Records why the three identities are separate, the DBC mechanic table each
[cc:type] now follows, and which DR pools the immunity safeguard may act on.
Every vMaNGOS and nampower claim behind the preceding four commits is cited
to the source line that establishes it, so the next person to touch this can
check the premises instead of re-deriving them: the DRTYPE_ALL set is three
stun pools and nothing else, Charge and Intercept are the only triggered
stuns forced into the controlled pool, and SPELL_MISS puts the spell ID in
arg3.
Also notes the mouseover learning gap as designed behavior rather than a bug,
since the failure on the other side is permanent bad SavedVariables data.
The DR safeguard keyed recentCCHits by ccType and never reset the counter,
so three stuns on a target left it at count >= 3 forever. Every later
IMMUNE on that target was read as diminishing returns and no permanent
immunity was ever learned from it again.
Count within a window instead: a hit more than DR_RESET_WINDOW seconds
after the last one starts over at 1, matching the 20s DR decay the skip
check already used. Both the hit path and the skip check now key on the DR
pool from GetSpellImmunityDRType rather than the mechanic, so only the three
groups that actually diminish against creatures can hold learning back.
Learning itself records immunityType, the exact mechanic, so a target that
resists Gouge is no longer written down as stun-immune.
CheckImmunity also resolves a spell name through GetSpellImmunityType now,
so bare [immune]/[noimmune] answers for mechanic immunity and not just
schools.
stun matched mechanics 12, 14 and 30 at once through CCMechanicGroups, so
[cc:stun] fired on a gouged or sapped target and there was no way to ask
about any of the three on its own.
Name each mechanic instead: knockout (14) and sap (30) join the table, with
fumble (6), turn (23) and interrupt (26) filling the remaining gaps, and
grip stays as an alias of fumble so existing macros keep working. The group
indirection had no entries left afterwards, so it and its lookup branch are
gone; aliases in CCMechanics already cover every name that shares a mechanic.
CCTypesLossOfControl is untouched, so bare [cc] and [cc:any] still span the
whole loss-of-control set.
TriggerSpellMissEvent signals casterGuid, targetGuid, spellId, missInfo for
both SPELL_MISS_SELF and SPELL_MISS_OTHER. Both handlers read arg1 as the
spell ID, so every miss dispatched a GUID string where an ID was expected
and no miss ever reached the immunity recorder through the Nampower path.
Name mechanic 14 knockout to match the 1.12 DBC, and normalise CC type input
through one place so incap and incapacitate stay accepted as aliases for it.
Learned immunity shared one coarse bucket with the [cc:*] conditionals, so
mechanic 13 (freeze), 14 (knockout) and 30 (sap) all collapsed into "stun".
An NPC immune to Gouge was then recorded as stun-immune, and every later
Kidney Shot was suppressed against a target that had never resisted one.
Split the two identities. MECHANIC_TO_IMMUNITY_TYPE keeps the mechanic
one-to-one for learning, and GetSpellImmunityType resolves it from the DBC
spell/effect mechanic fields, falling back to the aura type only where a
spell leaves those empty.
The DR safeguard needs the real Vanilla DR pool rather than the mechanic:
GetSpellImmunityDRType separates controlled stun, triggered stun and Kidney
Shot, since vMaNGOS marks only those three DRTYPE_ALL against creatures.
The remaining staged groups are player-only and must not hold back learning.
[button:N] was a held-state modifier: IsMouseButtonDown(name), true while you
physically hold that button. Retail's is a property of the activation -- which
button clicked the action button -- and a keybind press counts as button 1
because it activates through the left-click path. Different in kind, and the
visible break was [button:1] Rejuvenation; Regrowth casting Regrowth from a
keybind, where retail casts Rejuvenation.
It now reads ClassicAPI's GetMouseButtonClicked, defaulting to LeftButton when
no click dispatch is running. That default IS the keybind rule: Button:Click()
with no argument reports LeftButton, and a macro run from anywhere else answers
button 1 the same way.
