential_artist_list)
|
||
if artist_name is not None:
|
||
if string_processing.unify(artist_name) not in unified_artist_list:
|
||
name = new_name
|
||
return
|
||
|
||
name = potential_name
|
||
return
|
||
|
||
name = new_name
|
||
|
||
album_status_id = album_card.get("data-type")
|
||
if album_status_id.isdigit():
|
||
album_status_id = int(album_status_id)
|
||
album_type = ALBUM_TYPE_MAP[album_status_id]
|
||
|
||
if album_status_id == 5:
|
||
album_status = AlbumStatus.BOOTLEG
|
||
|
||
def parse_release_anchor(_anchor: BeautifulSoup, text_is_name=False):
|
||
nonlocal _id
|
||
nonlocal name
|
||
nonlocal source_list
|
||
|
||
if _anchor is None:
|
||
return
|
||
|
||
href = _anchor.get("href")
|
||
if href is not None:
|
||
# add url to sources
|
||
source_list.append(Source(
|
||
self.SOURCE_TYPE,
|
||
self.HOST + href
|
||
))
|
||
|
||
# split id from url
|
||
split_href = href.split("-")
|
||
if len(split_href) > 1:
|
||
_id = split_href[-1]
|
||
|
||
if not text_is_name:
|
||
return
|
||
|
||
set_name(_anchor.text)
|
||
|
||
anchor_list = album_card.find_all("a", recursive=False)
|
||
if len(anchor_list) > 0:
|
||
anchor = anchor_list[0]
|
||
parse_release_anchor(anchor)
|
||
|
||
thumbnail: BeautifulSoup = anchor.find("img")
|
||
if thumbnail is not None:
|
||
alt = thumbnail.get("alt")
|
||
if alt is not None:
|
||
set_name(alt)
|
||
|
||
image_url = thumbnail.get("src")
|
||
else:
|
||
self.LOGGER.debug("the card has no thumbnail or url")
|
||
|
||
card_body = album_card.find("div", {"class": "card-body"})
|
||
if card_body is not None:
|
||
parse_release_anchor(card_body.find("a"), text_is_name=True)
|
||
|
||
def parse_small_date(small_soup: BeautifulSoup):
|
||
"""
|
||
<small>
|
||
<i class="zmdi zmdi-calendar" title="Добавлено"></i>
|
||
13.11.2021
|
||
</small>
|
||
"""
|
||
nonlocal timestamp
|
||
|
||
italic_tagging_soup: BeautifulSoup = small_soup.find("i")
|
||
if italic_tagging_soup is None:
|
||
return
|
||
if italic_tagging_soup.get("title") != "Добавлено":
|
||
# "Добавлено" can be translated to "Added (at)"
|
||
return
|
||
|
||
raw_time = small_soup.text.strip()
|
||
timestamp = ID3Timestamp.strptime(raw_time, "%d.%m.%Y")
|
||
|
||
# parse small date
|
||
card_footer_list = album_card.find_all("div", {"class": "card-footer"})
|
||
if len(card_footer_list) != 3:
|
||
self.LOGGER.debug("there are not exactly 3 card footers in a card")
|
||
|
||
if len(card_footer_list) > 0:
|
||
for any_small_soup in card_footer_list[-1].find_all("small"):
|
||
parse_small_date(any_small_soup)
|
||
else:
|
||
self.LOGGER.debug("there is not even 1 footer in the album card")
|
||
|
||
return Album(
|
||
_id=_id,
|
||
title=name,
|
||
source_list=source_list,
|
||
date=timestamp,
|
||
album_type=album_type,
|
||
album_status=album_status
|
||
)
|
||
|
||
def _parse_album(self, soup: BeautifulSoup) -> Album:
|
||
name: str = None
|
||
source_list: List[Source] = []
|
||
artist_list: List[Artist] = []
|
||
date: ID3Timestamp = None
|
||
|
||
"""
|
||
if breadcrumb list has 4 elements, then
|
||
the -2 is the artist link,
|
||
the -1 is the album
|
||
"""
|
||
# breadcrumb
|
||
breadcrumb_soup: BeautifulSoup = soup.find("ol", {"class", "breadcrumb"})
|
||
breadcrumb_elements: List[BeautifulSoup] = breadcrumb_soup.find_all("li", {"class": "breadcrumb-item"})
|
||
if len(breadcrumb_elements) == 4:
|
||
# album
|
||
album_crumb: BeautifulSoup = breadcrumb_elements[-1]
|
||
name = album_crumb.text.strip()
|
||
|
||
# artist
|
||
artist_crumb: BeautifulSoup = breadcrumb_elements[-2]
|
||
anchor: BeautifulSoup = artist_crumb.find("a")
|
||
if anchor is not None:
|
||
href = anchor.get("href")
|
||
artist_source_list: List[Source] = []
|
||
|
||
if href is not None:
|
||
