from typing import List, Iterable, Dict from collections import defaultdict from dataclasses import dataclass from .parents import DatabaseObject @dataclass class AppendResult: was_in_collection: bool current_element: DatabaseObject was_the_same: bool class Collection: """ This a class for the iterables like tracklist or discography """ _data: List[DatabaseObject] _by_url: dict _by_attribute: dict def __init__(self, data: List[DatabaseObject] = None, element_type=None, *args, **kwargs) -> None: # Attribute needs to point to self.element_type = element_type self._data: List[DatabaseObject] = list() """ example of attribute_to_object_map the song objects are references pointing to objects in _data ```python { 'id': {323: song_1, 563: song_2, 666: song_3}, 'url': {'www.song_2.com': song_2} } ``` """ self._attribute_to_object_map: Dict[str, Dict[object, DatabaseObject]] = defaultdict(dict) self._used_ids: set = set() if data is not None: self.extend(data, merge_on_conflict=True) def sort(self, reverse: bool = False, **kwargs): self._data.sort(reverse=reverse, **kwargs) def map_element(self, element: DatabaseObject): for name, value in element.indexing_values: if value is None: continue self._attribute_to_object_map[name][value] = element self._used_ids.add(element.id) def unmap_element(self, element: DatabaseObject): for name, value in element.indexing_values: if value is None: continue if value in self._attribute_to_object_map[name]: if element is self._attribute_to_object_map[name][value]: try: self._attribute_to_object_map[name].pop(value) except KeyError: pass def append(self, element: DatabaseObject, merge_on_conflict: bool = True, merge_into_existing: bool = True) -> AppendResult: """ :param element: :param merge_on_conflict: :param merge_into_existing: :return did_not_exist: """ # if the element type has been defined in the initializer it checks if the type matches if self.element_type is not None and not isinstance(element, self.element_type): raise TypeError(f"{type(element)} is not the set type {self.element_type}") # return if the same instance of the object is in the list for existing in self._data: if element is existing: return AppendResult(True, element, True) for name, value in element.indexing_values: if value in self._attribute_to_object_map[name]: existing_object = self._attribute_to_object_map[name][value] if not merge_on_conflict: return AppendResult(True, existing_object, False) # if the object does already exist # thus merging and don't add it afterwards if merge_into_existing: existing_object.merge(element) # in case any relevant data has been added (e.g. it remaps the old object) self.map_element(existing_object) return AppendResult(True, existing_object, False) element.merge(existing_object) exists_at = self._data.index(existing_object) self._data[exists_at] = element self.unmap_element(existing_object) self.map_element(element) return AppendResult(True, existing_object, False) self._data.append(element) self.map_element(element) return AppendResult(False, element, False) def extend(self, element_list: Iterable[DatabaseObject], merge_on_conflict: bool = True, merge_into_existing: bool = True): for element in element_list: self.append(element, merge_on_conflict=merge_on_conflict, merge_into_existing=merge_into_existing) def __iter__(self): for element in self.shallow_list: yield element def __str__(self) -> str: return "\n".join([f"{str(j).zfill(2)}: {i.__repr__()}" for j, i in enumerate(self._data)]) def __len__(self) -> int: return len(self._data) def __getitem__(self, key): if type(key) != int: return ValueError("key needs to be an integer") return self._data[key] def __setitem__(self, key, value: DatabaseObject): if type(key) != int: return ValueError("key needs to be an integer") old_item = self._data[key] self.unmap_element(old_item) self.map_element(value) self._data[key] = value @property def shallow_list(self) -> List[DatabaseObject]: """ returns a shallow copy of the data list """ return self._data.copy() @property def empty(self) -> bool: return len(self._data) == 0