"""Doctor dashboard read models (#12): clinical DB rows → frozen API schemas (`app/api/schemas.py`).

Everything shown to the doctor is derived from stored structured data — no LLM decides colours, trends or the
discontinuation risk (ADR-007). Time comes in as `now` (from `app.core.clock`, ADR-005).

Identity data (guardrail #5): the API is the doctor's display boundary, so it reads exactly two identity fields —
`display_name` (minimal identification for the treating doctor) and `birth_year` (age). SNILS, phone and chat
id are never selected into a response model.
"""

from collections.abc import Iterable, Sequence
from dataclasses import dataclass
from datetime import date, datetime, timedelta
from typing import Any, Literal, cast, get_args

from sqlalchemy import select
from sqlalchemy.ext.asyncio import AsyncSession

from app.api.schemas import (
    AlertOut,
    Color,
    DiscontinuationRisk,
    DoseOut,
    LevelName,
    Metric,
    Module,
    PatientCard,
    PatientListItem,
    SeriesPoint,
)
from app.core.rules import color, patient_status
from app.core.types import AlertLevel
from app.db import repo
from app.db.clinical import Alert, DailyCheckin, Dose, PatientProfile, Prescription, SafetyPlan, ScaleResult
from app.db.identity import Patient
from app.integrations.base import Prescription as PrescriptionOut
from app.services.anonymous_care import display_name as anonymous_name

COLOR_ORDER: dict[str, int] = {"red": 0, "yellow": 1, "grey": 2, "green": 3}
_MODULES = frozenset(get_args(Module))
_METRICS = frozenset(get_args(Metric))
_SCALE_LABELS = {"phq9": "PHQ-9", "pcl5": "PCL-5", "gad7": "GAD-7", "auditc": "AUDIT-C", "dar5": "DAR-5"}

# Daily check-in buckets → numbers for charts. Sleep: bucket midpoints in hours ("<4" plotted as 3, ">8" as 9).
SLEEP_HOURS = {"<4": 3.0, "4-6": 5.0, "6-8": 7.0, ">8": 9.0}
# Alcohol: 0 = none, 1 = 1–2 drinks, 2 = more (same coding as app/llm/summary.py expects).
ALCOHOL_LEVEL = {"none": 0.0, "1-2": 1.0, "more": 2.0}

# Trend: the last two complete results of one scale. A change smaller than about half of the minimal clinically
# important difference (PHQ-9: 5 points, PCL-5: 10 points) is reported as "flat" so noise does not flip arrows.
TREND_MIN_CHANGE = {"phq9": 3, "pcl5": 5}

CARD_DAYS = 60  # how far back the card series / dose log reach


# --------------------------------------------------------------------------- discontinuation risk index
#
# Rule-based index (NOT a trained model) of the risk that the patient drops the therapy, 0–100, for sorting the
# list and for the "talk about adherence" hint. Each factor is a deterministic check over the clinical DB with a
# fixed weight; the score is the capped sum. Weights reflect the known predictors of early antidepressant
# discontinuation (non-adherence, side effects in the first weeks, expressed intent to stop, low mood, alcohol):
#
#   factor                                            weight   label for the doctor
#   missed/skipped doses on >= 3 of the last 7 days     30     "пропуски приёма N/7 дн."
#   ... on 1–2 of the last 7 days                       15     "пропуски приёма N/7 дн."
#   side effects reported >= 3 days in a row            20     "побочные эффекты N дн. подряд"
#   ... 1–2 days in a row                               10     "побочные эффекты N дн. подряд"
#   intent to stop medication voiced in last 14 days    25     "хотел бросить (день N)"
#   open, unacknowledged quit_intent alert             +15     "не разобран сигнал «хочет бросить»"
#   mood <= 3 for >= 3 days in a row                    15     "настроение ≤ 3 N дн. подряд"
#   alcohol on >= 3 of the last 7 days                  15     "алкоголь N дн. из 7"
#   ... on 1–2 of the last 7 days                        5     "алкоголь N дн. из 7"
#
# Level: >= 60 high, >= 30 medium, otherwise low. "Day" = calendar day in the clinic time zone; streaks end today
# if today's check-in exists, otherwise yesterday (same anchor as app.core.rules), and a day without a check-in
# breaks a streak — we never assume a symptom the patient did not report.

