thermograph/web/app.py

"""Thermograph API — grade recent local weather against ~45 years of climatology."""
import contextlib
import datetime
import functools
import hashlib
import hmac
import json
import mimetypes
import os
import queue
import threading
import time

from fastapi import APIRouter, FastAPI, HTTPException, Query, Request, Response
from fastapi.middleware.gzip import GZipMiddleware
from fastapi.responses import RedirectResponse
from fastapi.staticfiles import StaticFiles

from accounts import api_accounts
from core import audit
from data import climate
from web import content
from notifications import digest
from notifications import discord_interactions as discord_interactions_mod
from notifications import discord_link
from accounts import db
from data import grid
from core import metrics
from notifications import notify
import paths
from data import places
from core import singleton
from data import store
from accounts import users
from web import views
from web.schemas import UserCreate, UserRead, UserUpdate

FRONTEND_DIR = paths.FRONTEND_DIR

# Everything (pages, assets, API) is served under this base path so the app can
# live at https://<host>/thermograph/ behind a shared domain. Override with the
# THERMOGRAPH_BASE env var. The frontend uses base-relative URLs, so it follows
# this automatically without hardcoding the prefix.
#
# THERMOGRAPH_BASE=/ (or empty) means the app owns the whole domain: BASE becomes
# "" so every route sits at the domain root ("/", "/calendar", "/api/v2/…") with
# no prefix and no leading double slash. A non-empty value keeps the sub-path form
# ("/thermograph").
_BASE = os.environ.get("THERMOGRAPH_BASE", "/thermograph").strip("/")
BASE = f"/{_BASE}" if _BASE else ""

# Which duties this process performs: "web" (serve requests, no notifier),
# "worker" (own the notifier, still serves requests today — see _lifespan), or
# "all" (both — today's single-process default, so an unset ROLE is unchanged
# behavior). Read once at import: toggling requires a process restart, same as
# every other env-driven constant here (BASE, WORKERS, ...).
ROLE = os.environ.get("THERMOGRAPH_ROLE", "all").strip().lower()

# The PWA manifest is served as a static file; register its media type so
# StaticFiles labels it correctly (not in Python's default mimetypes map).
mimetypes.add_type("application/manifest+json", ".webmanifest")


def _weather_fetch_error(e) -> HTTPException:
    """A clean, retryable 503 when weather data is temporarily unavailable; the
    raw error as a 502 otherwise. The fetch layer classifies its own failures
    (climate.WeatherUnavailable carries the user-facing text) so no message
    parsing happens here; the is_rate_limit fallback covers a raw upstream 429
    from a fetcher that didn't classify it."""
    if isinstance(e, climate.WeatherUnavailable):
        return HTTPException(status_code=503, detail=str(e))
    if climate.is_rate_limit(e):
        return HTTPException(status_code=503, detail=climate.limit_message(False))
    return HTTPException(status_code=502, detail=f"weather data fetch failed: {e}")


# --- background neighbor warming ---------------------------------------------
# /cell?neighbors=1 asks the server to warm the 8 grid cells around the request
# (the frontend used to fire 8 staggered prefetch requests, mirroring grid.py's
# snapping math in JS). One worker drains the queue so warms never stack; the
# hard guarantee matches prefetch=1 — a cell with no cached archive is skipped,
# so no weather-API quota is ever spent, and reverse_geocode itself paces the
# at-most-one Nominatim call per never-labeled cell (~1/s policy).

_WARM_TTL = 3600.0                 # don't re-enqueue a cell within the hour
_WARM_SEEN: dict[str, float] = {}  # cell_id -> last enqueue time
_warm_queue: "queue.Queue[dict]" = queue.Queue()


def _warm_cell(cell: dict) -> None:
    """Materialize the history-derived slices for one cell — the same rows the
    prefetch=1 bundle serves. Never fetches weather upstream."""
    history = climate.load_cached_history(cell)
    if history is None or history.is_empty():
        return
    place = climate.reverse_geocode(cell["center_lat"], cell["center_lon"])
    token = views.history_token(history)
    start_ts, end_ts = views.cal_span(history, None, None, 24)
    cal_key = views.calendar_key(start_ts, end_ts, 24)
    if store.get_payload("calendar", cell["id"], cal_key, token) is None:
        store.put_payload("calendar", cell["id"], cal_key, token,
                          views.build_calendar(cell, history, start_ts, end_ts, 24, place))
    last = history["date"].max()
    day_key = views.day_key(last)
    if store.get_payload("day", cell["id"], day_key, token) is None:
        store.put_payload("day", cell["id"], day_key, token,
                          views.build_day(cell, history, last, place))


