#4437 feat(service,features): an inbox mark, answer, or close answers the transaction it committed in

The service half of the inbox's txid wait: each member write on the inbox returns the txid of the transaction that changed its row, as Neon's mutations guide describes. The host PR that waits on it with confirmWrite (#4423) follows.

What changes

• inbox-store.ts: pg_current_xact_id()::text is read in the RETURNING of the write that changes the row. No extra transaction or statement is added.
  • upsertRead, upsertUnread, and upsertAllRead (mark read, mark unread, mark all read).
  • answerItem (answer by revision, and answer to an ask by its message).
  • memberCloseItem (a member's close).
• No-op writes carry no txid:
  • A mark already standing.
  • Mark all read with nothing unread.
  • An answer or close sent again under the same key.
• Wire (inbox-wire.ts): inboxAnsweredSchema, inboxMessagedSchema (an answer by reply), inboxReadStateSchema, and inboxReadAllSchema each get an optional txid. A desktop that doesn't know the field drops it.
• Out of scope: a plain thread message carries no txid here. It isn't one of the inbox area's writes (mark, answer, close).

Old work

#3264 merged without a txid half. Its head (refs/pull/3264/head) has no txid hunks, and no *inbox*txid* branch exists, so this is new code. It follows the chat txid PR (#4251) as its pattern.

Tests

• inbox-store.test.ts:
  • mark read and mark unread return a txid the first time and none when repeated
  • mark all read returns one txid, and { marked: 0 } with no txid when nothing was unread
  • an answer and a close each return their own txid
• inbox-threads.test.ts: an answer to an ask is taken with a txid, and the same answer sent again (REPEATED) has none.
• coding-agents-app.test.ts: the PUT and DELETE read routes and mark all read carry a txid on the wire.

Made with Luke

---