streamlined/server/entertainment_decider/models/entities.py

from __future__ import annotations

from abc import abstractproperty
from dataclasses import dataclass
from datetime import datetime, timedelta
from functools import cache
import itertools
import logging
import re
from typing import (
    Iterable,
    List,
    Mapping,
    Optional,
    Sequence,
    Set,
    Tuple,
    TypeVar,
)

import magic
import requests
from pony import orm

from .custom_types import Query, SafeStr
from .thumbnails import THUMBNAIL_ALLOWED_TYPES, THUMBNAIL_HEADERS
from .extras import UriHolder
from ..common import trim

db = orm.Database()


T = TypeVar("T")


####
## Model Extensions
####


class Tagable:

    ## abstracted

    @abstractproperty
    def orm_assigned_tags(self) -> Query[Tag]:
        """
        Tags which are directly assigned to this object by the user or automatic actions.
        """

    @property
    def orm_inherited_tags(self) -> Query[Tag]:
        """
        Tags, which are inherited by any other means than super/sub-tag relationships.
        This relationship does not declare a distance between this tags and assigned tags.
        """
        return []  # TODO to orm compatible

    @property
    def orm_super_tags(self) -> Query[Tag]:
        """
        Tags, which are inherited only by super/sub-tag relationships.
        This relationship does declare a distance between this tags and assigned tags.
        """
        return []  # TODO to orm compatible

    ## implemented

    @property
    def assigned_tags(self) -> Set[Tag]:
        return set(self.orm_assigned_tags)

    @property
    def inherited_tags(self) -> Set[Tag]:
        return set(self.orm_inherited_tags)

    @property
    def super_tags(self) -> Set[Tag]:
        return set(self.orm_super_tags)

    @property
    def orm_direct_tags(self) -> Query[Tag]:
        return self.orm_assigned_tags + self.orm_inherited_tags

    @property
    def direct_tags(self) -> Set[Tag]:
        return set(self.orm_direct_tags)

    @property
    def all_tags(self) -> Set[Tag]:
        queue: List[Tag] = list(self.direct_tags)
        used: Set[Tag] = set(queue)
        while queue:
            tag = queue.pop(0)
            new_tags = tag.super_tags - used
            queue.extend(new_tags)
            used |= new_tags
        return used


@dataclass
class CollectionStats:

    to_watch_count: int
    ignored_count: int  # but not watched
    watched_count: int

    to_watch_seconds: int
    ignored_seconds: int  # but not watched
    watched_seconds: int

    @property
    def full_count(self) -> int:
        return self.to_watch_count + self.ignored_count + self.watched_count

    @property
    def full_seconds(self) -> int:
        return self.to_watch_seconds + self.ignored_seconds + self.watched_seconds

    @classmethod
    def from_collection(cls, collection: MediaCollection) -> CollectionStats:
        to_watch_count = 0
        ignored_count = 0
        watched_count = 0
        to_watch_seconds = 0
        ignored_seconds = 0
        watched_seconds = 0
        for link in collection.media_links:
            media = link.element
            if media.watched:
                watched_count += 1
                watched_seconds += media.length
            else:
                watched_seconds += media.progress
                if media.ignored:
                    ignored_count += 1
                    ignored_seconds += media.left_length
                else:
                    to_watch_count += 1
                    to_watch_seconds += media.left_length
        return CollectionStats(
            to_watch_count=to_watch_count,
            ignored_count=ignored_count,
            watched_count=watched_count,
            to_watch_seconds=to_watch_seconds,
            ignored_seconds=ignored_seconds,
            watched_seconds=watched_seconds,
        )


####
## Models
####


## Tag & Selection Score's


TEMPORARY_TAGS_IDENTIFIER = (
    "automatic_temporary_tag:82e4509f-e262-463f-8ee5-140ca400ea79"
)
"""random static UUID for identification

This string shall not be parsed and only used as a whole.
"""


class Tag(db.Entity, Tagable):
    @classmethod
    def gen_temporary_tag(cls, hint: str) -> Tag:
        """Generates a new, unique and temporary tag. Required for some algorithms."""
        return Tag(
            title=f"[A] {hint}",
            notes=TEMPORARY_TAGS_IDENTIFIER,
            use_for_preferences=True,
        )

