app/services/redaction_upload.py

"""Upload-time redaction strategy parsing and session PII context assembly.

Builds the ``db_context`` map forwarded to :func:`sanitise_text_for_rag_sync`
for exact-string wipes in the deterministic redaction path. Values originate
from optional multipart form fields and, when absent, from the tenant's most
recent report section text (best-effort identity extraction).
"""

from __future__ import annotations

import json
import logging
from typing import TYPE_CHECKING

from app.services.document_sanitiser import RedactionStrategy

if TYPE_CHECKING:
    from sqlalchemy.ext.asyncio import AsyncSession

logger = logging.getLogger(__name__)


class InvalidRedactionStrategyError(ValueError):
    """Raised when an upload ``redaction_strategy`` value cannot be mapped."""


_STRATEGY_ALIASES: dict[str, RedactionStrategy] = {
    "ai_hybrid": RedactionStrategy.AI_HYBRID,
    "ai": RedactionStrategy.AI_HYBRID,
    "hybrid": RedactionStrategy.AI_HYBRID,
    "deterministic": RedactionStrategy.DETERMINISTIC_CODE,
    "deterministic_code": RedactionStrategy.DETERMINISTIC_CODE,
    "code": RedactionStrategy.DETERMINISTIC_CODE,
}


def parse_redaction_strategy(raw: str | None) -> RedactionStrategy:
    """Map an upload form ``redaction_strategy`` string to :class:`RedactionStrategy`.

    Args:
        raw: User-supplied strategy name. ``None`` or blank defaults to
            ``AI_HYBRID``.

    Returns:
        The resolved enum member.

    Raises:
        InvalidRedactionStrategyError: When ``raw`` is non-empty but not recognised.
    """
    if raw is None or not str(raw).strip():
        return RedactionStrategy.AI_HYBRID
    key = str(raw).strip().lower().replace("-", "_")
    resolved = _STRATEGY_ALIASES.get(key)
    if resolved is None:
        allowed = sorted({"AI_HYBRID", "DETERMINISTIC_CODE", *_STRATEGY_ALIASES})
        raise InvalidRedactionStrategyError(
            f"Invalid redaction_strategy {raw!r}. "
            f"Allowed values: {', '.join(allowed)}."
        )
    return resolved


def merge_db_context(*parts: dict[str, str] | None) -> dict[str, str]:
    """Merge context dicts; later keys do not overwrite earlier non-empty values."""
    merged: dict[str, str] = {}
    for part in parts:
        if not part:
            continue
        for key, value in part.items():
            v = (value or "").strip()
            if v and key not in merged:
                merged[key] = v
    return merged


def db_context_from_form_fields(
    *,
    client_name: str | None = None,
    property_address: str | None = None,
    surveyor_name: str | None = None,
) -> dict[str, str]:
    """Build a ``db_context`` fragment from optional multipart form fields."""
    ctx: dict[str, str] = {}
    if client_name and client_name.strip():
        ctx["client_name"] = client_name.strip()
    if property_address and property_address.strip():
        ctx["property_address"] = property_address.strip()
    if surveyor_name and surveyor_name.strip():
        ctx["surveyor_name"] = surveyor_name.strip()
    return ctx


async def fetch_tenant_db_context(
    db: AsyncSession,
    tenant_id: str,
) -> dict[str, str]:
    """Best-effort PII context from the tenant's recent report section text.

    Uses :func:`app.services.generation._extract_property_identity` on the
    latest report sections so uploads can wipe names/addresses already present
    in the tenant's active reporting session without re-entering them manually.

    Args:
        db: Active async database session.
        tenant_id: Authenticated tenant identifier.

    Returns:
        A (possibly empty) ``{label: sensitive_value}`` map suitable for
        ``db_context`` in the deterministic redaction path.
    """
    ctx: dict[str, str] = {}
    try:
        from sqlalchemy import select

        from app.db.models import Report, ReportSection, ReportStatus
        from app.services.generation import _extract_property_identity

        res = await db.execute(
            select(ReportSection.text)
            .join(Report, Report.id == ReportSection.report_id)
            .where(Report.tenant_id == tenant_id)
            .where(
                Report.status.in_(
                    (
                        ReportStatus.pending,
                        ReportStatus.generating,
                        ReportStatus.complete,
                        ReportStatus.partial,
                    )
                )
            )
            .order_by(Report.updated_at.desc())
            .limit(24)
        )
        lines = [row[0] for row in res.all() if row and row[0]]
        if not lines:
            return ctx

        identity = _extract_property_identity(lines)
        addr = (identity.get("address") or "").strip()
        pc = (identity.get("postcode") or "").strip()
        if addr:
            if pc and pc.upper() not in addr.upper():
                addr = f"{addr}, {pc}"
            ctx["property_address"] = addr
        elif pc:
            ctx["property_address"] = pc
        if identity.get("surveyor_name"):
            ctx["surveyor_name"] = str(identity["surveyor_name"]).strip()
        if identity.get("firm"):
            ctx["company_name"] = str(identity["firm"]).strip()
    except Exception as exc:  # noqa: BLE001 — context fetch must never block upload
        logger.debug("Tenant db_context backfill skipped tenant=%s: %s", tenant_id, exc)
    return ctx


async def build_upload_db_context(
    db: AsyncSession,
    tenant_id: str,
    *,
    client_name: str | None = None,
    property_address: str | None = None,
    surveyor_name: str | None = None,
) -> dict[str, str]:
    """Merge upload form PII fields with tenant session backfill from reports."""
    form_ctx = db_context_from_form_fields(
        client_name=client_name,
        property_address=property_address,
        surveyor_name=surveyor_name,
    )
    tenant_ctx = await fetch_tenant_db_context(db, tenant_id)
    return merge_db_context(tenant_ctx, form_ctx)


def serialize_db_context(ctx: dict[str, str] | None) -> str | None:
    """Persist ``db_context`` on a :class:`~app.db.models.Document` row."""
    if not ctx:
        return None
    cleaned = {k: v.strip() for k, v in ctx.items() if k and (v or "").strip()}
    return json.dumps(cleaned, ensure_ascii=False) if cleaned else None


def deserialize_db_context(raw: str | None) -> dict[str, str] | None:
    """Load persisted ``db_context`` JSON; returns ``None`` when absent/invalid."""
    if not raw or not str(raw).strip():
        return None
    try:
        data = json.loads(raw)
    except json.JSONDecodeError:
        logger.warning("Invalid redaction_context_json on document; ignoring.")
        return None
    if not isinstance(data, dict):
        return None
    return {
        str(k): str(v).strip()
        for k, v in data.items()
        if k and isinstance(v, (str, int, float)) and str(v).strip()
    }


def load_document_redaction_strategy(raw: str | None) -> RedactionStrategy:
    """Resolve a stored document strategy string (never raises — defaults safely)."""
    try:
        return parse_redaction_strategy(raw)
    except InvalidRedactionStrategyError:
        logger.warning("Unknown stored redaction_strategy %r; defaulting to AI_HYBRID.", raw)
        return RedactionStrategy.AI_HYBRID