This also reverts 891467c. That commit refreshed the action bar on
GLOBAL_MOUSE_DOWN/UP so the icon could track the held state -- under the real
semantics there is no such state to track, and the icon correctly sits on
whatever button 1 resolves to. PLAYER_TOTEM_UPDATE stays; it shared the comment
and inherits it.
Requires ClassicAPI v1.15.8, up from v1.15.0. v1.15.8 scoped
GetMouseButtonClicked to the click dispatch; before it a HELD button kept the
value alive (by design -- OnDragStart reads it), so turning the camera with
right-click down made every keybind press read RightButton. The function exists
below v1.15.8 and answers wrongly, and a conditional that is quietly wrong is
worse than one that warns, so the floor moved rather than the conditional being
gated.
Bare [button] / [nobutton] survive as an extension retail has no equivalent for,
now meaning "a click dispatch is running" -- the practical "activated by a
keybind, not a click" test, and no longer leaning on anything but the documented
nil.
The LOSS_OF_CONTROL_ADDED/UPDATE registration was wrapped in a
`type(C_LossOfControl) == "table"` check, to keep an older ClassicAPI without
the namespace from erroring on an unknown event. That can't happen:
C_LossOfControl ships in v1.10.0, and the addon refuses to finish loading below
v1.15.0. Same shape as the health/power fallbacks removed just before this.
GetSchoolLockout's comment claimed it "Returns 0 for a client without
C_LossOfControl", which was not merely obsolete but wrong -- the function
indexes C_LossOfControl unguarded on its first line, so a missing namespace
raises rather than returning 0. The sentence describes a guard that was never
there.
Left alone: the C_Macro.SetMacroDisplay feature-detect in Init.lua looks like
the same pattern but isn't. SetMacroDisplay lands in v1.15.0, exactly the floor,
so a nil there means the client mod is absent outright, and the flag doubles as
the internal "may we publish" guard that ReleaseDisplays clears.
UnitHealthMissing, UnitPower, UnitPowerMax and UnitPowerMissing each carried an
`X or function(...)` fallback computing the value from UnitHealth/UnitMana. None
could run: the addon refuses to finish loading below ClassicAPI v1.15.0, and all
four globals ship at or below v1.10.0. The file header already promised as much
-- "the wrappers below call the API directly -- no fallbacks" -- and these four
were the only ones that didn't.
The UnitPower fallback was worse than dead. It dropped powerType and returned
UnitMana(unit), so a rage or energy read would have quietly answered with mana;
UnitPowerMax did the same with UnitManaMax. Only unreachability kept that from
being a bug.
They were also the file's only plain `API.X =` assignments, bound to the global
at load time. Now all 38 wrappers are `function API.X(...)`, resolving per call
like their neighbours -- no practical difference, since the DLL registers these
globals before any addon Lua runs.
Every caller passes (unit) or (unit, powerType); both pass straight through.
A totem is a summoned creature, not an aura, so nothing in the addon could see
one. Strength of Earth and Mana Spring could be faked through [nomybuff], but
Searing, Magma, Fire Nova, Tremor and Grounding put nothing on the player at
all, and those are the ones worth conditioning a recast on. There was no
workaround.
ClassicAPI backports the TBC totem bar (src/totem/Tracker.cpp), tracking the
player's own summons from their SMSG_SPELL_GO and classifying them by the
summon spell's SUMMON_TOTEM_SLOT1..4 effect, so Turtle's custom totems
self-classify. Ungated like GLOBAL_MOUSE_*: the tracker landed in v1.10.0, far
below the v1.15.0 floor the addon already refuses to run below.
[totem:X] takes an element (fire/earth/water/air), a slot number, or the totem's
own name -- ResolveTotemSlot tries the slot table first, then matches X against
the name standing in each slot, so [totem:Searing_Totem] asks for that totem
rather than whatever holds the fire slot. Bare [totem] takes the action as its
argument the way [mybuff] does, making /cast [nototem] Searing Totem a complete
recast macro.