    @classmethod
    def scrub_temporary_tags(cls) -> int:
        """Scrubs all temporary tags, which where left over because of errors."""
        count = orm.delete(tag for tag in cls if tag.notes == TEMPORARY_TAGS_IDENTIFIER)
        return count if isinstance(count, int) else 0

    id: int = orm.PrimaryKey(int, auto=True)

    title: str = orm.Required(str)
    notes: Optional[str] = orm.Optional(
        str,
        nullable=True,
    )

    use_for_preferences: bool = orm.Required(bool, default=True)

    super_tag_list: Iterable[Tag] = orm.Set(lambda: Tag, reverse="sub_tag_list")
    sub_tag_list: Iterable[Tag] = orm.Set(lambda: Tag, reverse="super_tag_list")

    _collection_list: Iterable[MediaCollection] = orm.Set(lambda: MediaCollection)
    _media_list: Iterable[MediaElement] = orm.Set(lambda: MediaElement)

    @property
    def orm_assigned_tags(self) -> Query[Tag]:
        return [self]

    @property
    def orm_super_tags(self) -> Query[Tag]:
        return self.super_tag_list


## Element <-> Collection Linking


class MediaCollectionLink(db.Entity):

    collection: MediaCollection = orm.Required(lambda: MediaCollection)
    element: MediaElement = orm.Required(lambda: MediaElement)
    orm.PrimaryKey(collection, element)
    season: int = orm.Required(int, default=0)
    episode: int = orm.Required(int, default=0)
    orm.composite_index(season, episode)

    @property
    def element_id(self) -> int:
        return self.element.id

    @property
    def element_release_date(self) -> datetime:
        return self.element.release_date

    @staticmethod
    def sort_key(link: MediaCollectionLink) -> Tuple:
        return (
            link.season,
            link.episode,
            link.element.release_date,
            link.element.id,
        )

    @staticmethod
    def desc_sort_key(link: MediaCollectionLink) -> Tuple:
        """
        Only working for Pony calls
        """
        return (
            orm.desc(link.season),
            orm.desc(link.episode),
            orm.desc(link.element.release_date),
            orm.desc(link.element.id),
        )

    @staticmethod
    def sorted(iterable: Iterable[MediaCollectionLink]) -> List[MediaCollectionLink]:
        return sorted(iterable, key=MediaCollectionLink.sort_key)


## Media Elements


class MediaElement(db.Entity, UriHolder, Tagable):

    ### columns

    id: int = orm.PrimaryKey(
        int,
        auto=True,
    )
    uri: str = orm.Required(
        str,
        unique=True,
    )

    title: str = orm.Optional(str)
    description: Optional[str] = orm.Optional(
        orm.LongStr,
        nullable=True,
    )
    thumbnail: Optional[MediaThumbnail] = orm.Optional(
        lambda: MediaThumbnail,
        nullable=True,
    )
    notes: str = orm.Optional(str)
    release_date: datetime = orm.Optional(
        datetime,
        index=True,
    )

    extractor_name: str = orm.Optional(str)
    extractor_key: str = orm.Optional(str)
    orm.composite_index(extractor_name, extractor_key)
    last_updated: datetime = orm.Optional(datetime)

    watched: bool = orm.Required(
        bool,
        column="watched",
        default=False,
        index=False,  # does not need index because of composite index below
    )
    ignored: bool = orm.Required(
        bool,
        column="ignored",
        default=False,
        index=True,
    )
    orm.composite_index(watched, ignored)
    progress: int = orm.Required(
        int,
        default=0,
    )
    length: int = orm.Optional(int)

    tag_list: Iterable[Tag] = orm.Set(
        lambda: Tag,
    )
    uris: Iterable[MediaUriMapping] = orm.Set(
        lambda: MediaUriMapping,
    )
    collection_links: Iterable[MediaCollectionLink] = orm.Set(
        lambda: MediaCollectionLink,
    )

    blocked_by: Set[MediaElement] = orm.Set(
        lambda: MediaElement,
        reverse="is_blocking",
    )
    is_blocking: Set[MediaElement] = orm.Set(
        lambda: MediaElement,
        reverse="blocked_by",
    )

    ### for UriHolder

    @property
    def _primary_uri(self) -> str:
        return self.uri

    def _set_primary_uri(self, uri: str) -> None:
        self.uri = uri

    @property
    def _get_uri_set(self) -> Set[str]:
        return {m.uri for m in self.uris}

    def _set_uri_set(self, uri_set: Set[str]) -> None:
        self.uris = set()
        self.add_uris(uri_set)

