breakpilot-compliance/backend-compliance/compliance/services/cookie_to_vendor_fallback.py

"""
Cookie-zu-Vendor-Fallback (P52 Lite).

Wenn weder cmp_payloads noch vendor_llm_extract Vendors lieferten,
matchen wir die im after_accept gesehenen Cookies gegen die
compliance.cookie_library und bauen Vendor-Records aus den Library-
Eintraegen (cookie_name → vendor_name, actual_category).

Typisches Szenario: VW nutzt ein Custom-CMP (cookiemgmt-Wrapper),
kein bekanntes IAB-Tool. cmp_payloads = leer, aber after_accept.cookies
hat 28 Eintraege. Diese 28 Cookies sind in der Library = ~15-20 Vendors.
"""

from __future__ import annotations

import logging
import re
from typing import Iterable

from sqlalchemy import text
from sqlalchemy.orm import Session

logger = logging.getLogger(__name__)


def _collect_cookie_names(banner_result: dict | None) -> set[str]:
    names: set[str] = set()
    if not isinstance(banner_result, dict):
        return names
    for ph in (banner_result.get("phases") or {}).values():
        if not isinstance(ph, dict):
            continue
        for ck in (ph.get("cookies") or []):
            if isinstance(ck, str):
                names.add(ck.strip())
            elif isinstance(ck, dict):
                n = (ck.get("name") or "").strip()
                if n:
                    names.add(n)
    return {n for n in names if n and len(n) <= 120}


def lookup_vendors_from_library(
    db: Session,
    cookie_names: Iterable[str],
) -> list[dict]:
    """Resolves cookie names to vendor records via cookie_library."""
    names = [n for n in cookie_names if n]
    if not names:
        return []
    rows = db.execute(text(
        """
        SELECT cookie_name, actual_category, vendor_name
        FROM compliance.cookie_library
        WHERE LOWER(cookie_name) = ANY(:lc)
        """
    ), {"lc": [n.lower() for n in names]}).fetchall()
    by_vendor: dict[str, dict] = {}
    for cname, cat, vendor in rows:
        if not vendor:
            continue
        entry = by_vendor.setdefault(vendor, {
            "name": vendor,
            "country": "",
            "purpose": "",
            "category": cat or "",
            "opt_out_url": "",
            "privacy_policy_url": "",
            "persistence": "",
            "cookies": [],
            "source": "library_fallback",
        })
        entry["cookies"].append({
            "name": cname, "purpose": "", "expiry": "",
            "is_third_party": True,
        })
    return list(by_vendor.values())


def fallback_vendors_for_run(
    db: Session,
    banner_result: dict | None,
    existing_vendor_count: int,
    cookie_doc_text: str | None = None,
) -> list[dict]:
    """Returns extra vendor records to merge with the run's cmp_vendors.

    VW-Lehre: cmp_vendors=6 (alle LLM-grob) reicht NICHT — die echte
    Cookie-Tabelle hat 30+ Eintraege. Wir fuehren den Lookup jetzt auch
    bei mid-tier-Counts aus, solange after_accept >= 15 Cookies hat
    ODER der Cookie-Doc-Text Cookie-Tabellen-Signale enthaelt.
    """
    names = _collect_cookie_names(banner_result)

    # Erweitere names um Cookie-Namen die im Cookie-Doc-Text als
    # Tabellen-Eintraege auftauchen (Pattern: NAME gefolgt von
    # "Tracking Cookies"/"Session Cookies"/"Funktional"/...).
    if cookie_doc_text:
        names |= _extract_cookie_names_from_doc(cookie_doc_text)

    # Skip-Bedingungen ueberarbeitet:
    #  - sehr wenige Cookies UND >= 5 Vendors schon vorhanden → skip
    #  - sonst IMMER versuchen
    if len(names) < 5 and existing_vendor_count >= 5:
        return []
    if not names:
        return []

    vendors = lookup_vendors_from_library(db, names)
    if vendors:
        logger.info(
            "Cookie-Library-Fallback: %d Vendors aus %d Cookies "
            "(existing cmp_vendors=%d)",
            len(vendors), len(names), existing_vendor_count,
        )
    return vendors


_TABLE_ROW_RE = re.compile(
    r"\b([A-Za-z_][A-Za-z0-9_\-\.]{2,40})\s+"
    r"(?:Tracking Cookies|Session Cookies|Funktional|Marketing|"
    r"Analytics|Performance|Notwendig|Strictly\s+Necessary|"
    r"Statistik|Werbung|Targeting|Personalisierung)",
    re.I,
)


def _extract_cookie_names_from_doc(text: str) -> set[str]:
    """Pattern-basiertes Erkennen von Cookie-Tabellen-Zeilen.

    VW-Cookie-Tabelle hat Form:
      'IDE Tracking Cookies (Marketing) Dieser Cookie ... 13 Monate'
    Das fangen wir mit einem Cookie-Name-vor-Category-Pattern.
    """
    out: set[str] = set()
    for m in _TABLE_ROW_RE.finditer(text):
        name = m.group(1).strip()
        # Filter offensichtliche Noise (Pronomen, Verben)
        nl = name.lower()
        if nl in ("dieser", "diese", "ein", "der", "die", "das",
                  "session", "permanent", "funktional", "notwendig",
                  "marketing", "analytics", "werbung", "anbieter",
                  "google", "facebook", "tracking", "cookie", "cookies"):
            continue
        if len(name) >= 3:
            out.add(name)
    return out