Comparisons follow the aura precedent exactly: an empty slot reads -1, the same
"missing counts as least" ValidateAura uses, so [totem:Searing_Totem<5] passes
while the totem is expiring AND while it is absent -- one clause, no [nototem]
needed. A totem is either up or it isn't, so #N stack comparisons read 1 for a
standing totem. The parser needed nothing: its name/operator/amount capture is
generic, so multi-comparisons ([totem:X>2&<8]), OR groups and the De Morgan
flip all came for free.
Occupancy is read from GetTotemInfo's totemName, never from the time left.
GetTotemInfo's first return is TOOL presence -- whether the player carries the
slot's Earth/Fire/Water/Air Totem item -- not whether a totem is out, and a
summon spell with no SpellDuration row reports 0 seconds, which is "up with no
timer", not "absent".
Icon refresh needs two feeds. PLAYER_TOTEM_UPDATE covers a slot changing hands
(dropped, expired, killed, recalled). It does not fire as a totem counts toward
zero, and unlike an aura there is no ambient event stream to ride -- UNIT_AURA
is what quietly keeps [mybuff:X<N] honest -- so a [totem:X<N] icon held its last
answer until some unrelated event queued a refresh: the macro sat on the
question mark through the window it should have lit up. Hence the once-a-second
re-test in OnUpdate while any totem timer runs, skipped in realtime mode, which
re-tests everything anyway. Only the icon was ever at stake; the click path
evaluates conditionals live.
A [button:N] macro's icon never moved. The conditional reads IsMouseButtonDown
live, but nothing ever noticed that state changing: the OnUpdate's input poll
samples alt/shift/ctrl only, and no event fired for a mouse button, so no
refresh was queued and the icon sat on whatever the macro resolved to with no
button held.
ClassicAPI fires GLOBAL_MOUSE_DOWN / GLOBAL_MOUSE_UP on every raw press and
release, whether or not the click lands on a frame. Registering both and queuing
an action update is the whole fix -- the existing path re-runs TestAction per
action, so the conditional is read fresh and PublishDisplay hands the new answer
to C_Macro.SetMacroDisplay.
Ungated, unlike the Nampower-backed input events next to it: GLOBAL_MOUSE_* has
been in ClassicAPI since v1.0.0, far below the v1.15.0 floor the addon already
refuses to run below.
The handlers set isActionUpdateQueued directly rather than calling
QueueActionUpdate, as KEY_DOWN does: these fire from the message pump, which can
beat VARIABLES_LOADED to the CleveRoidMacros table QueueActionUpdate reads.
arg1 carries the button name and goes deliberately unused. The mask behind
IsMouseButtonDown is maintained by the same hook that fires the event, so
mirroring the payload into Lua would copy state the API already exposes -- and
the refresh covers every macro regardless, since any of them may test any
button.
A paladin's white swing refreshes their own judgement debuff on the victim, and
the server's edit is packet-silent, so pfUI#45 was the visible symptom: the
timer ran to zero while the debuff stayed up. This addon worked around it three
ways at once -- a UNIT_CASTEVENT MAINHAND/OFFHAND handler in Core.lua, an
AUTO_ATTACK_OTHER fallback in Conditionals.lua, and a CHAT_MSG_COMBAT_SELF_HITS
parser in Utility.lua that string-matched "hit"/"crit" and guessed at melee vs
spell by looking for a parenthesis -- each rewriting rec.start and pushing the
result into pfUI by hand.
ClassicAPI mirrors the server rule in the DLL (src/aura/JudgementRefresh.cpp),
keyed off SMSG_ATTACKERSTATEUPDATE, which is the visible trigger for the
invisible edit. It matches judgements by their DBC marker -- SPELLFAMILY_PALADIN
with SPELL_ATTR_EX3_ALWAYS_HIT -- for any paladin's judgements, not only ours.
