ohm_streaming/app/services/discover.py

"""Découverte : nouveautés des sources, incontournables et recommandations.

Trois sections :
- **latest** — « récemment ajoutés » scrapés sur chaque source activée (cliquables
  directement vers la fiche) ;
- **must_watch** — titres les plus populaires du catalogue Kitsu (tous temps) ;
- **for_you** — recommandations par genres : les genres des titres téléchargés
  (serveur), des favoris (par utilisateur) et des séries téléchargées sur Sonarr
  sont agrégés, puis Kitsu est interrogé sur ces catégories en excluant le
  déjà-possédé.

Toutes les sources externes sont optionnelles : un échec (réseau, scraping, API)
laisse la section vide et n'est jamais remonté au caller (dégradation gracieuse).
Un cache mémoire TTL évite de re-scaper à chaque chargement de page.
"""

import asyncio
import dataclasses
import json
import logging
import re
import time
import unicodedata

import httpx

from app.config import get_settings
from app.db import db
from app.scrapers.base import (
    ScrapeError,
    SourceScraper,
    all_sources,
    import_all_scrapers,
)
from app.services.kitsu import KitsuService, normalize_title
from app.services.settings import is_source_enabled
from app.services.sonarr import sonarr

logger = logging.getLogger(__name__)

import_all_scrapers()

# Bornes de l'algorithme
_MAX_HISTORY_TITLES = 12  # titres récents analysés (téléchargements + favoris)
_MAX_GENRES = 4           # genres retenus pour la requête Kitsu
_KITSU_PAGE_MAX = 20      # limite dure de l'API Kitsu (page[limit] > 20 → 400)
_ENRICH_CONCURRENCY = 6   # enrichissements Kitsu parallèles max (nouveautés)

_LATEST_TTL_SECONDS = 600        # nouveautés : re-scrape au bout de 10 min
_MUST_WATCH_TTL_SECONDS = 21600  # incontournables : quasi statique, 6 h
_FOR_YOU_TTL_SECONDS = 3600      # recommandations : 1 h (l'historique évolue lentement)


class _TTLCache:
    """Cache mémoire minimal avec expiration (mono-processus, suffisant ici)."""

    def __init__(self) -> None:
        self._data: dict[str, tuple[float, object]] = {}

    def get(self, key: str) -> object | None:
        entry = self._data.get(key)
        if entry is None:
            return None
        expires_at, value = entry
        if expires_at <= time.monotonic():
            del self._data[key]
            return None
        return value

    def set(self, key: str, value: object, ttl_seconds: float) -> None:
        self._data[key] = (time.monotonic() + ttl_seconds, value)

    def clear(self, prefix: str = "") -> None:
        """Invalide les clés commençant par prefix (vide = tout le cache)."""
        for key in [k for k in self._data if k.startswith(prefix)]:
            del self._data[key]


def category_slug(name: str) -> str:
    """Nom de genre → slug de catégorie Kitsu (« Slice of Life » → « slice-of-life »)."""
    decomposed = unicodedata.normalize("NFKD", name)
    ascii_only = "".join(char for char in decomposed if not unicodedata.combining(char))
    return re.sub(r"[^a-z0-9]+", "-", ascii_only.casefold()).strip("-")


class DiscoverService:
    """Agrégation des trois sections de découverte, avec cache mémoire."""

    def __init__(self) -> None:
        self._cache = _TTLCache()
        self._kitsu = KitsuService()

    # ------------------------------------------------------------ nouveautés

    async def latest(self, limit: int = 24) -> list[dict]:
        """Nouveautés toutes sources confondues, triées par date de sortie réelle.

        Les « récemment ajoutés » de chaque source sont fusionnés (doublons retirés),
        enrichis via Kitsu (date de début, statut de diffusion) puis triés du plus
        récent au plus ancien — ce qui sort / vient de sortir en premier.
        """
        cached = self._cache.get(f"latest:{limit}")
        if cached is not None:
            return cached  # type: ignore[return-value]

        sources = [s for s in all_sources() if await is_source_enabled(s.name)]
        outcomes = await asyncio.gather(*(self._latest_of(source) for source in sources))
        items = [item for outcome in outcomes for item in (outcome or [])]

        merged: dict[str, dict] = {}
        for item in items:
            key = normalize_title(item["title"]).casefold()
            existing = merged.get(key)
            if existing is None or (not existing.get("image_url") and item.get("image_url")):
                merged[key] = item
        semaphore = asyncio.Semaphore(_ENRICH_CONCURRENCY)

        async def bounded(item: dict) -> dict:
            async with semaphore:
                return await self._with_release_info(item)

