galaxy-game/backend/internal/diplomail/README.md

# diplomail

`diplomail` owns the diplomatic-mail subsystem of the Galaxy backend
service. Messages live in the lobby-side domain (their storage and
lifecycle are tied to a game), but they are surfaced inside the game UI
— the lobby exposes only an unread-count badge per game.

## Stages

The package ships in four staged increments. Stage A is the surface
described below; the remaining stages add admin / system mail,
lifecycle hooks, paid-tier broadcast, multi-game broadcast, bulk
purge, and the language-detection / translation cache.

| Stage | Scope | Status |
|-------|-------|--------|
| A | Schema, personal single-recipient send / read / delete, unread badge, push event with body-language `und` | shipped |
| B | Owner / admin sends + lifecycle hooks (paused, cancelled, kick); strict soft-access for kicked players | shipped |
| C | Paid-tier personal broadcast + admin multi-game broadcast + bulk purge + admin observability | shipped |
| D | Body-language detection (whatlanggo) + translation cache + lazy per-read translator dispatch | shipped |
| E | LibreTranslate HTTP client + async translation worker with exponential backoff + delivery gating on translation completion | shipped |

## Tables

Three Postgres tables in the `backend` schema:

- `diplomail_messages` — one row per send (personal, admin, or
  system). Captures `game_name` and IP at insert time so audit
  rendering survives renames and purges.
- `diplomail_recipients` — one row per (message, recipient). Holds
  per-user `read_at`, `deleted_at`, `delivered_at`, `notified_at`
  state. Snapshot fields (`recipient_user_name`,
  `recipient_race_name`) are captured at insert time and survive
  membership revocation.
- `diplomail_translations` — cached per (message, target_lang)
  rendering. One translation is reused across every recipient that
  asks for that language.

## Permissions

| Action | Caller | Pre-conditions |
|--------|--------|----------------|
| Send personal | user | active membership in game; recipient is active member |
| Paid-tier broadcast | paid-tier user | active membership; recipients = every other active member |
| Send admin (single user) | game owner OR site admin | recipient is any-status member of the game |
| Send admin (broadcast) | game owner OR site admin | recipient scope ∈ `active` / `active_and_removed` / `all_members`; sender excluded |
| Multi-game admin broadcast | site admin | scope `selected` (with `game_ids`) or `all_running` |
| Bulk purge | site admin | `older_than_years >= 1`; targets games with terminal status finished more than N years ago |
| Read message | the recipient | row exists in `diplomail_recipients(message_id, user_id)`; non-active members see admin-kind only |
| Mark read | the recipient | row exists; idempotent if already marked |
| Soft delete | the recipient | `read_at IS NOT NULL` (open-then-delete, item 10) |

Stage D will add body-language detection (whatlanggo) and the
translation cache + async worker.

System mail is produced internally by lobby lifecycle hooks:
`Service.transition()` emits `game.paused` / `game.cancelled` system
mail to every active member; `Service.changeMembershipStatus` /
`Service.AdminBanMember` emit `membership.removed` /
`membership.blocked` system mail addressed to the affected user.

## Content rules

- Body is plain UTF-8 text. The server does **not** parse, sanitise,
  or escape HTML — the UI renders messages via `textContent`.
- Body length is capped by `BACKEND_DIPLOMAIL_MAX_BODY_BYTES` (default
  4096). Subject length is capped by
  `BACKEND_DIPLOMAIL_MAX_SUBJECT_BYTES` (default 256). Both limits
  live in the service layer so they can be tuned without a schema
  migration.
- `body_lang` is filled at send time by the configured
  `detector.LanguageDetector` (default: `whatlanggo`, body-only,
  ≥ 25 runes; shorter bodies stay `und`).

## Translation

Stage D adds a lazy translation cache. When a recipient reads a
message through `GET /api/v1/user/games/{game_id}/mail/messages/{id}`,
the handler resolves the caller's `accounts.preferred_language` and
asks `Service.GetMessage(…, targetLang)` to attach a translation:

- on cache hit (row in `diplomail_translations`), the rendering is
  returned directly under `translated_subject` / `translated_body`;
- on cache miss, the configured `translator.Translator` is invoked.
  A non-noop result is persisted and returned to the caller; the
  noop translator that ships with Stage D returns `engine == "noop"`,
  which is treated as "translation unavailable" and the caller falls
  back to the original body.

The inbox listing (`/inbox`) reuses cached translations but never
calls the translator on miss — bulk listings stay fast even when a
real translator (LibreTranslate, SaaS engine) introduces I/O cost.