That last detail retires the auto-detection too. The hardcoded judgementSpells
list could not cover Turtle's custom judgement ranks, so a cast queued a pending
scan that read the target's debuffs 0.5s later, matched names against
"^Judgement", and grew the list at runtime. Matching on the DBC attribute needs
no list to grow. Gone with it: judgementSpells, pendingJudgements,
detectedJudgementDebuffIDs, the pending-scan pass and its queue buffer, and both
cast-site hooks.
The scan also registered each ID it discovered into lib.sharedDebuffs so
[debuff] fallback matching worked for custom IDs. That mattered when detection
went through the tracking store; the read path resolves auras through
C_UnitAuras now, which returns a debuff whether or not we have a row for it.
Same shape as Carnage, opposite direction: Dark Harvest is a channel that makes
the caster's DoTs on the target tick 30% faster, and the server never tells an
observer the remaining durations changed. So this addon stamped dhStartTime on
every tracked record when the channel began, dhEndTime when it stopped, and
subtracted 30% of the elapsed window from each remaining time on read.
ClassicAPI compresses the ticks in the DLL (src/turtle/DarkHarvest.cpp), so
expirationTime already reflects it.
Removed: GetDarkHarvestReduction, ApplyDarkHarvestStart, ApplyDarkHarvestEnd,
GetTimeRemainingWithDarkHarvest, the channel-start capture on both the
UNIT_CASTEVENT and nampower paths, the channel-stop finalizer in
SPELLCAST_CHANNEL_STOP, the darkHarvestData state, and DarkHarvestSpellIDs.
The darkHarvest flag in Extensions/CursiveCustomSpells.lua stays -- that is
spell metadata this addon exports to Cursive, not tracking of our own.
Carnage refreshes the caster's Rip and Rake when Ferocious Bite procs it, and
1.12 tells an observing caster nothing about it -- no packet carries a debuff's
new remaining duration on another unit. So this addon inferred the proc: track
every Ferocious Bite cast with its target, watch PLAYER_COMBO_POINTS for combo
points failing to drop to 0 within 0.5s, then rewrite the stored Rip and Rake
records and push the new durations into pfUI's tables by hand. Roughly 570 lines
across three files, and it could only ever be a guess, because the roll that
decides whether Carnage procs is server-side.
ClassicAPI does it in the DLL (src/turtle/Carnage.cpp), roll-gated, so
expirationTime is simply correct and there is nothing to infer.
Removed with it: ApplyCarnageRefresh, the Ferocious Bite cast tracking on both
the UNIT_CASTEVENT and nampower paths, the avoided-Bite clearing in the miss
handler, the Rip/Rake cast snapshots on the UNIT_CASTEVENT and SPELL_GO paths,
the Rake landed/failed verification hooks, the proc detector in
ComboPointTracker, and lastRipCast / lastRakeCast / RipSpellIDs / RakeSpellIDs,
which had no readers left afterwards.
carnageDurationOverrides goes too, and it never worked: every reference to it
was a write, a clear, or the 30-second sweep in Core.lua's cleanup. Nothing ever
read it to change a duration. FerociousBiteSpellIDs stays -- the combo-point
snapshot still needs to know a finisher when it sees one.
C_UnitAuras.UnitDebuff / UnitBuff return the same fifteen values as
GetAuraDataByIndex without building a table, and the two index readers are
called in a loop over every aura on a unit -- one table per slot per refresh is
exactly the garbage the positional form exists to avoid. Only the first seven
values are wanted (name, icon, count, dispelType, duration, expirationTime,
source), and the range-locked variants keep the HARMFUL/HELPFUL split without a
filter string.
FindPlayerDebuff / FindPlayerBuff stay on GetUnitAuraBySpellID: one lookup
rather than a scan, so the AuraData table costs a single allocation and buys the
PLAYER filter. The -1-means-unknown rule moves into RemainingFrom so both shapes
share it.