        enriched = await asyncio.gather(*(bounded(item) for item in merged.values()))
        result = sorted(enriched, key=lambda it: it.get("start_date") or "", reverse=True)
        result = result[:limit]
        self._cache.set(f"latest:{limit}", result, _LATEST_TTL_SECONDS)
        return result

    async def _latest_of(self, source: SourceScraper) -> list[dict] | None:
        """Items latest() d'une source, aplatis avec les infos de source ([] si KO)."""
        try:
            results = await source.latest()
        except ScrapeError as exc:
            logger.warning("Nouveautés indisponibles pour %s : %s", source.name, exc)
            return None
        return [
            {**dataclasses.asdict(r), "source": source.name, "label": source.label}
            for r in results
        ]

    async def _with_release_info(self, item: dict) -> dict:
        """Complète un item de nouveauté avec sa date de sortie Kitsu (None si absent)."""
        item.setdefault("start_date", None)
        item.setdefault("status", None)
        item.setdefault("rating", None)
        match = await self._kitsu_match_for_title(item["title"])
        if match is None:
            return item
        attrs = match.get("attributes", {})
        item["start_date"] = attrs.get("startDate")
        item["status"] = attrs.get("status")
        item["rating"] = KitsuService._to_rating_10(attrs.get("averageRating"))
        return item

    # --------------------------------------------------------- incontournables

    async def must_watch(self, limit: int = _KITSU_PAGE_MAX) -> list[dict]:
        """Titres les plus populaires du catalogue Kitsu (tous temps)."""
        limit = min(limit, _KITSU_PAGE_MAX)
        key = f"must_watch:{limit}"
        cached = self._cache.get(key)
        if cached is not None:
            return cached  # type: ignore[return-value]
        items = await self._kitsu_anime({"sort": "-userCount", "page[limit]": limit})
        self._cache.set(key, items, _MUST_WATCH_TTL_SECONDS)
        return items

    # ------------------------------------------------------------ pour toi

    async def for_you(self, user_id: int, limit: int = _KITSU_PAGE_MAX) -> dict:
        """Recommandations par genres, à partir de l'historique de l'utilisateur.

        Genres = téléchargements du serveur (titres → Kitsu) + favoris de l'utilisateur
        (genres du payload) + séries téléchargées sur Sonarr (genres fournis par
        Sonarr). On exclut les titres déjà possédés (local et Sonarr).
        """
        limit = min(limit, _KITSU_PAGE_MAX)
        cache_key = f"for_you:{user_id}:{limit}"
        cached = self._cache.get(cache_key)
        if cached is not None:
            return cached  # type: ignore[return-value]

        owned, favorite_genres = await self._owned(user_id)
        genre_counts = await self._genres_from_downloads(owned)
        for genre, count in favorite_genres.items():
            genre_counts[genre] = genre_counts.get(genre, 0) + count
        sonarr_owned, sonarr_genres = await sonarr.profile()
        for genre, count in sonarr_genres.items():
            genre_counts[genre] = genre_counts.get(genre, 0) + count
        owned |= sonarr_owned
        if not genre_counts:
            result: dict = {"based_on": [], "items": []}
            self._cache.set(cache_key, result, _FOR_YOU_TTL_SECONDS)
            return result

        top_genres = sorted(genre_counts, key=genre_counts.get, reverse=True)[:_MAX_GENRES]
        slugs = [category_slug(genre) for genre in top_genres]
        items = await self._kitsu_anime(
            {
                "filter[categories]": ",".join(slugs),
                "sort": "-userCount",
                "page[limit]": limit,  # le déjà-possédé est filtré après
            }
        )
        kept = [item for item in items if item["title"] and item["title"].casefold() not in owned]
        result = {"based_on": top_genres, "items": kept}
        self._cache.set(cache_key, result, _FOR_YOU_TTL_SECONDS)
        return result

    def invalidate_for_you(self) -> None:
        """Recommandations recalculées au prochain appel (réglages Sonarr modifiés)."""
        self._cache.clear("for_you:")

    async def _owned(self, user_id: int) -> tuple[set[str], dict[str, int]]:
        """Titres possédés (normalisés) + genres directement connus via les favoris."""
        rows = await db.fetchall(
            "SELECT DISTINCT title FROM downloads ORDER BY created_at DESC LIMIT ?",
            (_MAX_HISTORY_TITLES,),
        )
        fav_rows = await db.fetchall(
            "SELECT payload FROM favorites WHERE user_id = ? ORDER BY created_at DESC LIMIT ?",
            (user_id, _MAX_HISTORY_TITLES),
        )
        owned = {normalize_title(row["title"]).casefold() for row in rows}
        owned.discard("")
        genre_counts: dict[str, int] = {}
        for row in fav_rows:
            try:
                payload = json.loads(row["payload"]) if row["payload"] else {}
            except (TypeError, ValueError):
                continue
            for genre in payload.get("genres") or []:
                if isinstance(genre, str) and genre.strip():
                    genre_counts[genre.strip()] = genre_counts.get(genre.strip(), 0) + 1
        return owned, genre_counts

