from collections import defaultdict from enum import Enum from typing import List, Dict, Set, Tuple, Optional, Iterable from urllib.parse import urlparse from ..utils.enums.source import SourcePages, SourceTypes from ..utils.config import youtube_settings from ..utils.string_processing import hash_url from .metadata import Mapping, Metadata from .parents import OuterProxy from .collection import Collection class Source(OuterProxy): url: str page_enum: SourcePages referer_page: SourcePages audio_url: str _default_factories = { "audio_url": lambda: None, } # This is automatically generated def __init__(self, page_enum: SourcePages, url: str, referer_page: SourcePages = None, audio_url: str = None, **kwargs) -> None: if referer_page is None: referer_page = page_enum super().__init__(url=url, page_enum=page_enum, referer_page=referer_page, audio_url=audio_url, **kwargs) @classmethod def match_url(cls, url: str, referer_page: SourcePages) -> Optional["Source"]: """ this shouldn't be used, unlesse you are not certain what the source is for the reason is that it is more inefficient """ parsed = urlparse(url) url = parsed.geturl() if "musify" in parsed.netloc: return cls(SourcePages.MUSIFY, url, referer_page=referer_page) if parsed.netloc in [_url.netloc for _url in youtube_settings['youtube_url']]: return cls(SourcePages.YOUTUBE, url, referer_page=referer_page) if url.startswith("https://www.deezer"): return cls(SourcePages.DEEZER, url, referer_page=referer_page) if url.startswith("https://open.spotify.com"): return cls(SourcePages.SPOTIFY, url, referer_page=referer_page) if "bandcamp" in url: return cls(SourcePages.BANDCAMP, url, referer_page=referer_page) if "wikipedia" in parsed.netloc: return cls(SourcePages.WIKIPEDIA, url, referer_page=referer_page) if url.startswith("https://www.metal-archives.com/"): return cls(SourcePages.ENCYCLOPAEDIA_METALLUM, url, referer_page=referer_page) # the less important once if url.startswith("https://www.facebook"): return cls(SourcePages.FACEBOOK, url, referer_page=referer_page) if url.startswith("https://www.instagram"): return cls(SourcePages.INSTAGRAM, url, referer_page=referer_page) if url.startswith("https://twitter"): return cls(SourcePages.TWITTER, url, referer_page=referer_page) if url.startswith("https://myspace.com"): return cls(SourcePages.MYSPACE, url, referer_page=referer_page) def get_song_metadata(self) -> Metadata: return Metadata({ Mapping.FILE_WEBPAGE_URL: [self.url], Mapping.SOURCE_WEBPAGE_URL: [self.homepage] }) def get_artist_metadata(self) -> Metadata: return Metadata({ Mapping.ARTIST_WEBPAGE_URL: [self.url] }) @property def hash_url(self) -> str: return hash_url(self.url) @property def metadata(self) -> Metadata: return self.get_song_metadata() @property def indexing_values(self) -> List[Tuple[str, object]]: return [ ('id', self.id), ('url', self.url), ('audio_url', self.audio_url), ] def __str__(self): return self.__repr__() def __repr__(self) -> str: return f"Src({self.page_enum.value}: {self.url}, {self.audio_url})" page_str = property(fget=lambda self: self.page_enum.value) type_str = property(fget=lambda self: self.type_enum.value) homepage = property(fget=lambda self: SourcePages.get_homepage(self.page_enum)) class SourceCollection(Collection): def __init__(self, data: Optional[Iterable[Source]] = None, **kwargs): self._page_to_source_list: Dict[SourcePages, List[Source]] = defaultdict(list) super().__init__(data=data, **kwargs) def _map_element(self, __object: Source, **kwargs): super()._map_element(__object, **kwargs) self._page_to_source_list[__object.page_enum].append(__object) @property def source_pages(self) -> Set[SourcePages]: return set(source.page_enum for source in self._data) def get_sources_from_page(self, source_page: SourcePages) -> List[Source]: """ getting the sources for a specific page like YouTube or musify """ return self._page_to_source_list[source_page].copy()