artist_source_list.append(Source(self.SOURCE_TYPE, self.HOST + href.strip()))
|
||
|
||
span: BeautifulSoup = anchor.find("span")
|
||
if span is not None:
|
||
artist_list.append(Artist(
|
||
name=span.get_text(strip=True),
|
||
source_list=artist_source_list
|
||
))
|
||
else:
|
||
self.LOGGER.debug("there are not 4 breadcrumb items, which shouldn't be the case")
|
||
|
||
# meta
|
||
meta_url: BeautifulSoup = soup.find("meta", {"itemprop": "url"})
|
||
if meta_url is not None:
|
||
url = meta_url.get("content")
|
||
if url is not None:
|
||
source_list.append(Source(self.SOURCE_TYPE, self.HOST + url))
|
||
|
||
meta_name: BeautifulSoup = soup.find("meta", {"itemprop": "name"})
|
||
if meta_name is not None:
|
||
_name = meta_name.get("content")
|
||
if _name is not None:
|
||
name = _name
|
||
|
||
# album info
|
||
album_info_ul: BeautifulSoup = soup.find("ul", {"class": "album-info"})
|
||
if album_info_ul is not None:
|
||
artist_anchor: BeautifulSoup
|
||
for artist_anchor in album_info_ul.find_all("a", {"itemprop": "byArtist"}):
|
||
# line 98
|
||
artist_source_list: List[Source] = []
|
||
|
||
artist_url_meta = artist_anchor.find("meta", {"itemprop": "url"})
|
||
if artist_url_meta is not None:
|
||
artist_href = artist_url_meta.get("content")
|
||
if artist_href is not None:
|
||
artist_source_list.append(Source(self.SOURCE_TYPE, url=self.HOST + artist_href))
|
||
|
||
artist_meta_name = artist_anchor.find("meta", {"itemprop": "name"})
|
||
if artist_meta_name is not None:
|
||
artist_name = artist_meta_name.get("content")
|
||
if artist_name is not None:
|
||
artist_list.append(Artist(
|
||
name=artist_name,
|
||
source_list=artist_source_list
|
||
))
|
||
|
||
time_soup: BeautifulSoup = album_info_ul.find("time", {"itemprop": "datePublished"})
|
||
if time_soup is not None:
|
||
raw_datetime = time_soup.get("datetime")
|
||
if raw_datetime is not None:
|
||
try:
|
||
date = ID3Timestamp.strptime(raw_datetime, "%Y-%m-%d")
|
||
except ValueError:
|
||
self.LOGGER.debug(f"Raw datetime doesn't match time format %Y-%m-%d: {raw_datetime}")
|
||
|
||
return Album(
|
||
title=name,
|
||
source_list=source_list,
|
||
artist_list=artist_list,
|
||
date=date
|
||
)
|
||
|
||
def _get_discography(self, url: MusifyUrl, artist_name: str = None, stop_at_level: int = 1) -> List[Album]:
|
||
"""
|
||
POST https://musify.club/artist/filteralbums
|
||
ArtistID: 280348
|
||
SortOrder.Property: dateCreated
|
||
SortOrder.IsAscending: false
|
||
X-Requested-With: XMLHttpRequest
|
||
"""
|
||
|
||
endpoint = self.HOST + "/" + url.source_type.value + "/filteralbums"
|
||
|
||
r = self.connection.post(url=endpoint, json={
|
||
"ArtistID": str(url.musify_id),
|
||
"SortOrder.Property": "dateCreated",
|
||
"SortOrder.IsAscending": False,
|
||
"X-Requested-With": "XMLHttpRequest"
|
||
})
|
||
if r is None:
|
||
return []
|
||
soup: BeautifulSoup = BeautifulSoup(r.content, features="html.parser")
|
||
|
||
discography: List[Album] = []
|
||
for card_soup in soup.find_all("div", {"class": "card"}):
|
||
new_album: Album = self._parse_album_card(card_soup, artist_name)
|
||
album_source: Source
|
||
|
||
if stop_at_level > 1:
|
||
for album_source in new_album.source_collection.get_sources_from_page(self.SOURCE_TYPE):
|
||
new_album.merge(self.fetch_album(album_source, stop_at_level=stop_at_level-1))
|
||
|
||
discography.append(new_album)
|
||
|
||
return discography
|
||
|
||
def fetch_artist(self, source: Source, stop_at_level: int = 1) -> Artist:
|
||
"""
|
||
fetches artist from source
|
||
|
||
[x] discography
|
||
[x] attributes
|
||
[] picture gallery
|
||
|
||
Args:
|
||
source (Source): the source to fetch
|
||
stop_at_level: int = 1: if it is false, every album from discograohy will be fetched. Defaults to False.