    ### for Tagable

    @property
    def orm_assigned_tags(self) -> Query[Tag]:
        return self.tag_list

    @property
    def orm_inherited_tags(self) -> Query[Tag]:
        # TODO: return orm.select(tag for link in self.collection_links for tag in link.collection.orm_direct_tags)
        return [
            tag for link in self.collection_links for tag in link.collection.direct_tags
        ]

    ### properties

    @property
    def was_extracted(self) -> bool:
        return self.last_updated is not None

    @property
    def _left_length(self) -> int:
        return self.length - self.progress

    @property
    def left_length(self) -> int:
        return 0 if self.watched else self._left_length

    @property
    def started(self) -> bool:
        return not self.skip_over and self.progress != 0

    @property
    def average_release_per_week(self) -> float:
        return (
            orm.avg(
                link.collection.average_release_per_week
                for link in self.collection_links
                if link.collection.watch_in_order
            )
            or 0
        )

    @property
    def average_release_per_week_now(self) -> float:
        return (
            orm.avg(
                link.collection.average_release_per_week_now
                for link in self.collection_links
                if link.collection.watch_in_order
            )
            or 0
        )

    @property
    def ignored_recursive(self) -> bool:
        return (
            orm.count(
                link
                for link in self.collection_links
                if link.collection.ignored == True
            )
            > 0
        )

    @property
    def ignored_any(self) -> bool:
        return self.ignored or self.ignored_recursive

    @property
    def skip_over(self) -> bool:
        return self.ignored or self.watched

    @property
    def can_considered(self) -> bool:
        DIRECT_SQL = True
        if DIRECT_SQL:
            return is_considered(self.id)
        if self.skip_over:
            return False
        if self.release_date > datetime.now():
            return False
        if orm.exists(e for e in self.blocked_by if not e.skip_over):
            return False
        ordered_collections: Query[MediaCollection] = orm.select(
            l.collection for l in self.collection_links if l.collection.watch_in_order
        )
        for collection in ordered_collections:
            next = collection.next_episode
            if next is not None and self != next.element:
                return False
        return True

    @property
    def detected_creators(self) -> Query[MediaCollectionLink]:
        return orm.select(
            link for link in self.collection_links if link.collection.is_creator
        )

    @property
    def detected_playlists(self) -> Query[MediaCollectionLink]:
        return orm.select(
            link for link in self.collection_links if link.collection.watch_in_order
        )

    @property
    def info_link(self) -> str:
        return f"/media/{self.id}"

    ### methods

    def merge_to(self, other: MediaElement) -> None:
        if self.watched:
            other.watched = True
        if self.ignored:
            other.ignored = True
        if self.progress >= 0 and other.progress <= 0:
            other.progress = self.progress
        for uri_map in self.uris:
            uri_map.element = other
        for link in self.collection_links:
            if not MediaCollectionLink.get(collection=link.collection, element=other):
                link.element = other
        self.delete()  # will also delete still existing uri mappings and collection links
        orm.flush()

    def add_single_uri(self, uri: str) -> bool:
        mapping: MediaUriMapping = MediaUriMapping.get(uri=uri)
        if not mapping:
            logging.debug(f"Add URI mapping {uri!r} to media {self.id!r}")
            MediaUriMapping(
                uri=uri,
                element=self,
            )
            return True
        if mapping.element != self:
            raise Exception(
                f"URI duplicated for two different media's: {uri}"
            )  # TODO may replace with merge call
        return False

    def before_insert(self) -> None:
        self.before_update()

    def before_update(self) -> None:
        self.add_single_uri(self.uri)


class MediaThumbnail(db.Entity):

    id: int = orm.PrimaryKey(
        int,
        auto=True,
    )
    uri: str = orm.Required(
        str,
        unique=True,
    )
    last_downloaded: datetime = orm.Optional(
        datetime,
        default=None,
        nullable=True,
    )
    last_accessed: datetime = orm.Optional(
        datetime,
        default=None,
        nullable=True,
    )
    mime_type: str = orm.Optional(
        str,
        default="",
    )
    data: bytes = orm.Optional(
        bytes,
        default=None,
        nullable=True,
        lazy=True,  # do not always preload huge image data
    )

    elements: Set[MediaElement] = orm.Set(lambda: MediaElement)

