The InjectLearnedData purge loop was deleting entire NPC entries from
Questie.dbLearner.global.npcs whenever data[7] (spawns) was empty,
even if the entry still had a name, home zone, recorded kills, or
quest references. This caused learner data to silently disappear on
/reload for NPCs that had been learned but never killed in a position
the learner could record (e.g. party-kill position attribution fix
now passes nil coords for uncredited kills).
Now only purges when ALL of name, zone, kills, and quest references
are missing. Entries with other useful state get data[7] set to nil
instead so the row stays.
- ElvUI tooltip style: new 'ElvUI tooltip style' option (General tab, on by
default) skins GameTooltip/WorldMapTooltip/ItemRefTooltip/shopping tooltips
and the secondary learner frame with ElvUI's transparent flat look (dark bg
+ thin 1px border) when ElvUI is not installed. Corrected the secondary
frame fallback that used the chunky WoW border. No-op when ElvUI is loaded.
- Source attribution now shows ONLY inside the secondary learner tooltip when
'Use secondary learner tooltip' is enabled; removed from the main NPC/object/
item tooltip and gated off entirely when the secondary tooltip is disabled
(map-pin source gated the same way).
- Fixed learner-learned pins mislabelled 'AscensionDB': GetPinDataSource is now
mode-aware and returns 'Learner' in learner mode when the entity has a learner
record, even if AscensionDB also curates it (curated coords are discarded by
GetNPC/GetObject in learner mode anyway).
Test-neutral (145 successes / same 7 pre-existing failures + 1 error).
In learner-only mode the map/minimap drew curated AscensionDB (and static)
spawns for every quest NPC/object the player had not personally recorded.
GetNPC/GetObject fall back to the npcDataOverrides/objectDataOverrides
entry for metadata when there is no learner record, and that entry's
spawns (AscensionDB-curated for un-recorded entities) were drawn as pins.
Both now track whether a real learner record exists and, in learner mode,
strip spawns when it does not -- so learner-only shows exclusively what the
learner recorded (metadata fallback for names/tooltips kept). GetObject
also now nils the override in learner mode to match GetNPC. Recorded
entities still show their learner spawns; auto/static/none unchanged.
Verified test-neutral on the full suite (144 successes / same 7 pre-existing
failures) with a local regression test that passes with the fix and fails
without it.
Quest-type filters (Available Dungeon/Raid/PvP/Repeatable/Event quests and
other ShouldBeHidden rules) were honored on the world map but not the
minimap. At toggle time HideQuestIcons FakeHides both map and minimap
filtered icons, but the minimap icon's per-frame FadeLogic re-showed any
hidden icon once in range via 'elseif self.hidden then self:FakeShow()',
undoing the filter on the minimap only (the world icon has no FadeLogic).
FadeLogic now re-checks ShouldBeHidden before re-showing, in both the
world-icon and manual-icon minimap fade paths, so filters apply to both.
A third-party addon (e.g. AtlasLoot) registers an AceComm prefix longer
than the client's 16-character limit. Because Questie's bundled AceComm is
the LibStub winner and its xpcall polyfill wraps AceAddon's OnEnable, the
upstream hard error() in RegisterComm surfaced as a recurring Lua error at
every login (seen on ChromieCraft).
The prefix can never work on the client regardless, so RegisterComm now
warns once and skips the registration for over-long prefixes instead of
throwing, leaving the calling addon's OnEnable intact and removing the
error popup.
Two crashes reported in #15 on Elune (1.6.3):
- TrackerUtils:IsVoiceOverLoaded only verified the VoiceOver addons were
loaded, not that VoiceOver.QuestOverlayUI.questPlayButtons exists. Some
VoiceOver builds expose a QuestOverlayUI without that table, so
UpdateVoiceOverPlayButtons and SetAllPlayButtonAlpha crashed with
'attempt to index field questPlayButtons (a nil value)'. Now verified in
IsVoiceOverLoaded, which all play-button call sites gate on.
- AceGUIContainer-TreeGroup crashed indexing a nil AceGUI.tooltip when a
conflicting addon registered a broken AceGUI-3.0 core (version 1.#INF)
that won LibStub but never created the shared tooltip frame. The tree
button handlers now lazily recreate it; widget version bumped 47->48 so
the fixed widget wins registration.