|
||
|
||
Returns:
|
||
Artist: the artist fetched
|
||
"""
|
||
|
||
url = parse_url(source.url)
|
||
|
||
artist = self._get_artist_attributes(url)
|
||
|
||
discography: List[Album] = self._get_discography(url, artist.name)
|
||
artist.main_album_collection.extend(discography)
|
||
|
||
return artist
|
||
|
||
def fetch_label(self, source: Source, stop_at_level: int = 1) -> Label:
|
||
return Label()
|
||
|
||
|
||
class OldMusify(Page):
|
||
API_SESSION: requests.Session = requests.Session()
|
||
API_SESSION.headers = {
|
||
"User-Agent": "Mozilla/5.0 (X11; Linux x86_64; rv:106.0) Gecko/20100101 Firefox/106.0",
|
||
"Connection": "keep-alive",
|
||
"Referer": "https://musify.club/"
|
||
}
|
||
API_SESSION.proxies = shared.proxies
|
||
TIMEOUT = 7
|
||
POST_TIMEOUT = 15
|
||
TRIES = 5
|
||
HOST = "https://musify.club"
|
||
|
||
CONNECTION = Connection(
|
||
host="https://musify.club/",
|
||
logger=MUSIFY_LOGGER
|
||
)
|
||
|
||
SOURCE_TYPE = SourcePages.MUSIFY
|
||
|
||
LOGGER = shared.MUSIFY_LOGGER
|
||
|
||
|
||
|
||
|
||
|
||
@classmethod
|
||
def get_plaintext_query(cls, query: Query) -> str:
|
||
if query.album is None:
|
||
return f"{query.artist or '*'} - {query.song or '*'}"
|
||
return f"{query.artist or '*'} - {query.album or '*'} - {query.song or '*'}"
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
|
||
@classmethod
|
||
def _get_type_of_url(cls, url: str) -> Optional[Union[Type[Song], Type[Album], Type[Artist], Type[Label]]]:
|
||
url: MusifyUrl = cls.parse_url(url)
|
||
|
||
if url.source_type == MusifyTypes.ARTIST:
|
||
return Artist
|
||
if url.source_type == MusifyTypes.RELEASE:
|
||
return Album
|
||
if url.source_type == MusifyTypes.SONG:
|
||
return Song
|
||
return None
|
||
|
||
@classmethod
|
||
def _download_song_to_targets(cls, source: Source, target: Target, desc: str = None) -> DownloadResult:
|
||
"""
|
||
https://musify.club/track/im-in-a-coffin-life-never-was-waste-of-skin-16360302
|
||
https://musify.club/track/dl/16360302/im-in-a-coffin-life-never-was-waste-of-skin.mp3
|
||
"""
|
||
endpoint = source.audio_url
|
||
|
||
if source.audio_url is None:
|
||
url: MusifyUrl = cls.parse_url(source.url)
|
||
if url.source_type != MusifyTypes.SONG:
|
||
return DownloadResult(error_message=f"The url is not of the type Song: {source.url}")
|
||
|
||
endpoint = f"https://musify.club/track/dl/{url.musify_id}/{url.name_without_id}.mp3"
|
||
|
||
cls.LOGGER.warning(f"The source has no audio link. Falling back to {endpoint}.")
|
||
|
||
r = cls.CONNECTION.get(endpoint, stream=True, raw_url=True)
|
||
if r is None:
|
||
return DownloadResult(error_message=f"couldn't connect to {endpoint}")
|
||
|
||
if target.stream_into(r, desc=desc):
|
||
return DownloadResult(total=1)
|
||
return DownloadResult(error_message=f"Streaming to the file went wrong: {endpoint}, {str(target.file_path)}")
|