The four readers that answered "what is on this unit, for how long, cast by
whom" spliced two sources together: the engine's aura array for existence and
icon, lib.objects for duration and caster. Everything the second half supplied
is now in the AuraData the first half already returns -- duration is the
caster-modified value when ClassicAPI saw the cast, expirationTime gives the
true remaining, sourceUnit names the caster -- so they read one source.
That deletes the overflow-slot rule with them. UnitDebuff used to split its
index space at 16 and map 17-48 onto buff slots, because a debuff can be parked
in a buff slot on an NPC, then filter the results through GetDuration to guess
which of those buffs were really debuffs. isHarmful is the aura's real polarity,
so the HARMFUL range already contains them and the guess is gone.
timeleft keeps its -1-means-unknown convention. ClassicAPI reports
expirationTime 0 for an aura whose cast it never observed -- one predating
login, or a max-stack refresh whose cache entry elapsed -- which is the same
"present but untimed" state the store expressed by having no record, so callers
testing `timeleft > 0` see no change.
GetDebuffCaster, IsOurDebuff and GetAllDebuffsOnTarget are deleted outright
rather than re-based: all three have no callers anywhere in the addon, and they
were the only readers of the slotOwnership and allSlots mirrors.
CheckImmunity asks "is this NPC immune to that school, and if the entry is
conditional, does it hold the buff that grants it" in two places. The
plain-school path was migrated to C_UnitAuras.GetAuraDataBySpellName -- one
by-name lookup across the unit's auras. The split-damage path was not, and was
still walking 32 buff slots calling C_Spell.GetSpellName on each to compare the
name back, twice over, once for the initial school and once for the DoT's.
Both copies were the same twenty lines with the variable names changed, so they
fold into a SchoolImmune(unitId, school, targetName) helper that the split path
calls twice. Same answer, same order of checks: a permanent entry is immune, a
table entry without a buff is immune, a table entry with one is immune only
while the aura is up.
The plain-school path keeps its own inline version -- it is not the same
predicate. It gates on the recorded spell name first for the "unknown" school,
and a table entry carrying neither buff nor spell falls through to false there
where the split path calls it immune. Unifying those is a behavior change, not a
refactor, so it is left alone.
Three loops walked 1..36 by hand, two of them re-arguing in comments why 36
is the right number. It is: the client addresses 18 account-wide macros at
1-18 and 18 character-specific ones at 19-36, and that range is what
GetMacroInfo and C_Macro.SetMacroDisplay index. Nothing in the API hands it
back -- GetNumMacros() returns how many of each tab are used, which cannot be
summed into a range, because the character block still starts at 19 when the
account block is empty. Blizzard's MAX_MACROS is no better: it lives in the
load-on-demand Blizzard_MacroUI and is nil until the macro window is opened.
So CleveRoids.MAX_MACRO_SLOTS in Init.lua, carrying that reasoning once.
ValidateAllMacros also dropped a numAccountMacros it computed and never read.
ReleaseDisplays keeps sweeping the full range rather than the macros it knows
it published: PublishAllDisplays clears that record on each re-parse, so a
macro claimed before one and gone after it would keep our last published value
forever. Releasing a slot we never claimed costs nothing, and it runs once.
An out-of-date ClassicAPI did not degrade the addon, it decapitated it.
Utility.lua calls ev:RegisterUnitEvent at file scope (v1.15.0), so on an
older build that call raises and Lua abandons the rest of the chunk --
every definition below line 4118 of a 10040-line file, some 6000 lines,
silently never happens. What the user sees is the first one a common path
reaches: PLAYER_TARGET_CHANGED calling a nil ClearResistState, once per
target swap, forever.
The requirement check still said v1.12.1 (the positional
C_UnitAuras.UnitAura), so the affected client sailed through it and the
warning it did print promised only that "dispel-type conditionals will be
unavailable" -- describing a graceful degradation that was not happening.