RISK_WINDOW_DAYS = 7
QUIT_WINDOW_DAYS = 14
MOOD_LOW = 3
W_MISSED_MANY, W_MISSED_SOME = 30, 15
W_SIDE_EFFECTS_LONG, W_SIDE_EFFECTS_SHORT = 20, 10
W_QUIT_INTENT, W_QUIT_ALERT_OPEN = 25, 15
W_MOOD_LOW = 15
W_ALCOHOL_MANY, W_ALCOHOL_SOME = 15, 5
STREAK_DAYS = 3
RISK_HIGH, RISK_MEDIUM = 60, 30


@dataclass
class PatientData:
    """Everything the list item and the card are computed from (one patient, loaded once)."""

    profile: PatientProfile
    checkins: Sequence[DailyCheckin]
    results: Sequence[ScaleResult]
    doses: Sequence[Dose]
    prescriptions: Sequence[Prescription]
    alerts: Sequence[Alert]  # newest first
    contact_requested_at: datetime | None = None


async def load_patient_data(session: AsyncSession, profile: PatientProfile) -> PatientData:
    token = profile.token
    doses = (await session.scalars(select(Dose).where(Dose.token == token).order_by(Dose.planned_at))).all()
    rx = (
        await session.scalars(select(Prescription).where(Prescription.token == token).order_by(Prescription.id))
    ).all()
    dialog = await repo.get_dialog_state(session, token)
    requests = (dialog.data.get("contact_requests") or []) if dialog else []
    contact_at = datetime.fromisoformat(requests[-1]["ts"]) if requests else None
    return PatientData(
        profile=profile,
        checkins=await repo.list_checkins(session, token),
        results=await repo.list_scale_results(session, token),
        doses=doses,
        prescriptions=rx,
        alerts=await repo.list_alerts(session, token=token),
        contact_requested_at=contact_at,
    )


def _local(ts: datetime, now: datetime) -> date:
    return ts.astimezone(now.tzinfo).date()


def _streak_days(checkins: Iterable[DailyCheckin], today: date, pred: Any) -> int:
    """Length of the run of consecutive check-in days satisfying `pred`, ending at the anchor day."""
    by_day = {c.day: c for c in checkins if c.day <= today}
    day = today if today in by_day else today - timedelta(days=1)
    n = 0
    while (c := by_day.get(day)) is not None and pred(c):
        n += 1
        day -= timedelta(days=1)
    return n


def _therapy_day(profile: PatientProfile, day: date) -> int:
    return (day - profile.therapy_start).days if profile.therapy_start else 0


def _missed_days(d: PatientData, now: datetime, first: date) -> set[date]:
    # Same sources as rule missed_doses_7d: dose records and the check-in answer "did not take".
    days = {_local(x.planned_at, now) for x in d.doses if x.status in ("missed", "skipped") and x.planned_at <= now}
    days |= {c.day for c in d.checkins if c.med_taken is False}
    return {x for x in days if first <= x <= now.date()}


def discontinuation_risk(d: PatientData, now: datetime) -> DiscontinuationRisk:
    """Deterministic discontinuation-risk index; see the weight table above."""
    today = now.date()
    first = today - timedelta(days=RISK_WINDOW_DAYS - 1)
    score = 0
    factors: list[str] = []

    missed = len(_missed_days(d, now, first))
    if missed:
        score += W_MISSED_MANY if missed >= STREAK_DAYS else W_MISSED_SOME
        factors.append(f"пропуски приёма {missed}/{RISK_WINDOW_DAYS} дн.")

    side = _streak_days(d.checkins, today, lambda c: bool(c.side_effects))
    if side:
        score += W_SIDE_EFFECTS_LONG if side >= STREAK_DAYS else W_SIDE_EFFECTS_SHORT
        factors.append(f"побочные эффекты {side} дн. подряд")

    quit_since = today - timedelta(days=QUIT_WINDOW_DAYS - 1)
    quit_days = [c.day for c in d.checkins if c.quit_intent and quit_since <= c.day <= today]
    if quit_days:
        score += W_QUIT_INTENT
        factors.append(f"хотел бросить (день {_therapy_day(d.profile, max(quit_days))})")
    if any(a.code == "quit_intent" and a.acked_at is None for a in d.alerts):
        score += W_QUIT_ALERT_OPEN
        factors.append("не разобран сигнал «хочет бросить»")