The 'Use secondary learner tooltip' option's off-path appended learner
spawn/kill lines directly to the main GameTooltip. The toggle now gates
the data entirely: on = shown in the separate secondary frame, off = not
shown anywhere. Learner spawn/kill data is never inline in the main NPC
tooltip under any setting.
Removed the now-unused _AddTooltipSeparator helper and updated the option
description.
Re-adds the 'Show data source' tooltip option (General tab, default off),
but accurate this time. The previous version was removed because it
guessed the source from the global data-source mode (showing Learner when
data was AscensionDB, etc.). This tags each pin with its real provenance
at creation and reads that tag.
- QuestieDB.GetPinDataSource(entityType, id, spawnData): resolves
Learner (per-spawn isLearned or learner record), AscensionDB (curated
ascensionOverrideKeys override), or base Questie DB. Never guesses.
- Objective pins tagged per-spawn in _DetermineIconsToDraw; available/
finisher pins defaulted by quest in DrawWorldIcon; manual notes tagged
Townsfolk in DrawManualIcon.
- World-map pins read the per-pin tag (MapIconTooltip); unit/object/item
hovers derive per-id at the render layer (TooltipHandler via
QuestieTooltips:GetDataSourceLine). Comms appended via KeyExists.
- No line shown when source is genuinely unknown (never misleading).
Hovering an NPC right after turning in one of its associated quests
spammed a debugstack trace in DEVELOP mode: the learner objective-
correlation block in GetTooltip called QuestLogCache.GetQuestObjectives
for a quest no longer in QuestLogCache (which gracefully returns {} but
logs the stack at line 306).
Guard the NPC and object correlation loops on
QuestiePlayer.currentQuestlog[questId] so live objective progress is only
looked up for quests the player is currently on. Not a crash and no
functional change for active quests.
Two parallel learner-tooltip systems both rendered on unit hover, so with
the 'Use secondary learner tooltip' option on, learner spawn/kill lines
still leaked into the main NPC tooltip instead of going solely to the
secondary frame.
QuestieLearner's OnTooltipSetUnit hook (toggle-aware, secondary-capable)
already owns unit-hover learner display, routing it to the main tooltip
(secondary off) or the separate secondary frame (secondary on). But
Tooltip.lua's _GetLearnerTooltipLines also injected learner lines into the
m_<npcId> tooltip data, redundantly.
Add a suppressLearnerLines flag to QuestieTooltips:GetTooltip and pass it
from the NPC unit-hover call in TooltipHandler so the inline lines are
omitted there. Map-pin and object tooltips (no secondary frame) keep their
learner lines. Also removes the latent duplicate present even with the
secondary frame disabled.
When a quest object node is opened/looted, that specific node's
map/minimap pin now disappears immediately even before the objective is
fully collected, instead of all node pins staying until turn-in.
On LOOT_OPENED, QuestieQuest:RemoveLootedObjectivePins finds the active
object-objective spawn nearest the player's world position (<=12 yd),
marks it consumed in Questie.db.char.lootedObjectSpawns, and repopulates
just that objective so clustering recomputes with the node skipped.
_DetermineIconsToDraw skips consumed object spawns, so removal persists
across redraws and /reload. ClearLootedSpawns runs on AcceptQuest, so
abandoning and re-doing the quest (including after an Ascension prestige)
shows every node again.
Works in every data source mode (auto/learner/static/none): suppression
is centralized in the shared _DetermineIconsToDraw that all redraw routes
funnel through, operating on the mode-resolved objective.spawnList.
Matching is radius-based (1.5 zone units) so a learner spawn re-added at
the player's position for the just-looted node is suppressed too.
Scoped to object-type objectives; monster/kill pins are unchanged.
Completed objectives could leave their map/minimap pins on screen until
turn-in. Pin removal relied solely on _UnloadAlreadySpawnedIcons, which
walks objective.AlreadySpawned; that table desyncs from the live frames
after learner spawn-list invalidation or complete->abandon->reaccept
cycles, so when reset to {} the frames leaked.
Add QuestieMap:UnloadQuestFramesForObjective(questId, objectiveIndex)
which unloads frames off the map's own questIdFrames registry by matching
frame.data.ObjectiveIndex, and call it from every completion path in
PopulateObjective. Restricted to positive standard-objective indices;
SpecialObjectives (sentinel index 0) keep using AlreadySpawned.
