Files
DmorphOcto/CLAUDE.md
T
2026-10-19 09:45:42 +01:00

92 KiB

DMorph for OctoWoW

A WoW 1.12 addon: a morph window (models, races, items, wardrobe) plus DMorph-style chat commands for the OctoWoW client (modified 1.12.1, build 5875). See README.md for features and install.

The live copy is the game's Interface\AddOns\DmorphOcto folder, a git checkout of main (github.com/bdot89/DmorphOcto). Edit there, /reload in game, commit and push.

How it fits together

  • Morphing is done by VanillaHelpers.dll (v1.1.0+, OctoLauncher's vanillaHelpers mod), through its Lua API: SetUnitDisplayID, SetUnitVisibleItemID, UnitDisplayInfo. The addon never touches game memory itself. The original DMorph DLL targets WotLK 3.3.5a and can't run on this client.
  • Weapon glows and real skin/face/hair (2.0.0) come from our own optional DLL, DmorphHelper (helper/, see 2.0.0 below). Without it the addon works as before.
  • DmorphOcto.lua is the core: state, wardrobe, catalogue search and chat commands. Public functions live on the DmorphOcto table and return ok, message; the chat commands and the window both call them.
  • UI.lua is the window and the minimap button.
  • Share.lua (1.8.0 protocol "DMO2", see its header): DMorph users see each other's morphs. Looks are base 36 (" <slot letter+item>..."), sent once they stand still 1 s (at most 3 s late), only to the audience (DMorph users seen in a scan, or who asked about us, in the last 180 s) by the quietest way: group/raid/BG addon message, GUILD addon message if every outside listener is a guild mate, else the hidden password channel "DmorphShare" (1.7's "DmorphOcto" is left). "M " goes out at once on (dis)mount. Questions "Q name:fingerprint" are answered on the way they came, with "H " if unchanged; a channel question someone else already asked is dropped from our queue. Looks are cached per realm in DmorphOctoAccount.peerCache (14 days): a remembered player shows at once and is asked with the fingerprint; no answer in 20 s = forgotten. ChatFrame_OnEvent is wrapped to drop every CHAT_MSG_CHANNEL* event for our channels. At most one message per 0.6 s on the channel (0.2 s hidden), mount first, then look, replies, questions. Receivers morph the sender through VanillaHelpers with any token whose UnitName matches: target, mouseover, party/raid, or the GUID with SuperWoW (UnitExists returns it). VanillaHelpers' source (gitea.com/D-C-M-Inc/VanillaHelpers, src/Morph.cpp): overrides are kept per GUID (g_overrides), any player unit works but CanAssist must pass ("Unit is not friendly."). Remote mount skins only go on through steady tokens (party/raid/GUID), because the override must come off when they dismount. Received display IDs must exist in the catalogue. Settings: DmorphOctoAccount.shareOff / othersOff.
  • Data/Creatures.lua, Data/Items.lua and Data/ClassicTags.lua are generated by tools/build_data.py from the cmangos classic-db 1.12.1 dump. Data/HiddenModels.lua and Data/HiddenItems.lua come from the game's files (see 1.9.0). Data/Models.lua is generated by tools/build_models.ps1 from the client's MPQs: every display ID in CreatureDisplayInfo.dbc that Creatures.lua doesn't name, flagged when it's not in stock 1.12.1 (patch.MPQ). Don't edit them by hand. Both creature files are sorted in Lua's byte order so GetCreatureCatalog merges them in one pass.
  • Saved variables: DmorphOctoDB (per character, the current look) and DmorphOctoAccount (shared: outfits, minimap position, ui = list/grid and model filter).

Data (1.3.0)

  • Data/Models.lua kind column: 0 Turtle creature name, 1 model-file name, 2/3 the same for displays not in stock patch.MPQ. Creatures.lua entries are kind 0. Data/ModelTags.lua: boss / rare per (display, name) from Turtle ranks, mount per display (NPC mount displays + model files matching riding|mount|... but not MountainGiant).
  • Data/Looks.lua: per "race:gender" the CreatureDisplayInfoExtra looks "id,skin,face,hair,hairColour,facial,pieces,outfit;" deduped, fewest clothes first. The creator morphs into these NPC displays: no DLL here can write PLAYER_BYTES (checked VanillaHelpers, ClassicAPI, SuperWoW, nampower, UnitXP strings), and item morphs don't show on NPC displays.
  • Data/ItemsExtra.lua (tools/build_items.ps1): Turtle wearables from tools/data/turtle_items.tsv (ShellLab export) + the game's WDB itemcache, per slot, plus DmorphOcto_ItemPending (Atlas-CFM loot IDs and set pieces with unknown slot). OctoWoW has items beyond Turtle (e.g. 33000+ "Prophecy" tier variants, Atlas-CFM's "Vanilla Plus" ranges). The core looks pending IDs up after login (GetItemInfo, or ClassicAPI C_Item.RequestLoadItemDataByID + ITEM_DATA_LOAD_RESULT), 4 in flight, saved in DmorphOctoAccount.itemInfo.
  • Data/ItemSets.lua: the client's ItemSet.dbc (patch-4), 352 sets with visible pieces.
  • Favourites (1.4.0): DmorphOctoAccount.favourites = { { id = displayID, mount = true|nil } }, sanitized in InitDB. Right-click a tile or models-list row toggles one for the current "Morph: you/mount" target; the Favourites tab wears (SetDisplay / SetMount) or unstars them.
  • 1.5.0, from the user's in-game report: SetUnitMountDisplayID held the mount model even on foot (the override rewrites the server's 0 on dismount), so mount skins are now applied by WatchMount only while riding: mount field > 0 without override = mounted (that moment's new buff textures = the mount buff), buff gone = dismounted, override cleared. db.mount = skin for all mounts, db.mountSkins[real] per mount, db.ridden. ClassicAPI IsMounted() reads the same field, so it can't see dismounts.
  • Races + Creator are one tab. 1.6.0: < > is strict, M.StepLook / M.GetLookChoices only accept looks with the other four features equal (same outfit number preferred), else nothing changes and the status says so. The user twice reported clicks changing other numbers ("random"). Only ~20% of looks have a second choice per feature (measured), so rows show "n of m" and go grey at 1. Looks.lua field 8 = outfit number per race:gender (same gear = same number).
  • 1.6.0 Mounts tab: list 290 px wide (List takes a width), "Your skin" box with a turning preview of MountPick() (ui.mountPick covers a click still at the gate), ticked/gold row, auto-scroll to it, up/down arrows (StepMount) = previous/next model in the filtered list. User asked for these after 1.5.0.
  • Item data strings are kept after parsing (catalogs are rebuilt when looked-up items arrive); creature data strings are dropped.

1.9.0 (user's screenshot of the Races tab, 11 Oct 2026)

  • The user asked to remove models/items that don't show, add every boss and mount, items and races "from all expansions", and make Races stop "working really randomly" (the screenshot showed the red 1.6.0 refusal "No Undead female NPC has another skin colour with this face, ..."). Later-expansion models, races, items and sets aren't in the 1.12 client's MPQs (and the server doesn't know the items): explained, not possible.
  • Races: < > always moves its row to the next value the race's NPCs have (M.StepLook returns k, pos, n, changed). FindLook keeps rows by weight skin 32 > hair style 16 > hair colour 8 > face 4 > facial hair 2 > outfit 1 (each beats all below), then the closest number of clothes. Measured over all 20 race/genders: whatever the weights, only ~4% of clicks can keep every other face/hair row, and the outfit almost always changes, because every NPC look is a unique combination. So the status names changed face/hair rows, not the outfit. Real per-row customisation needs a DLL that overrides PLAYER_BYTES / PLAYER_BYTES_2 (none here can; VanillaHelpers' Morph.cpp only handles display, native display, mount, BYTES_0 and visible items).
  • Data/ClassicTags.lua (build_data.py, cmangos classic-db): boss = Rank 3, ScriptName boss_*, or an elite spawned at most twice with HealthMultiplier >= 4 and either MechanicImmuneMask bit 0x800000 (most dungeon bosses: VanCleef, Rivendare, Pythas...) or only spawned in instance maps with HealthMultiplier >= 5; rare = Rank 2/4 otherwise. Mounts = every spell_template aura 78 (MOUNTED) EffectMiscValue creature's models + creature_template_addon.mount. Merged with ModelTags.lua in GetCreatureCatalog (boss wins). 174 -> 564 bosses.
  • Data/HiddenModels.lua / Data/HiddenItems.lua (shipped empty; written by build_models.ps1 / build_items.ps1, both run by tools/update_from_game.ps1): the core skips those IDs in the model list, item lists, learned items and NPC looks. Models are hidden when the display isn't in CreatureDisplayInfo.dbc (Creatures.lua names from cmangos weren't checked before), the model file is missing, its M2 header isn't "MD20" version 256-257, M2Array vertices (0x44) or views (0x4C) are empty, or its first skin (display field 6, <model dir>\<skin>.blp) is missing. Items are hidden when their display (tools/data/classic_item_displays.tsv, or the item cache) isn't in ItemDisplayInfo.dbc or none of its strings (read column-agnostically with Dbc.TryStr, which only accepts offsets right after a NUL; INV_ icons skipped) names a file under Item\ObjectComponents{Weapon,Shield,Shoulder,Head,Cape,Quiver,Ammo} (helmets with race suffixes _HuM...) or Item\TextureComponents<8 regions><name>_U/_M/_F.blp. Safety valves: a model check flagging > 1/4 of what it looks at is switched off, > 1/4 of named displays hidden = none, per item slot > 1/4 = none; each prints a warning. MpqDbc.cs gained ReadStart (first sector only), TryStr and GameFiles (hash each name once, cached; fast enough for every item).
  • tests/make_game_fixture.py builds a fake game folder with real-format MPQs and DBCs; the cloud session ran both PowerShell scripts on it with PowerShell 7 (see the file's docstring for the expected output).

