ClassicAPI advertises WOW_PROJECT_ID = WOW_PROJECT_CLASSIC, which
steers modern addon ports onto their Classic Era code paths -- where
they also assume Era's C-level frame behaviors. Supply the missing
ones (surfaced by AI_VoiceOver's Era path, but generic to every port):
- region:SetPoint("RIGHT") -- the one-arg form. Vanilla's parser
accepts every other modern shape but raises the usage error when
arg 3 is fully omitted (LUA_TNONE fails its type validation); a
co-hook on Script_SetPoint (0x007A2540, the ONE implementation in
the Region base table) normalizes it to (point, 0, 0) -- vanilla's
parent-relative form, exactly the modern semantic.
- region:SetSize / region:GetSize -- registered on the REGION base
method registry (ctx 0x00CF54B4), so they resolve on frames,
buttons, textures, fontstrings -- including engine-created textures
(GetNormalTexture():SetSize(...)).
- frame:SetShown -- Show/Hide are per-branch script functions with
their own object typechecks, so it's registered separately on
Frame / Texture / FontString with matching delegates.
- frame:SetResizeBounds -- modern rename of SetMinResize/SetMaxResize.
- frame:HookScript -- absent in vanilla entirely. Chains the old
handler (invoked vanilla-style, reading the this/event/argN
globals), then invokes the hook handler with modern positional args
((frame, event, arg1..arg9) for OnEvent, (frame, arg1..arg9)
otherwise) via a C closure over (handler, old, isOnEvent).
All methods delegate to the engine's own Script_* implementations by
reshaping the Lua stack -- no re-derived setters, so UI-scale
conversion, object resolution, and type checking stay engine code.
Method-registry contexts beyond GameTooltip's (Region / Frame /
Texture / FontString) are now in Offsets.h per the
docs/BlizzardScriptAPI.md catalogue.
Verified in-game: GameTooltip:GetSize() returns real dimensions,
CreateFrame + SetSize(50,50) reads back 50, and AI_VoiceOver's Era
path runs through its entire frame construction (SetPoint one-arg,
SetSize, SetShown, SetResizeBounds, 14 HookScript sites) where it
previously errored on the first call.
Rupture at 4 combo points reported duration=6 (the SpellDuration.dbc
base) instead of the real 14s: finisher durations are computed
server-side at cast time and 1.12 never transmits them for debuffs the
player puts on other units. Verified with a whole-protocol sniff --
SMSG_UPDATE_AURA_DURATION (0x137) is self-scoped (slot+duration for the
player's own aura array), 0x2DD has no client handler and never
arrives, and nothing else carries the value. (pfUI issue #17.)
Mirror the server's computation instead (new Aura::ComboDuration):
- Snapshot combo points ([player+0xE68]+0x1029, the field behind
Script_GetComboPoints) the instant CMSG_CAST_SPELL leaves, via a
co-hook on the NetClient send funnel (FUN_NET_SEND 0x005379A0). The
server consumes CP before SMSG_SPELL_GO returns, so reading at
SpellGo time is too late; the send funnel also covers
nampower-queued casts, which bypass any Spell_C_CastSpell co-hook
via trampoline re-entry.
- Compute base + (max - base) * CP / 5 from the spell's own
SpellDuration.dbc row ((v)mangos CalculateSpellDuration), then apply
the player's duration SpellMods, matching server order. DBC-driven
and spell-agnostic: finishers are exactly the rows with base != max
(Rupture 6/16s, Kidney Shot 1/6s, r1 0/5s, Slice and Dice 6/21s;
Expose Armor 30/30 self-disables). Consumed by Aura::Source's
SpellGo hook; 0 falls through to the existing base path.
- Turtle's reworked Rip (8s + 2s per CP) has NO client data: all six
ranks' DurationIndex points at SpellDuration row 87, which Turtle
added server-side only -- the client table jumps 86 -> 105. Built-in
values cover it, gated on the dangling-row condition itself (the
only dangling cross-reference in the entire client Spell.dbc, per a
full DurationIndex/CastingTimeIndex/RangeIndex sweep) rather than on
realm detection: stock clients resolve Rip's index normally, and if
Turtle ships the missing row the DBC takes over automatically.
- C_UnitAuras.RegisterComboDuration(spellID, baseSec, maxSec) as an
escape hatch for other servers with the same class of client-data
gap; a registered override beats DBC and built-ins.
Verified in-game: Rupture at 4 CP now reads duration=14.
Turtle's core doesn't broadcast SMSG_SPELL_FAILURE or
SMSG_SPELL_FAILED_OTHER on interrupt (verified by logging both
handlers through live tests), so the SPELL_FAILED_OTHER co-hook never
fired and interrupted enemy casts (Kick / Earth Shock / Counterspell)
lingered as ghost bars. What actually stops the cast visual on Turtle
is SuperWoW calling CGUnit_C::ClearCastingSpell (0x0060D040) directly.
Co-hook that choke point instead: every unit-stopped-casting path
funnels through it (failure packets, SPELL_GO completion, movement,
death, animation events, local cancel), gated on the unit's live
current-cast field at +0xC8C, so eviction is independent of which
signal the server sends. Both failure-packet handlers stay co-hooked
as backstops for servers that do broadcast them. Verified in-game:
all three interrupts evict; non-interrupting damage doesn't
false-evict.
Also confirmed via a whole-protocol sniff on the NetClient message
dispatch (FUN_NET_MESSAGE_DISPATCH = 0x00537AA0, documented for
future use): SMSG_SPELL_DELAYED goes only to the affected caster --
observers receive no packet of any kind for pushback, so remote bars
can't stretch. Documented as an inherent 1.12 protocol limitation.
Fires (no payload) when any equipped item's durability changes -
combat damage, death, repairs - via a co-hook on the engine's
inventory-alerts recompute (FUN_004C7EE0, the routine behind vanilla's
UPDATE_INVENTORY_ALERTS). The engine calls it whenever an owned item's
fields change (the item post-update handler FUN_005D9A90 routes every
item values-update through it) plus equip/enter-world paths, so it is
the engine's own durability-may-have-changed signal. Post-original we
diff a per-slot {itemGuid, durability} snapshot and fire once per
recompute with a real change - a death dinging all armor produces one
event, not nineteen.
The snapshot seeds silently on the first recompute of the process and
is deliberately NOT re-seeded per enter-world: death/release/graveyard
teleports re-run the setup, and a per-setup reset swallowed the
death's own durability loss (verified in-game; repairs fired, deaths
didn't until this was fixed).