def _enqueue_neighbor_warming(cell: dict) -> None:
    now = time.time()
    for n in grid.neighbors(cell):
        if now - _WARM_SEEN.get(n["id"], 0.0) < _WARM_TTL:
            continue
        _WARM_SEEN[n["id"]] = now
        _warm_queue.put(n)


def _neighbor_warmer() -> None:
    while True:
        cell = _warm_queue.get()
        try:
            _warm_cell(cell)
        except Exception:  # noqa: BLE001 - warming is best-effort, never fatal
            pass
        finally:
            _warm_queue.task_done()


def _should_run_notifier() -> bool:
    """Whether this process should start the subscription-notifier thread: ROLE must
    permit it, and it must win the leader election. Split out from _lifespan so this
    decision is unit-testable without booting the full app (DB init, places index,
    neighbor warmer) — see tests/web/test_role.py."""
    return ROLE in ("worker", "all") and singleton.claim_leader()


@contextlib.asynccontextmanager
async def _lifespan(app):
    # Warm the local place-name index (/suggest's typo tolerance) in the
    # background; the app boots and serves fine without it — suggestions just
    # fall back to the upstream geocoder until it's ready. A server-startup
    # hook (not import time) so offline importers (tests, the migrate script's
    # dependencies) don't kick off a GeoNames download. The neighbor warmer is
    # also a server concern: enqueues before startup just wait in the queue.
    # Create the account DB tables (users/sessions/subscriptions/notifications)
    # before serving — a fast no-op once they exist.
    await db.create_db_and_tables()
    places.start_loading()
    threading.Thread(target=_neighbor_warmer, name="neighbor-warmer", daemon=True).start()
    # Background subscription evaluator. Its periodic sweep fetches upstream on a timer
    # (not per request), so with multiple workers — or multiple *hosts* under Swarm —
    # it must run in exactly one: elect a leader (see singleton.claim_leader, which
    # picks a host-local flock or a cluster-wide Postgres advisory lock from env).
    # Unset (single worker / dev / tests) => always the leader. The neighbor warmer
    # above stays per-worker: each drains its own request-fed queue. ROLE further
    # restricts this to processes that are allowed to own it at all — "web" replicas
    # never start it even if they'd otherwise win the leader election, so a stateless
    # N-replica web tier can scale without also scaling notifier instances.
    if _should_run_notifier():
        notify.start()
    yield
    notify.stop()


app = FastAPI(title="Thermograph", version="0.2.0", lifespan=_lifespan)

# Compress every sizeable response (the 2-year calendar JSON shrinks ~6-8×).
# Applies to API JSON and static assets alike; tiny responses are left alone.
app.add_middleware(GZipMiddleware, minimum_size=1024)


@app.get("/healthz")
def healthz():
    """Liveness probe: the process is up and serving. Deliberately does no I/O (no
    DB query, no upstream call) so it stays cheap and reliable enough for a tight
    Swarm/Docker healthcheck interval — readiness (DB reachable, etc.) is a
    separate concern the ingress health-gate covers by hitting a real route. Not
    under BASE: a fixed, unauthenticated path a healthcheck can hit without
    knowing THERMOGRAPH_BASE, on both the web and worker roles."""
    return {"status": "ok", "role": ROLE}


def _client_ip(request) -> "str | None":
    """The real client IP: the left-most X-Forwarded-For hop when a proxy fronts us
    (Caddy on prod sets it), otherwise the direct peer address (LAN dev)."""
    xff = request.headers.get("x-forwarded-for")
    if xff:
        return xff.split(",")[0].strip()
    return request.client.host if request.client else None


@app.middleware("http")
async def revalidate_static(request, call_next):
    """Serve the frontend with no-cache (NOT no-store): browsers may keep a copy
    but must revalidate it on every use. Starlette's FileResponse/StaticFiles
    already emit ETag + Last-Modified, so an unchanged asset costs one conditional
    request answered with an empty 304 instead of a full re-download — page-to-page
    navigation stops re-transferring the JS/CSS/HTML while still picking up every
    deploy immediately (no "my change isn't showing" bugs)."""
    response = await call_next(request)
    path = request.url.path
    try:
        cat = metrics.classify_inbound(path, BASE)
        metrics.record_inbound(cat, response.status_code)
        # Retain per-request client IPs for later analysis; skip static assets and the
        # dashboard's own metrics polling to keep the log to real, meaningful traffic.
        if cat not in ("static", "metrics", "event"):
            audit.log_access({"ip": _client_ip(request), "method": request.method,
                              "path": path, "status": response.status_code, "cat": cat})
    except Exception:  # noqa: BLE001 - never let instrumentation break a response
        pass
    pages = (BASE, f"{BASE}/", f"{BASE}/calendar", f"{BASE}/day", f"{BASE}/score",
             f"{BASE}/compare", f"{BASE}/legend", f"{BASE}/alerts", f"{BASE}/privacy")
    if path.endswith((".js", ".css", ".html")) or path in pages:
        response.headers["Cache-Control"] = "no-cache"
    return response


