The migration consolidation (27fb0fa) merged 33 incremental patches
into 0001_init.sql and marks the baseline as applied for any existing
database. Users who only ran some of the 33 patches have schema gaps
that cause runtime errors.
0002_catch_up_patches.sql replays all 33 patches (skipping 15 and 20,
which are destructive data resets) with idempotency guards so it is a
no-op on fresh or fully-patched databases and fills gaps otherwise.
Replace 4 independent schema management code paths (Docker shell
script, setup wizard pg_restore, test helpers, manual psql) with a
single migration runner embedded in the server binary.
The new server/migrations/ package uses Go embed to bundle all SQL
schemas. On startup, Migrate() creates a schema_version tracking
table, detects existing databases (auto-marks baseline as applied),
and runs pending migrations in transactions.
Key changes:
- Consolidated init.sql + 9.2-update + 33 patches into 0001_init.sql
- Setup wizard simplified to single "Apply schema" checkbox
- Test helpers use migrations.Migrate() instead of pg_restore
- Docker no longer needs schema volume mounts or init script
- Seed data (shops, events, gacha) embedded and applied via API
- Future migrations just add 0002_*.sql files — no manual steps
When config.json is missing, Erupe now launches a temporary HTTP server
on port 8080 serving a guided setup wizard instead of exiting with a
cryptic error. The wizard walks users through database connection,
schema initialization (pg_restore + SQL migrations), and server settings,
then writes config.json and continues normal startup without restart.
Codecov requires an account and token to function. Replace it with
a self-contained `go tool cover` step that fails the build if total
coverage drops below 50% (currently ~58%). This catches test
regressions without external service dependencies.
Users without a Go toolchain can now download pre-built binaries
directly from GitHub Releases (closes#161). The workflow triggers
on v* tags, builds Linux and Windows amd64 archives, and attaches
them along with SCHEMA.sql to an auto-generated release.
Also fixes the README badge URL (go-improved.yml → go.yml).
Players could never claim monthly guild items because the handler
always returned 0x01 (claimed). Now tracks per-character per-type
(standard/HLC/EXC) claim timestamps in the stamps table, comparing
against the current month boundary to determine claim eligibility.
Adds MonthStart() to gametime, extends StampRepo with
GetMonthlyClaimed/SetMonthlyClaimed, and includes schema migration
31-monthly-items.sql.
Replace vague "Alpelo object system backport" references in CHANGELOG
and AUTHORS with accurate descriptions of the per-session object ID
allocation rework. Remove stale test comment referencing the deleted
NextObjectID method.
When a quest or scenario file was missing, handleMsgSysGetFile sent nil
data via doAckBufSucceed, which crashed the game client. Now sends
doAckBufFail so the client can handle the missing file gracefully.
Closes#109
Wrap *rand.Rand in a mutex-protected SafeRand type to make the global
RNG safe for concurrent use across goroutines. The previous bare
*rand.Rand caused data races detected by go test -race.
UTF8ToSJIS panicked when encountering characters outside the Shift-JIS
range (emoji, Lenny faces, cuneiform, etc.), crashing the server when
such characters were sent via the Discord relay channel.
Replace the panic with graceful filtering that drops unmappable runes
and preserves valid content. Also fix ToNGWord index-out-of-range panic
on empty encoder output.
Closes#116
