- Serve /swf and /nitro-assets (~2.8GB) with 7-day Cache-Control plus
stale-while-revalidate so client opens stop re-fetching hundreds of
files while asset updates still propagate in the background
- Issue the SSO ticket and the online count query in parallel on the
client page to shave a round-trip off the critical render path
- Send the verification email after the response via after() so it
never blocks sign-up
- Invalidate the cached login lookup right after account creation so
the automatic sign-in always finds the fresh row
- Auto sign in with the submitted credentials and go straight to /me,
with a fallback to /login?registered=1 if sign-in is refused (e.g.
email verification required)
- Cache the register page's online/latest user queries to cut DB load
under traffic
- Fix terms checkbox label double-toggle cancelling the selection
- Add pages.register.redirecting translation to all locales
Recent optimization commits accidentally gutted next.config.ts, removing
the createNextIntlPlugin wrapper. This caused every page to crash at
runtime with 'Couldn't find next-intl config file', showing the error
page after a successful build.
Restores:
- next-intl plugin (./src/i18n/request.ts)
- Security headers (HSTS, X-Frame-Options, nosniff, etc.)
- Redirects from /admin/import/* to /admin/studio/*
- Cache headers for /assets and /images, AVIF/WebP image formats
- compress and productionBrowserSourceMaps
Keeps recent improvements: reactStrictMode and optimizePackageImports.
Per single furni import, patchLocalizedFurniDataEntries ran for every
language and did a full 100k-entry JSON.parse + scan + JSON.stringify on the
main thread. That CPU spike is now offloaded to the translation worker
(init/prepare/finalize reuses the same pure core helpers), so importing
never blocks the event loop. The main thread only does async file I/O and the
network LibreTranslate calls. Falls back to the same helpers on the main
thread if no worker is available.
Also cache readFurniData() by file mtime+size so the 100k-entry JSON is
parsed once per change instead of on every import/fix/reconcile read
(invalidated on write).
The translation build's CPU-heavy work (deep clone of the 100k+ entry
master, the two full scans and JSON.stringify per language) now runs in a
worker thread, so "Alles vertalen" no longer blocks the main event loop.
Network/DB parts (official Habbo fetch, LibreTranslate, file writes) stay
on the main thread. The pure helpers were extracted to furni-data-i18n-core
(no server-only deps) so the worker can import them. Falls back to the same
helpers on the main thread when no worker is available or in tests.
The synchronous convertSwfToNitro / extractIconFromSwf calls (pure CPU,
no DB/network) now run in a small worker pool so the main Node event
loop stays free during imports — the dominant import-time CPU spike is
moved off the request thread. The pool degrades gracefully to a
synchronous fallback if workers can't be created, and is skipped in the
test environment.
- patchLocalizedFurniDataEntries now only touches the wanted classnames
instead of scanning the whole 100k file, reads the master once (was
twice), and uses the batched translator instead of one HTTP call per
text.
- Remove the now-unused serial translateWithLibre.
- buildLocalizedFurniDataFiles yields between the two section scans and
before stringify so a single language no longer blocks the event loop
for the whole pass at once.
Lower LibreTranslate concurrency to 2 and yield the event loop between
languages so the 13x deep-clone + stringify loop no longer pins a core
continuously. The build is also already opt-in (off by default on import).
The maintenance "Fix alles" now has a checkbox to include or skip the
duplicate catalog_pages merge, so admins can choose whether to run it.
removeDuplicates takes an includePages flag and fixEverything threads it
through to the action.
Furni imports can now choose whether to translate (per import) and which
languages to build, instead of always rebuilding all 13 FurnitureData_<lang>
files. The studio header also has a global switch that persists the
furnidata_translate_enabled setting, seeding the per-import default.
The maintenance "Fix alles" now collapses catalog_pages that share the same
caption_save + parent_id into a single survivor (moving catalog_items and
reparenting child pages before deleting duplicates), so the catalog tree
never renders double links. The duplicate-page count is also surfaced in the
health panel.
- Move /admin/import/* tools under /admin/studio/* and add studio nav, layout and tabs
- Add shared catalog maintenance panel plus a /admin/studio/maintenance tab
- Make furnidata reconciliation overwrite conflicting entries with the
DB-authoritative items_base classname/id (surfaced via fixedConflicts)
- Add furnidata_translate_enabled setting to skip the heavy localized
furnidata build during import (manual build-languages still forces it)
- Update staff smoke contract test for the studio hub
Translate deduplicated texts in batches of 48 with 4 parallel workers
instead of one request per string, serve cached hits locally, and
persist results so rebuilds never re-call the API for the same text.
Every write path now sets offer_id = catalog_items.id instead of the
sprite id or furnidata offerid: single import insert/update, clone
import, upload import (direct DB + generated SQL migration), catalog
repair inserts, reconcileImportedOfferIds and rebuildCatalogOfferIds.
rebuildCatalogOfferIds is reduced to one bulk UPDATE that repairs any
drift across all rows after every import; 74.838 legacy rows were
repaired on the live database with this change.
- Change skipFurniDataWrite from true to false in both import routes
- Previously: FurnitureData.json entries were deferred/skipped during import
- Now: Entries are written immediately after each furniture item import
- Result: FurnitureData.json is updated directly, translations work again, offer_id fix preserves correctly set values
- Change condition from to
- Prevents overwriting offer_id that was correctly set to sprite_id during furniture import
- Fixes catalog showing wrong furniture when offer_ids and sprite_ids don't match
- Switch default EMULATOR_MODE from 'rcon' to 'api'
- Remove RCON_HOST/PORT/timeout/maxRetries env vars
- Add EMULATOR_API_URL to .env and .env.example
- Update createEmulatorTransport() to use HTTP API only
- Remove rconHost/port from offline alert context
- All 827 tests pass, TypeScript compiles cleanly
- Add EMULATOR_MODE=api option with HTTP transport mirroring RCON payload
- buildLocalizedFurniDataFiles now falls back to LibreTranslate for customs (incl. pl/ru/ar/ja)
- Remove Organiseer import button from studio and Reorganize Catalog from import page
- Delete organize API route and PUT handler; remove organizeCatalogAll from catalog-repair.ts
- Increase imported-furni list perPage from 50 to 500
tryDownloadCandidates now fires every candidate URL concurrently and keeps
the first valid response, removing the old 15s×retry sequential waits on
slow/unreachable sources that made import speed uneven. Batch concurrency
raised 8 -> 12 to absorb the now-faster downloads.
AddSize32ToBundle downscales the largest existing sprite (usually size 64)
to generate a 32 visualization, so every bundle — including the
source-provided gamedata furniture — carries a 32 frame without needing the
original SWF. AddSize32AllNitros rewrites every existing .nitro across all
nitro directories (incl. /var/www/Gamedata/bundled/furniture) and is
idempotent (bundles that already have 32 are skipped). Expose it via a Studio
button and the background admin endpoint POST /api/admin/import/furni/add-size32.
The SWF-to-Nitro converter previously dropped the size 32 visualization
and any sprite image whose name contained '_32_', so bundles lacked the
small/catalog render size. Keep them so each furniture bundle renders
correctly at every scale the client requests.
- xml-processor: stop skipping visualization size === 32
- index.ts: stop excluding '_32_' sprite assets/images from the sheet
Run the independent icon and swf downloads in parallel via Promise.all
instead of sequentially. The nitro fallback still runs after the swf check
since it depends on the swf result.
Combined with batch concurrency and the single FurnitureData write, this
further cuts per-item import time (roughly halves the download phase).
openhands
Simo