# --- derived-store plumbing --------------------------------------------------
# Every graded payload is cached in SQLite under (kind, cell, key) and validated
# by a token that encodes what it was computed from (see store.py). The freshness
# drivers stay where they were — climate.get_history tops up the archive tail
# hourly and get_recent_forecast refetches hourly — and the tokens are derived
# from what those return, so a cached payload expires exactly when its inputs
# change and never before. The same tokens double as ETags: a client sending
# If-None-Match gets an empty 304 without the payload even being loaded.
# The payloads themselves are assembled in views.py, shared with the offline
# migrate script.

def _etag_for(kind: str, cell_id: str, key: str, token: str) -> str:
    """Deterministic weak ETag from a payload's identity + validity token. Weak
    because the same content may be served under different encodings (gzip)."""
    h = hashlib.sha1(f"{kind}:{cell_id}:{key}:{token}".encode()).hexdigest()[:20]
    return f'W/"{h}"'


def _not_modified(request: Request, etag: str) -> bool:
    """Does the client already hold this exact payload? Answerable from the token
    alone — no payload load needed."""
    inm = request.headers.get("if-none-match")
    if not inm:
        return False
    tags = {t.strip() for t in inm.split(",")}
    return "*" in tags or etag in tags or etag.removeprefix("W/") in tags


def _encode(payload: dict) -> bytes:
    """Compact JSON bytes, matching store.put_payload's encoding (allow_nan=False
    mirrors Starlette — a NaN from grading should fail loudly, not reach a client)."""
    return json.dumps(payload, default=str, allow_nan=False, separators=(",", ":")).encode()


def _json_response(body: bytes, etag: str | None = None) -> Response:
    headers = {"ETag": etag} if etag else {}
    return Response(content=body, media_type="application/json", headers=headers)


def _fetch_history(run, cell, recent_too=False, recent_phase="recent"):
    """The shared fetch preamble for every data route: the archive record (and
    optionally the hourly recent/forecast bundle) with upstream failures mapped
    to clean HTTP errors and an empty record to a 404."""
    try:
        with run.phase("history"):
            history, cache_meta = climate.get_history(cell)
        recent = None
        if recent_too:
            with run.phase(recent_phase):
                recent = climate.get_recent_forecast(cell)
    except Exception as e:  # noqa: BLE001
        raise _weather_fetch_error(e)
    if history.is_empty():
        raise HTTPException(status_code=404, detail="No historical data for this cell.")
    return history, cache_meta, recent


def _cached_response(request, run, kind, cell, key, token, build, cache_placeless=True):
    """The shared derived-store flow behind every per-view endpoint:
    If-None-Match -> empty 304; a token-valid store row -> replay its exact
    bytes; otherwise resolve the place label, build the payload, persist, serve.

    ``build(place) -> payload`` does any endpoint-specific audit tagging itself.
    ``cache_placeless=False`` skips persisting a payload whose place label
    failed to resolve (a transient reverse-geocode miss) — otherwise the
    bare-coordinates fallback would stick for the life of the token."""
    cid = cell["id"]
    etag = _etag_for(kind, cid, key, token)
    if _not_modified(request, etag):
        run.set(run_type="cache", not_modified=True)
        return Response(status_code=304, headers={"ETag": etag})
    body = store.get_payload(kind, cid, key, token)
    if body is not None:
        run.set(run_type="cache", history_source="store")
        return _json_response(body, etag)
    with run.phase("reverse_geocode"):
        place = climate.reverse_geocode(cell["center_lat"], cell["center_lon"])
    payload = build(place)
    return _json_response(
        store.put_payload(kind, cid, key, token, payload,
                          cache=cache_placeless or place is not None), etag)



# --- endpoints ----------------------------------------------------------------

def api_geocode(q: str = Query(..., min_length=1)):
    try:
        return {"results": climate.geocode(q)}
    except Exception as e:  # noqa: BLE001 - surface upstream failures to the client
        raise HTTPException(status_code=502, detail=f"geocoding failed: {e}")