1.9.1 (user's screenshot of the in-game Octo Shop, 11 Oct 2026)

  • Shop mounts (Crimson / Green Spectral Tiger, spectral horses) were in the model list but not tagged as mounts: the old rule only matched "riding|mount..." in the model file name. build_models.ps1 now tags a mount when (a) it's known: NPC riders (turtle_ranks.tsv), the 1.12.1 mount spells (Data/ClassicTags.lua), or its model file or any skin (display fields 6-8) is named like a mount; (b) its model is a known mount's model (every skin of it); (c) its creature model has M2 attachment point 0, where the rider sits (GameFiles.MountPoint: the attachment lookup table, an M2Array of int16 at 0x10C in version 256-257 headers, read with Mpq.ReadRange without decompressing the whole file). Rule (c) only counts if >= 5 known creature mount models were readable, >= 80% of them have the point, and at most half of all creature models do; else it prints a warning and only (a)/(b) count. It prints example names of the new mounts: check them.
  • Weapon enchant glows (asked for, "every enchant from all expansions"): not possible with VanillaHelpers. The glow is the enchant ID in PLAYER_VISIBLE_ITEM__0 + 1 (permanent) / + 2 (temporary) -> SpellItemEnchantment.dbc -> ItemVisuals.dbc, and VanillaHelpers' Morph.cpp only overrides the item ID field. Needs a DLL (same as real skin/face/hair: PLAYER_BYTES). Only the glows in this 1.12 client's ItemVisuals.dbc exist; later expansions' effect models aren't in its MPQs. Offered to the user, not built.

2.0.0 (asked for after 1.9.1: "a weapon enchant/glow tab", and Races rows that just change the skin)

  • helper/DmorphHelper.c (C11, MinGW-w64 i686, sh helper/build.sh; MinHook vendored in helper/minhook, BSD 2-clause): hooks LoadScriptFunctions (0x490250, chained after VanillaHelpers' MinHook detour) and registers DmorphHelper_Version, _GetItemEnchants(unit, slot), _SetItemEnchants(unit, slot, permanent, temporary), _GetLook(unit) -> skin, face, hairStyle, hairColour, facialHair, _SetLook(unit, ...). Units: "player" or a friendly player (CanAssist), like VanillaHelpers. It writes descriptor fields directly (PLAYER_VISIBLE_ITEM_1_0 0x104 + (slot-1)*0xC + 1/2; PLAYER_BYTES 0xC1, low byte of PLAYER_BYTES_2 0xC2) and rebuilds the model by flipping UNIT_FIELD_DISPLAYID to 15435 and back with UpdateDisplayInfo (0x60ABE0) twice. Every address is from VanillaHelpers' Offsets.h (proven on the Octo exe by VanillaHelpers working there). It starts only when VanillaHelpers.dll is loaded and both 0x42A320 (InvalidFunctionPtrCheck, which VanillaHelpers disables so Lua C functions outside WoW.exe can be called) and 0x490250 start with a jump (VanillaHelpers' hooks), waiting up to 60 s in a thread if VanillaHelpers comes later in dlls.txt. DllMain always returns TRUE. Built DLL committed (31 KB, imports KERNEL32 + msvcrt only). tools/install_helper.ps1 copies it to the game folder and adds it to dlls.txt after VanillaHelpers.dll (backup dlls.txt.dmorph-backup; -Remove); OctoLauncher keeps foreign dlls.txt lines (its dllsTxt.ts).
  • tools/build_options.ps1 (run by update_from_game.ps1) writes Data/Glows.lua (DmorphOcto_GlowData: visualenchantnames;...effect models, one line per ItemVisuals entry an enchant uses whose effect models exist; the lowest enchant ID shows it) and Data/CharOptions.lua (DmorphOcto_CharOptions["race:gender"] = skin, face[skin], hairStyle, hairColour[style] / anyHairColour, facialHair, npc* sets). Layouts from WoWDBDefs for 1.12: SpellItemEnchantment 0 ID, 13-21 name locstring, 22 ItemVisual, 23 flags; ItemVisuals 0 ID 1-5 effects; ItemVisualEffects 0 ID 1 model; CharSections 0 ID, 1 race, 2 sex, 3 section (0 skin, 1 face, 2 facial hair, 3 hair, 4 underwear), 4 variation, 5 colour, 6-8 textures, 9 flags; CharHairGeosets 0 ID 1 race 2 sex 3 variation; CharacterFacialHairStyles has no ID (0 race 1 sex 2 variation). The tool still checks columns against what they hold: text columns are found by how many different strings they hold (a small number like 1 also points at a string start), the visual column is the first after the last text column whose values are all ItemVisuals IDs; CharSections' skin rows must have variation 0 (else variation and colour are swapped). Flag 1 = offered at character creation only if most skin-colour-0 rows have it (TrinityCore's SECTION_FLAG_PLAYER); else no NPC marks. Rows whose textures aren't in the MPQs are left out (1/4 safety valve). Both ship empty.
  • Core (Glows and your own look section): db.glows[slot] (16/17/18; enchant ID, 0 = none, nil = own), db.looks["race:gender"] (five values per body), db.glowReal/glowWritten, db.lookReal/lookWritten. WatchHelper runs every second (and right after a change): what it reads that differs from what it last wrote is the server's real value; it writes the picked values (glow = permanent enchant, temporary 0 so an oil doesn't hide it) or, once, the real ones back. It holds off 0.5 s after any body change (ApplyBody) so a look isn't written into a model the game is still building. Looks are per body (your race, or the race morph); picking clamps every row to what that race has, row by row (a face is per skin colour, a hair colour per hair style) and the status names rows that had to change. Outfits store glows and the body's look; DescribeOutfit says "own face", "glows".
  • UI: an 8th tab "Glows" (tabs now 56 px, gap 3): slot buttons, search, list, No glow / My own glow. Races: with the helper and options for the race, rows 1-5 call M.StepOwnLook (from an NPC look they first go back to your model with that NPC's values), row 6 is "NPC look", dice = M.RandomOwnLook, My own look = M.ResetOwnLook; otherwise the 1.9.0 NPC-look behaviour. .glow/.enchant <slot> <id|0|own>; /dmorph status reports the helper.
  • Tests: a fake DmorphHelper in run_tests.lua (survives /reload like the DLL; data = what make_game_fixture.py's game gives build_options.ps1). Mutation-checked: dropping the restore, the clamp, the real-value capture, temp 0, the hold, the NPC mark or the reset each fail a check.
  • Not done: other DMorph users don't see glows or picked faces (Share.lua protocol unchanged).
  • 2.0.1, from the user's first run of update_from_game.ps1 on 2.0.0: both tables came out empty ("couldn't find the ItemVisual field", "CharSections.dbc ... doesn't look as expected"). The fixture reproduced it with two realistic quirks the first detection was too strict for: an enchant pointing at a visual ItemVisuals.dbc doesn't have (it required all of them), and hair rows sharing a texture (it required half the rows to have different texture names). Now: the WoWDBDefs 1.12 layouts are tried first and validated loosely (visual field: >= 4 in 5 non-zero values are ItemVisuals IDs, >= 3 different; CharSections: >= 9 in 10 rows read right, >= 3 in 10 have a texture, >= 10 different texture names, skin rows variation 0), then the later layout, then detection. If a table still can't be read, the tool prints "report ..." lines (per field: range, different values, texts, ItemVisuals hits; the first two rows) for the user to send.
  • 2.0.1 mounts: on the user's client the rider point rule switched itself off (attachment 0 on 166 of 169 known mount models but also 398 of 708 creature models: it is the shield point on humanoids). New seed in build_models.ps1: Spell.dbc mount spells (aura 78 in fields 91-93, English name 120; layout checked on 27 known 1.12 mount spell IDs, >= 8 present and >= 60% mounting) -> displays with that name (Creatures.lua or Models.lua, "(skin)" part dropped, "Summon " dropped, names shared by > 12 displays skipped) whose creature model has the rider point. These then spread to every display of the same model like the other known mounts. The Mounts line reports "N mount spells in the game, M more displays by their names".
  • 2.0.1 on the user's PC (11 Oct 2026): 19 glows from 108 enchants, options for all 20 race/genders, 484 mount spells -> 17 more mounts (869 in all; the spectral tigers and horses show under Mount models). DmorphHelper v1 loads ("/dmorph status"), with Crusader set on the main hand; whether the glow and the looks actually show wasn't reported yet. But 738 CharSections rows were dropped for a missing texture, just under the 1/4 valve.
  • 2.0.2: Mpq.Hs now reads "/" as "" like Storm does (names written with forward slashes were "missing" to every tool). build_options.ps1 prints "N of M looks left out (texture missing, e.g. ...)" with three example names, to see what the rest are.
  • 2.0.2 on the user's PC: glows work in game (user: "glows are working"); 738 of 6309 CharSections rows still dropped, the examples all Character\Orc\FacialUpperHair02_0x.blp: the table lists extra layers that were never made, and the game skips them. 2.0.3: a row is left out only if none of its textures exist, or it's a skin row without its first (body) texture; the tool reports drops per section and how many rows were kept without an extra layer. Facial hair choices come from CharacterFacialHairStyles anyway, so those drops barely mattered; faces and hair colours could have lost choices.
  • Asked (2.0.2): TBC and later glows. Not possible: their effect models aren't in the 1.12 MPQs, are newer M2 versions (260+), and OctoWoW's patch-1..5 outrank any extra MPQ, so new ItemVisuals rows couldn't be added; Blizzard files from other clients can't be shipped. Offered instead: a second glow in the temporary enchant slot, if the client draws both (unknown; the watcher writes temporary 0 now).
  • 2.1.0 tried it (db.glows2 written into the temporary enchant field, a "Picking: 2nd glow" toggle). User, in game: "wont show 2 glows": 1.12 draws one enchant visual per weapon. Reverted in 2.2.0 (the watcher still writes temporary 0 under a picked glow so an oil's glow doesn't cover it). The same report confirmed the Races rows change the user's own character ("yes it changes my own character").