    @classmethod
    def from_uri(cls, uri: str) -> MediaThumbnail:
        return cls.get(uri=uri) or MediaThumbnail(uri=uri)

    def access(self) -> None:
        self.last_accessed = datetime.now()

    def download(self) -> bool:
        res = requests.get(url=self.uri, headers=THUMBNAIL_HEADERS)
        mime = magic.from_buffer(res.content, mime=True)
        if mime not in THUMBNAIL_ALLOWED_TYPES:
            return False
        self.mime_type = mime
        self.data = res.content
        self.last_downloaded = datetime.now()
        return True

    def prepare(self) -> bool:
        if self.last_downloaded is not None:
            return True
        return self.download()

    def access_data(self) -> None:
        self.prepare()
        self.access()

    def receive_data(self) -> bytes:
        self.access_data()
        return self.data


class MediaUriMapping(db.Entity):

    id: int = orm.PrimaryKey(int, auto=True)
    uri: str = orm.Required(str, unique=True)
    element: MediaElement = orm.Required(MediaElement)


## Media Collections


class MediaCollection(db.Entity, UriHolder, Tagable):

    ### columns

    id: int = orm.PrimaryKey(
        int,
        auto=True,
    )
    uri: str = orm.Required(
        str,
        unique=True,
    )

    title: str = orm.Optional(str)
    description: str = orm.Optional(
        orm.LongStr,
        nullable=True,
    )
    notes: str = orm.Optional(str)
    release_date: Optional[datetime] = orm.Optional(
        datetime,
        nullable=True,
    )
    creator: Optional[MediaCollection] = orm.Optional(
        lambda: MediaCollection,
        nullable=True,
    )

    extractor_name: str = orm.Optional(str)
    extractor_key: str = orm.Optional(str)
    orm.composite_index(extractor_name, extractor_key)
    last_updated: datetime = orm.Optional(datetime)

    keep_updated: bool = orm.Required(
        bool,
        default=False,
        index=True,
    )
    watch_in_order_auto: bool = orm.Required(
        bool,
        default=True,
    )
    sorting_method: int = orm.Required(
        int,
        default=0,
    )

    pinned: bool = orm.Required(
        bool,
        default=False,
        index=True,
    )
    ignored: bool = orm.Required(
        bool,
        default=False,
    )
    watch_in_order: bool = orm.Required(
        bool,
        default=True,
        index=True,
    )

    tag_list: Iterable[Tag] = orm.Set(
        lambda: Tag,
    )
    uris: Iterable[CollectionUriMapping] = orm.Set(
        lambda: CollectionUriMapping,
    )
    media_links: Iterable[MediaCollectionLink] = orm.Set(
        MediaCollectionLink,
    )
    created_collections: Set[MediaCollection] = orm.Set(
        lambda: MediaCollection,
    )

    ### for UriHolder

    @property
    def _primary_uri(self) -> str:
        return self.uri

    def _set_primary_uri(self, uri: str) -> None:
        self.uri = uri

    @property
    def _get_uri_set(self) -> Set[str]:
        return {m.uri for m in self.uris}

    def _set_uri_set(self, uri_set: Set[str]) -> None:
        self.uris = set()
        self.add_uris(uri_set)

    ### for Tagable

    @property
    def orm_assigned_tags(self) -> Query[Tag]:
        return self.tag_list

    ### properties

    @property
    def is_creator(self) -> bool:
        return self == self.creator

    @property
    def has_creator(self) -> bool:
        return self.creator is not None

    @property
    def is_root_collection(self) -> bool:
        return self.is_creator or not self.has_creator

    @property
    def was_extracted(self) -> bool:
        return self.last_updated is not None

    @property
    def count(self) -> int:
        return orm.count(self.media_links)

    @property
    def full_length(self) -> int:
        return orm.sum(link.element.length for link in self.media_links)

    @property
    def first_episode(self) -> Optional[MediaCollectionLink]:
        return (
            orm.select(l for l in self.media_links)
            .order_by(MediaCollectionLink.sort_key)
            .first()
        )

    @property
    def last_episode(self) -> Optional[MediaCollectionLink]:
        return (
            orm.select(l for l in self.media_links)
            .order_by(MediaCollectionLink.desc_sort_key)
            .first()
        )