_SUGGEST_LIMIT = 5


@functools.lru_cache(maxsize=1024)
def _suggest_upstream(q: str) -> tuple:
    """Open-Meteo lookup for /suggest, memoized per query string — type-ahead
    re-asks the same prefixes constantly (backspacing, retyping). Failures
    raise and are not cached, so a transient upstream error doesn't stick."""
    return tuple(climate.geocode(q, count=_SUGGEST_LIMIT))


def api_suggest(q: str = Query(..., min_length=1)):
    """Type-ahead location suggestions: the top 5 places for a (possibly
    typo'd) query prefix; `corrected` reports the respelling that produced the
    results ("pest seattle" → "west seattle"), or null. The ranking/typo policy
    lives in places.suggest."""
    try:
        results, corrected = places.suggest(q, _suggest_upstream, _SUGGEST_LIMIT)
    except Exception as e:  # noqa: BLE001 - nothing at all to serve
        raise HTTPException(status_code=502, detail=f"geocoding failed: {e}")
    return {"results": results, "corrected": corrected}


def api_place(
    lat: float = Query(..., ge=-90, le=90),
    lon: float = Query(..., ge=-180, le=180),
):
    """Best-effort neighbourhood/city label for a point — the same string the view
    endpoints expose as ``place`` (snapped to the grid cell, reverse-geocoded and
    cached). Resolved on its own so the compare page can show a location's name as
    soon as it's added, before its full series loads."""
    cell = grid.snap(lat, lon)
    with audit.RunAudit(endpoint="place", lat=round(lat, 4), lon=round(lon, 4),
                        cell_id=cell["id"]) as run:
        with run.phase("reverse_geocode"):
            place = climate.reverse_geocode(cell["center_lat"], cell["center_lon"])
    return {"place": place,
            "cell": {"center_lat": cell["center_lat"], "center_lon": cell["center_lon"]}}


def api_grade(
    request: Request,
    lat: float = Query(..., ge=-90, le=90),
    lon: float = Query(..., ge=-180, le=180),
    date: str | None = Query(None, description="target date YYYY-MM-DD (default today)"),
    days: int = Query(14, ge=1, le=60, description="days of history to grade before the target"),
    after: int = Query(14, ge=0, le=14,
                       description="days to grade after the target (observed, or forecast when future; "
                                   "the forecast reaches ~7 days out so future days cap there)"),
):
    target = datetime.date.fromisoformat(date[:10]) if date else datetime.date.today()
    cell = grid.snap(lat, lon)

    with audit.RunAudit(
        endpoint="grade",
        lat=round(lat, 4),
        lon=round(lon, 4),
        target_date=target.isoformat(),
        recent_days=days,
        cell_id=cell["id"],
    ) as run:
        history, cache_meta, recent = _fetch_history(run, cell, recent_too=True)
        return _cached_response(
            request, run, "grade", cell,
            views.grade_key(target, days, after), views.recent_token(history, cell["id"]),
            lambda place: views.build_grade(cell, target, days, history, recent,
                                            cache_meta, place, run, after=after))


def api_calendar(
    request: Request,
    lat: float = Query(..., ge=-90, le=90),
    lon: float = Query(..., ge=-180, le=180),
    start: str | None = Query(None, description="first date YYYY-MM-DD (overrides months)"),
    end: str | None = Query(None, description="last date YYYY-MM-DD (default = latest available)"),
    months: int = Query(24, ge=1, le=24, description="months back to grade when no start given"),
):
    """Grade every day over a date range (default: last 2 years) for the calendar view.

    A custom [start, end] range is supported and capped at ~2 years per request;
    the frontend splits longer spans into successive 2-year chunks. Grades against
    the already-cached history record (which reaches to ~6 days ago), so no extra
    fetch is needed. The graded payload is cached in the derived store keyed by
    the clamped span and validated by the record's end date, so a repeat span is
    a database read — across restarts — until new archive days arrive. Returns a
    compact per-day shape (see grading.grade_range). New in API v2.
    """
    cell = grid.snap(lat, lon)

    with audit.RunAudit(
        endpoint="calendar",
        lat=round(lat, 4),
        lon=round(lon, 4),
        recent_days=months * 31,  # approximate span graded
        cell_id=cell["id"],
    ) as run:
        history, cache_meta, _ = _fetch_history(run, cell)
        start_ts, end_ts = views.cal_span(history, start, end, months)

        def build(place):
            payload = views.build_calendar(cell, history, start_ts, end_ts, months, place, run)
            full = not cache_meta.get("cached", False)
            run.set(run_type="full" if full else "partial",
                    history_source="fetch" if full else "cache")
            return payload