2.2.0 (same report: "enable zoom in and out on character", "should be able to pick epic cloth head, or rare plate legs", "make the layout neater"; screenshot of the Items tab full of OctoWoW "1000 Test dagger 60 blue" items)

  • Item types: Data/ItemTypes.lua (DmorphOcto_ItemTypeData, "kindid,id,..." with kind = class * 100 + subclass) from tools/build_data.py <dump> --shelllab <oplancelot/ShellLab data/shelllab.db>: the cmangos item_template first, ShellLab's (Turtle WoW) for the rest; every turtle_items.tsv item has one. build_items.ps1 adds the item cache's kinds (WDB record: class at +0, subclass at +4) as DmorphOcto_ItemTypeExtra in ItemsExtra.lua. In game M.GetItemType falls back to GetItemInfo's English type/sub-type (5th/6th returns in 1.12). M.ITEM_TYPES names and orders them (armour by material, then weapons). Item catalogues carry types[i]; sets get the armour material most of their pieces have. M.FilterCatalog(..., byID, kind).
  • Items and Sets tabs: one row of three buttons, Type (only the kinds the slot has, with counts; remembered per slot in Prefs().itemTypes, sets in Prefs().setType) / Quality / Source, replacing the combined filter button. Items list 14 rows from y -90, bottom row at -352; Sets list 14 rows from -60. Menus hold 20 lines. Lists draw a dark track behind their scroll bar.
  • Test items: names with "test" as a word are skipped in build_data.py (148 left Items.lua), build_items.ps1, and at runtime (IsTestItem in ParseItems/LearnedItems, for the shipped ItemsExtra.lua and looked-up items).
  • Zoom: the dressing room's drag layer is DmorphOctoModelDrag: wheel = Zoom (0.3 per notch, -1.5..3), right-drag pans, left-drag turns; + / - / Reset view buttons at the top of the stage. ApplyView sets SetPosition(zoom, panX, panY - zoom * 0.28) (closer also lower, towards the face) on the loading model and again on swap. The mock's SetPosition now records the position. Pan direction (+y = right) and the zoom limits are unverified in game.
  • Layout was checked with a wireframe drawn from the mock's anchors (single-anchor approximation), not on screen.

Previews (ClassicAPI)

  • Model:SetDisplayInfo(id) comes from ClassicAPI.dll (installed here, optional in code: ui.canPreview). Source: brues-code/ClassicAPI src/model/DisplayInfo.cpp. It loads the model async; dressed NPCs (displays with a CreatureDisplayInfoExtra) are dressed through at most 8 jobs at once, so the grid has 8 tiles. NPC weapons never show.
  • A model frame draws white until its textures load (the "white flash"). Every model loads at a frame level below an opaque layer and is raised above it after REVEAL (0.45 s): tiles have a cover frame; the character preview has two PlayerModels behind/in front of the backdrop art and swaps them. Never call SetUnit on the visible preview model.
  • WoW.exe's method names: plain Model has SetFacing/GetFacing, SetCamera, SetLight, SetModelScale, SetPosition, GetModel; SetRotation/SetUnit are PlayerModel-only (Blizzard's FrameXML only rotates PlayerModels with SetRotation). The mock enforces this.

Rules for this code

  • Lua 5.0: no #, no % (use math.mod), no string.match/gmatch (use strfind/string.gfind), no select(), no ... in functions. table.getn counts element by element, so long lists keep their own count. At most 32 upvalues per function.
  • 1.12 UI API: script handlers take no arguments and read the globals this, event, arg1. There's no SetSize (use SetWidth/SetHeight), no SetNumeric, no hooksecurefunc. Lists use FauxScrollFrameTemplate. Templated frames need a name.
  • Regions that share a draw layer have no fixed order in this client (learned in OctoBG): anything that must stack goes on child frames with explicit SetFrameLevel.
  • Never use GameTooltip:SetHyperlink on an item the client may not have cached: it disconnects 1.12 clients.
  • The trig functions' units depend on how the client was built, so avoid them (the minimap button uses plain vectors).
  • Don't name custom widget fields name, kind, text (on frames that get SetText), level: the mock stores its own state there.

Testing

  • powershell -ExecutionPolicy Bypass -File tools\run_tests.ps1 on this PC (no Lua or Python installed): builds tools/LuaHost (.NET 8) and runs tests/run_tests.lua on OBS Studio's lua51.dll (LuaJIT) with Lua 5.0 shims. Elsewhere: lua tests/run_tests.lua with Lua 5.0.
  • LuaJIT accepts #/%/..., so also run the Lua 5.0 checker: ..\OctoBG\tools\LuaCheck.ps1 -Globals -Files <full paths> (pass an array of paths).
  • tests/mock_wow.lua is a strict 1.12 UI mock: it errors on methods, templates, fonts and script names the client doesn't have. Add any new widget method there only after confirming it exists in 1.12 (Blizzard's 1.12 FrameXML via ..\OctoBG\tools\MpqRead.ps1, the method-name strings in WoW.exe, or a 1.12 addon here such as pfUI). ClassicAPI's SetDisplayInfo exists only while Mock.classicAPI is set.
  • In game: /console scriptErrors 1 to see Lua errors, and /reload after edits. /dmorph status shows whether VanillaHelpers is loaded.

Seen in game

  • 1.2.0 (user's screenshots, 10 Oct 2026): all tabs work, race icons incl. Goblin/High Elf are right. The user then reported textures mixing "if you click around too much": 1.3.0 adds the click gate (0.35 s, latest click wins), never reloads a loading preview model (p.loading / p.again), and ClearModel before SetDisplayInfo.

Not verified in game yet (1.3.0)