    low = _streak_days(d.checkins, today, lambda c: c.mood is not None and c.mood <= MOOD_LOW)
    if low >= STREAK_DAYS:
        score += W_MOOD_LOW
        factors.append(f"настроение ≤ {MOOD_LOW} {low} дн. подряд")

    drinking = sum(1 for c in d.checkins if first <= c.day <= today and c.alcohol_bucket not in (None, "none"))
    if drinking:
        score += W_ALCOHOL_MANY if drinking >= STREAK_DAYS else W_ALCOHOL_SOME
        factors.append(f"алкоголь {drinking} дн. из {RISK_WINDOW_DAYS}")

    score = min(score, 100)
    level: Literal["low", "medium", "high"] = (
        "high" if score >= RISK_HIGH else "medium" if score >= RISK_MEDIUM else "low"
    )
    return DiscontinuationRisk(score=score, level=level, factors=factors)


# --------------------------------------------------------------------------- list item


def _complete(d: PatientData, scale: str) -> list[ScaleResult]:
    return [r for r in d.results if r.scale == scale and r.complete]  # repo returns them oldest first


def trend(d: PatientData) -> Literal["up", "down", "flat", "unknown"]:
    """Direction of symptom scores (up = worse) from the scale with the most recent pair of results."""
    pairs = [(s, rs[-2:]) for s in TREND_MIN_CHANGE if len(rs := _complete(d, s)) >= 2]
    if not pairs:
        return "unknown"
    scale, (prev, last) = max(pairs, key=lambda p: p[1][-1].ts)
    delta = last.total - prev.total
    if abs(delta) < TREND_MIN_CHANGE[scale]:
        return "flat"
    return "up" if delta > 0 else "down"


def adherence_7d(d: PatientData, now: datetime) -> float | None:
    """Taken / planned doses of the last 7 days; doses still `pending` (not resolved yet) are not counted."""
    since = now - timedelta(days=RISK_WINDOW_DAYS)
    resolved = [x for x in d.doses if since < x.planned_at <= now and x.status != "pending"]
    if not resolved:
        return None
    return round(sum(1 for x in resolved if x.status == "taken") / len(resolved), 2)


def _drug_label(rx: Prescription) -> str:
    return f"{rx.drug} {rx.dose}".strip()


def _active_rx(d: PatientData, today: date) -> Prescription | None:
    active = [
        r
        for r in d.prescriptions
        if (r.start_date is None or r.start_date <= today) and (r.end_date is None or r.end_date >= today)
    ]
    return (active or list(d.prescriptions) or [None])[-1]


def _last_event(d: PatientData, now: datetime) -> str:
    alert = d.alerts[0] if d.alerts else None
    scales = [r for r in d.results if r.complete and r.scale in _SCALE_LABELS]
    scale = scales[-1] if scales else None
    if d.contact_requested_at is not None and all(
        d.contact_requested_at >= ts for ts in [x.ts for x in (alert, scale) if x is not None]
    ):
        return "Пациент просит врача связаться"
    if scale is not None and (alert is None or scale.ts > alert.ts):
        day = _therapy_day(d.profile, _local(scale.ts, now))
        return f"{_SCALE_LABELS[scale.scale]} = {scale.total} (день {day})"
    return alert.reason if alert is not None else ""


def _age(patient: Patient | None, profile: PatientProfile, today: date) -> int:
    birth_year = (patient.birth_year if patient else None) or profile.birth_year
    return today.year - birth_year if birth_year else 0


def list_item(d: PatientData, patient: Patient | None, now: datetime) -> PatientListItem:
    p = d.profile
    today = now.date()
    open_alerts = [a for a in d.alerts if a.acked_at is None]
    # ADR-006: the colour is derived from open alert levels (the source of truth), not from the cached column.
    status = patient_status(AlertLevel[a.level.upper()] for a in open_alerts)
    phq9, pcl5 = _complete(d, "phq9"), _complete(d, "pcl5")
    rx = _active_rx(d, today)
    return PatientListItem(
        token=p.token,
        display_name=patient.display_name if patient else anonymous_name(p.token),
        age=_age(patient, p, today),
        modules=[cast(Module, m) for m in p.modules or [] if m in _MODULES],
        drug=_drug_label(rx) if rx else None,
        therapy_day=_therapy_day(p, today),
        next_visit=p.next_visit,
        status_color=status.color,
        top_alert_level=cast(LevelName, status.top_level.name.lower()) if status.top_level is not None else None,
        open_alerts=len(open_alerts),
        phq9_last=phq9[-1].total if phq9 else None,
        pcl5_last=pcl5[-1].total if pcl5 else None,
        trend=trend(d),
        adherence_7d=adherence_7d(d, now),
        discontinuation_risk=discontinuation_risk(d, now),
        last_event=_last_event(d, now),
    )