    async def _genres_from_downloads(self, titles: set[str]) -> dict[str, int]:
        """Genres Kitsu des titres téléchargés (cache DB puis recherche)."""
        semaphore = asyncio.Semaphore(_MAX_HISTORY_TITLES)

        async def genres_of(title: str) -> list[str]:
            async with semaphore:
                return await self._kitsu_genres_for_title(title)

        outcomes = await asyncio.gather(*(genres_of(t) for t in list(titles)[:_MAX_HISTORY_TITLES]))
        counts: dict[str, int] = {}
        for genres in outcomes:
            for genre in genres:
                counts[genre] = counts.get(genre, 0) + 1
        return counts

    async def _kitsu_match_for_title(self, title: str) -> dict | None:
        """Match Kitsu d'un titre scrapé (cache DB 72 h via metadata_cache)."""
        query = normalize_title(title)
        if not query:
            return None
        cache_key = f"kitsu:anime:{query.casefold()}"
        match = await self._kitsu.get_cached(cache_key)
        if match is None:
            match = await self._kitsu.search_anime(title)
            if match is not None:
                await self._kitsu.set_cached(cache_key, match)
        return match

    async def _kitsu_genres_for_title(self, title: str) -> list[str]:
        """Genres Kitsu d'un titre scrapé (cache DB 72 h via metadata_cache).

        La recherche Kitsu ne renvoie plus les genres (`include=genres` vide) :
        on complète avec l'endpoint /anime/<id>/categories.
        """
        match = await self._kitsu_match_for_title(title)
        if match is None:
            return []
        genres = [g for g in match.get("genres", []) if isinstance(g, str)]
        if not genres:
            genres = await self._kitsu_categories(match.get("id"))
            match["genres"] = genres
            query = normalize_title(title)
            await self._kitsu.set_cached(  # refresh avec les genres
                f"kitsu:anime:{query.casefold()}", match
            )
        return genres

    # -------------------------------------------------------------- Kitsu

    async def _kitsu_anime(self, params: dict) -> list[dict]:
        """Requête générique liste Kitsu → items normalisés ([] si échec)."""
        settings = get_settings()
        try:
            async with httpx.AsyncClient(
                timeout=settings.http_timeout,
                headers={
                    "User-Agent": settings.user_agent,
                    "Accept": "application/vnd.api+json",
                },
            ) as client:
                response = await client.get(f"{settings.kitsu_base_url}/anime", params=params)
                response.raise_for_status()
                payload = response.json()
        except (httpx.HTTPError, ValueError) as exc:
            logger.warning("Liste Kitsu échouée (%s) : %s", params, exc)
            return []
        return [
            self._normalize_anime(item)
            for item in payload.get("data", [])
            if item.get("type") == "anime"
        ]

    async def _kitsu_categories(self, anime_id: object) -> list[str]:
        """Titres des catégories Kitsu d'un anime ([] si échec)."""
        if not anime_id:
            return []
        settings = get_settings()
        try:
            async with httpx.AsyncClient(
                timeout=settings.http_timeout,
                headers={
                    "User-Agent": settings.user_agent,
                    "Accept": "application/vnd.api+json",
                },
            ) as client:
                response = await client.get(
                    f"{settings.kitsu_base_url}/anime/{anime_id}/categories",
                    params={"page[limit]": _KITSU_PAGE_MAX},
                )
                response.raise_for_status()
                payload = response.json()
        except (httpx.HTTPError, ValueError) as exc:
            logger.warning("Catégories Kitsu échouées (anime %s) : %s", anime_id, exc)
            return []
        return [
            attrs["title"]
            for item in payload.get("data", [])
            if isinstance(attrs := item.get("attributes", {}), dict) and attrs.get("title")
        ]

    @staticmethod
    def _normalize_anime(item: dict) -> dict:
        attrs = item.get("attributes", {})
        titles = attrs.get("titles") or {}
        poster = attrs.get("posterImage") or {}
        return {
            "kitsu_id": item.get("id"),
            "title": attrs.get("canonicalTitle") or titles.get("en_jp"),
            "image_url": poster.get("large") or poster.get("medium") or poster.get("tiny"),
            "rating": KitsuService._to_rating_10(attrs.get("averageRating")),
            "year": KitsuService._extract_year(attrs.get("startDate")),
            "subtype": attrs.get("subtype"),
            "user_count": attrs.get("userCount"),
        }


discover = DiscoverService()