  • Whether the gate stops the texture mix-ups; the grid tiles' portrait framing; mount buff detection (UnitBuff textures) and that clearing the override dismounts you visually.
  • The new tabs' layout (6 tabs of 74 px), the drop-down menu, the resize grip (SetScale keeping the top-left corner).
  • RegisterEvent("ITEM_DATA_LOAD_RESULT") and the server's answers for unknown item IDs (pcall'd; timeouts retry next session).
  • 1.6.0: the Mounts tab layout (290 px list + Your skin box, scroll-arrow textures as 30 px buttons), the hover card on the list's left.
  • 1.7.0 sharing: whether the server lets the channel messages through at this rate, VanillaHelpers morphs on other players, SuperWoW GUID tokens inside VanillaHelpers (it resolves tokens with the client's GetGUIDFromName), the Sharing button/menu above the footer.
  • 1.7.1, from the user's report ("flash on and off, mounts sometimes don't load"): the mount watcher didn't Notify when the skin went on, so the mount message only went out with a later change (fixed, tested). Flashing is taken to be players rebuilt on re-entering sight without the GUID override (unverified how 1.12 creates objects): Share.lua now scans every 0.2 s, marks anyone not seen in a scan as away (UnitIsVisible false counts) and re-applies everything when they reappear. Sends 0.5 s after a change, mount changes at once, Q every 3 s. .share reports whether GUID tokens worked.
  • 2.0.0: all of DmorphHelper (never run on Windows or in game): that the game starts with it in dlls.txt, that the hook check passes (else /dmorph status says it isn't loaded), that the descriptor offsets and the display flip rebuild the weapon glow and the face/skin/hair, that the server's resends are caught by the 1 s watcher, how the flip looks while mounted or shapeshifted, CharSections flag meaning on 1.12, NPC-only values on a player model, build_options.ps1 on the real MPQs (check the "Glows: N" line), the 8-tab layout.
  • 1.9.1: the rider point rule on the real client (does it pass its own check, and are the example mounts it prints real mounts?).
  • 1.9.0 (ran on the user's PC: 207 of 13,448 displays left out, 25 no model file, 8 empty, 174 missing skin; 142 named displays and 28 items hidden; no check switched itself off): tools/update_from_game.ps1 on the real client (the M2 header offsets, the skin path, the item file folders: the safety valves cover misreads; check its printed counts), the new boss list's quality, how the always-moving Races rows feel.
  • 1.8.0, from the user's report (channel spam, slow updates while changing gear, mounts not always showing): the DMO2 protocol above. Not seen in game yet: the channel password, the ChatFrame_OnEvent filter with the user's chat addons, GUILD addon messages and GuildRoster on OctoWoW, how many messages the server lets through.
  • 11 Oct 2026: the data files written by the user's local update_from_game.ps1 run on 2.2.0 (Models, ModelTags, Looks, ItemsExtra, ItemSets, Hidden*, Glows, CharOptions) are committed, so downloads ship real OctoWoW data instead of empty Glows/CharOptions. Rerun and commit after a client patch.
  • 2.2.1, from the user's screenshot ("You don't know that language" after a race morph to the other faction): VanillaHelpers rewrites UNIT_FIELD_BYTES_0 race for racial displays, so the client's default chat language follows the morph. The SendChatMessage hook now picks a known language (GetLanguageByIndex loop; 1.12 has GetNumLaguages with Blizzard's typo, not GetNumLanguages) for spoken chat types: asked-for if known, else GetDefaultLanguage if known, else Orcish/Common by UnitFactionGroup, else the first known.
  • 2.2.2, user: "the chat channel shouldn't be spammed, only send when in range of another DMorph user". Share.lua: messages only to DMorph users within CheckInteractDistance(token, 4) (~28 yd) or in the group, AUDIENCE 60 s; questions only for target/mouseover in range or group tokens (not targettarget/partyNtarget/GUID); a player who doesn't answer within CONFIRM_WAIT goes into DmorphOctoAccount.noDmorph[realm] for a day (NO_DMORPH_KEEP), also when someone else asked on the channel; QUERY_GAP 10, RECHECK 120, ASK_AGAIN 3600. Looks still go on at any distance.
  • 2.3.0, a friend with the same launcher/VanillaHelpers saw "not loaded": VanillaHelpers registers its morph functions unconditionally (LoadScriptFunctions hook) and sets VANILLAHELPERS_VERSION on FrameScript init, so missing functions = the DLL wasn't injected (game not started via OctoLauncher, or not restarted). Versions < 1.1.0 (10100) set the version but have no morph functions. MissingAPI() now says which case, using other client mods (SUPERWOW_VERSION, C_Item, GetNampowerVersion, UnitXP) to tell "nothing injects" from "only VanillaHelpers missing". M.GetSetupReport() feeds the footer Setup button (green/red, opens DmorphOctoSetup panel, auto-opens when broken) and /dmorph check.
  • 2.3.1, a user who updated expected the DLL to install itself: an addon can't write outside its sandbox, so "Install DmorphHelper.bat" / "Remove DmorphHelper.bat" (CRLF, .gitattributes) run tools\install_helper.ps1, which now finds the game folder by walking up to WoW.exe, copies only when the hash differs, Unblock-Files the DLL, refuses only when a WoW.exe from this folder is visibly running (Windows hides the path of the elevated game; a locked DLL fails the copy instead), and prints one-line errors (trap). Tested on a fake game folder. HELPER_VERSION (core) flags an older DmorphHelper_Version() in the Setup check: bump it with helper/DmorphHelper.c's HELPER_VERSION.
  • 2.4.0, user report: Ashes of Al'ar mount "gone" (it was display 18179, Creature\Phoenix\Phoenix.mdx, named "Phoenix Hatchling" by Turtle, not mount-tagged). build_models.ps1 now has $MOUNT_MODEL (later-expansion rideable models by path: phoenix, winged/celestial horses, Northrend drakes, victorious steed, armoured raptor, lava/fel horses, Antonidas horse, nether rays, taxi wyverns/hippogryphs, ghost gryphon, riding turtle: 66 displays) and $MOUNT_NAMES (Phoenix -> "Ashes of Al'ar (Phoenix)"). Race morphs: VanillaHelpers rewrites BYTES_0 race, so TaxiNodeGetType hides the other faction's nodes and the spellbook re-sorts. M.HoldRace(reason) clears the display override (native race) while TaxiFrame (TAXIMAP_OPENED, redrawn via TaxiFrame_OnEvent) or SpellBookFrame (OnShow/OnHide wrapped at PLAYER_LOGIN) is open, pauses the DmorphHelper look watcher and shows the real look; ApplyBody waits while held. Untested in game: whether TaxiNodeGetType/spellbook re-read the race live. Scale slider asked: no installed DLL can set OBJECT_FIELD_SCALE_X; needs DmorphHelper_SetScale (C) and a rebuild (no compiler on this PC).
  • 2.5.0 size: DmorphHelper v2 (HELPER_VERSION 2 in C and Lua) adds DmorphHelper_GetScale/SetScale(unit, 0.1-10) writing OBJECT_FIELD_SCALE_X (field 4, a float) + the display flip. Built on this PC with winlibs gcc 16.2 i686 msvcrt, unzipped to C:\Users\bdot1\tools\winlibs\mingw32 (downloaded with the user's OK): PATH=/c/Users/bdot1/tools/winlibs/mingw32/bin:$PATH sh helper/build.sh from Git Bash. Core: db.scale/scaleReal/scaleWritten, WatchScale inside WatchHelper, M.SetScale/GetScale, .scale, Reset everything clears it. UI: DmorphOctoSize slider (0.25-3, step 0.05, applied 0.3 s after it stops, right-click resets) on the name plate. Not shared with other DMorph users yet. Untested in game: whether the client draws a changed OBJECT_FIELD_SCALE_X on your own model after the flip, and how mounts and the camera react.
  • 2.5.1, user report: "Gnome still can't speak that language, Goblin speaks Orcish". GetLanguageByIndex/GetDefaultLanguage follow the race VanillaHelpers writes (a Gnome morph lists Common, Gnomish), so 2.2.1's check passed Common. CaptureNative (run by EnsureSession right after clearing the display override, i.e. with the native race) now stores db.languages/db.defaultLanguage and SpeakableLanguage uses those. Goblin -> Orcish is correct for a Horde character. Also: flight-path birds got the mount skin (taxi = mount field > 0, no new buff -> the "any buff" fallback), and after landing the override held the bird ("can't cast spells"). WatchMount now skips UnitOnTaxi, uses the any-buff fallback only without an on-foot snapshot (login/reload mounted), and PLAYER_CONTROL_LOST/GAINED take the skin off (re-applied next tick if still riding).
  • 2.6.0, user reports: (1) race morph still "speaks Orcish" = the client garbles chat (own and faction's) by the race in UNIT_FIELD_BYTES_0 (0x24, low byte), which VanillaHelpers sets to the morph race. DmorphHelper v3 hooks CGUnit_C::UpdateDisplayInfo (0x60ABE0, thiscall(unit), prologue push esi/push edi/mov esi,ecx/call rel32): for the active player, while DmorphHelper_SetRaces(build, own) is on, it writes the build race before the original and the own race after, so the model is built as the morph race but the byte holds your own race (languages, taxi nodes, spellbook). Lua KeepRace (core) switches it on for race morphs of another race, rechecks every 0.1 s (VanillaHelpers rewrites the override on server writes) and DmorphHelper_Rebuild on UNIT_INVENTORY_CHANGED(player); HoldRace is skipped while raceKept ~= 0. (2) Size moved only the camera: the camera reads OBJECT_FIELD_SCALE_X but the model is drawn at unit+0x90, which the per-frame code at 0x614a90 only eases towards the field (start time unit+0xA0, from +0xA4, 2 s) when the server changes it; 0x614640 is the game's instant path (+0x90 = field, +0xA0 = 0). SetScale now writes the field, +0x90 and clears +0xA0. Found by disassembling WoW.exe.original with winlibs objdump (8 places read field 4). Sizes are shared: "s" token after the guid in L and H messages (none = 100; old clients ignore it), "S " when only the size changes; receivers put it on with DmorphHelper_SetScale and restore the size they read first. All of this is untested in game.
  • 2.6.1, user report (screenshots): race morph heads/hair green (missing texture) with 2.6.0. The game reads the race byte for textures after UpdateDisplayInfo returns (deferred), and 48 inline reads of race exist (no getter to hook; unit code reads it via the unit-fields pointer at unit+0x110, race = +0x78, display = +0x1F4). So v3's "own race except while building" broke textures. DmorphHelper v4 drops the UpdateDisplayInfo hook: the byte keeps the morph race (as VanillaHelpers sets it), and the SMSG_MESSAGECHAT (0x96) handler 0x49D560 (found via its registration: mov $0x49d560,%edx; mov $0x96,%ecx; call 0x5AB650 SetMessageHandler; fastcall(param, msgId) + 2 stack args, ret 8) is hooked to hold DmorphHelper_SetChatRace's race while it runs (languages; the user confirmed languages worked with 2.6.0). HoldRace with DmorphHelper_SetRace swaps only the byte (no rebuild) while the flight map/spellbook is open. Also: a portrait's own look is fitted to the race (M.FittedLook/ClampLook in GetLookState and WatchHelper), so values a race lacks don't go green.
  • 2.7.0, user asked for item previews on hover: Items and Sets rows show DmorphOctoItemCard (a DressUpModel, the Dressing Room's model: SetUnit("player") once per window opening, Dress() then TryOn(itemID) per hover, as OctoWoW's DressUpFrame.lua does with item numbers) left of the row. Items the client hasn't cached are asked for with C_Item.RequestLoadItemDataByID and put on when GetItemInfo knows them (retry every 0.25 s while shown); TryOn is never called on unknown items. Untested in game: framing of the card, TryOn on a creature morph (shows nothing useful).