Dead end, kept documented at OFF_DESC_ITEM_DURABILITY: bank-1 (ITEM
field) descriptor observers register without error but are never
invoked - item values-updates demonstrably arrive and player-bank
observers fire in the same session; registration arithmetic fully
verified (bank base FUN_00465690(1)=0x18, bank walk item 0->1->2).
Prime suspect is the registrar's internal wrapper lookup
(FUN_00464890, the observer registry, distinct from the object
manager) silently no-op'ing for item objects.
Also in Player::Equipment: resolve the player via the non-throwing
GUID resolver (FUN_00468460, what the engine's own bag observer uses)
instead of FUN_RESOLVE_UNIT_TOKEN, which raises a Lua error when the
token doesn't resolve - unsafe inside SMSG_UPDATE_OBJECT processing
where the player can be mid-create.
The quest's objective count — the length GetQuestDetails(questID)
.requirements would have — read straight from the cache block with no
Lua tables or strings materialized. nil when the quest isn't cached
(same contract as GetQuestDetails).
Exists because materializing full detail tables in a login-path loop
is expensive at warm cache: GetQuestDetails copies description /
objectives / completion text into Lua strings (~5 KB per quest).
TurtleGuide's guide-load OIDX validation did exactly that for every
tagged step — fine on the first (cold-cache) session, but once the
quest cache was warm from questcache.wdb every subsequent login
exhausted vanilla's fixed Lua memory pool and crashed the client in
lmemPool.cpp:53. With this accessor the same validation costs a few
dozen byte reads regardless of cache temperature.
The earlier investigation tried to hook a firer of
UNIT_INVENTORY_CHANGED; none of the three known 0xBA fire sites
trigger on equip because equipment descriptor writes are silent — the
engine registers descriptor-field observers (the __AUCMirrorHandler__
system: registrar 0x00467E70, dispatcher 0x00465570 with per-node
mirror + memcmp change detection) for the bag/backpack/bank/keyring
GUID ranges but never for equipment 0x4A8..0x538.
So register our own: co-hook the engine's inventory observer setup
(FUN_004F8CC0, once per enter-world) and, post-original, register one
observer per equipment GUID field through the engine's registrar —
same bank/size/GUID shape as the engine's bag observers, engine-
managed lifetime, no polling, no hot-function hook. The dispatcher
invokes us only on a real change.
Callback ABI verified from the dispatcher call site at 0x004655DC and
the engine callback FUN_004F8DB0's RET 0x10:
__fastcall(fieldOffset /*ecx*/, size /*edx*/, guidLo, guidHi,
oldValue*, userArg) -> ret 0x10
equipmentSlot is the 1-based Lua inventory slot (1 head .. 19 tabard);
hasCurrent (1/nil) is read from the player invMgr GUID array — the
same post-change truth the engine's bag observer reads.
Vanilla has two disjoint questgiver paths. Gossip NPCs send
SMSG_GOSSIP_MESSAGE into the gossip arrays this namespace reads; but
quest-only NPCs send SMSG_QUESTGIVER_QUEST_LIST (the QUEST_GREETING
event / QuestFrameGreetingPanel), whose data lands in entirely
separate storage C_GossipInfo never read — every getter returned
empty for those NPCs. GetText was also ungated: the gossip text
buffer persists after close, so it served the previous gossip NPC's
greeting.
- Gate all gossip reads on the live gossip NPC GUID (0x00BC3F58,
cleared by the close worker at 0x004E1FA0 before GOSSIP_CLOSED
fires). Stale text/options/quests are no longer visible, and
selectors can't send a CMSG from stale slots.
- Serve the quest-greeting session when one is live instead: the
questgiver GUID (0x00BE0810) must match the greeting-fill GUID
(0x00C4D740) so leftovers can't leak when a quest session starts
from a gossip-menu click. GetText returns the greeting text;
Get{Available,Active}Quests / GetNum* read the greeting arrays
(32 x 0x4C entries {questID, level, title}) with isComplete
recovered from the packet handler's scratch status array
(status 4 = turn-in-ready); Select{Available,Active}Quest route to
the engine's greeting selectors (0x005012A0 / 0x00501320).
- Neither session live: empty tables, 0 counts, empty text.
GetOptions/GetNumOptions stay gossip-only — greetings have no
options. Retail keeps the two APIs disjoint; merging is deliberate so
modern-addon ports work on greeting-path NPCs (common on 1.12
servers).
Introduce QUEST_REMOVED (fires with questID) and document it in API.md and README. Rename Quest::Accepted → Quest::LogEvents and update the quest-log rebuild hook to compute pre/post deltas for both QUEST_ACCEPTED (questLogIndex, questID) and QUEST_REMOVED (questID). Add suppression rules to avoid firing on login/resync bulk changes. Update Offsets.h commentary to reflect the new LogEvents module and its responsibilities.
Mind Control is a 3 s cast on a spell flagged CHANNELED with a 60 s
duration. HandleSpellStart branched on the channel attribute alone, so
at cast start it stamped g_channel with the 60 s duration and killed
the client hook's correct 3 s cast stamp — the bar showed a premature
60 s countdown (pfUI #9). Same story remotely: an observed MC stored a
60 s channel entry instead of the 3 s cast.
- Treat a channeled spell with castTime > 0 as its CAST phase in
HandleSpellStart (player and remote paths); instant channels keep
the immediate stamp.
- Co-hook the MSG_CHANNEL_START handler (0x006E7550, layout from
nampower's SpellChannelStartHandlerHook: spellId u32, duration u32).
The server sends it to the caster the moment the channel begins —
after MC's cast completes — with the server-authoritative duration.
Stamp g_channel there, before the original fires
SPELLCAST_CHANNEL_START, so pollers see fresh data. For instant
channels it re-stamps what the SpellStart path just wrote.
Verified: MC shows 3 s cast then 60 s channel; Mind Flay-style instant
channels unaffected.
tinycbor's grow-and-retry contract: on buffer overflow the encoder
switches to counting mode and encoding must CONTINUE for
cbor_encoder_get_extra_bytes_needed() to be accurate — a nested
container's tally only reaches the parent via close_container. The old
code returned false on the first CborErrorOutOfMemory, so a mid-
container overflow could report extra == 0 and be misread as a genuine
"value not representable" failure (returning nil), or undercount and
churn through multiple resize passes.