The gate is now v1.15.0 and says the addon cannot finish loading, with the
installed version in the message the way the Nampower branch does it.
ClassicAPI.lua gains GetVersion() to decode CLASSIC_API_VERSION, keeping
the X*10000 + Y*100 + Z encoding inside that module, and its header notes
that the minimum has to rise with any ClassicAPI call adopted at file
scope -- that is the property that turns a version mismatch into a silent
half-load rather than a missing feature.
C_Macro.SetMacroDisplay also ships in v1.15.0, which retires the last
pre-ClassicAPI display path:
useClassicAPIDisplay folds into ClassicAPIMacroDisplay. The two were
always equal -- set together at load, cleared together in
ReleaseDisplays -- and one of them is a handshake ClassicAPI reads, so
a second name for the same state could only ever drift.
TestForAllActiveActions loses the per-slot ACTIONBAR_SLOT_CHANGED
fan-out. Publishing repaints every slot holding the macro through the
client's own notifier; the fan-out was what ran when SetMacroDisplay
was unavailable, and its one remaining reachable state was after
DisableAddon, where ReleaseDisplays has already handed all 36 macros
back and ClassicAPI repaints them itself.
The load-time feature detect stays, with a new reason: below the minimum
the addon is already broken, so a nil SetMacroDisplay now means ClassicAPI
is absent outright -- the case the requirement check warns about and then
keeps running.
Two handlers opened by testing arg1 against a fixed set of unit tokens, so
every other unit's copy of UNIT_AURA reached Lua only to be compared away
-- in a raid that is the bulk of them. ClassicAPI's RegisterUnitEvent
pushes that filter into the client, so the handler only runs for the units
named. Filtering is per (frame, event), so the other events sharing the
libdebuff frame are unaffected.
Utility.lua UNIT_AURA -> "target" (libdebuff seeding)
ComboPointTracker UNIT_AURA -> "target", "player"
Extensions register through CleveRoids.RegisterEvent, so ExtensionsManager
gains a RegisterUnitEvent alongside it.
Separately, the main frame's unit state streams (UNIT_AURA / UNIT_HEALTH /
UNIT_POWER, or the _GUID variants under Nampower v2.39+) are now registered
only in event-driven mode. Their handlers are wholly wrapped in
`realtime == 0`, so with realtime on every one of them crossed into Lua and
returned immediately while the OnUpdate did the refreshing. Registration is
applied at load, again at VARIABLES_LOADED once the saved setting is
readable, and whenever /cleveroid realtime flips it.
Those three stay plain RegisterEvent calls: their handlers ignore the unit
and refresh every macro, and a conditional may name any unit
([@party3,hp:<50], @focus, @mouseover), so narrowing the token set would
leave those icons stale. The _GUID variants could not be filtered anyway --
their first argument is a GUID, not a unit token.
Squash of the classicapi_next branch (10 commits). ClassicAPI is now a hard
requirement, and the Lua-side reimplementations it supersedes are gone:
net -1087 lines across 12 files.
Macro display: resolved macro actions are published through
C_Macro.SetMacroDisplay, so the client draws macro buttons and the
action-bar function overrides this addon used to install are removed.
Ownership is per macro -- ClassicAPI keeps the macros we never claim.
Auras and timing: ClassicAPI is the source of truth for non-player aura
timing via the positional C_UnitAuras.UnitAura, replacing the write-only
buff tracking tables and the overflow-slot fallback.
Cleanup: the hasPfUI76 flag and the pfUI 7.6 branches it gated, two
silently-shadowed code paths, and per-call allocations in the event and
publish paths (SPELL_CAST_EVENT merged). Packaging moves to
brues-code/packager@vCAPI.
Macro syntax: a clause may carry a leading run of [group] blocks sharing
one action, Blizzard-style -- groups are OR'd, first pass wins, and []
always passes. The ;-separated form is unchanged and mixes freely. @focus
clauses with no focus set now fail quietly instead of printing
"Invalid target" on the way past.