  • 2.8.0, user report: "Bank NPC is hostile, I can't use NPCs on my faction" during a race morph. VanillaHelpers never touches the faction template; the client works out reactions in CGUnit_C::UnitReaction 0x6061E0 (thiscall, ecx = unit, 1 stack arg = other unit, ret 4; 33 callers incl. CanAssist 0x6066F0), which for player vs NPC goes through reputation (forced reactions 0x4d6490, at-war table 0xb73290 via 0x4d61b0/0x4d5600) whose base values come from the race byte (0x4d2cb0-0x4d5430). DmorphHelper v5 (HELPER_VERSION 5 in C and Lua) hooks 0x6061E0: while unit or other is you and an own race is set, your own race is in the byte for the call. DmorphHelper_SetChatRace was renamed DmorphHelper_SetOwnRace (chat + reactions). ReputationFrame is held like the spellbook (HoldWhileShown) so standings use your race. Untested in game: NPC reactions, name colours, the reputation list.
  • 2.8.1, user asked to see which helper version is loaded: M.GetHelperVersion() (running version or nil, and HELPER_VERSION). The Setup button (160 wide) reads "Setup OK Helper vN", yellow "Old helper vN: fix" or "Setup OK No helper"; its tooltip and the Setup panel row say "Running vN (this DMorph comes with vM)"; .dmorph status flags an older one.
  • 2.8.2, user screenshot: Bat Handler (Undercity) red/hostile with 2.8.0 (DLL v5 installed). The reaction hook was the wrong place: player-vs-NPC reactions read standing = server value (0xb7329c) + cached base (0xb73298, 16-byte rows per reputation index at 0xb73290) via 0x4d6370/0x4d63a0; at-war flags come from the server. The base is built once from the race byte (+0x78, class +0x79) by 0x4D5430 (no args, plain ret; called on entering the world from 0x4909d3 and after SMSG_INITIALIZE_FACTIONS at 0x4d5694), so a loading screen or reload during a race morph caches the morph race's bases. DmorphHelper v6 drops the 0x6061E0 hook and hooks 0x4D5430 to build with your own race; DmorphHelper_SetOwnRace rebuilds when the cache was built with another race (g_builtRace) and is tied to the character GUID (g_ownGuid). The addon now sends your own race whenever the session is ready, not only during race morphs. Untested in game.
  • 2.8.3, user: "crashes more and more". Errors today: 10x UnitXP_SP3.dll -> 0x672184 -> 0x69C056 -> write past a heap page in 0x69C6DA..0x69C8A9 (world collision; addons calling UnitXP("inSight"): Spy List.lua (any distance), UpsiRinse, Cursive-Raid); not DMorph. 4x 0x4D38AE read [0x10] = GetSkillLineInfo: SkillRaceClassInfo lookup 0x6DDF90 (fastcall ecx race, edx class, 1 stack arg skill, ret 4; 11 callers, all pass the unit's race byte +0x78) returns NULL for the morph race. DmorphHelper v6 also hooks 0x6DDF90: when the race is your byte's and differs from OwnRace(), it looks up your own race (also covers the spellbook's skill lookups and trainers). Untested in game.
  • 2.8.4, crash without UnitXP_SP3 at 0x69C8A9, stack 0x672184 <- 0x480E42 (0x480DF0: mouse-pointer world pick, ray to farClip) <- 0x481351 <- 0x48184F. 0x69C600/0x69C780/0x69C4E0/0x69C570 rasterize a line into map cells (~4.17 yd) appending to TSGrowableArray 0xC962C0 {alloc, count, data 0xC962C8, chunk} via counter 0xC89F40 with no bounds check; sized once by 0x691B77 to 0x800 (0x4291F0 reserve: thiscall(array, more, roundUp), ret 8). User's Config.wtf farClip "3000" (game max 777) makes the pick ray ~720 cells, up to 4 entries each: overflow at EBX 0xBFE. UnitXP inSight crashes are the same overflow. DmorphHelper v7 (HELPER_VERSION 7) calls 0x4291F0 from LoadScriptFunctions_h to reserve 0x10000 entries (whole map diagonal). HIGHEST_ADDRESS raised to 0xC962D0. Untested in game.
  • 2.9.0, user asked for a weather changer tab. SMSG_WEATHER (0x2F4) is handled by the multi-opcode handler 0x48F690 (registered at 0x48F55A), case at 0x48FA5F: reads uint32 type, float intensity, uint32 sound, uint8; calls CWeather::SetWeather 0x67BAF0 (thiscall on *0xC6326C, args type 0-3, float 0-1, blend = (byte == 0); ret 0xC; ignored while 0xC7B584 "force-weather-type-on" is set) then the weather sound 0x4586B0 (fastcall sound ID, caches at 0xB05FBC). Types: 0 clear, 1 rain, 2 snow, 3 sandstorm. Sounds (SoundEntries.dbc): rain 8533-8535, snow 8536-8538, sandstorm 8556-8558 (light <0.4, medium <0.7, heavy). DmorphHelper v8 hooks both and, only for calls returning into the packet handler (0x48FAAA, 0x48FAB2), records the server's weather and substitutes the picked one; DmorphHelper_SetWeather/ClearWeather/GetWeather. Addon: M.SetWeather/GetWeather/DescribeWeather, db.weather {kind, intensity} per character, re-applied in ApplySaved and on ZONE_CHANGED_NEW_AREA; .weather command; Weather tab (Clear + 3x3 grid, Strength slider, Server's weather); tabs now sized from label:GetStringWidth(). Untested in game.
  • 2.10.0, user: "nothing to change to night; the skies don't change" (rain/snow worked). Clear skies only means no precipitation; storm skies come from Light.dbc storm params blended by the weather intensity (world env struct +0x40 = CWeather +0x14). Time of day: game clock CGameTime at 0xCE8538 (minute int, +4 hour int); 0x6424C0 minutes of day (thiscall, 15 callers), 0x643A70 day progress 0-1 float (thiscall; 2 callers). The world env builder (0x482Bxx, per frame) stores minutes (call returns to 0x482BD6) and progress (0x482BE2); 0x483DD0 (sky/celestial) uses minutes+720 (returns to 0x483DE1). DmorphHelper v9 hooks both getters and answers only those three callers with the picked minutes, so GetGameTime and the clock stay real. DmorphHelper_SetTime(minutes or -1)/GetTime(). Addon: M.SetTime/GetTime/DescribeTime, db.time, .time command, Time of day row (Dawn 6:00, Noon, Dusk 19:30, Midnight) + 15-minute slider on the Weather tab, Real time button. Untested in game.
  • 2.10.1, user asked for more mount models ("deep search", like a big walker robot). Only models in the client's MPQs can show. GameFiles.Attachments(path) (MpqDbc.cs) returns the attachment-lookup bits of an M2. Attachment 0 (rider) is on 182 of 191 known mount models but also on shield-bearing humanoids (398 of 708 creature models), and most mount models also have hand points 1-2, so no automatic rule is clean (rider point without shoulders 5/6 and helmet 11 still lets through gnolls, trolls, murlocs). Hand-picked additions to $MOUNT_MODEL: Undercity bat taxi, gnome/goblin rocket cars, flying machine vehicle, motorcycle, skiff, Tuskarr boat, fel hound mount (its display is skipped: missing skin), nether-gorged greatwyrm, Darkshade drake, Dragonmaw (orc) drake, druid stag, druid cats, pterrordax, mistfox, tigon, unicorn, pet wyvern and hippogryph: 935 -> 982 mount displays; readable names via $MOUNT_NAMES. Big machines without a rider point (shredder, spider tank, fel reaver, harvest golem, silithid tank, arcane golem) left out: the rider would sit at their feet.
  • 2.11.0, model porting (user asked for Wrath/MoP models, accepted the risks): a separate C# tool in C:\Users\bdot1\tools\wowport (not in this repo: it works on Blizzard's files) reads a CASC install (Retail by default, env WOWPORT_PRODUCT picks another), converts modern M2 (MD21 v272-274, with .anim files) to v256 and writes OctoWoW\Data\patch-.mpq (models 6001+, displays 60001+) plus this addon's Data\Ported.lua (gitignored; "idnametags"). DMorph: optional Data\Ported.lua in the toc (the client skips a missing file), KIND_PORTED 4 and a "Ported (later expansions)" model filter, ported mounts tagged mount, M.IsPorted, GetSharedLook never sends ported IDs (others would see nothing). build_items/models/options.ps1 skip the tool's archives so public data never lists ported models. Lessons from the in-game tests: textures over 512 px draw nothing; modern models leave the texture-unit lookup empty (1.12 reads it per batch, needs [0]); each view needs its own arrays (shared arrays broke the client); key bone lookup max 27; OctoWoW's converted Quilin (Creature\Quilin\QuilinMount.m2) is the reference to diff against.
  • 2.12.0, ported weapons: DmorphHelper v10 (HELPER_VERSION 10) hooks DBCache<ItemStats_C>::GetRecord 0x55BA30 (thiscall: cache, itemID, GUID*, callback, userData, askServer; ret 0x14; 190 callers; ItemStats_C display ID at 0x18, 468 bytes as VanillaHelpers' Game.h): DmorphHelper_SetVirtualItem(id, templateID, displayID) makes an item ID the server doesn't have answer with a copy of the template's record (same kind, so held/sheathed the same) showing a ported ItemDisplayInfo (90001+). VanillaHelpers' SetUnitVisibleItemID writes any ID. DataPorted.lua's DmorphOcto_PortedItems "virtualID display kind template name" (virtual IDs 900001+, template = first item of that kind in DataItemTypes.lua); DMorph registers them at session start, asks the server for templates (C_Item.RequestLoadItemDataByID), lists them in the slots their kind fits (PORTED_SLOTS) as epic KIND_PORTED with an items "Ported" filter, never shares them.
  • 2.12.1: the porting tool splits its output over several patch-.mpq (each under 2 GB; DBCs in patch-Z, loaded last; letters OctoWoW doesn't use) and marks each with a DmorphPort.txt file; build_*.ps1 skip archives that have it. DmorphHelper v10 virtual items: up to 32768, open-addressing hash, records allocated on first use (Retail gave 12,862 weapon looks).
  • 2.13.0, ported armour and sets: Data\Ported.lua item lines may end with a slot ("virtualIDdisplaykindtemplatenameslot"); armour (helmets with race/gender models, shoulders, cloaks, body pieces as 1.12 region textures) is listed only in that slot, weapons still by kind (PORTED_SLOTS). DmorphOcto_PortedSets ("namevirtual,items") adds sets numbered 900001+ after the real ones, kind KIND_PORTED, with a "Ported (later expansions)" Sets filter. Mounts are named from Retail's Mount/MountXDisplay; legendaries whose looks OctoWoW already has (Warglaives, Ashbringer...) get virtual items pointing at OctoWoW's own display. Untested in game: helmets on every race, body textures fitted to 1.12's region sizes.
  • 2.13.1: Retail gave 33,397 armour looks + 13,196 weapons, over v10's 32768 virtual items. DmorphHelper v11 (HELPER_VERSION 11 in C and Lua): up to 131072 virtual items, 262144-slot hash with uint32 slots.
  • 2.13.2, user report (screenshot: checkered boxes on the shoulders and in the hand, a helmet floating above the head; Warglaives missing under Legendary; "once you start walking with models they don't stop the walking animation"). Found in WoW.exe and the files: (1) Walking: the v256 playable animation lookup (header 0x2C) is per model: each animation ID -> (animation to play, 0 normal / 1 backwards / 3 frozen), and it must name an animation the model has, else nothing new plays and the model keeps its last one. The porting tool copied the Quilin's table into every model (2741 of 2800 had entries pointing at missing animations, e.g. Walkbackwards, shuffles, jumps); now rebuilt per model like Blizzard's (own animation, else the stock Raptor's choice, else Walk/Run for each other, else Stand). The client's M2 code only tests sequence flag 0x1 (play once) and never reads the alias field. (2) Checkered boxes = Spells\ErrorCube.mdx, loaded by 0x707350 when 0x706a50 can't open the model file. The game opened every patch-?.MPQ (0x403740: all of them, sorted, patch-1 lowest, Z highest) but couldn't read any file from the tool's archives with 33,000-42,000 files (131072-entry hash tables, patch-R/S/T: weapons and shoulders); those with up to 17,000 files (65536 entries, like Blizzard's patch.MPQ) worked. Archives now hold at most 30000 files. (3) 887 helmets and 120 Blizzard 1.12 shoulders had the same paths as OctoWoW's files and replaced them; the tool now leaves out every file OctoWoW has. (4) Retail helmets sit 10-27 cm higher (made for the newer character models): measured per race and gender from those 887 pairs (bounding boxes, medians) and applied to vertices, pivots, attachments and bounds. (5) Shoulders: both sides took the left model (Retail resources hold _l and _r). (6) Ported items carry their Retail quality (item line "...nameslot (0 = by kind)quality"); the Legendary filter also lists quality 6 (the Warglaives of Azzinoth, 18583/18584, are quality 6 in the 1.12 database). Untested in game.