def sort_key(item: PatientListItem) -> tuple[int, int, str]:
    """red → yellow → grey → green, then discontinuation risk (desc), then name for a stable order."""
    return (COLOR_ORDER[item.status_color], -item.discontinuation_risk.score, item.display_name)


# --------------------------------------------------------------------------- card


def alert_out(a: Alert, patient_name: str) -> AlertOut:
    level = cast(LevelName, a.level)
    return AlertOut(
        id=a.id,
        token=a.token,
        patient_name=patient_name,
        ts=a.ts,
        level=level,
        color=cast(Color, color([AlertLevel[a.level.upper()]])),
        code=a.code,
        reason=a.reason,
        acked_by=a.acked_by,
        acked_at=a.acked_at,
    )


def series(d: PatientData, now: datetime) -> list[SeriesPoint]:
    since = now.date() - timedelta(days=CARD_DAYS)
    out: list[SeriesPoint] = []
    for c in d.checkins:
        if not since <= c.day <= now.date():
            continue
        if c.mood is not None:
            out.append(SeriesPoint(day=c.day, metric="mood", value=float(c.mood)))
        if c.sleep_bucket in SLEEP_HOURS:
            out.append(SeriesPoint(day=c.day, metric="sleep_hours", value=SLEEP_HOURS[c.sleep_bucket]))
        if c.alcohol_bucket in ALCOHOL_LEVEL:
            out.append(SeriesPoint(day=c.day, metric="alcohol", value=ALCOHOL_LEVEL[c.alcohol_bucket]))
    for r in d.results:
        day = _local(r.ts, now)
        if r.complete and r.scale in _METRICS and since <= day and r.ts <= now:
            out.append(SeriesPoint(day=day, metric=cast(Metric, r.scale), value=float(r.total), scheduled=r.scheduled))
    return sorted(out, key=lambda p: (p.day, p.metric))


def doses(d: PatientData, now: datetime) -> list[DoseOut]:
    labels = {r.id: _drug_label(r) for r in d.prescriptions}
    since = now - timedelta(days=CARD_DAYS)
    return [
        DoseOut(planned_at=x.planned_at, drug=labels.get(x.prescription_id, ""), status=x.status)  # type: ignore[arg-type]
        for x in d.doses
        if since <= x.planned_at <= now + timedelta(days=1)
    ]


def prescriptions(d: PatientData) -> list[PrescriptionOut]:
    out = []
    for r in d.prescriptions:
        schedule = dict(r.schedule or {})
        out.append(
            PrescriptionOut(
                drug=r.drug,
                dose=r.dose,
                schedule={"times": [str(t) for t in schedule.get("times", [])]},
                instructions=" ".join(str(t) for t in _as_list(schedule.get("instructions"))),
                start_date=r.start_date,
                end_date=r.end_date,
            )
        )
    return out


def _as_list(value: Any) -> list[Any]:
    if value is None:
        return []
    return list(value) if isinstance(value, list) else [value]


def _plan_line(step: Any) -> str:
    # Same rendering as the dialog's safety-plan reply (app/dialog/actions.py).
    if isinstance(step, dict):
        return ": ".join(str(v) for v in step.values() if v)
    return str(step)


async def safety_plan(session: AsyncSession, token: str) -> list[str] | None:
    row = await session.get(SafetyPlan, token)
    return [_plan_line(s) for s in row.steps] if row is not None and row.steps else None


async def build_card(session: AsyncSession, d: PatientData, patient: Patient | None, now: datetime) -> PatientCard:
    """Card without the summary (the caller adds it: the LLM call must not hold the DB session)."""
    item = list_item(d, patient, now)
    return PatientCard(
        patient=item,
        prescriptions=prescriptions(d),
        series=series(d, now),
        doses=doses(d, now),
        alerts=[alert_out(a, item.display_name) for a in d.alerts],
        safety_plan=await safety_plan(session, d.profile.token),
    )