        # Compare loads several cells at once, so a transient reverse-geocode miss
        # is most likely here; cache_placeless=False keeps that miss un-persisted.
        return _cached_response(request, run, "calendar", cell,
                                views.calendar_key(start_ts, end_ts, months),
                                views.history_token(history), build,
                                cache_placeless=False)


def api_day(
    request: Request,
    lat: float = Query(..., ge=-90, le=90),
    lon: float = Query(..., ge=-180, le=180),
    date: str | None = Query(None, description="target date YYYY-MM-DD (default = latest available)"),
):
    """Full percentile breakdown for a single day: the value at every tier boundary
    in that day-of-year's ±7-day window, plus where the observed values land.

    Powers the single-day detail page. Grades against the cached history record;
    for a date newer than the cache it pulls the recent window for the observation
    (those payloads expire hourly — the recent bundle's own cadence — while fully
    archived days stay valid until the record itself advances). New in API v2.
    """
    cell = grid.snap(lat, lon)

    with audit.RunAudit(
        endpoint="day",
        lat=round(lat, 4),
        lon=round(lon, 4),
        cell_id=cell["id"],
    ) as run:
        history, cache_meta, _ = _fetch_history(run, cell)
        last = history["date"].max()
        target = datetime.date.fromisoformat(date[:10]) if date else last
        run.set(target_date=target.isoformat())

        def build(place):
            payload = views.build_day(cell, history, target, place, run)
            full = not cache_meta.get("cached", False)
            run.set(run_type="full" if full else "partial",
                    history_source="fetch" if full else "cache")
            return payload

        return _cached_response(request, run, "day", cell,
                                views.day_key(target), views.day_token(history, target),
                                build)


def api_score(
    request: Request,
    lat: float = Query(..., ge=-90, le=90),
    lon: float = Query(..., ge=-180, le=180),
):
    """Climate-drift score for a location: how far the last 6 years have moved
    from the full 45-year baseline, per metric, percentile category and season,
    rolled into per-metric and overall scores.

    Derived purely from the archive record, so it is cached until the record's
    tail advances (the hourly top-up); the scoring-math version is baked into the
    cache key so a math change invalidates only score rows. New in API v2.
    """
    cell = grid.snap(lat, lon)

    with audit.RunAudit(endpoint="score", lat=round(lat, 4), lon=round(lon, 4),
                        cell_id=cell["id"]) as run:
        history, cache_meta, _ = _fetch_history(run, cell)

        def build(place):
            payload = views.build_score(cell, history, place, run)
            full = not cache_meta.get("cached", False)
            run.set(run_type="full" if full else "partial",
                    history_source="fetch" if full else "cache")
            return payload

        return _cached_response(request, run, "score", cell,
                                views.score_key(), views.history_token(history), build)


def api_forecast(
    request: Request,
    lat: float = Query(..., ge=-90, le=90),
    lon: float = Query(..., ge=-180, le=180),
    days: int = Query(7, ge=1, le=14, description="how many forecast days ahead to grade"),
):
    """Grade the next `days` of forecast weather against local climatology.

    Same shape as /grade (so the frontend renders it identically), but `recent`
    holds the forward forecast — furthest-out day first. The forecast is fetched
    fresh and cached only ~1 hour (see climate.get_recent_forecast) to track
    updates; the graded payload's validity is tied to that fetch stamp, so it
    expires exactly when a new forecast lands. New in API v2.
    """
    cell = grid.snap(lat, lon)

    with audit.RunAudit(endpoint="forecast", lat=round(lat, 4), lon=round(lon, 4),
                        recent_days=days, cell_id=cell["id"]) as run:
        history, _, fc = _fetch_history(run, cell, recent_too=True, recent_phase="forecast")
        today = datetime.date.today()

        def build(place):
            payload = views.build_forecast(cell, days, history, fc, today, place, run)
            run.set(run_type="partial")
            return payload

        return _cached_response(request, run, "forecast", cell,
                                views.forecast_key(today, days),
                                views.recent_token(history, cell["id"]), build)


def api_cell(
    request: Request,
    lat: float = Query(..., ge=-90, le=90),
    lon: float = Query(..., ge=-180, le=180),
    prefetch: int = Query(0, ge=0, le=1,
                          description="1 = warm-only: never fetch weather upstream; 204 for a cold cell"),
    neighbors: int = Query(0, ge=0, le=1,
                           description="1 = also warm the 8 surrounding cells in the background "
                                       "(warm-only: cold neighbors are skipped, no upstream quota)"),
):
    """One bundle carrying every view's payload for a cell, so the frontend warms
    all views with a single request instead of four.