Add EncOk() (NoError or OutOfMemory = keep going) around every encode
call; the retry loop now distinguishes real failures (return nil) from
overflow (single complete-tally resize, retry).
Remote casts cached from SMSG_SPELL_START were expired only by their
computed end time — an interrupted remote cast lingered as a ghost bar
for every UnitCastingInfo consumer (pfUI #11). The interrupt signal
does exist at the packet layer: the server broadcasts
SMSG_SPELL_FAILED_OTHER to observers whenever a started cast aborts
(interrupt, death, movement, fizzle).
Co-hook its handler (0x006E8E40, same __stdcall shape as
FUN_SPELL_DELAYED; address + layout mirrored from nampower's
SpellFailedOtherHandlerHook — body is plain casterGuid(u64),
spellId(u32)) and:
- evict the matching g_remoteCasts entry (guid + spellID), so remote
cast/channel bars drop the same tick the cast dies;
- clear the player's own g_cast on a guid match — covers the
server-stamped chained recast that the WorldTick VAR==0 poll can't
clear (self-receipt is server-dependent; the endMs self-expiry
backstop remains if it never fires).
Update the stale 'no per-unit interrupt signal in 1.12' notes in
Cast.cpp and docs/API.md.
Re-add the C_VoiceChat / C_TTSSettings SAPI module removed in 6f543ec,
gated so it coexists with the standalone VanillaTTS DLL: when
VanillaTTS.dll is loaded in the process, the module stands down
entirely — no Lua functions, no ttsVoice/ttsSpeed/ttsVolume cvars, no
VOICE_CHAT_TTS_* event reservations, no SAPI init. Without the gate,
each DLL would claim its own event-table slot for the same event names
(listeners bind to one slot while the other DLL fires the other) and
the cvar registrar would replace the first DLL's clamp callbacks.
The check (GetModuleHandleA at RegisterLuaFunctions time) is
load-order-independent — VanillaFixes loads every DLL at process
attach, long before FUN_LOAD_SCRIPT_FUNCTIONS fires. The event
AutoReserves move from file scope into the gated registration as
function-local statics so a stand-down leaves no reservations behind.
Restore the ole32 link, the README VoiceChat/API rows + event table,
and the docs/API.md VoiceChat section (with a coexistence note).
GameTooltip:GetItem() returned the previous item after SetUnitAura /
SetEquipmentSet (and any tooltip that shows no item). The engine's per-tooltip
clear FUN_00530050 zeroes the unit / gameobject / itemID / spell identity fields
but leaves the item GUID (+0x380/+0x384) intact -- the field CGItem tooltips
(SetBagItem/SetInventoryItem) set and that Item::TooltipItem::CurrentID resolves
when the itemID field is 0. So a stale GUID lingered across a switch to non-item
content and GetItem/HasItem/IsEquippedItem resolved the old item.
Co-hook FUN_00530050 and zero +0x380/+0x384 after the original. All 14
Set*/builder paths route through this clear (verified via xrefs; the aura
builder FUN_0052f880 and the item builder FUN_0052b650 both call it), and the
item builder calls it before writing the GUID -- so every fresh tooltip resets
the item GUID and only a real item build re-sets it. The hook runs no Lua, just
zeroes 8 bytes; the CurrentID unit/GO/spell gating is now belt-and-suspenders.
Verified in-game: bag/equipped items still resolve; SetUnitAura / SetEquipmentSet
no longer leak the previous item.
The initial fire used FUN_00702690, whose exec-context mutation (DAT_00ceeac0)
is only valid from the engine's top-level frame-script dispatch; firing it from
inside the item builder (mid-Script_Set*, and in the wild under a stack of addon
Set* wrappers) raised through the outer wrappers. Two fixes make the fire safe
for normal handlers:
- Call the inner invoker FUN_00704d50(handler, tooltip) directly. It binds the
global `this` = tooltip and runs the handler under its own protected pcall
(0 args), self-contained and context-independent.
- Snapshot lua_gettop before the fire and lua_settop back after, so the still-
running outer C code sees exactly the stack it left. Without this the invoker
left the stack shifted and the outer code / other DLLs' hooks indexed a stale
slot -> ACCESS_VIOLATION.
Verified in-game: SetScript / HookScript / GetScript all work, `this` is the
tooltip (a1 is nil -- vanilla convention, not modern `self`), and no-op /
print / AddLine handlers fire cleanly on bag / equipped / merchant / auction.
Docs: correct the handler example to use `this` (not a `self` arg), and add the
caveat that re-entrant tooltip rebuilds from the handler (ShowCompareItem, Set*
on another tooltip) can crash in a hook-heavy client and must be deferred
(C_Timer.After(0, ...)) -- the event fires mid-build.
Backports the modern OnTooltipSetItem tooltip script to 1.12 as a real
frame script, settable via the standard SetScript / GetScript / HookScript
and firing whenever a tooltip's item is set (with the tooltip as self). Lets
addons annotate item tooltips by hooking one script instead of wrapping every
Set* method -- the destructive wrap-and-rebuild pattern (AtlasLoot's
AddSourceLine) that clobbers the tooltip's item linkage.
Fully native, two Game::HookAutoRegister co-hooks in Tooltip::SetItemEvent:
- Resolver (FUN_005295d0, the CGGameTooltip vtable script-name -> slot
vmethod): for "OnTooltipSetItem" (a name the engine doesn't know) return a
pointer to a C-side per-tooltip 8-byte {handler, context} cell. The tooltip
object (alloc size 0x460) has no free slot, and the resolver's only contract
is "return where the handler lives", so external storage behaves exactly
like the engine's own slots -- proven by OnTooltipCleared (+0x44c).
- Item builder (FUN_0052B650, the shared funnel for SetBagItem /
SetInventoryItem / SetHyperlink / SetMerchantItem / SetAuctionItem / ...):
after the build, fire the handler via FUN_00702690 (the no-arg runner the
clear uses for OnTooltipCleared). Recursion-guarded, and suppressed via
Tooltip::SetItemEvent::Suppressor during Tooltip::Compare's internal
equipped-item build so a handler doesn't run mid line-shift.
Item only for now; OnTooltipSetSpell / OnTooltipSetUnit follow the same
pattern with their own builders.
Introduce Item::TooltipItem::CurrentID -- the single "what item is this
tooltip showing" resolver. It trusts the link-path itemID (+0x398), and
otherwise resolves the CGItem GUID (+0x380) but only when no unit /
gameobject / spell id is set, since the clear does not zero the item GUID
and it goes stale after the tooltip moves on.
