Vorrania/backend/app/services/master_data.py

"""Export/Import der Stammdaten als JSON – mit ID-Erhalt für Sicherung/Umzug.

Umfasst Kategorien (inkl. eigener Felder), Lagerorte, Einheiten und Gebinde. Die
IDs werden bewusst mitgeführt: nur so passen z.B. gedruckte Lagerort-QR-Codes
(`/l/<id>`) nach einem Wiederherstellen noch, und Verweise (parent_id,
category_id) bleiben gültig.

Beim Import entscheidet der Modus, was bei bereits vergebener ID passiert:
- ``skip``      – vorhandene Einträge unangetastet lassen, nur neue anlegen.
- ``overwrite`` – vorhandene per ID überschreiben, fehlende anlegen.

Einzelne Zeilen, die an einer Eindeutigkeit scheitern (z.B. gleicher Name unter
anderer ID), werden per Savepoint übersprungen statt den ganzen Import
abzubrechen.
"""

from __future__ import annotations

import json

from sqlalchemy import text
from sqlalchemy.orm import Session

from ..models import Category, FieldDefinition, Location, PackageType, Unit, UnitKind
from .fields import options_list

FORMAT_KEY = "vorrania_master_data"
FORMAT_VERSION = 1
# Nur Tabellen mit fortlaufender Integer-id brauchen die Sequenz-Korrektur.
# Lagerorte tragen einen zufälligen Code als id – dort gibt es keine Sequenz.
_TABLES = ("units", "package_types", "categories", "field_definitions")


def export_master_data(db: Session) -> dict:
    return {
        FORMAT_KEY: FORMAT_VERSION,
        "categories": [
            {"id": c.id, "name": c.name, "parent_id": c.parent_id, "tracking": c.tracking}
            for c in db.query(Category).order_by(Category.id).all()
        ],
        "field_definitions": [
            {
                "id": f.id, "category_id": f.category_id, "label": f.label, "key": f.key,
                "field_type": f.field_type, "unit": f.unit, "options": options_list(f),
                "required": f.required, "position": f.position, "is_builtin": f.is_builtin,
            }
            for f in db.query(FieldDefinition).order_by(FieldDefinition.id).all()
        ],
        "locations": [
            {"id": l.id, "name": l.name, "parent_id": l.parent_id}
            for l in db.query(Location).order_by(Location.name).all()
        ],
        "units": [
            {"id": u.id, "name": u.name, "kind": u.kind.value, "factor": u.factor,
             "is_builtin": u.is_builtin}
            for u in db.query(Unit).order_by(Unit.id).all()
        ],
        "package_types": [
            {"id": p.id, "singular": p.singular, "plural": p.plural, "is_builtin": p.is_builtin}
            for p in db.query(PackageType).order_by(PackageType.id).all()
        ],
    }


class _Report:
    def __init__(self) -> None:
        self.created: dict[str, int] = {}
        self.updated: dict[str, int] = {}
        self.skipped: dict[str, int] = {}

    def add(self, bucket: dict[str, int], key: str) -> None:
        bucket[key] = bucket.get(key, 0) + 1

    def as_dict(self) -> dict:
        return {"created": self.created, "updated": self.updated, "skipped": self.skipped}


def _fix_sequence(db: Session, table: str) -> None:
    """Postgres: die id-Sequenz auf MAX(id) heben, sonst kollidieren spätere
    Auto-IDs mit importierten. SQLite braucht das nicht."""
    if db.get_bind().dialect.name != "postgresql":
        return
    db.execute(text(
        f"SELECT setval(pg_get_serial_sequence('{table}', 'id'), "
        f"GREATEST((SELECT COALESCE(MAX(id), 1) FROM {table}), 1))"
    ))


def import_master_data(db: Session, data: dict, mode: str = "skip") -> dict:
    if not isinstance(data, dict) or FORMAT_KEY not in data:
        raise ValueError("Das ist keine Vorrania-Stammdaten-Datei.")
    overwrite = mode == "overwrite"
    rep = _Report()

    def upsert(model, row, key, build, *, is_tree=False, conflict=None) -> None:
        rid = row.get("id")
        existing = db.get(model, rid) if rid is not None else None
        if existing is not None and not overwrite:
            rep.add(rep.skipped, key)
            return
        # Vorab-Check statt IntegrityError abzufangen: würde die Zeile eine
        # Eindeutigkeit unter ANDERER id verletzen (z.B. gleicher Name), lieber
        # überspringen als den Import abbrechen.
        if conflict is not None and conflict(row):
            rep.add(rep.skipped, key)
            return
        if existing is not None:
            for k, v in build(row).items():
                setattr(existing, k, v)
            db.flush()
            rep.add(rep.updated, key)
            return
        kwargs = build(row)
        if rid is not None:
            kwargs["id"] = rid
        if is_tree:
            kwargs["parent_id"] = None  # Eltern erst im 2. Durchgang
        db.add(model(**kwargs))
        db.flush()
        rep.add(rep.created, key)

    def set_parents(model, rows) -> None:
        for row in rows or []:
            rid = row.get("id")
            obj = db.get(model, rid) if rid is not None else None
            if obj is None:
                continue
            pid = row.get("parent_id")
            ziel = db.get(model, pid) if pid is not None else None
            obj.parent_id = ziel.id if ziel is not None else None
        db.flush()

    def name_clash(model, field, r):
        val = r.get(field)
        return db.query(model.id).filter(
            getattr(model, field) == val, model.id != r.get("id")
        ).first() is not None

    for row in data.get("units") or []:
        upsert(Unit, row, "units", lambda r: {
            "name": r["name"], "kind": UnitKind(r["kind"]),
            "factor": float(r.get("factor", 1.0)),
            "is_builtin": bool(r.get("is_builtin", False)),
        }, conflict=lambda r: name_clash(Unit, "name", r))
    for row in data.get("package_types") or []:
        upsert(PackageType, row, "package_types", lambda r: {
            "singular": r["singular"], "plural": r.get("plural") or r["singular"],
            "is_builtin": bool(r.get("is_builtin", False)),
        }, conflict=lambda r: name_clash(PackageType, "singular", r))

    cats = data.get("categories") or []
    for row in cats:
        upsert(Category, row, "categories", lambda r: {
            "name": r["name"], "tracking": r.get("tracking") or "food",
        }, is_tree=True)
    set_parents(Category, cats)

    for row in data.get("field_definitions") or []:
        if db.get(Category, row.get("category_id")) is None:
            rep.add(rep.skipped, "field_definitions")
            continue
        upsert(FieldDefinition, row, "field_definitions", lambda r: {
            "category_id": r["category_id"], "label": r["label"],
            "key": r.get("key") or r["label"],
            "field_type": r.get("field_type", "text"),
            "unit": r.get("unit"),
            "options": json.dumps(r.get("options") or []) if r.get("field_type") == "select" else None,
            "required": bool(r.get("required", False)),
            "position": int(r.get("position", 0)),
            "is_builtin": bool(r.get("is_builtin", False)),
        })

    locs = data.get("locations") or []
    for row in locs:
        upsert(Location, row, "locations", lambda r: {"name": r["name"]}, is_tree=True)
    set_parents(Location, locs)

    db.commit()
    for table in _TABLES:
        _fix_sequence(db, table)
    db.commit()
    return rep.as_dict()