    @property
    def first_released_episode(self) -> Optional[MediaCollectionLink]:
        return (
            orm.select(l for l in self.media_links)
            .order_by(lambda l: l.element.release_date)
            .first()
        )

    @property
    def last_released_episode(self) -> Optional[MediaCollectionLink]:
        # return orm \
        #    .select(l for l in self.media_links) \
        #    .order_by(lambda l: orm.desc(l.element.release_date)) \
        #    .first()
        return (
            orm.select(l for l in self.media_links)
            .order_by(lambda l: orm.desc(l.element.release_date))
            .first()
        )

    @property
    def last_release_date_to_watch(self) -> Optional[datetime]:
        return orm.max(
            (
                l.element.release_date
                for l in self.media_links
                if not l.element.skip_over
            ),
            default=None,
        )

    def __to_watch_episodes(self) -> Query | Iterable[MediaCollectionLink]:
        return orm.select(
            link for link in self.media_links if not link.element.skip_over
        )

    @property
    def next_episode(self) -> Optional[MediaCollectionLink]:
        return (
            orm.select(link for link in self.media_links if not link.element.skip_over)
            .order_by(MediaCollectionLink.sort_key)
            .first()
        )

    @property
    def to_watch_count(self) -> int:
        return self.__to_watch_episodes().count()

    @property
    def completed(self) -> bool:
        return self.to_watch_count <= 0

    @property
    def average_release_per_week(self) -> float:
        return (
            (
                self.full_length
                / (
                    (
                        (
                            (
                                self.last_released_episode.element.release_date
                                - self.first_released_episode.element.release_date
                            )
                            * (self.count / (self.count - 1))
                        )
                        / timedelta(days=7)
                    )
                    or 1
                )
            )
            if self.count >= 2
            else self.full_length
        )

    @property
    def average_release_per_week_now(self) -> float:
        return self.full_length / (
            (
                (datetime.now() - self.first_released_episode.element.release_date)
                / timedelta(days=7)
            )
            or 1
        )

    @property
    def stats(self) -> CollectionStats:
        return CollectionStats.from_collection(self)

    @property
    def info_link(self) -> str:
        return f"/collection/{self.id}"

    ### methods

    def set_watch_in_order_auto(self, watch_in_order: bool) -> None:
        if self.watch_in_order_auto:
            self.watch_in_order = watch_in_order

    def add_episode(
        self,
        media: MediaElement,
        season: int = 0,
        episode: int = 0,
    ) -> Optional[MediaCollectionLink]:
        link: MediaCollectionLink = MediaCollectionLink.get(
            collection=self, element=media
        )
        change = False
        if link is None:
            change = True
            link = MediaCollectionLink(collection=self, element=media)
        if (link.season, link.episode) != (season, episode):
            change = True
            link.season, link.episode = season, episode
        if self.ignored and not media.skip_over:
            change = True
            media.ignored = True
        if change:
            orm.flush()
            return link
        return None

    def add_single_uri(self, uri: str) -> bool:
        mapping: CollectionUriMapping = CollectionUriMapping.get(uri=uri)
        if not mapping:
            logging.debug(f"Add URI mapping {uri!r} to collection {self.id!r}")
            CollectionUriMapping(
                uri=uri,
                element=self,
            )
            return True
        if mapping.element != self:
            raise Exception(
                f"URI duplicated for two different collections's: {uri}"
            )  # TODO may replace with merge call
        return False

    def before_insert(self) -> None:
        self.before_update()

    def before_update(self) -> None:
        self.add_single_uri(self.uri)


class CollectionUriMapping(db.Entity):

    id: int = orm.PrimaryKey(int, auto=True)
    uri: str = orm.Required(str, unique=True)
    element: MediaCollection = orm.Required(MediaCollection)


MEDIAELEMENT_BLOCKING_LOOKUP_CACHE_TABLE = "element_lookup_cache"


SQL_WHITESPACE_PATTERN = re.compile(r"(\s|\n)+")


def sql_cleanup(sql: str) -> str:
    return SQL_WHITESPACE_PATTERN.sub(" ", sql).strip()


def sql_where_in(id: str, id_list: Iterable[str | int]) -> str:
    return f"{id} IN ({','.join(str(i) for i in id_list)})"


