Root cause: _LoadCorrections writes to QuestieDB.objectData (binary DB companion) but GetObject reads from QueryObject binary handle + objectDataOverrides. Items in Load() path are patching the binary DB so never reach GetObject's override check. Moving 176213 to LoadFactionFixes puts it directly in objectDataOverrides, the path GetObject actually reads.
The previous patch hoisted _CollectFinisherSpawns and _CollectObjective to
module level but failed to remove the original inner function bodies from
_CollectQuestTargets, leaving 100+ lines of orphaned code and a phantom
bare 'end' that caused '<eof> expected near end' on load.
_CollectQuestTargets is now clean: calls the hoisted module-level functions
directly, no inner local function declarations.
- HBD.lua: Cache GetPlayerWorldPosition/GetPlayerZonePosition at 50ms
intervals instead of hammering GetPlayerMapPosition() every frame.
Invalidate cache on PLAYER_ENTERING_WORLD and ZONE_CHANGED_* events.
Expected ~97% reduction in C API position calls (45,570 -> ~1,200 / 10min).
- zoneDB.lua: Replace O(n) linear scan in GetAreaIdByUiMapId with an
O(1) reverse lookup cache (uiMapIdToAreaIdCache) built at Initialize().
Cache is kept in sync by ApplyCustomZones and name-match fallback now
caches its result so subsequent calls are also O(1).
- Tooltip.lua: Throttle GameTooltip OnUpdate hook to 100ms intervals
(was firing every frame at 60-144 Hz). Added _tooltipLastText cache
to avoid redundant GetText() + CountTooltip() calls when nothing changed.
- QuestieArrow.lua: Hoist _HasMissingCompletedFlag, _GetCompleteIconType,
_CollectFinisherSpawns, and _CollectObjective out of _CollectQuestTargets
to module-level functions. These were re-created as closures on every
UpdateNearestTargets call (1 Hz). Shared per-cycle context is published
via _arrow_* module upvalues to avoid closure capture overhead.
- QuestieLearnerComms.lua: Four micro-optimizations:
(1) Reduce ProcessQueues ticker 0.2s -> 0.5s (still 7x faster than
minChatInterval of 3.5s).
(2) Cache hidden channel ID at init; lazy refresh on disconnect.
(3) Drop LibDeflate compress level 9 -> 1 (fraction of CPU cost).
(4) O(1) messageCacheCount counter replaces O(n) pairs() size scan.
Xurkon