    Each slice is the EXACT payload its per-view endpoint returns — built by the
    same builders and cached under the same derived-store keys/tokens — paired
    with the etag that endpoint would emit. The client seeds its per-view cache
    from the slices and later revalidates each view individually with
    If-None-Match, so the bundle and the per-view endpoints stay one cache.

    prefetch=1 is the neighbor-warming mode with a hard guarantee: it never
    spends weather-API quota. A cell with no cached archive answers 204 (no
    body), and only the history-derived slices (calendar + latest-day detail)
    are built — grading recent/forecast days needs the hourly upstream bundle.
    Reverse geocoding makes at most one Nominatim call for a never-labeled cell;
    the client staggers neighbor prefetches to respect that service. New in API v2.
    """
    cell = grid.snap(lat, lon)
    today = datetime.date.today()

    with audit.RunAudit(endpoint="cell", lat=round(lat, 4), lon=round(lon, 4),
                        cell_id=cell["id"], prefetch=bool(prefetch)) as run:
        recent = None
        if prefetch:
            history = climate.load_cached_history(cell)
            if history is None or history.is_empty():
                run.set(run_type="cold-skip")
                return Response(status_code=204)
            cache_meta = {"cached": True}
        else:
            history, cache_meta, recent = _fetch_history(run, cell, recent_too=True)

        cid = cell["id"]
        last = history["date"].max()

        if neighbors:
            _enqueue_neighbor_warming(cell)

        with run.phase("reverse_geocode"):
            place = climate.reverse_geocode(cell["center_lat"], cell["center_lon"])

        def slice_for(kind: str, key: str, token: str, build) -> dict:
            """The endpoint's derived row (or build + store it), paired with the
            etag that endpoint would emit for the same request."""
            etag = _etag_for(kind, cid, key, token)
            data = store.get_json(kind, cid, key, token)
            if data is None:
                data = build()
                store.put_payload(kind, cid, key, token, data)
            return {"etag": etag, "data": data}

        slices = {}
        hist_token = views.history_token(history)
        # Calendar: the default last-24-months span — what the calendar view (and
        # the cross-view prefetch) requests first.
        start_ts, end_ts = views.cal_span(history, None, None, 24)
        slices["calendar"] = slice_for(
            "calendar", views.calendar_key(start_ts, end_ts, 24), hist_token,
            lambda: views.build_calendar(cell, history, start_ts, end_ts, 24, place, run))

        if prefetch:
            # Latest archived day: its observation comes from history alone, so
            # it's buildable without the (skipped) recent bundle.
            slices["day"] = slice_for(
                "day", views.day_key(last), hist_token,
                lambda: views.build_day(cell, history, last, place, run))
        else:
            rf_token = views.recent_token(history, cid)
            slices["grade"] = slice_for(
                "grade", views.grade_key(today, 14, 14), rf_token,
                lambda: views.build_grade(cell, today, 14, history, recent, cache_meta, place, run, after=14))
            slices["forecast"] = slice_for(
                "forecast", views.forecast_key(today, 7), rf_token,
                lambda: views.build_forecast(cell, 7, history, recent, today, place, run))
            slices["day"] = slice_for(
                "day", views.day_key(today), views.day_token(history, today),
                lambda: views.build_day(cell, history, today, place, run, recent=recent))

        # The bundle's identity is the combination of its slices' identities.
        etag = _etag_for("cell", cid, "bundle" + (":p" if prefetch else ""),
                         "|".join(s["etag"] for s in slices.values()))
        if _not_modified(request, etag):
            run.set(run_type="cache", not_modified=True)
            return Response(status_code=304, headers={"ETag": etag})
        run.set(run_type="bundle", slices=sorted(slices))
        payload = {
            "api_version": "v2",
            "cell": cell,
            "place": place,
            "today": today.isoformat(),
            "slices": slices,
        }
        # Not stored as its own derived row — the slices already are.
        return _json_response(_encode(payload), etag)