* feat: add /stopchannel command
* review: give /stopchanneling a real handler and warn when nampower can't honor it
---------
Co-authored-by: Brues <5278969+brues-code@users.noreply.github.com>
IsRelicSlot no longer tests playerClass against PALADIN/DRUID/SHAMAN. It now
reads CleveRoids.hasRelicSlot, sampled once in PLAYER_LOGIN.
Also folds PerformEquipSwap's slot-18 check into the same helper. Its class list
(HUNTER/WARRIOR/ROGUE/MAGE/WARLOCK/PRIEST) was the exact complement of the relic
classes, so it becomes `not IsRelicSlot(inventoryId)` and the two lists can no
longer drift apart.
* add support for 'button' conditional
* Route [button] through Multi, add [nobutton], register as static
- [button] now reads conditionals.button via Multi(...) instead of indexing
_groups.button[1].values[1] directly, so OR/AND lists and repeated groups
behave like every other argument conditional. [button:1/2] now matches left
or right; previously only the first value was ever checked and the rest were
silently dropped. This also removes the unguarded _groups reach-in, which
Multi itself defends against (it checks _groups is present and falls back to
_operators).
- Adds [nobutton:N] via NegatedMulti, following the convention that every
conditional ships its negation (nomod, nostance, nokeydown, ...).
- Bare [button] / [nobutton] now mean "any / no mapped mouse button held",
mirroring bare [mod]. Bare [nobutton] doubles as an "activated by a keybind
rather than a click" test.
- Registers button/nobutton in STATIC_CONDITIONALS next to mod and keydown, so
they are checked once up front rather than re-evaluated for every candidate
unit during multiscan target scanning.
Still open: IsMouseButtonDown reports live physical state, so a keybound macro
has no button held and [button:N] is false for it, and whether the button still
reads down at macro-execution time depends on whether the action button fires on
mouse-down or mouse-up. Needs in-game verification.
---------
Co-authored-by: Brues <5278969+brues-code@users.noreply.github.com>
#2 corrected IsItemInSlot's native path, but the same 0-based assumption was
still present in the two Lua fallbacks, which run on nampower < v2.18. Per
nampower's SCRIPTS.md the slot is 1-19 (16=MainHand, 17=OffHand), and
ClassicAPI's GetInventoryItemID is 1-based as well:
- GetEquippedItem's fallback called GetInventoryItemID(slot + 1), contradicting
both its own native path and its doc comment.
- GetEquippedItems' fallback built 0-18 keys while the native path returns
1-19, so the two branches handed back differently-keyed tables.
Neither wrapper currently has any callers, so this was latent rather than
user-visible. Also records the 1-indexed convention at the IsItemInSlot call
site, where the misleading "0-indexed" comment used to be.
Adds CleveRoids.conditionalAliases, applied in ParseMsg so a deprecated name
is rewritten to its current keyword before anything downstream (evaluation,
_groups) sees it. This is the mechanism for renamed modifiers, which -- unlike
predicates such as [stl] -> [stealth] -- have no Keywords function an alias
could simply point at.
[mouseuse] now resolves to [cursor], routing old macros through ClassicAPI's
C_Spell.CastAtCursor / C_Item.UseAtCursor. Drops the post-cast block that cast
normally and then faked a mouse click via CameraOrSelectOrMoveStart/Stop to
place the AoE circle, along with its workaround for that call spuriously
starting auto-attack.
MacroErrorChecker seeds VALID_CONDITIONALS from the alias table, so the old
names stay valid syntax without needing a placeholder ignoreKeywords entry.
scanDispel now reads each slot through the positional C_UnitAuras.UnitAura
(added in ClassicAPI v1.12.1) instead of the table-return GetAuraDataByIndex,
dropping a throwaway AuraData table per slot on the up-to-48-slot scan behind
the [magic]/[curse]/[disease]/[poison]/[dispellable] conditionals.