- GameTooltip:GetItem / HasItem now route through CurrentID: fixes GetItem
returning nil for CGItem tooltips (bag / equipped / merchant / loot) and
HasItem false-positiving on a unit tooltip after an item hover.
- Add GameTooltip:IsEquippedItem() (3.3.5 backport, registry index 63),
resolving the displayed item via CurrentID + the equipment-slot walk.
- SetHyperlinkCompareItem: the comparisonTooltip (4th arg) fallback uses
CurrentID; stat deltas now show by default and hide only for an explicit
false / 0 (was: show only for a truthy shift).
- GetItem round-trips the random-suffix id in the link for auction/compare
items (no CGItem instance) via the compare descriptor.
- FrameXML GameTooltip_ShowCompareItem falls back to self.compareLink, so a
caller can hand it an explicit link when the tooltip's item state was
rebuilt out from under it (e.g. an AH addon appending source lines).
- Offsets: compare-descriptor + FontString offsets; drop the duplicate
tooltip item-id/guid offsets in favor of the canonical OFF_TOOLTIP_*.
Backports the 3.3.5+ shopping-tooltip method, fully native (no lua_pcall):
fills a tooltip with the item currently equipped in the given item's slot
— a grey "Currently Equipped" header, the equipped item's tooltip, then a
green/red per-stat delta breakdown gated on shiftButton.
- Extract the stat engine (base record + on-equip-spell auras + random
suffix) out of item/Stats.cpp into the shared Item::StatAccum, reused by
C_Item.GetItemStats/GetItemStatDelta and the compare deltas.
- Build via the engine's own item-tooltip builder; prepend the grey header
by shifting the line FontStrings, replicating the show/hide protocol,
keying on the visible flag (no stale text) and resetting cell color.
- Delta colors sourced from ColorData.h via a new constexpr ByTag helper.
- offset selects the slot for two-slot items (rings/trinkets/1H weapons);
shiftButton gates the deltas (true/1 vs false/0/nil); a comparisonTooltip
4th arg lets it run link-free, 1:1 with the retail call.
- FrameXML GameTooltip_ShowCompareItem consumer with a dynamic,
backdrop-derived anchor gap.
Vanilla GameTooltip:AddLine silently drops any line past the 30
TextLeftN/TextRightN FontStrings the FrameXML template declares, so
stat-heavy tooltips (pfUI eqcompare, AtlasLoot) lose their tail.
Tooltip::LinePool co-hooks the tooltip's XML-OnLoad region scan
(FUN_00529650, CGameTooltip vtable slot 9) via HookAutoRegister. After
the engine sets up the template's 30 lines it creates TextLeft/Right
31..60 in C++ (alloc from the CSimpleFontString pool -> ctor(parent) ->
name -> font copied from the previous line -> anchor) and appends each
pair to the tooltip's three line arrays, bumping numLinesAllocated. Runs
once per GameTooltip-family frame at creation and per /reload, so it
auto-covers GameTooltip, ShoppingTooltip1/2, ItemRefTooltip, AtlasLoot,
WorldMapTooltip, etc. No addon Lua and no per-line AddLine hook.
SetPoint stores offsets in internal coordinates, not pixels, so the
anchor offsets are scaled the way Script_SetPoint does
(pixel * [0x00832A44] / ([0x00832A4C] * 1024)) — without that a -2px gap
rendered ~3950px off.
Adds the tooltip line-pool, FontString-creation, and UI-scale offsets to
Offsets.h; documents the behavior in the README and marks TODO #97 done.
Cross-checked against Turtle's server source (patch_1172): ItemModType
is unchanged vanilla (0-7) and ApplyItemEquipSpell applies only
ON_EQUIP, confirming special stats are all equip-spell auras.
- Value now uses the engine's CalculateSimpleValue = EffectBasePoints +
EffectBaseDice (new OFF_SPELL_RECORD_EFFECT_BASE_DICE = 0x10C) instead
of a hardcoded +1; equal for the fixed-die stat auras but exact.
- Decode the additional stat auras Turtle puts on custom gear: dodge %
(49), parry % (47), block chance % (51), spell crit % (57/71), and
flat health (34) / mana (35) folded into the existing keys. Adds shield
block value from the record's m_block. New keys ITEM_MOD_DODGE_RATING,
ITEM_MOD_PARRY_RATING, ITEM_MOD_BLOCK_RATING, ITEM_MOD_CRIT_SPELL_RATING,
ITEM_MOD_BLOCK_VALUE (percent stats keep vanilla-native values).
Documents the expanded key set in docs/API.md.
Weapon DPS (avg damage across the slots / swing time) is emitted under
ITEM_MOD_DAMAGE_PER_SECOND_SHORT, matching 3.3.5's GetItemStats.
Vanilla stores the "special" gear stats (crit, attack power, spell
power/healing, hit, mp5, defense) as on-equip spells rather than stat
slots, so decode each item's SpellTrigger == ON_EQUIP spells - and the
random-suffix equip enchants - through the SPELL_AURA_* effects and fold
them into the modern GetItemStats keys. Percent stats (crit/hit/defense)
are reported as the raw vanilla percentage under the *_RATING key, since
vanilla has no (level-dependent) rating conversion; e.g. Krol Blade's
+1% crit is CRIT_MELEE_RATING = 1, not the 13 rating TBC/Era would show.
Generic attack power subsumes its paired ranged-AP aura; rarer auras
with no clean key (percent AP, per-weapon-line skill, procs) are skipped.
Documents the full key set and its sources in docs/API.md.
Read an item's stats out of the cached ItemStats_C record and return
them keyed by the FrameXML global-string names WoW's GetItemStats uses
(ITEM_MOD_*_SHORT, RESISTANCE0..6_NAME). GetItemStatDelta returns the
per-stat difference (item2 - item1).
Random suffixes are resolved from the link's third
item:id:enchant:SUFFIX:unique field through ItemRandomProperties.dbc ->
SpellItemEnchantment.dbc: type-3 (equip-spell) enchants defer to the
spell's SPELL_AURA_MOD_STAT / MOD_RESISTANCE effect (value =
EffectBasePoints + 1), type-4 add a direct resistance. Reported set
stays consistent with the base record - base attributes, armor, and the
six resistance schools - so a plain item and its suffixed variant report
the same categories.