# TODO reducing cache table to only contain videos not watched/ignored (not huge speedup)
# TODO add bool for (not)? blocking to direct dependencies (similar to above) (not huge speedup)
def sql_is_considered(elem_id: str, use_cache: bool = True) -> str:
    # NOT EXISTS seems worlds better then making a OUTER JOIN
    return sql_cleanup(
        f"""
            NOT EXISTS (
        """
        + (
            f"""
                SELECT c.element2
                FROM {MEDIAELEMENT_BLOCKING_LOOKUP_CACHE_TABLE} c
                        INNER JOIN {MediaElement._table_} m2 ON c.element1 = m2.id
                WHERE c.element2 = {elem_id} AND NOT (m2.watched OR m2.ignored)
            """
            if use_cache
            else f"""
                SELECT *
                FROM {MediaElement._table_} look_elem
                        INNER JOIN {MediaCollectionLink._table_} link ON link.element = look_elem.id
                        INNER JOIN {MediaCollection._table_} coll ON coll.id = link.collection
                        INNER JOIN {MediaCollectionLink._table_} coll_link ON coll_link.collection = coll.id
                        INNER JOIN {MediaElement._table_} coll_elem ON coll_elem.id = coll_link.element
                WHERE look_elem.id = {elem_id}
                    AND coll.watch_in_order
                    AND NOT (coll_elem.watched OR coll_elem.ignored)
                    AND (coll_link.season, coll_link.episode, coll_elem.release_date, coll_elem.id) < (link.season, link.episode, look_elem.release_date, look_elem.id)
            """
        )
        + f"""
            ) AND NOT EXISTS (
                SELECT *
                FROM mediaelement_mediaelement m_m
                        INNER JOIN {MediaElement._table_} m ON m_m.mediaelement = m.id
                WHERE m_m.mediaelement_2 = {elem_id} AND NOT (m.watched OR m.ignored)
            )
        """
    )


def is_considered(elem_id: int) -> bool:
    return db.exists(
        sql_cleanup(
            f"""
        SELECT elem.id
        FROM {MediaElement._table_} elem
        WHERE elem.id = {elem_id}
            AND NOT (elem.watched OR elem.ignored)
            AND elem.release_date <= NOW()
            AND ({sql_is_considered('elem.id')})
    """
        )
    )


def are_multiple_considered(elem_ids: Iterable[int]) -> Mapping[int, bool]:
    res = {
        r[0]
        for r in db.execute(
            sql_cleanup(
                f"""
        SELECT elem.id
        FROM {MediaElement._table_} elem
        WHERE NOT (elem.watched OR elem.ignored)
            AND elem.release_date <= NOW()
            AND ({sql_is_considered("elem.id")})
    """
            )
        )
    }
    return {elem_id: elem_id in res for elem_id in elem_ids}


def get_all_considered(
    order_by: str = "NULL", filter_by: str = "true"
) -> List[MediaElement]:
    return MediaElement.select_by_sql(
        sql_cleanup(
            f"""
        SELECT elem.*
        FROM {MediaElement._table_} elem
        WHERE NOT (elem.watched OR elem.ignored)
            AND elem.release_date <= NOW()
            AND {filter_by}
            AND ({sql_is_considered("elem.id")})
        ORDER BY {order_by}
    """
        )
    )


def get_all_elements_tags_recursive() -> Mapping[int, Sequence[Tag]]:
    elem_tag_query: Iterable[Tuple[int, int]] = db.execute(
        """
        WITH RECURSIVE found_tag (mediaelement_id, tag_id) AS
        (
                SELECT mediaelement_tag.mediaelement, mediaelement_tag.tag
                FROM mediaelement_tag
            UNION
                SELECT mediacollectionlink.element, mediacollection_tag.tag
                FROM mediacollectionlink
                JOIN mediacollection_tag ON mediacollectionlink.collection = mediacollection_tag.mediacollection
            UNION
                SELECT found_tag.mediaelement_id, tag_tag.tag_2
                FROM found_tag
                JOIN tag_tag ON found_tag.tag_id = tag_tag.tag
        )
        SELECT found_tag.mediaelement_id, found_tag.tag_id
        FROM found_tag
        JOIN tag ON found_tag.tag_id = tag.id
        WHERE tag.use_for_preferences
        ORDER BY mediaelement_id, tag_id;
    """
    )

