goon/app/normalize/scenes.py

"""Mapowanie RawScene/RawPerformer/RawStudio/RawTag → znormalizowane DTO gotowe do upsertu.

Normalizacja = wyliczenie pól indeksujących (`*_normalized`, `slug`). Surowe pola jak
`title`, `release_date` przekazujemy bez zmian, kanon wybiera resolver.
"""
from __future__ import annotations

from dataclasses import dataclass, field
from datetime import date

from app.connectors.base import (
    RawPerformer,
    RawPlaybackSource,
    RawScene,
    RawStudio,
    RawTag,
)
from app.normalize.tag_inference import infer_tag_slugs
from app.normalize.text import (
    clean_description,
    normalize,
    normalize_person,
    slugify,
    unescape_html,
)


@dataclass
class NormalizedTag:
    name: str
    slug: str
    # Optional, `resolve_tag` keys on slug and `Tag` has no external_id column.
    external_id: str | None = None


@dataclass
class NormalizedStudio:
    name: str
    name_normalized: str
    slug: str
    external_id: str | None
    parent_external_id: str | None
    parent_name: str | None
    network: str | None
    homepage_url: str | None


@dataclass
class NormalizedPerformer:
    canonical_name: str
    name_normalized: str
    slug: str
    external_id: str | None
    aliases: list[str]
    aliases_normalized: list[str]
    gender: str | None
    birth_date: date | None
    country: str | None
    as_alias_in_scene: str | None


@dataclass
class NormalizedScene:
    external_id: str
    title: str
    title_normalized: str
    slug: str
    release_date: date | None
    description: str | None
    duration_sec: int | None
    code: str | None
    director: str | None
    url: str | None

    studio: NormalizedStudio | None = None
    performers: list[NormalizedPerformer] = field(default_factory=list)
    tags: list[NormalizedTag] = field(default_factory=list)

    fingerprints: list[tuple[str, str]] = field(default_factory=list)
    """Lista (kind, value), np. [('phash', 'abc...'), ('oshash', '...')]."""

    playback_sources: list[RawPlaybackSource] = field(default_factory=list)
    """Linki do odtwarzania (passthrough, resolver dorzuca do tabeli playback_sources)."""

    cross_source_refs: dict[str, str] = field(default_factory=dict)
    """Mapowanie source_name → external_id deklarowane przez źródło z którego pochodzi
    ta scena (np. StashDB ujawnia tpdb_id w `urls`)."""


def normalize_tag(raw: RawTag) -> NormalizedTag:
    return NormalizedTag(
        name=raw.name,
        slug=raw.slug or slugify(raw.name),
        external_id=raw.external_id,
    )


def _merge_inferred_tags(raw_tags: list[RawTag], *, title: str) -> list[NormalizedTag]:
    """Tagi z connectora + tagi wywnioskowane z tytułu (`infer_tag_slugs`).

    Tube'y (96% grywalnego katalogu) przychodzą prawie bez tagów, tylko 16%
    tube-only scen ma jakikolwiek tag, vs 99% scen zmatchowanych z TPDB/StashDB.
    `infer_tag_slugs` mapuje phrasy z tytułu na canonical slugi (zgrane z DB), więc
    resolve_tag trafia w istniejące Tag rows. Union, nie nadpisanie: gdy scena potem
    zmergeuje się z TPDB, tagi się sumują. Inference odpalamy dla KAŻDEJ sceny,
    dla canonical to no-op (już mają komplet), dla tube to wypełnienie braku.
    """
    tags = [normalize_tag(t) for t in raw_tags]
    seen = {t.slug for t in tags}
    for slug in infer_tag_slugs(title):
        if slug in seen:
            continue
        seen.add(slug)
        tags.append(NormalizedTag(name=slug.replace("-", " ").title(), slug=slug, external_id=None))
    return tags


def normalize_studio(raw: RawStudio) -> NormalizedStudio:
    return NormalizedStudio(
        name=raw.name,
        name_normalized=normalize(raw.name),
        slug=raw.slug or slugify(raw.name),
        external_id=raw.external_id,
        parent_external_id=raw.parent_external_id,
        parent_name=raw.parent_name,
        network=raw.network,
        homepage_url=raw.homepage_url,
    )


_GENDER_ALIASES = {
    "female": "female",
    "f": "female",
    "male": "male",
    "m": "male",
    "transgender female": "transgender_female",
    "trans female": "transgender_female",
    "trans-female": "transgender_female",
    "transgender_female": "transgender_female",
    "transgender male": "transgender_male",
    "trans male": "transgender_male",
    "trans-male": "transgender_male",
    "transgender_male": "transgender_male",
    "non binary": "non_binary",
    "non-binary": "non_binary",
    "nonbinary": "non_binary",
    "non_binary": "non_binary",
    "intersex": "intersex",
    "unknown": "unknown",
}


def _normalize_gender(value: str | None) -> str | None:
    """TPDB/StashDB zwracają warianty z spacjami/myślnikami. Normalizujemy do enum
    `performer_gender` w bazie. Wartość niezmapowana → None (NULL w DB)."""
    if not value:
        return None
    return _GENDER_ALIASES.get(value.strip().lower())


def normalize_performer(raw: RawPerformer) -> NormalizedPerformer:
    aliases = list(dict.fromkeys(raw.aliases))  # de-dup zachowując kolejność
    return NormalizedPerformer(
        canonical_name=raw.name,
        name_normalized=normalize_person(raw.name),
        slug=slugify(raw.name),
        external_id=raw.external_id,
        aliases=aliases,
        aliases_normalized=[normalize_person(a) for a in aliases],
        gender=_normalize_gender(raw.gender),
        birth_date=raw.birth_date,
        country=raw.country,
        as_alias_in_scene=raw.as_alias_in_scene,
    )


#: Ile dni w przod jeszcze uznajemy date premiery za wiarygodna. Studia zapowiadaja
#: sceny z niewielkim wyprzedzeniem, ale data odlegla o miesiace to literowka w roku
#: albo blad parsera. Takie daty sa szkodliwe, bo listy sortowane po premierze pinuja
#: je na stale na gorze: dwie sceny porndish (najdalsza 2026-12-02, dodana w maju)
#: siedzialy nad wszystkim, przez co serwis wygladal, jakby przestal ingestowac
#: (zgloszenie de732da2). Odrzucamy je, czyli scena sortuje sie po dacie dodania.
_MAX_FUTURE_RELEASE_DAYS = 14


def _sane_release_date(value):
    if value is None:
        return None
    from datetime import date, timedelta

    return None if value > date.today() + timedelta(days=_MAX_FUTURE_RELEASE_DAYS) else value


def normalize_scene(raw: RawScene) -> NormalizedScene:
    # Tuby oddaja tytuly z encjami HTML (`we're`), patrz `unescape_html`.
    title = unescape_html(raw.title)
    return NormalizedScene(
        external_id=raw.external_id,
        title=title,
        title_normalized=normalize(title),
        slug=slugify(title),
        release_date=_sane_release_date(raw.release_date),
        description=clean_description(raw.description),
        duration_sec=raw.duration_sec,
        code=raw.code,
        director=raw.director,
        url=raw.url,
        studio=normalize_studio(raw.studio) if raw.studio else None,
        performers=[normalize_performer(p) for p in raw.performers],
        tags=_merge_inferred_tags(raw.tags, title=title),
        fingerprints=[(fp.kind, fp.value) for fp in raw.fingerprints],
        playback_sources=list(raw.playback_sources),
        cross_source_refs=dict(raw.cross_source_refs),
    )