Adds OFF_SPELL_RECORD_EFFECT_BASE_POINTS, VAR_ITEMRANDOMPROP_RECORDS/
COUNT, and OFF_ITEMRANDOMPROP_ENCHANT to Offsets.h.
Returns whether an item with the given "0x..." GUID is currently carried
by the player. Parses the GUID, then scans the 19 equipment slots plus
bags 0..4, comparing each item's instance GUID (+0x08 -> +0x00, the same
read GetItemGUID uses). Reuses the existing ResolveEquipmentSlot /
ResolveBag / GetBagSlotCount iteration.
Scope matches retail: equipment slots count (verified in-game —
IsItemGUIDInInventory(GetItemGUID({equipmentSlotIndex=1})) is true), bags
count, bank does not (GetItemCount(itemID, true) covers the bank). Returns
false for a malformed / zero GUID.
Factored the CGItem GUID read into a shared ReadItemGuid helper (also used
by GetItemGUID). Documented in docs/API.md and the README Item row.
Registers Enum.InventoryType (equip-type enum) as a hardcoded mirror of
Blizzard's 0..34 spec, via the same RegisterIntegerEnum path as
Enum.PowerType / Enum.AddOnSecurityStatus. Values 0..28 are the ones
vanilla items report; 29..34 (profession / equippable-spell types) are
post-vanilla, included so backport code referencing those keys resolves.
The field names (IndexHeadType, Index2HweaponType, …) aren't present in
engine data and don't derive cleanly from the INVTYPE key strings, so a
static mirror is the correct approach — it's a fixed API contract, not
engine-driven. For the live set of slot keys the engine actually has, use
C_Item.GetItemInventorySlotKey.
Documented in docs/API.md and the README enum row.
Three item-taxonomy lookups backported to the C_Item namespace, reusing
the ItemSubClass.dbc scan and INVTYPE key table already in the tree.
- GetItemSubClassInfo(classID, subClassID) -> subClassName,
subClassUsesInvType. Name is the localized ItemSubClass.dbc verbose
form (short-name fallback). subClassUsesInvType comes from the flags
field at record +0x10, bit 0x200 — set on exactly the armor material
types (Misc/Cloth/Leather/Mail/Plate), clear on weapons/shields/etc.
Bit derived by dumping ItemSubClass.dbc and verified against the 1.15
client (Cloth true, One-Handed Swords false, Shields false).
- GetItemInventorySlotKey(inventorySlot) -> the equipLoc key
("INVTYPE_HEAD"), straight from the engine's INVTYPE table.
- GetItemInventorySlotInfo(inventorySlot) -> the localized slot name
("Head"), resolved via the FrameXML global of that key so it's
localized on non-English clients.
All three return nil for out-of-range / missing input. Documented in
docs/API.md and the README Item row.
C_VoiceChat / C_TTSSettings (the SAPI text-to-speech surface) is moving to
its own injected DLL (C:\Git\VanillaTTS) so it can grow a cross-platform
software-synthesis backend (Flite under Wine) without bloating ClassicAPI.
- Delete src/voicechat/Tts.cpp.
- Drop the ole32 link from CMakeLists (it was only for SAPI/COM).
- Strip the VoiceChat section + TOC from docs/API.md and the VoiceChat
row + VOICE_CHAT_TTS_* events from the README.
CVar::Factory stays — it's general-purpose infra now, not TTS-specific
(reworded its header comment to drop the TTS example).
The repo was internally inconsistent: the LICENSE file is GPL v3 but every
src/ header carried the LGPL v3 notice and the README said "LGPL v3 or
later". LGPL was also the wrong fit — this is an injected mod plus a
bundled addon, not a library other code links against, so LGPL's linking
exception buys nothing.
It's also required: src/voicechat/Tts.cpp is ported from awesome_wotlk's
VoiceChat.cpp, which is GPL v3, and a derivative of GPL code cannot be
relicensed under more permissive terms. Dependencies stay compatible —
MinHook (BSD-2) and the nampower-mirrored packet shapes (BSD) are
GPL-compatible, and SAPI is covered by the system-library exception.
Swept "GNU Lesser General Public License" -> "GNU General Public License"
across all 257 src/ headers and changed the README license line to match
the GPL v3 LICENSE file. Comment-only; no build impact.
Backports modern WoW's TTS Lua surface, ported from awesome_wotlk's
3.3.5 VoiceChat.cpp. The SAPI core is client-version-agnostic; only the
three integration seams are ClassicAPI-native:
- namespaces via Game::Lua::RegisterTableFunction
- the five VOICE_CHAT_TTS_* events via Event::Custom (AutoReserve + Fire)
- ttsVoice / ttsSpeed / ttsVolume settings via CVar::Factory, persisted
to Config.wtf and clamped through change callbacks
C_VoiceChat: GetTtsVoices, GetRemoteTtsVoices (stub == local; vanilla has
no voice chat), SpeakText(voiceID, text[, dest, rate, volume]),
StopSpeakingText. C_TTSSettings: Get/SetSpeechRate, Get/SetSpeechVolume,
GetSpeechVoiceID, GetVoiceOptionName, SetVoiceOption[ByName],
SetDefaultSettings, RefreshVoices.
SAPI is created lazily on first use under an STA; SetNotifyCallbackFunction
marshals playback events back to the main thread's message pump (WoW pumps
every frame), so OnSpeechNotify fires PLAYBACK_STARTED/FINISHED on the main
thread — verified in-game. Voice enumeration uses CLSID_SpObjectTokenCategory
directly so the build needs no sphelper.h; GUIDs are instantiated via
<initguid.h>, so only ole32 is linked (no sapi.lib). Degrades gracefully
without SAPI (e.g. Wine): CoCreateInstance fails, SpeakText fires
PLAYBACK_FAILED and returns, voice list comes back empty, no crash.
Vanilla 1.12 exposes CVar reading and Lua-side RegisterCVar(name, default)
but no way to register a cvar with a change-callback or react to value
changes from C. CVar::Factory wraps the engine internals so modules can
publish persisted, validated settings.
- FUN_REGISTER_CVAR (0x0063DB90), __fastcall(name, edx, flags, default,
callback, categoryId, hiddenBool, userData). Dedups by name (idempotent
across /reload) and forces the archive flag bit on, so registered cvars
persist to Config.wtf.
- FUN_SET_CVAR_VALUE (0x0063DF50), __thiscall(cvar, value, a3..a6) — NOT
__fastcall (that routes value through EDX and leaves the callee popping
4 bytes too many off the stack, corrupting ESP). Fires the change
callback.