# --- ops metrics (gated) ---------------------------------------------------
def _metrics_allowed(request: Request) -> bool:
    """Guard the ops metrics: a valid token allows from anywhere (for SSH tunnels);
    with no token it's direct-loopback-only, and any forwarded header (i.e. proxied
    via Caddy) is refused — so prod ops data never leaks through the public domain."""
    token = os.environ.get("THERMOGRAPH_METRICS_TOKEN", "").strip()
    if token:
        supplied = request.headers.get("x-metrics-token") or request.query_params.get("token") or ""
        if hmac.compare_digest(supplied, token):
            return True
    client = request.client.host if request.client else None
    forwarded = any(h in request.headers for h in
                    ("x-forwarded-for", "x-forwarded-host", "x-forwarded-proto"))
    return client in ("127.0.0.1", "::1") and not forwarded


def api_metrics(request: Request):
    """Live in-process traffic counters + worker facts for the ops dashboard."""
    if not _metrics_allowed(request):
        raise HTTPException(status_code=404)   # 404, not 403 — don't reveal the route
    snap = metrics.snapshot()
    snap["warm_queue_depth"] = _warm_queue.qsize()
    snap["warm_seen"] = len(_WARM_SEEN)
    snap["threads"] = threading.active_count()
    snap["thread_names"] = sorted(t.name for t in threading.enumerate())
    return snap


async def api_event(request: Request) -> Response:
    """Record one product event (a tap on "use my location", a digest signup, …).

    Deliberately minimal: the body carries only an allowlisted event name. The
    referrer is read from this request's own Referer header — a client-supplied
    one would be forgeable and buys nothing — and is reduced to a bare domain.
    No cookies, no per-visitor id, no full URLs.

    Always answers 204, whether the event was counted, rejected by the allowlist,
    or dropped by the rate limiter, so probing the endpoint reveals nothing. That
    also suits navigator.sendBeacon, which ignores the response.
    """
    name = ""
    try:
        body = await request.json()
        if isinstance(body, dict):
            name = str(body.get("event") or "")
    except Exception:  # noqa: BLE001 - a junk body is just a no-op event
        pass
    if name:
        metrics.record_event(
            name,
            referer=request.headers.get("referer"),
            host=request.headers.get("host"),
            ip=_client_ip(request),
        )
    return Response(status_code=204)


# --- API versioning --------------------------------------------------------
# Non-backward-compatible additions land in a new version; every version stays
# mounted and served simultaneously. v1 is the original contract (grade/geocode);
# v2 adds the calendar. The unversioned /api/* paths are kept as aliases of v1 so
# existing clients keep working.
v1 = APIRouter(tags=["v1"])
v1.add_api_route("/geocode", api_geocode, methods=["GET"])
v1.add_api_route("/grade", api_grade, methods=["GET"])

v2 = APIRouter(tags=["v2"])
v2.add_api_route("/geocode", api_geocode, methods=["GET"])
v2.add_api_route("/suggest", api_suggest, methods=["GET"])
v2.add_api_route("/place", api_place, methods=["GET"])
v2.add_api_route("/grade", api_grade, methods=["GET"])
v2.add_api_route("/calendar", api_calendar, methods=["GET"])
v2.add_api_route("/day", api_day, methods=["GET"])
v2.add_api_route("/score", api_score, methods=["GET"])
v2.add_api_route("/forecast", api_forecast, methods=["GET"])
v2.add_api_route("/cell", api_cell, methods=["GET"])
v2.add_api_route("/metrics", api_metrics, methods=["GET"], include_in_schema=False)
v2.add_api_route("/event", api_event, methods=["POST"], include_in_schema=False)

app.include_router(v1, prefix=f"{BASE}/api")        # legacy unversioned == v1 (backward compatible)
app.include_router(v1, prefix=f"{BASE}/api/v1")
app.include_router(v2, prefix=f"{BASE}/api/v2")


# --- Discord slash commands (HTTP interactions) ----------------------------
async def discord_interactions(request: Request) -> Response:
    """Discord posts each slash-command interaction here. Verification runs over the
    RAW body, so this must read bytes, not request.json()."""
    raw = await request.body()
    status, payload = discord_interactions_mod.handle(
        raw,
        request.headers.get("x-signature-ed25519", ""),
        request.headers.get("x-signature-timestamp", ""),
    )
    return Response(json.dumps(payload), status_code=status, media_type="application/json")

app.add_api_route(f"{BASE}/discord/interactions", discord_interactions,
                  methods=["POST"], include_in_schema=False)