    @cache
    def get_tag(tag_id: int) -> Tag:
        return Tag[tag_id]

    return {
        elem_id: [get_tag(tag_id) for _, tag_id in group_iter]
        for elem_id, group_iter in itertools.groupby(
            elem_tag_query, key=lambda row: row[0]
        )
    }


def update_element_lookup_cache(collection_ids: List[int] = []):
    logging.info(
        f"Rebuild {MEDIAELEMENT_BLOCKING_LOOKUP_CACHE_TABLE} for {len(collection_ids) if collection_ids else 'all'} collections …"
    )

    def filter_clause(c_id: str):
        return sql_where_in(c_id, collection_ids) if collection_ids else "true"

    orm.flush()
    sql = [
        f"""
            DELETE QUICK FROM {MEDIAELEMENT_BLOCKING_LOOKUP_CACHE_TABLE}
            WHERE {filter_clause("collection")};
        """,
        f"""
            INSERT INTO {MEDIAELEMENT_BLOCKING_LOOKUP_CACHE_TABLE} (collection, element1, element2) SELECT
                c.id AS collection,
                l1.element AS element1,
                l2.element AS element2
            FROM
                {MediaCollection._table_} c
            INNER JOIN {MediaCollectionLink._table_} l1 ON
                c.id = l1.collection
            INNER JOIN {MediaCollectionLink._table_} l2 ON
                c.id = l2.collection
            INNER JOIN {MediaElement._table_} e1 ON
                l1.element = e1.id
            INNER JOIN {MediaElement._table_} e2 ON
                l2.element = e2.id
            WHERE
                (
                    l1.season,
                    l1.episode,
                    e1.release_date,
                    e1.id
                ) <(
                    l2.season,
                    l2.episode,
                    e2.release_date,
                    e2.id
                ) AND c.watch_in_order
                AND {filter_clause("c.id")}
            GROUP BY
                collection,
                element1,
                element2
        """,
    ]
    for q in sql:
        db.execute(sql_cleanup(q))


####
## Custom Table Framework
####


CUSTOM_TABLE_DEFINITIONS: Mapping[SafeStr, str] = {
    SafeStr(table_name): trim(table_sql)
    for table_name, table_sql in {
        MEDIAELEMENT_BLOCKING_LOOKUP_CACHE_TABLE: f"""
            CREATE TABLE {MEDIAELEMENT_BLOCKING_LOOKUP_CACHE_TABLE}(
                collection INT(11) NOT NULL,
                element1 INT(11) NOT NULL,
                element2 INT(11) NOT NULL
            ) SELECT
                c.id AS collection,
                l1.element AS element1,
                l2.element AS element2
            FROM
                {MediaCollection._table_} c
            INNER JOIN {MediaCollectionLink._table_} l1 ON
                c.id = l1.collection
            INNER JOIN {MediaCollectionLink._table_} l2 ON
                c.id = l2.collection
            INNER JOIN {MediaElement._table_} e1 ON
                l1.element = e1.id
            INNER JOIN {MediaElement._table_} e2 ON
                l2.element = e2.id
            WHERE
                (
                    l1.season,
                    l1.episode,
                    e1.release_date,
                    e1.id
                ) <(
                    l2.season,
                    l2.episode,
                    e2.release_date,
                    e2.id
                ) AND c.watch_in_order
            GROUP BY
                collection,
                element1,
                element2;
            ALTER TABLE
                `{MEDIAELEMENT_BLOCKING_LOOKUP_CACHE_TABLE}` ADD PRIMARY KEY(`element1`, `element2`, `collection`);
            ALTER TABLE
                `{MEDIAELEMENT_BLOCKING_LOOKUP_CACHE_TABLE}` ADD INDEX(`collection`);
        """,
    }.items()
}


def table_exists(table_name: SafeStr) -> bool:
    return db.provider.table_exists(
        connection=db.get_connection(),
        table_name=table_name,
    )


@orm.db_session
def setup_custom_tables() -> None:
    """
    Creates & fills custom tables (especially cache tables) if they do not exist.
    This should not destroy already existing data and should behave indempotent.
    """
    for table_name, table_sql in CUSTOM_TABLE_DEFINITIONS.items():
        if not table_exists(table_name):
            db.execute(table_sql)