- Change callback ABI verified at both the set path (0x0063DF7D) and the
re-register path (0x0063DC51): __fastcall(cvar, prevValue, newValue,
userData) returning char — 0 rejects the change, nonzero accepts. Clamp
by setting the clamped value from inside the callback and returning 0.
Offsets + struct fields (value +0x20, flags +0x1C, callback +0xBC,
userData +0xC0) documented with their call-site derivations.
UnitClassBase("player") and UnitRaceBase("player") returned nil until
~PLAYER_ENTERING_WORLD because they always read the class/race byte from
the in-world unit descriptor ([unit+0x110]+0x79/+0x78), which doesn't
exist until the player object spawns.
Vanilla's Script_UnitClass (0x00518350) avoids this: for the literal
"player" token it reads a login-time global (FUN_005abde0 ->
DAT_00c27e81) instead of the descriptor. That global mirrors the local
player's UNIT_FIELD_BYTES_0 race/class/sex and is populated at
character-enter, so UnitClass("player") resolves at addon-load.
Mirror that fast path: the "player" token now reads the login-time byte
globals (race 0x00C27E80, class 0x00C27E81, sex 0x00C27E82); all other
tokens keep the descriptor path, so remote-unit behavior is unchanged.
The "player" test is factored into Unit::Identity::IsPlayerToken, shared
by both modules (and GuidForToken). It routes through the engine's own
SStrCmpI (FUN_SSTR_CMP_I) to match Script_UnitClass's comparator exactly,
rather than the CRT _stricmp.
Vanilla doesn't populate the local player object (and its +0xC0 GUID) until
SMSG_UPDATE_OBJECT in the main loop, after FrameScript_Initialize and the
addon load pass — so UnitGUID("player") / PlayerGuid() returned nil/0 at
addon file-load on first login.
Co-hook the glue enter-world worker (FUN_0046B500, behind Script_EnterWorld)
to snapshot the selected character's GUID at the moment of entry: read from
the selected-character struct (VAR_GLUE_SELECTED_CHAR; GUID at offset 0,
verified against Script_GetCharacterInfo), gated on the "entering world"
state code so the rename/locked bail paths don't capture. PlayerGuid()
returns the live engine value when present and falls back to the captured
GUID otherwise — the same value, just available from addon-load time.
Captured fresh per entry, so a character switch can't serve a stale GUID.
This makes the GUID *value* available early (keying SavedVariables,
comparisons, the aura/cast caster checks); the player *object* still doesn't
exist until +0xC0 populates.
Also simplify GetPlayerGuid() to a passthrough now that UnitGUID("player")
is reliably available early and cheap — the memo saved only a trivial
C-call + snprintf and the GUID is session-stable.
Add FUN_GLUE_ENTER_WORLD / VAR_GLUE_SELECTED_CHAR offsets.
The 1.12 engine drops a unit's UNIT_FIELD_AURA slots when it loses sync with
the unit (rogue stealth, party range fluctuation) even though the auras are
still active server-side, so C_UnitAuras.GetAuraDataByIndex /
GetBuffDataByIndex returned nil for auras that are still up.
Fall back to the Aura::Source cache (caster/timing captured from
SMSG_SPELL_GO at cast time, which survives descriptor clears) when the
descriptor has no slot:
- Aura::Source: entries gain a helpful/harmful kind, recorded from the
descriptor slot at application time (OnAuraAdded/OnAuraStacksChanged);
SpellGo-only entries stay unknown and are excluded. New Enumerate() returns
non-expired entries of a given kind for a unit.
- Aura::Data: a shared BuildTable feeds both the descriptor Push and a new
cache-fed path; PushNthCacheFallback / AppendCacheFallbacks append cache
entries after the descriptor ones, deduped against populated slots.
- Aura::Api: the index path and GetUnitAuras consult the fallback.
To keep the fallback from resurrecting auras that are genuinely gone
(replaced by a higher rank, dispelled, fell off early), co-hook OnAuraRemoved
(FUN_00612320) and evict the cache entry. A rogue stealthing does NOT route
through OnAuraRemoved (object-visibility drop, not per-aura removal), so
eviction and stealth-recovery coexist: real removals evict, visibility drops
don't.
Add FUN_ON_AURA_REMOVED offset.
Fixes#5
The 1.12 engine short-circuits a chained same-spell recast: Spell_C_CastSpell
bails at `spellID == VAR_CURRENT_CAST_SPELL` before setting the global or
firing SPELLCAST_START, so the client never runs its cast path for the 2nd of
a back-to-back same-spell cast (common with nampower spell queueing). And the
player channel field (+0x228) is server-pushed, landing ~1 tick after the
channel starts, so polling it shows the bar ~1s late.
Rework the player UnitCastingInfo / UnitChannelInfo tracking into a hybrid:
- Normal casts: co-hook FUN_CAST_START_SET, stamp g_cast from the spellID
arg before the original (zero latency); cleared by the VAR==0 WorldTick
poll.
- Chained same-spell recasts: stamp g_cast from the SMSG_SPELL_START co-hook
(server castTime), deduped against a fresh client stamp, exempt from the
VAR clear, self-expiring on endMs.
- Channels: stamp g_channel from the same packet (engine-computed duration)
instead of the lagging +0x228 poll; self-expire on endMs.
- Cast and channel are mutually exclusive for the player.
- Co-hook SMSG_SPELL_DELAYED for cast pushback (extends endMs + delayMs).
Add FUN_SPELL_DELAYED and FUN_CAST_START_SET offsets.
Vanilla stores no cast record for other units, so UnitCastingInfo(other)
returned nil and remote channels had no timing. Co-hook the SMSG_SPELL_START
handler (0x006E7640, gated on opcode 0x131) -- the only packet carrying
another unit's cast with a server-authoritative cast time -- and cache it
per caster GUID (cast vs. channel split by AttributesEx channel bits).
- UnitCastingInfo(unit): full timing for any unit within its cast window;
enemy/target/focus/nameplate cast bars now work, not just the player.
- UnitChannelInfo(unit): remote channels get real start/end times when the
channel's begin was observed, validated against the live +0x228 field so a
stale entry never sticks; falls back to the old no-times path otherwise.
- Player path unchanged (self-tracked); the cache skips caster==PlayerGuid.
Factor the CDataStore packet reader out of Aura::Source into a shared
Net::PacketReader header (vtable-correct) used by both packet hooks.