Lua requires local functions to be defined before they are called.
_GetLearnerTooltipLines was defined after GetTooltip, causing
'attempt to call global _GetLearnerTooltipLines (a nil value)'.
Moved the function definition above GetTooltip.
Adds learned spawn position (most-visited GUID), total distinct spawns
learned, and total kills recorded to both unit hover tooltips and
world-map pin tooltips. Replaces the removed Source: attribution line
with factual learner data that was actually recorded.
On PARTY_KILL and unengaged kills, the learner was recording the local
player's position (GetCurrentPlayerPosition) as the spawn location.
This polluted the learner's spawn map with the wrong coords. Now only
records position for credited kills (your own kills or mobs you damaged
within the last 60 seconds).
Three-mode spawn-merge semantics:
- Learner: discard npcDataOverrides (which AscensionDB's _Asc_MergeInto
populates with curated coords). GetNPC returns learnerRecord only.
_ApplyNpcLiveUpdate and object-spawn injection use IsAscensionProtected
(mode-dependent) so learner coords pass through in learner mode.
- Auto: static DB primary, learner fills gaps. Curated spawns protected.
- Static: base DB only.
Also removes the allowSpawnMerge bypass that was the original hole
letting learner coords leak into curated spawns in auto mode.
The Source: line cannot accurately reflect per-pin provenance without
changing every spawn registration site. Until that infrastructure
exists, the label is misleading — showing Questie DB even when data
came from the learner, or Learner when it came from AscensionDB.
Remove from both MapIconTooltip (world-map pins) and Tooltip
(unit/object hover).
Some Area 52 quest records store qData[2][3] as a flat number instead of
a table-of-tables. The inner loop's entry[1] index then crashes with
'attempt to index local entry (a number value)'. Normalize with a
type check before comparing.
Mana Wyrm 15274 (quest 8325) pins rendered in the NE corner because the
learner's Sunstrider spawns leaked into the curated AscensionDB coords.
Root causes:
- In learner mode GetNPC uses rawdata = learner record, and _MergeOverride
deep-merges spawn tables, so the learner's bad zone (1445 @ 54,12)
survived since AscensionDB had no entry to displace it.
- The injection guards used IsAscensionProtected(), which returns false in
learner mode, so the learner overrode AscensionDB-owned spawns. Commit
0f20ea8 had also added a (not learnerLiveMode) bypass to the
_MergeSpawnEvidence Sunstrider guard, re-enabling the e80a008 regression.
Fixes:
- QuestieDB.GetNPC: for AscensionDB-owned NPC spawns (direct
ascensionOverrideKeys check), use ONLY the curated override spawns and
discard the merged learner zones.
- New mode-independent AscensionOwnsNpcSpawns() helper; used by the
InjectLearnedData spawn restore and the _MergeSpawnEvidence guard
(removed the (not learnerLiveMode) bypass).
Regression test asserts GetNPC returns only the curated 1241 coords and
drops the learner 1445 zone.
Follow-up to da5546f: the previous restore read the POST-migration data[7]
keys and ran them through NormalizeSpawnZoneKey. On Sunstrider that round-
tripped uiMapId 1241 -> parent areaId 3430 -> Eversong map 1941, placing
learner pins in the wrong (NE) corner of the map.
Now snapshot each NPC's [7] spawns in their NATIVE uiMapId space BEFORE the
uiMapId->areaId migration runs, and restore from that snapshot with no key
conversion. This keys the override exactly like the live _MergeSpawnEvidence
kill path (e.g. 1241 for Sunstrider), so pins render on the correct map and
HBD isSameZoneSpace handles Eversong<->Sunstrider visibility. Still gated by
IsAscensionProtected. Regression test extended to assert spawns land on 1241
and NOT 1941/3430/3431.
Document why stripping [7] in InjectLearnedData is only safe because the
guarded restore block below re-merges saved spawns (normalized to uiMapId).
Prevents re-introducing bug 7ce0cdc (learner quests losing pins on /reload
until the mob is re-killed).
InjectLearnedData stripped each learned NPC's spawn field [7] when injecting
into npcDataOverrides (CopyWithoutField(data,7) / k~=7), deferring to
_MergeSpawnEvidence. That promoter only runs on live kill evidence, so spawns
learned in a prior session never returned to the queryable DB on /reload, and
a freshly accepted quest (8325 -> Mana Wyrm 15274 on Sunstrider) had no pins
until the mob was re-killed.