Since UnitAura is called directly with no fallback, gate on it: the load-time
check in Core.lua warns (matching its warn-don't-disable design) when
ClassicAPI is older than v1.12.1, wiring up the previously-unused
ClassicAPI.HasMinimumVersion.
- Resolve Feed Pet from spell ID 6991 (C_Spell.GetSpellName) so /feedpet works on non-enUS clients without a Localized.Spells entry
- Fix CleveRoids.IsReactive referencing undefined global spellName instead of its name parameter
UPDATE_MACROS and SPELLS_CHANGED each did a full spell + talent + pet +
120-slot action-bar rebuild inline, and both fire several times during
login as the spellbook and macros populate - so the rebuild ran 4-6
times in the first ~2s, most of it duplicated work fanning 120 button
updates out to Blizzard/pfUI/Bongos each pass.
Defer the rebuild instead: UPDATE_MACROS, SPELLS_CHANGED and PLAYER_LOGIN
arm a 0.3s debounce (macroRebuildTime); the update loop runs RebuildMacros
once when the burst settles. RebuildMacros skips the action-bar pass until
`ready` (GetAction early-returns before then, and the +1.5s init timer
builds the bars once anyway), and BAG_UPDATE_DELAYED's bar rebuild is
gated on `ready` too. Net: IndexActionBars runs once at login (the
suppressed init-timer pass) instead of 4-6 times, IndexSpells ~2x instead
of ~5x. Runtime macro edits and bag changes still do a full unsuppressed
rebuild, just debounced.
ClassicAPI hooks the global GetItemInfo to auto-warm the item cache on a
miss and fires GET_ITEM_INFO_RECEIVED when the async fill lands, so the
tooltip-scan warmup is obsolete: IndexItems's own GetItemInfo calls
already trigger the same warmup, and owned items (bags + equipped) are
priority-prefetched by the engine. The old warmup also assumed the fill
was synchronous, which no longer holds.
Drop DoWDBWarmup and its login scheduling; instead listen for
GET_ITEM_INFO_RECEIVED and run a debounced re-index. IndexItems records
owned itemIDs it could not resolve into pendingItemInfo, and the handler
ignores any fill not in that set (quest DB scans, AH sweeps, chat-link
hovers, inspects) in O(1) so unrelated bursts do not cause reindex churn.
GetSpell/GetItem now resolve explicit spell:/item: ID forms. spell:<id>
prefers the player's spellbook entry via FindSpellBookSlotByID (per-rank
and pet aware) for full cost/cooldown/usability, falling back to an
id-only entry rendered via SetSpellByID for spells not in the book.
item:<id> reuses the existing numeric lookup for location-aware tooltips.
Guard the id-only path against nil spellSlot/cost in TestForActiveAction
and GetActionCooldown so display-only spell references don't crash.
Remove the dead, no-op GetSpellSlotByID stub.
Delete unnecessary local 'i' declarations in Extensions/Mouseover/pfUI.lua (ResolvePfUnit, RegisterPartyScripts, RegisterRaidScripts, RegisterRaidMarkScripts). The for-loop headers already provide a local loop variable, so the explicit locals were redundant and could shadow variables. No functional change.
HasGearEquipped now wraps ClassicAPI's native slot-walk, which
short-circuits on first match. Removes BuildEquipmentCache, the
_equipped* cache tables/invalidation, and the InvalidateEquipmentCache
call from PLAYER_EQUIPMENT_CHANGED. [equipped] reads live engine state,
so no cache staleness surface remains; also gains item-link support.
Replace the full IndexItems + action-bar rebuild with a single-slot
IndexEquipSlot using the event's arg1 (slot) and arg2 (hasCurrent),
and drop the now-pointless throttle/deferral. Bag-side deltas remain
covered by BAG_UPDATE_DELAYED; [equipped] runs off its own cache.
Brues