Best-effort: only casts begun while the unit was in range are seen, and an
interrupted remote cast lingers until its computed end (1.12 sends no
per-unit interrupt packet to hook).
SpellGo only sees directly-cast auras, so proc-applied debuffs (Shadow
Weaving etc.) had no expiration. Co-hook two more engine functions that
fire for every aura application:
- CGUnit_C::OnAuraAdded (0x006123F0) - any aura newly occupying a slot,
procs included; stamps now + base duration (no caster).
- CGUnit_C::OnAuraStacksChanged (0x00612450) - re-stamps on a stack-count
change so a climbing stacking debuff (Shadow Weaving 1->5) refreshes.
Both pass fromCast=false to Store, which skips entries SpellGo already owns,
so directly-cast auras keep their talented timing + caster regardless of
hook order. Consolidate the object-GUID read into Unit::Identity::
GuidForObject so the lookup and the new hooks key the cache identically.
Known limitation (documented): a refresh at MAX stacks (5->5) changes no
client-visible aura field, so the server sends no UpdateField and no
callback fires - there's no packet to hook. The countdown runs out and
expirationTime falls back to 0 (unknown) while the aura still shows.
duration came from a base Spell.dbc/SpellDuration.dbc lookup while
expirationTime came from the Aura::Source cache computed with the caster's
duration modifiers. For a talented DoT (e.g. 2/2 Improved Shadow Word: Pain,
18s -> 24s) the two disagreed, so remaining (expirationTime - GetTime())
exceeded duration -- impossible for a single aura.
Cache the applied durationMs alongside expirationMs in Aura::Source and have
Push prefer it for the duration field, keeping the two consistent
(remaining <= duration). Falls back to the base DBC value on a cache miss.
With the Aura::Source caster cache in place, the PLAYER filter token is now
derivable: restrict results to auras whose cached caster GUID matches the
local player (Unit::Identity::PlayerGuid). Wired into every getter that
takes a filter — GetAuraDataByIndex, GetUnitAuraBySpellID,
GetAuraDataBySpellName, and GetUnitAuras — via a playerOnly flag threaded
into the slot finders and the new Data::IsPlayerCast predicate.
Auras whose cast we didn't observe (predate login / no SMSG_SPELL_GO) count
as not-player-cast and are excluded, matching the best-effort caster model.
Also fix GetUnitAuras range selection so a bare "PLAYER" (no HELPFUL/HARMFUL)
returns both ranges restricted to player casts, instead of defaulting to
helpful-only.
RAID / CANCELABLE / nameplate filter tokens remain accepted-but-no-op.
Populate AuraData sourceUnit/sourceGUID and non-player expirationTime by
co-hooking SMSG_SPELL_GO (FUN_SPELL_GO) and caching (targetGuid, spellId)
-> caster + expiration. The vanilla unit aura array stores only spell IDs;
the SpellGo packet is the only place the client sees an aura's caster and a
server-authoritative duration. We parse the engine packet ourselves, so
this is self-sufficient (no nampower dependency) -- when nampower is loaded
we simply co-hook the same site.
sourceGUID is a ClassicAPI extension (raw "0x..." string), set whenever a
caster is known including when sourceUnit is nil; stable across the session,
unlike the volatile nameplate token, and usable as a unit token under
SuperWoW.
Also consolidate the local-player-GUID read into Unit::Identity::PlayerGuid()
(was duplicated inline in four files) and expose Unit::Identity GuidForToken
/ TokenFromGUID for reuse.
GameObject analog of C_CreatureInfo, reading the gameobject cache
(gameobjectcache.wdb, SMSG_GAMEOBJECT_QUERY_RESPONSE):
- GetGameObjectInfoByID -> { gameObjectID, name, type, displayID } | nil
(synchronous peek)
- RequestLoadGameObjectByID -> bool, fires GAMEOBJECT_DATA_LOAD_RESULT
when an uncached GO arrives.
The GO cache is a sibling class of the creature cache: same hash-map
shape but a different vtable and a larger entry (loaded flag at
entry+0x98 vs +0x50), so it has its own _GetRecord (FUN_00558560) and
response parser (FUN_005588E0). Generalized Cache::QueryLoad to take a
per-cache _GetRecord via Register and to hook BOTH parsers (creature +
gameobject), dispatching completion by cache instance. Renamed the
creature offsets FUN_CACHE_* -> FUN_CREATURE_* now that "generic" was a
misnomer.
GO field offsets: name@+0x08 verified (FUN_0062dc10 reads [block+0x08]),
type@+0x00 / displayID@+0x04 from the wire/WDB order. Verified in-game:
Mesa Elevator 47297 type 11 displayID 360, Windrunner 176250 type 15
displayID 7087, and RequestLoadGameObjectByID populated an uncached GO
("Lighting Impact Large", type 5). Creature path re-verified after the
refactor (GetCreatureInfoByID + the load event still work).
RequestLoadCreatureByID asynchronously fetches an uncached creature:
issues SMSG_CREATURE_QUERY and fires CREATURE_DATA_LOAD_RESULT(id,
success) when the response lands (or on timeout); fires success
synchronously on a cache hit. Mirrors C_Item.RequestLoadItemData.
New Cache::QueryLoad dispatcher generalizes Item::Data's
pending/timeout/event machinery so multiple query caches share one
response-handler hook. MinHook allows only one hook per target, and
the creature/gameobject (and other non-item/quest) caches all route
their SMSG_*_QUERY_RESPONSE through the generic parser FUN_00556E20 —
so we hook it once and dispatch by cache instance to the right pending
set + completion event. Item/quest use their own parsers, so no
collision. The gameobject cache will register on this same dispatcher.
Verified in-game: RequestLoadCreatureByID(10184) on an uncached Onyxia
returned true, fired CREATURE_DATA_LOAD_RESULT, and populated the cache
(type 2 Dragonkin, rank 3 world boss, displayID 8570).
Reads the client-side creature cache (creaturecache.wdb, fed by
SMSG_CREATURE_QUERY_RESPONSE) for an NPC by ID and returns
{ creatureID, name, subName, type, family, rank, displayID } — or nil
if not cached. A synchronous peek (no network query): answers for any
cached creatureID, unlike UnitCreatureType/UnitClassification which
need a live unit token.
Cache machinery is the same generic query-cache class as Item/Quest:
instance 0x00C0E354, generic _GetRecord FUN_00556AA0 (callback=0 peeks).