  • Ported packs (12 Oct 2026, cloud session): "Pack Ported.bat" -> tools/pack_ported.ps1 copies every Data\patch-?.MPQ with DmorphPort.txt inside + Data\Ported.lua to Desktop\DmorphPorted (or -Out; only empties a folder that has manifest.txt) with manifest.txt ("bundle12 hex of SHA-256 over the file lines", "dmorphtoc version", "made", then "namesizeSHA-256"), plus install_ported.ps1, MpqDbc.cs and Install/Remove Ported.bat. tools/install_ported.ps1 runs from the pack: finds the game (-GameDir, a folder above the pack, the running WoW.exe, \Games\OctoWoW etc., else asks), refuses an older DMorph than the pack's, checks every file's size and hash before changing anything, refuses a same-named patch without the marker (OctoWoW's own), skips identical ones, removes marked patches the pack doesn't have, checks free space, copies, then runs the addon's install_helper.ps1. -Remove deletes marked patches and Ported.lua. Tested with PowerShell 7 on fake game folders (install, re-install, damaged file, clash, old DMorph, remove, repack). The bundle ID isn't yet the in-game set ID of the ported-sharing work (done in a local session, not pushed when this was written).
  • 2.13.3, user report: guns and thrown weapons said "Loading from the server..." and turned the player white until a relog. A ported item copies its template's ItemStats record (DmorphHelper v10+); the porting tool picked the first item of each kind in ItemTypes.lua as template (guns 1046, thrown 2945), which the server may not have. ApplyItem now uses PortedTemplate(p): the item's own template if GetItemInfo knows it, else any cached item of the same kind (knownTemplate[kind]; re-registered with DmorphHelper_SetVirtualItem), else it refuses ("hasn't loaded an item to copy ... wait a few seconds") and asks the server for the template. The porting tool now uses fixed vendor templates for ranged kinds. Tests: unknown template refused, same-kind fallback (mutation-checked).
  • 2.14.0, user: friends with the same patches and version saw none of each other's ported morphs (GetSharedLook leaves ported parts out; the set-ID sharing begun in a local session was never pushed). Now: M.GetPortedPack() = DmorphOcto_PortedPack if Ported.lua defines it, else "Lxx" (read at load: Ported.lua loads before the core and its strings are dropped once parsed; same file = same ID). M.GetSharedPorted() returns the ported display/items/mount. Share.lua adds a fourth look part "P_" (same base-36 encoding), last in L messages so older DMorph (pattern "^(%S+) (%S+) (%S+) ?(%S*) ?(%S*)") reads only the first three parts and keeps the normal look; ReadLook reads s and P tokens in any order after the guid. look.look and fingerprints include the ported part (cache, H/Q comparisons); OwnFull() for our own. Receivers overlay it only when the pack matches (Overlay), with Display() checks and M.CanShowPortedItem (helper + a known template). A ported change sends the whole look at once (no M message). Tests: sent part, same pack shows, other pack doesn't, after a size token, fingerprint with/without the part (mutation-checked). Untested in game.
  • 2.14.1, user: characters/models turn white "after fast morphing" and the game sometimes crashes. Errors\2026-10-12 21.06.43: ERROR #124 Memory Invalid Block, Texture.cpp line 2149, stack through 0x76DBF0 = the Model:ClearModel script method (0x76DB20; found by matching {name, function} pointer pairs in WoW.exe). Only LoadView (tiles, hover card, mount box) calls ClearModel, before every SetDisplayInfo: clearing a model whose textures are still loading frees them twice. LoadView now changes a view's model at most once per SETTLE (0.6 s); newer IDs wait in view.pending (only the latest) and TickView starts it after. Tests: fast paging loads 8, then only the page you ended on (mutation-checked). Untested in game. The 09:39-12:02 crashes (0x70CD8F) predate the 30000-file archive fix.
  • 2.14.2 (13 Oct 2026): DmorphHelper v12 adds DmorphHelper_SetAmmo(itemID)/GetAmmo(): your own PLAYER_AMMO_ID (field 0x4C9 = UNIT_END 0xBC + 0x40D; only the active player's fields have it). Experiment for the user (Undead priest with a wand): bow/gun/crossbow morphs fire no missile because the missile is the ammo item and the character has none; thrown morphs too. Not yet seen in game; not yet used by the addon. Also found in game: Blizzard's own string-animated bows (Heartstriker 19350, Bow of Taut Sinew 21478, Huhuran's Stinger 21616) pull the string on the first wand shot only, then hold still while shooting continues; ported bows behave the same (not a porting bug). Virtual items (DmorphHelper_SetVirtualItem with a stock template and display) loop shooting normally.
  • 2.14.3, user: logging in during a race morph left own-faction NPCs hostile and no flight paths until a /reload. CaptureNative could read the morph's race as your own (VanillaHelpers' overrides outlive a logout and can still show the morph's model/race at that moment), and DmorphHelper_SetOwnRace / HoldRace then used it. Now db.ownRace/ownGender (per character) are saved whenever a read is clean; a read that equals the saved morph's race (db.race, or a racial db.display), or whose faction differs from UnitFactionGroup while a morph is on, uses the kept race instead (no kept race + wrong faction: Human/Orc by faction) and keeps the earlier languages. RACES entries carry their faction. Tests: login as a Human morph and as an Orc morph keeps a Troll (mutation-checked). Untested in game.
  • 2.15.0, user asked for TBC-MoP weapon glows from the Retail files: the porting tool (wowport Glows.cs, in "all") takes every Retail ItemVisuals entry an enchant below 5330 (WoD's first) uses, skips those whose effect models OctoWoW already has, converts the rest (spells...*.m2 + textures, 21 models, none failed) and adds rows to patch-Z's ItemVisualEffects / ItemVisuals / SpellItemEnchantment (copies of Crusader 1900 without its effects). Visual and enchant IDs are Retail's (Wowhead's) where OctoWoW doesn't use them, else after the highest. Names come from the spells that put the enchant on (db\SpellEffect.csv effects 53/54/92 + SpellName.csv: "Savagery", "Sunfire", "Mighty Agility"...). 33 glows: Mongoose, Executioner, Berserking, Battlemaster, Sunfire, Soulfrost, Spellsurge, Deathfrost, Blade Ward, Blood Draining, Frozen Rune Weapon / Razorice, Avalanche, Hurricane, Heartsong, Power Torrent, Landslide, Windsong, Jade Spirit, Dancing Steel, Colossus, River's Song... Ported.lua's DmorphOcto_PortedGlows (Glows.lua's line format) is appended by M.GetGlows after the game's glows, marked ported (row info "TBC-MoP", search word "ported", tooltip line). Not shared (glows never are). Untested in game: whether 1.12 draws the converted effect models on weapons (particles, scale).
  • 13 Oct, user: 31 of 33 new glows work (old ones too); Crackling Lightning and Jade Spirit drew nothing. Retail uses a constant twinkle scale (8-8, every particle) as the real size of 3 cm particles, and texture cells past the sheet (126 -> 111 on a 4x4 sheet, wrapped); 1.12 does neither (its own models only use twinkle ranges near 1, stock and Retail share those). wowport ParticleFix.cs moves constant twinkle scales into the sizes and cells onto the sheet, for every ported model (in place before packing).
  • 2.15.1, user (screenshot: a purple helmet left on, hanging off the shoulder, while paging through sets; "some sets only 3 items, most should have 6-8"): WearSet first takes off any set piece (catalogue.isPiece, built lazily) in a slot the new set has no piece for, so sets don't mix; items that aren't set pieces stay. Ported sets were short because the porting tool skipped a display whose look equals an earlier one (tier sets reuse textures per difficulty) before linking its items: Armour.cs now links them to that look's virtual item (virtual IDs unchanged). The game's own 2-3 piece sets (ZG, AQ20, Naxx Undead Slaying) really are that small in 1.12.
  • Sets, later the same day: most remaining short sets were Shadowlands+ helmets (files named _hu_m or unnamed) and unnamed shoulders; the porting tool now reads Retail's ComponentModelFileData (race 1-10 = OctoWoW's races, gender, side 0 left / 1 right): 898 of 915 ported sets complete, 36,014 armour looks (+2,618, numbered after the old ones so saved outfits keep their items). Helmets missing a race get their gender's human model (else any) instead of the error cube.
  • 2.15.2, user (screenshot: purple pieces around the head and shoulders with some sets, own robe still showing): a ported item copies its template's record, and the record says where the game draws it. The porting tool's armour templates were the first item per slot in Items.lua (OctoWoW's "63 Green Frost" test set, unknown to the server), so PortedTemplate fell back to any known item of the same material: a cloth shoulder could copy a cloth hat and be drawn on the head. Now armour falls back only to a known item whose equip location fits the slot (SLOT_EQUIP, GetItemInfo's 8th return), weapons still by kind; the tool picks a plain 1.12 item per Retail inventory type (chest 5 and robe 20 apart): Hooded Cowl, Blackrock Pauldrons, Layered Tunic, Well-Stitched Robe, Red Linen Sash, Dwarven Leather Pants, Recruit's Boots/Shirt, Bracers of the People's Militia, Thick Cloth Gloves, Old Blanchy's Blanket, Guild Tabard. Untested in game.
  • 2.15.3, from the user's Sets screenshot: the set tooltip named ported pieces by their template ("Layered Tunic", "Hooded Cowl"): the game answers GetItemInfo for a virtual item with the template's record. M.DescribeItem now names ported items from Ported.lua first. Same session: the user confirmed glows, helmets and sets work with 2.15.2; docs refreshed with their 2.15 screenshots (docs/src/banner.html: ported pack callout with the pack's counts; crops made without chat, other players' names, own name blanked).
  • 2.15.4 / DmorphHelper v13, user report (Khijazi, 14 Oct): while tanking, morphed head, waist, legs, feet and sometimes hands went back to the real items; shoulders, chest and weapons stayed; only removing and re-adding every morph fixed it. VanillaHelpers (gitea D-C-M-Inc/VanillaHelpers, src/Morph.cpp) keeps overrides by hooking CGObject_C::SetBlock (0x6142E0) and rewriting server writes; its SetUnitVisibleItemID -> ApplyField only writes and refreshes when the field differs from the override, so asking again does nothing while it thinks the morph is on. The slots went back some other way (likely the client refreshing a slot from the real item when its durability changes; not confirmed). v13's DmorphHelper_GetItemEnchants returns a third value, the item the slot shows (PLAYER_VISIBLE_ITEM__0); CheckItems re-applies a morphed slot that shows another item, in the 1 s watcher and at once plus 0.25 s after UNIT_INVENTORY_CHANGED(player) / UPDATE_INVENTORY_DURABILITY. Untested in game: if the field keeps the morph while the model shows the real item, this can't see it (next step: DmorphHelper_Rebuild after those events).
