mirror of
https://github.com/djdevin/recflare.git
synced 2026-09-08 14:41:28 -07:00
56 lines
2.4 KiB
SQL
56 lines
2.4 KiB
SQL
-- Subrooms as first-class entities. In the original game a subroom has its own
|
|
-- globally-unique, autoincrementing `SubRoomId` (minted from a single sequence, not
|
|
-- per-room), so they can't live inside the room JSON blob — cloning/dorm creation
|
|
-- would otherwise reuse ids and collide across rooms. This moves them into their own
|
|
-- `subroom` table keyed by an AUTOINCREMENT `sub_room_id`, backfills from each room's
|
|
-- embedded `SubRooms`, and drops `SubRooms` from the room blob. Rooms re-embed their
|
|
-- `SubRooms` array on read. Generated from packages/domain/src/rooms-db.ts
|
|
-- (SUBROOM_SCHEMA_DDL) — keep in sync.
|
|
|
|
CREATE TABLE IF NOT EXISTS subroom (
|
|
sub_room_id INTEGER PRIMARY KEY AUTOINCREMENT,
|
|
room_id INTEGER NOT NULL,
|
|
data TEXT NOT NULL
|
|
);
|
|
CREATE INDEX IF NOT EXISTS idx_subroom_room ON subroom (room_id);
|
|
|
|
-- Backfill. The live blob already contains collisions (the bug this migration fixes:
|
|
-- every dorm carries SubRoomId 1, clones reuse per-room ids), so we can't preserve all
|
|
-- ids. We keep the FIRST occurrence of each distinct SubRoomId at its original id (so the
|
|
-- seeded rooms keep their canonical ids) and mint fresh globally-unique ids for the rest.
|
|
|
|
-- 1) First occurrence of each non-null SubRoomId keeps its id (lowest room_id wins).
|
|
INSERT INTO subroom (sub_room_id, room_id, data)
|
|
SELECT sid, room_id, data FROM (
|
|
SELECT
|
|
CAST(json_extract(je.value, '$.SubRoomId') AS INTEGER) AS sid,
|
|
r.room_id AS room_id,
|
|
je.value AS data,
|
|
ROW_NUMBER() OVER (
|
|
PARTITION BY json_extract(je.value, '$.SubRoomId')
|
|
ORDER BY r.room_id
|
|
) AS rn
|
|
FROM room r, json_each(r.data, '$.SubRooms') je
|
|
)
|
|
WHERE rn = 1 AND sid IS NOT NULL;
|
|
|
|
-- 2) Everything else (the duplicates, and any null ids) gets a fresh autoincrement id.
|
|
-- Inserting the explicit ids above advanced sqlite_sequence, so these continue past
|
|
-- the highest kept id and never collide.
|
|
INSERT INTO subroom (room_id, data)
|
|
SELECT room_id, data FROM (
|
|
SELECT
|
|
CAST(json_extract(je.value, '$.SubRoomId') AS INTEGER) AS sid,
|
|
r.room_id AS room_id,
|
|
je.value AS data,
|
|
ROW_NUMBER() OVER (
|
|
PARTITION BY json_extract(je.value, '$.SubRoomId')
|
|
ORDER BY r.room_id
|
|
) AS rn
|
|
FROM room r, json_each(r.data, '$.SubRooms') je
|
|
)
|
|
WHERE NOT (rn = 1 AND sid IS NOT NULL);
|
|
|
|
-- Single source of truth: drop the now-migrated SubRooms array from the room blob.
|
|
UPDATE room SET data = json_remove(data, '$.SubRooms');
|