Field offsets verified against the binary — rank read by the engine's
classification helper FUN_00605620 as [block+0x20], anchoring the
wire/WDB field order (Name[4], SubName, typeFlags, type, family, rank,
…) — and cross-checked in-game against real creaturecache.wdb rows
(Misthoof Stag 61332 type 1; Nordrassil Nymph 61582 type 7 rank 1;
Greathorn Hunter 61337 family 26). All match.
RequestLoadCreatureByID (async fetch for uncached creatures) + the
gameobject cache follow.
Companion to ExportInterfaceFiles: dumps every .dbc table the client
loads to DBFilesClient\ next to the exe. Reuses the same MPQ
enumeration + file-read pipeline (factored into a shared WriteFiles),
and unions two sources for a complete set:
1. MPQ (listfile) under DBFilesClient\ — catches archive files with
no loader (e.g. wowerror_strings.dbc).
2. A .text scan for the DBC path-getter pattern
(mov eax, &"DBFilesClient\X.dbc"; ret) — the authoritative list of
what the build loads. The master DBC init (FUN_0053f8b0) is
straight-line with no iterable name table, so the per-DBC getters
are the registry; scanning for them is the technique that built
docs/DBCs.md. Bounded reads, can't fault.
Listfile alone yielded 137; the union yields 155 (+18 cosmetic
anim/sound/lookup DBCs the listfile omits, 0 lost). Results deduped
case-insensitively. Verified in-game: old vs new extract diff is
exactly the 18 expected additions.
UnitCastingInfo / CastingInfo / UnitChannelInfo / ChannelInfo were
registered as globals (mirroring modern WoW), but that clobbers addons
that ship their own vanilla cast-tracking via the
`_G.UnitCastingInfo or <fallback>` idiom — notably ShaguTweaks'
libcast.lua, which scrapes the combat log for remote/enemy casts the
1.12 engine never exposes. With our global present those addons adopt
our player-only version and drop their (superior, for enemies)
fallback, killing enemy cast bars. We can't match the combat-log
scraper in the DLL (the engine carries no remote regular-cast data),
so cede the global names and register under C_Spell instead.
Tradeoff: addons calling the global UnitCastingInfo/UnitChannelInfo
must use the C_Spell.-prefixed names now; enemy-cast consumers got nil
from the global anyway.
Resolves an item-enchantment ID (the enchantID from
C_Item.GetWeaponEnchantInfo, and item permanent-enchant IDs) into its
SpellItemEnchantment.dbc record:
{ enchantID, name, effects = {{type,amount,arg},...}, spellID }
- name: localized display name ("Crusader", "Brilliant Mana Oil").
- effects: each active slot's ITEM_ENCHANTMENT_TYPE / amount / arg.
- spellID: convenience for spell-type enchants (1/3/7) — the proc/
equip/use spellID, e.g. Crusader -> 20007 "Holy Strength", to chain
into C_Spell.GetSpellDescription.
Record layout verified by parsing the on-disk DBC against known
records (Crusader 1900: type 1, arg 20007; Sharpened+3 id 13: type 2,
amount 3; Reinforced Armor+8 id 15: type 4, amount 8): Type[3]@+0x04,
Amount[3]@+0x10, EffectArg[3]@+0x28, Name[8]@+0x34. The engine's
by-ID readers only touch name/flags/condition, so the effect columns
couldn't be pinned from disassembly — the extracted dbc/ dump was the
oracle. Verified in-game across all three example enchants.
Placed in C_Item (not C_Spell): the id originates from
C_Item.GetWeaponEnchantInfo and the concept is an item enchantment.
Source-item is not derivable (no back-reference, no client reverse
index).
The console registrar (FUN_0063F9E0) dedups by name, and we pass the
same string literal so its pointer-equality fast path makes a repeat
call a hard no-op. Verified in-game: after a logout→login cycle
(which re-fires the glue hook), ExportInterfaceFiles stays listed once
in `help debug` and still runs. The static latch was unnecessary.
Console commands are process-global and persist for the session, so a
single registration at glue boot (the earliest hook, fires at launch)
covers both the login console and in-world — confirmed in-game. Drop
the redundant in-game ModuleAutoRegister; it only ever hit the latch.
The AddOns\ skip dropped Blizzard's own MPQ-baked UI addons
(Blizzard_AuctionUI, Blizzard_TalentUI, …) — part of the stock UI
source. Loose on-disk addons are never in an archive listfile, so
nothing surfaces them (verified: both branches of FUN_00401470 walk
MPQ listfiles / an in-memory registered-file list, not the loose disk
dir); the skip was both wrong and unnecessary. Matches 4.3.4
(extension filter only, no path exclusion).
Backport of the 4.3.4 developer-console command: dumps Blizzard's UI
files out of the mounted MPQs to disk (`ExportInterfaceFiles code`
writes .lua/.xml/.toc/.xsd to BlizzardInterfaceCode\; `art` writes
.blp/.tga to BlizzardInterfaceArt\), preserving the Interface\ subtree
and skipping the user's own AddOns\.
- New src/interface/Export.cpp: enumerates each archive's (listfile)
via the MPQ walker FUN_00401470, reads each entry through
FUN_FILE_READ, and writes it out with Win32.
- New generic Game::Console facility (RegisterCommand / Write / Category
enum) wrapping the engine's console-command registrar and output
paths, so future commands register in one line.
- Offsets.h: FUN_MPQ_ENUM_FILES, FUN_CONSOLE_COMMAND_REGISTER,
FUN_CONSOLE_WRITE; corrected FUN_FILE_READ to __stdcall and documented
its archive-handle arg0.
- Docs: README "Console commands" section + docs/API.md "Console" entry.
Holding both mouse buttons to run forward never fired
PLAYER_STARTED_MOVING. Ghidra trace of the input button handler
(FUN_00514840) shows that gesture drives the forward move directly at
the movement layer (FUN_005103e0) and explicitly clears the autorun
bit on the both-button press — so it sets none of the
INPUT_FLAGS_MOVING_ANY bits the mover-detection reads. (The Offsets.h
comment claiming the both-button combo sets autorun was wrong;
corrected.)
Detect it from the bit combo instead: both INPUT_FLAG_MOUSELOOK (RMB)
and INPUT_FLAG_FREE_LOOK (LMB) held means the engine is mouse-steering
the character forward. ORed into `moving`. Composes cleanly with the
TURNING/LOOKING latches (unchanged).