  • 2.15.5, user: a friend's ported pack "had gone" after updating DMorph in the morning. Updating by replacing the DmorphOcto folder (the README's way) deleted Data\Ported.lua, which isn't in the download. install_ported.ps1 now puts the list in its own addon, Interface\AddOns\DMorph_Ported (Ported.lua + DMorph_Ported.toc written by the installer: title "DMorph ported pack", Version = the pack's bundle ID, UTF-8 without BOM), removes an old DmorphOcto\Data\Ported.lua (it would load later and win), and -Remove deletes the folder. DmorphOcto.toc has "## OptionalDeps: DMorph_Ported" (load order) and still lists Data\Ported.lua for older installs. pack_ported.ps1 takes the list from DMorph_Ported first. Tested on a fake game folder (install, reinstall, remove). On the user's PC the list was moved to DMorph_Ported (Version "own build"): the porting tool's install step must copy newpatch\Ported.lua there now. Untested in game: OptionalDeps load order, the new folder showing as enabled after a game restart.
  • Same day, the friend's screenshots: he ran Pack Ported.bat (only for the PC that made the models) and the template Install Ported.bat in DmorphOcto\tools (no manifest). The pack's files now live in tools\pack (install_ported.ps1, Install/Remove Ported.bat; pack_ported.ps1 copies them from there, MpqDbc.cs stays in tools); install_ported.ps1 checks for the template case before loading MpqDbc.cs and says to open the downloaded pack folder; pack_ported.ps1's errors and Pack Ported.bat's banner say it's only for the model maker. Tested on the fake game: template message, install, pack on a friend's PC, pack + install from the new pack.
  • Then (friend's screenshot): his game is C:\WOW, not in the old list of usual places, and an empty answer to "Paste the OctoWoW folder" ended in "The path is not of a legal form". install_ported.ps1 now searches every fixed drive two folders deep (skipping Windows, ProgramData, $Recycle.Bin...) plus Desktop, Documents and Downloads, keeps game folders with DmorphOcto in them if any, asks Y/n for one or a number for several, and asks for the folder up to three times (empty: a tip to drag the folder in; quotes trimmed; not a game folder: says why); Ask() returns $null when Read-Host has no input (it throws), which ends the asking with a clear message. -SearchRoots (tests only) replaces the drive list. Tested: one found, several (pick 2), DMorph preferred, empty, wrong then right, quoted, and the real search on this PC (0.5 s, finds C:\Games\OctoWoW). Reaches friends with the next pack (the installer travels in the pack).
  • Ported legs/boots/belts wrong on most sets (user screenshots): Retail geoset values past 1.12's (boots 4-5, foot shapes, knee pads, belt buckles). The porting tool caps them per slot at 1.12's own highest (Armour.GeosetMax, from casc geosetcheck); casc fixgeosets rewrote the installed patch-Z in place (2,881 values, numbers unchanged). Not yet confirmed in game: the next screenshots still showed the user's real robe (pink diamond) on the legs under a set, so something else draws real gear: open questions to the user (does /run DmorphHelper_Rebuild("player") fix it; does it go away with the real robe taken off).
  • 2.15.6, user: "my character runs on the spot when I stop moving" (after 2.15.4's item repairs). Each repair reloads the model (VanillaHelpers' ApplyField -> ReloadUnitDisplay), so a slot the game kept switching back would reload it every second. CheckItems now pauses a slot for 60 s after 5 repairs in 10 s (clock from the helper frame's OnUpdate; prints why), counts repairs per slot for /dmorph status, and /dmorph itemcheck off|on|log turns it off or names each repair. Whether the repairs cause the running is unconfirmed (asked the user to try itemcheck off/log).
  • Black/white squares on ported shoulders/helms/weapons (user screenshots, Baleful Grave-Knight shoulders): their models use texture types 3/4/24 (Retail's 2nd/3rd item materials, ItemDisplayInfoModelMatRes), which 1.12 never fills (OctoWoW's own 4,749 helmet/shoulder models use only types 0 and 2). wowport fixtextypes bakes them as type-0 files into the models (2,883 models, 4,143 textures, into patch-Z; installed 14 Oct); the full build does it after packing.
  • 2.16.0, user ("purple filler" = their own pink robe under 5-piece sets; asked for an option, on by default): WearSet hides your own gear (item 0, which VanillaHelpers shows as nothing) in armour slots the set has no piece for (HIDE_UNDER_SETS: head, shoulder, shirt, chest, waist, legs, feet, wrist, hands, back, tabard), only where your own gear shows (an item picked on its own stays); a slot a set hid counts as a set piece for the next set. DmorphOctoAccount.setsKeepOwnGear (nil = hide), M.Get/SetHideGearUnderSets, "Own gear: hidden/shown" button in the Sets tab (DmorphOctoSetGear, x 316 y -313). DescribeItem(0) "nothing (your own gear hidden)"; outfits don't count 0 as an item. Share.lua accepts item 0 from others (older DMorph ignores it and shows the real gear). Untested in game: item 0 through VanillaHelpers on the player and on others.
  • 2.16.1, users on the Octo Discord (Jurki, 15 Oct): "every time the durability on the items changes the original skin is visible again"; "the only way to get the tmog back is to switch the item or switch between set profiles". So the reverts are visual only (the visible item fields keep the morphs, which is why 2.15.4's CheckItems saw nothing): the game redraws a slot from the real item when its durability changes (CGPlayer_C::UpdateInventoryComponent 0x5DEE30, per VanillaHelpers' own comment the proper per-slot refresh), while switching an item makes VanillaHelpers rebuild the whole model from the fields (ReloadUnitDisplay: display 15435 and back, UpdateDisplayInfo twice = DmorphHelper_Rebuild). RebuildForItems: after UPDATE_INVENTORY_DURABILITY / UNIT_INVENTORY_CHANGED(player), 0.2 s later, if any item morph is on, DmorphHelper_Rebuild("player"); at most once a second (a later change waits, rebuildPending), gear events within 0.5 s after our own rebuild ignored (no loop); off with /dmorph itemcheck off, named with log. Untested in game; the user's "running on the spot" report is still unexplained (rebuilds while moving are a suspect: watch for it).
  • 2.16.2, user: clicking the same set again "toggles around set pieces randomly"; "all sets ONLY equip these items, nothing else, one click". WearSet now works out the target look for every slot first and changes only the slots that differ (each change is a VanillaHelpers model reload; a second click used to clear the set-hidden slots and hide them again). Own gear hidden (default): every armour slot (HIDE_UNDER_SETS) the set has no piece for shows nothing, also items picked on their own; weapons keep theirs. Already wearing it (nothing changes): one DmorphHelper_Rebuild. Shown: set pieces and set-hidden slots cleared, other items stay.
  • 2.16.3, user: "models in the mount list that are not mounts and vice versa... go over every database you need and match IDs to names". Checked against Blizzard's Mount / MountXDisplay tables: wowport mountbase matches each OctoWoW display to Retail displays with the same model and skin files (by path) and wrote tools/data/retail_mounts.tsv (look: a Retail mount shows exactly that, with its name; model: another skin of a Retail mount's model). build_models.ps1 tags both, names a look display after its mount too (a Models.lua row unless it has that name already: "Hogrus, Swine of Good Fortune" for Encrypted06, 190 new rows), and drops $NOT_MOUNT models whatever the other rules say (sabre cubs, hatchlings/pet wyvern and hippogryph, baby hawkstrider, models with the rider built in: the MountedDeathKnight* / MountedKnight / MountedDemonKnight / KodoRider / chess knight / WC2 death knight, the unmounted death knights, druid cat form and the lynxes on it, the C'Thun tentacle, InvisibleStalker; careful: RidingUndeadWarHorse sits in the MountedDeathKnight folder and is a mount). Mount rows in ModelTags.lua now may carry a name (the Retail name or $MOUNT_NAMES): BuildCatalog tags only the catalogue entry with that name, so a pet sharing a mount's display (Phoenix Hatchling on Al'ar's 18179) or every NPC on a mount display isn't listed in Mounts; if no entry has the name, every entry of the display is tagged as before. Result: 940 mount displays (46 removed, 16 added), 265 with a name. The ported list (DMorph_Ported\Ported.lua) got the same check with wowport mountcheck (tag = a Retail mount look, another skin of a Retail mount model, or a file named like a mount with the rider point; Retail mount looks named after the mount): 334 tagged, 10 untagged (loose parts), 139 renamed, numbers unchanged; on the user's PC and in wowport\newpatch, held back from friends until the final pack.
  • 2.17.0 / DmorphHelper v14, user's screenshot: on Archmage's Felscorned Disc the rider sat "should be stood". Retail sets the rider's pose per mount through the mount spell's visual: SpellXSpellVisual -> SpellVisualEvent with StartEvent 7 (aura on) -> kit -> SpellVisualKitEffect type 6 -> SpellVisualAnim LoopAnimID (wowport riderposes; the DB2 reader needed the copy table for these). 120 (StealthStand, a crouched stand) on the mage discs, boards, kites, carpets, quilts, surfboards, Golden Discus, Sky Surfer, tempests and deathwalkers; 97 (SitGround) on Surf Jelly, Pond Nettle, Fathom Dweller, the Grand Expedition Yak; the rest use anims 1.12's characters don't have (1490 drakes and bears, 1776 phoenixes, 484 trikes...) and keep the riding pose. Ported.lua tags them "rider120" / "rider97" (wowport mountcheck; a look's own Retail mount, else the pose every Retail mount of its model shares): 62 ported mounts; no OctoWoW display matches one. DMorph reads the tags at file load (before the list is parsed) and at session start calls DmorphHelper_SetRiderPose(display, anim) for each. The helper patches the two places 1.12 picks the riding pose (91): 0x5FE809 mov eax,0x5B in the unit's animation picker 0x5FE2F0 (unit in ebx, mounted when +0xDC, the mount model, is set) becomes a call to a stub returning RiderAnim(ebx); 0x607B3C push 0; push -1; push 0x5B (right after the rider model is attached to the mount, before the set-animation call 0x7121A0 on the rider's model, unit in esi) becomes call + nop to a stub that pushes the same with the pose. RiderAnim reads UNIT_FIELD_MOUNTDISPLAYID (0x85, a mount skin too) and looks it up; patched once on the game's thread in LoadScriptFunctions_h, only if both places hold the stock bytes (same in WoW.exe and WoW.exe.original); DmorphHelper_GetRiderPose's second value says whether it worked, and /dmorph check says so. Stubs checked by disassembling an unstripped build. Untested in game: the pose itself, other players on a disc skin, the rider's height on the disc.