# --- accounts (fastapi-users) ----------------------------------------------
# Library-provided routers: /auth/login + /auth/logout (cookie session),
# /auth/register (signup), and /users/me (session check / profile). All under the
# same v2 prefix as the rest of the API, so the frontend calls them base-relative.
app.include_router(
    users.fastapi_users.get_auth_router(users.auth_backend),
    prefix=f"{BASE}/api/v2/auth", tags=["auth"],
)
app.include_router(
    users.fastapi_users.get_register_router(UserRead, UserCreate),
    prefix=f"{BASE}/api/v2/auth", tags=["auth"],
)
app.include_router(
    users.fastapi_users.get_users_router(UserRead, UserUpdate),
    prefix=f"{BASE}/api/v2/users", tags=["users"],
)
# Subscriptions + notifications (authed, user-scoped).
app.include_router(api_accounts.router, prefix=f"{BASE}/api/v2")
# Discord account linking (OAuth2, authed).
app.include_router(discord_link.router, prefix=f"{BASE}/api/v2")


# --- static frontend (served under BASE) -----------------------------------
def _page(name):
    """Route handler for one HTML page. Serves the file with its __ORIGIN__
    placeholders (the link-preview/Open Graph tags) filled in as the request's
    scheme://host + BASE — preview crawlers (Discord, Slack, …) need absolute
    URLs, and the host differs between LAN and prod. The scheme comes from
    X-Forwarded-Proto when a reverse proxy (Caddy) fronts the plain-HTTP
    uvicorn; the proxy passes the original Host header through untouched."""
    path = os.path.join(FRONTEND_DIR, name)

    def route(request: Request):
        with open(path, encoding="utf-8") as f:
            html = f.read()
        proto = request.headers.get("x-forwarded-proto") or request.url.scheme
        host = request.headers.get("host") or request.url.netloc
        html = html.replace("__ORIGIN__", f"{proto}://{host}{BASE}")
        # Search-engine verification <meta> tags (same source as the SEO pages), so
        # the homepage Google/Bing verify against carries them too.
        verify = content.head_verify_html()
        if verify:
            html = html.replace("<head>", f"<head>\n  {verify}", 1)
        etag = f'W/"{hashlib.sha1(html.encode()).hexdigest()[:20]}"'
        if _not_modified(request, etag):
            return Response(status_code=304, headers={"ETag": etag})
        return Response(html, media_type="text/html", headers={"ETag": etag})
    return route


# The un-slashed base redirects to BASE/ so the frontend's relative asset URLs
# resolve correctly. Under a sub-path this keeps the app scoped to BASE (never
# claiming "/", so the domain root stays free for another app via the proxy). When
# BASE is "" the app owns the root: "/" is already the index route below, so there
# is no bare base to redirect (and "" is not a valid route path).
# Pages also answer HEAD — link-preview crawlers probe with it before fetching.
if BASE:
    app.add_api_route(BASE, lambda: RedirectResponse(url=f"{BASE}/"), methods=["GET", "HEAD"], include_in_schema=False)
# NOTE: "/" is not here — the homepage is server-rendered by content.register()
# below, so its hero, records strip and city links reach crawlers and no-JS
# readers as real HTML.
app.add_api_route(f"{BASE}/calendar", _page("calendar.html"), methods=["GET", "HEAD"], include_in_schema=False)
app.add_api_route(f"{BASE}/day", _page("day.html"), methods=["GET", "HEAD"], include_in_schema=False)
app.add_api_route(f"{BASE}/score", _page("score.html"), methods=["GET", "HEAD"], include_in_schema=False)
app.add_api_route(f"{BASE}/compare", _page("compare.html"), methods=["GET", "HEAD"], include_in_schema=False)
app.add_api_route(f"{BASE}/legend", _page("legend.html"), methods=["GET", "HEAD"], include_in_schema=False)
app.add_api_route(f"{BASE}/alerts", _page("subscriptions.html"), methods=["GET", "HEAD"], include_in_schema=False)

# Monthly-digest signup (the footer form on every page). Answers the same generic
# message whatever the outcome, so it can't be probed for whether an address is
# already subscribed.
app.add_api_route(f"{BASE}/digest", digest.signup, methods=["POST"], include_in_schema=False)

# Crawlable, server-rendered SEO pages (the homepage, climate hub / per-city /
# month / records / glossary / about / privacy) + robots.txt + sitemap.xml.
# Registered before the static mount so these explicit routes win.
content.register(app)

# Everything else under BASE (app.js, style.css, nav.js, …) is a static asset.
# Registered last so the explicit page routes above win. Mounts must start with
# "/", so at the root (BASE == "") mount at "/" — the earlier routes still win
# because Starlette matches in registration order.
app.mount(BASE or "/", StaticFiles(directory=FRONTEND_DIR), name="static")