Re-merge the saved [7] spawns into the override at injection time, normalized
to the canonical uiMapId via NormalizeSpawnZoneKey (so an areaId left by the
uiMapId->areaId migration, e.g. 3431/3430, maps back to 1241 for the renderer/
HBD), deep-merged with InsertIfNewBucket, and gated by IsAscensionProtected so
curated AscensionDB coords are never overwritten (always restored in learner
mode; in auto mode only non-curated NPCs). Regression from 7ce0cdc.
_l10n:translate now short-circuits the common no-argument case
(select('#', ...) == 0), returning the translation directly and skipping
the {...} allocation, the tostring loop, and safeFormat. Behavior matches
the slow path: a successful no-arg lookup already returned the raw value,
and missing/invalid entries fall back to the key. Cherry-picked from
phase2-lua50-sweep.
Five Options files seeded their tab tables with { ... } instead of {}.
At chunk scope WoW passes ... = (addonName, addonTable), so each table
got two stray junk entries instead of being empty; { ... } is also a
strict Lua 5.0 parse error. Fixed QuestieOptions.tabs and the Arrow,
General, Keybinds, and Tracker tab tables. Cherry-picked from
phase2-lua50-sweep (the only version-independent fix in that branch).
OnQuestAccepted now seeds objective pins directly from the SavedVariables
learner payload (objIndex) on accept in learner mode, instead of waiting
for quest-log text sync — so quests like 8325 spawn pins immediately when
the objective mapping already exists. The accept-time ID lookup is widened
from killcredit to also cover monster objectives, and falls back to
objData.Id when no IdList is present.
In 'auto' data-source mode, QuestieDB.GetQuest now overlays the
SavedVariables learner record on top of the static DB quest, filling only
nil fields and deep-merging objectives/objIndex. Fixes quests like 8325
where the shipped DB has only a stub record but QuestieLearner holds the
real objective mapping that the stub was blocking. Learner data never
overwrites a present static value, so it strictly enriches.
Objective tooltips (Tooltip.lua) and world-map icon tooltips
(MapIconTooltip.lua) now append a greyed-out 'Source:' line listing the
data origin: Static DB or Learner per the active data-source mode, plus
Comms when remote player data exists for the key. Also corrects an
internal QuestieDB:GetQuest colon call to the dot form QuestieDB.GetQuest.
AddItemDataToTooltip no longer gates the Item ID line on the item having
a registered quest-objective tooltip, so the ID shows for every item hover
when enabled (matching NPC/Object behavior). Switched to self:AddDoubleLine
so the line lands on the frame that fired OnTooltipSetItem (e.g. ItemRefTooltip
for chat links). Also surfaces a 'Drops a quest !' line with the colored quest
title when the item has a non-zero startQuest and the player lacks the quest,
read from QueryItemSingle at hover time.
Tests are for local dev only. .gitignore already excludes Tests/ and
Modules/*_spec.lua from new adds, but the files were still tracked from
prior commits. Untrack them with --cached so the working copies stay
on disk for development but no longer ship in the live repo.
- SmoothReset now accepts { skipCacheTest = true } to short-circuit the game cache wait and draw-queue drain
- ClusterRedraw passes the flag so icon-cluster and dense-pin-clustering sliders redraw immediately
- QuestieOptionsAdvanced wires the spawn-pin-dedup-radius slider directly to ClusterRedraw instead of the delayed SmoothReset path
- QuestieLib.GetColoredQuestName now nil-guards QuestieDB.GetQuest() before reading .isComplete
- TooltipHandler.AddItemDataToTooltip re-keyed on itemId only (was firing on every ElvUI refresh)
- Removed unconditional ResizeTooltip call from item tooltip path (caused first-hover pop)
- OnQuestDetail no longer writes objectives text to data[6] or quest body to data[17] or zoneId to data[8]
- OnQuestAccepted documents why requiredLevel/requiredRaces/requiredClasses are intentionally not captured
- Reused logIdx from data build pass instead of redeclaring it for the objective scan
- Secondary learner tooltip now matches GameTooltip style with or without ElvUI (defers to ElvUI when loaded, replicates its Transparent template otherwise)
- Combined-tooltip learner stats now framed by invisible spacer lines for visual separation
Xurkon