Future work plugs a real `translator.Translator` (LibreTranslate
HTTP client is the documented next step) without touching the rest
of the system.

## Async translation (Stage E)

Stage E switches the translation pipeline from "lazy at read" to
"async at send". The send path stays synchronous from the
caller's perspective: the message and recipient rows are inserted
in one transaction. What changes is delivery semantics:

- Recipients whose `preferred_language` matches the detected
  `body_lang` (or whose body language is `und`) get
  `available_at = now()` straight away and the push event fires
  during the request.
- Recipients whose `preferred_language` differs are inserted with
  `available_at IS NULL`. They are **not** visible in inbox, unread
  count, or push events until the worker translates the message.

The worker (`internal/diplomail.Worker`, started as an
`app.Component` in `cmd/backend/main`) ticks once every
`BACKEND_DIPLOMAIL_WORKER_INTERVAL` (default `2s`). Each tick:

1. Picks one distinct `(message_id, recipient_preferred_language)`
   pair from `diplomail_recipients` where `available_at IS NULL`
   and `next_translation_attempt_at` is unset or due.
2. Loads the source message, checks the translation cache.
3. On cache hit → marks every pending recipient of the pair
   delivered and emits push.
4. On cache miss → asks the configured `Translator`:
   - success → caches the translation, marks delivered, push;
   - HTTP 400 (unsupported pair) → marks delivered without a
     translation (fallback to original);
   - other failure → bumps `translation_attempts`, schedules the
     retry via `next_translation_attempt_at`, leaves pending.
5. After `BACKEND_DIPLOMAIL_TRANSLATOR_MAX_ATTEMPTS` (default `5`)
   the worker falls back to delivering the original body so a
   prolonged LibreTranslate outage does not strand messages.

Retry backoff is exponential `1s → 2s → 4s → 8s → 16s` (capped at
60s) per pair. Operators monitor the LibreTranslate dependency
through standard OpenTelemetry export — translation outcomes
surface in `diplomail.worker` logs at Info / Warn levels;
Grafana / Prometheus dashboards live outside this package.

### Multi-instance posture (known limitation)

`PickPendingTranslationPair` intentionally drops `FOR UPDATE`: the
worker is single-threaded per process, and we did not want a slow
LibreTranslate HTTP call to keep a row-lock open. The cost is a
small window where two backend instances pulling at the same
moment can both claim the same pair: the cache-write side stays
clean (`INSERT … ON CONFLICT DO NOTHING`), but each instance will
publish its own push event to every recipient of the pair, so the
duplicate push is the visible failure mode.

The current deployment runs a single backend instance and the
window does not exist. When the platform scales to multiple
instances, we will revisit the pickup query — either by holding
the lock through the HTTP call (with a short timeout to bound the
worst case) or by introducing a `claimed_at` column and a
short-lived advisory lease. The change is local to this package
and does not affect callers.

For the LibreTranslate operational recipe — installing, wiring,
manual smoke test — see
[`backend/docs/diplomail-translator-setup.md`](../../docs/diplomail-translator-setup.md).

## Push integration

Every successful send emits a `diplomail.message.received` push
intent through the existing notification pipeline. The catalog entry
limits delivery to the push channel — email is intentionally absent;
the inbox endpoint is the durable fallback for offline users. The
payload includes the recipient's freshly recomputed unread count for
the lobby badge and for the in-game header.

## Lifecycle hooks (Stage B)

The lobby module is the producer of system mail. Stage B will add a
`DiplomailPublisher` collaborator on `lobby.Service` and call it on
`paused` / `cancelled` transitions and on `BlockMembership` /
`AdminBanMember`. The publisher constructs a
`kind='admin', sender_kind='system'` message with a templated body;
the recipient receives the durable copy in their inbox even after the
membership is revoked.

If a future stage adds inactivity-based player removal at the lobby
sweeper, that path **must** call the same publisher so the kicked
player has the explanation in their inbox.

## Wiring

`cmd/backend/main.go` constructs `*diplomail.Service` with three
collaborators:

- `*Store` over the shared Postgres pool;
- `MembershipLookup` adapter that walks the lobby cache for the
  active `(game_id, user_id)` row and stitches in the immutable
  `accounts.user_name`;
- `NotificationPublisher` adapter that translates each
  `DiplomailNotification` into a `notification.Intent` and routes it
  through `*notification.Service.Submit`.