  • 2.17.0 SEEN IN GAME (user, 16 Oct): standing on Archmage's Felscorned Disc in the crouched pose; "mounts seem fine". Same report: set and item previews stuck on "Loading from the server..." (Wrath of Spellfire, a ported set). The card (UI.lua TryOnAll) waited for GetItemInfo(virtual item), which only answers once the template Ported.lua names is cached, and asked the server for the virtual ID itself (meaningless to it); wearing already let a known item of the same slot stand in (PortedTemplate). TryOnAll now calls M.CanShowPortedItem for a ported piece the game can't describe yet (same stand-in), never asks the server for ported IDs, retries every 0.25 s for 2 s then every second, and after CARD_WAIT (8 s) says "The server didn't send this." / "N pieces didn't load." The test mock's GetItemInfo answers DmorphHelper's virtual items with their template's info (Mock.virtualItems = the helper mock's game.virtual).
  • 2.17.1, user: a friend who updated only saw old models (favourites "Display 604xx": no ported list) and the user only saw his old transmog. Two causes. (1) The friends' pack (Desktop\DmorphPorted, 13 Oct 18:01, bundle 2893DEAD89E1, made with DMorph 2.15.2) has the pre-2.15.5 installer, which puts Ported.lua in DmorphOcto\Data: replacing the DmorphOcto folder to update deletes it; re-running the pack's Install Ported.bat puts it back (patches already there are only checked). (2) The pack ID was the three lists' text sizes (L185873x2669198x55495 for that pack); the mount names and rider tags written into the user's own list on 15-16 Oct changed it, so the user no longer matched anyone. It's now "Pxx" (models, items, sets: P6399x48795x915 for both lists), so name/tag fixes keep it; every PC needs 2.17.1+ to match (older DMorph sends "L..."). DmorphOcto_PortedPack, if a list sets it, still wins.
  • 2.17.2, user: a hunter friend's ported bows, guns, crossbows and thrown weapons had broken shooting animations, a wand sound, and the character "blinking like going invisible then back really quick". Two causes. (1) 2.16.1's RebuildForItems ran after every UNIT_INVENTORY_CHANGED(player), and every arrow, bullet or throw changes the inventory: the model was rebuilt (display 15435 and back) on nearly every shot, which is the blink and restarts the shooting animation (likely also the "running on the spot" report). Now GearState() keeps the item links of slots 1-19 and the ammo (slot 0) and ranged (18) counts, snapshotted at PLAYER_ENTERING_WORLD; an event where only the counts changed is a shot and is ignored; UPDATE_INVENTORY_DURABILITY and real gear changes still rebuild. Tests mock GetInventoryItemLink/Count via Mock.gear (mutation-checked). (2) In the pack: the ported ranged weapons' missile SpellVisuals (65350-65368, ItemDisplayInfo field 10) were copies of the fire wand's 225 with only the cast kit cleared; the precast kit 372 (wand sound 745, character animation 111) played on every shot. wowport fixmissilekits clears kits 1-2 on SpellVisual rows after the stock ones (patch-Z in place, only SpellVisual.dbc changes); Pack2.cs clears both for new builds. Untested in game. Hunters with ammo may see the ammo's arrow and the visual's arrow together.
  • 2.17.3, user's screenshot right after 2.17.2 and the new patch-Z: logging in left the character invisible in the world and white in DMorph's preview (9 items morphed, outfit with ported pieces). The game builds the model before it has every item the morphs show (a ported piece needs its template from the server); re-applying the morphs changes no field, so nothing rebuilt it, and 2.17.2 had removed the stray inventory-event rebuilds that used to fix it by chance. After every PLAYER_ENTERING_WORLD, itemCheck.loginWait now waits until sessionReady and MorphedItemsLoaded() (GetItemInfo knows every db.items entry) or LOGIN_WAIT (12 s), then RebuildForItems() once. Test + 2 mutations. Not yet confirmed in game that this was the cause (a /reload with 2.17.3 should show it).
  • 2.17.4, same report (user: "reload and the command does not work, it's something with the patch"). Not the patch: Upsilon's 9 ported pieces have byte-identical ItemDisplayInfo rows in all three patch-Z versions of 15-16 Oct. The pieces were wrong: the user's outfit "Dope" (and the current db.items) was saved with the 12 Oct list (wowport\Ported.lua: 920417 = Dark Coven Hood), and the 13 Oct build renumbered the virtual items (920417 = Floating Belt now; Dark Coven Hood = 920002, an offset of 415 for those), so the head showed a belt, the shoulders gloves, the chest boots... The game draws a ported piece where its template's record says, and the clash left the character invisible and white in previews. ApplyItem now refuses a ported armour piece outside its Ported.lua slot (db.items cleared, own gear shown, message "is a piece (from an outfit saved with an older ported pack?)"). Old outfits can be mapped to the new numbers by name + slot from the 12 Oct list. Test + mutation.
  • 2.17.5 (17 Oct), user: "character keeps running on the spot or freezing, only just started doing it again"; and the Mounts tab's Ported filter listed all 6,399 ported models, like the Models tab. (1) 1.12 has no UPDATE_INVENTORY_DURABILITY (not in WoW.exe; UPDATE_INVENTORY_ALERTS is) and no durability function (GetInventoryItemDurability is 2.x), so every rebuild came from UNIT_INVENTORY_CHANGED(player), which also fires for bags, loot, food and ammo: the model was rebuilt (at most once a second) whenever anything in the inventory moved, and a rebuild while moving leaves you running on the spot. GearState() now holds the item links of slots 1-19 plus the durability of each morphed slot, read from a hidden GameTooltipTemplate tooltip (DmorphOctoDurabilityTip, pattern from DURABILITY_TEMPLATE); only a change of that rebuilds (counts kept but unused). A real change inside the 0.5 s after our own rebuild sets rebuildPending instead of being lost. The login rebuild (2.17.3) only waits when a morphed item wasn't loaded at PLAYER_ENTERING_WORLD. Tests: the mock has GameTooltip/GameTooltipTemplate (SetInventoryItem fills "Durability x / y" from Mock.gear[slot][3..4]); old UPDATE_INVENTORY_DURABILITY test events replaced; 2 mutations. (2) MOUNT_FILTERS 3 and 4 add tag "mount": "OctoWoW's new mounts", "Ported mounts" (same indexes, so saved choices keep their meaning); "All models" still lists any model as a skin.
  • Same day: hair showed through every ported helmet (user's screenshot): the porting tool zeroed ItemDisplayInfo 12-13 (helmet geoset visibility, male/female). wowport fixhelmhair sets them from the Retail helmets' HelmetGeosetVis (Retail kept 1.12's 40 HelmetGeosetVisData IDs; newer IDs become the 1.12 ID whose Retail masks are closest, hair weighted 4x): 3,870 rows, all matched. Built into wowport\newpatch\patch-Z.hair.mpq (only ItemDisplayInfo.dbc differs, only fields 12-13), not installed: the user holds every patch change until the final pack.
  • 18 Oct pack fixes (wowport only, no addon change; user confirmed sets, glows, ranged weapons, names work): set pieces looked random per click because 34,000 ported body textures (Item\TextureComponents) were DXT3, which 1.12's character painter doesn't read (it pasted stale buffer contents); now BLP2 palette + 8-bit alpha like Blizzard's (fixregions). Glow cards got the display's main texture (squares): voted Retail material or a black see-through one (fixtextypes refix). Ranged: stock shot visuals per type (bow 5, gun 224, crossbow 743, thrown 98) and the $BMD/$WTT/$WTB markers 1.12 draws the bowstring/arrow/swing trail from (fixranged). 6,421 file-named weapons got Retail names (renameweapons); 150 sets got their transmog-set belts/bracers/boots/cloaks (setfill). Drakes missing neck/tail/legs: key bone IDs past the 27-entry table (fixkeybones) and, the bigger cause, one view with 106-bone submeshes and boneCountMax 0 (Blizzard ships views cut to 256/75/53/21 bones and the client picks what the card's palette fits: fixviews). Mounts too big: Retail CreatureModelScale was never written (fixscale: mounts only, never larger, < 0.2 kept).
  • 2.18.0, user: "a Slutmog section, look only at what people use; make it possible to have items invisible (OctoWoW's Summer Collection of Unseen Fashion)". Data/Slutmog.lua (generated by wowport skinshow from the body textures: share of the slot's regions with alpha >= 128, female copy first): chests/shirts covering < 40% of the torso (or < 50% with a bare belly, torso lower < 5%), legs < 35%, feet < 25% (robes measured over the legs); 219 stock + ported items (Chainkinis, tropical halter tops/skirts/sandals, loincloths, harnesses). ITEM_FILTERS[5] "Slutmog" (appended: saved indexes keep their meaning) keeps those IDs plus any item named "...unseen fashion..." (M.IsSlutmogItem). StartLookup also asks for items 100-160 once (the Unseen Fashion tabard is 109, others likely nearby; unverified). M.HideItem(slot) puts item 0 (nothing shows) in an armour slot; the Items tab's "Invisible" button (x 352). Tests: filter keeps only listed legs (mutation-checked), Unseen name counts, HideItem/SetItem 0. Untested in game: the list's quality, the Unseen Fashion lookups.
  • 2.19.0, user (screenshots of 2.18.0): "move the Slutmog option, it should have its own tab"; "a Birthday Suit button at the top next to Reset view: makes your character naked, no items, then build up from there". Slutmog left the Items source menu (ITEM_FILTERS back to 4; a saved 5 falls back to 1 via the existing sanitizer) and is a tab after Items: M.GetSlutmogCatalog() = every Slutmog item of the armour slots (M.SLUTMOG_SLOTS = HIDE_UNDER_SETS' slots), name-sorted, catalog.slots[i] its slot, rebuilt when GetItemsVersion changes; buttons All / Chest / Shirt / Legs / Feet / Unseen narrow it (DmorphOctoSlutmogShow1-6), search, rows show the slot, click wears it there, hover = item card; a Birthday suit button at the bottom too. M.BirthdaySuit(): item 0 in every armour slot (weapons stay). The stage's "-", "+", "Reset view" moved 51 px left for "Birthday suit" (92 wide, DmorphOctoBirthdaySuit). 10 tabs now (widths from the labels). Netherwing drakes: the 0.6 seat move made the rider stand behind the saddle, reverted (0.81; the missing neck was the bone-palette bug). Tests: tab list, slot narrowing, click wears in slot, Unseen note, Birthday suit (UI and core). Untested in game: the 10-tab layout, the stage buttons' fit.
  • 2.19.1 / DmorphHelper v15 (18 Oct), user (race-morphed Undead, screenshot "[BG] [Upsilon]: n AG KAZRETH", OctoBG's "3 AT STABLES" call): your own typed chat scrambled too. Not the sending side (DMorph's SendChatMessage hook picks Orcish correctly; teammates read it): the client decides per received message whether you understand its language. Traced in WoW.exe.original (winlibs objdump): the SMSG_MESSAGECHAT handler 0x49D560 often doesn't show the message itself: 0x49CAE0 queues it (list 0xB4FDC0) until the sender's name is in the name cache (DBCache GetRecord 0x55F080 on 0xC0E228 with callback 0x49CCC0), so v4's "own race while the handler runs" missed it. The decision: 0x49AA61 calls 0x5EC720 (thiscall on your player, language, &skill), which reads race (+0x78) and class (+0x79) from the unit fields (+0x110) and asks the SkillRaceClassInfo table's second index 0x6DE040 (fastcall ecx race, edx class, 1 stack arg skill line, ret 4, 14 callers; exact race+class+skill hash, table at 0xCEAC08/0xCEAC10) for the language's skill; no row with the morph's race = skill 0, and the scrambler 0x49B560 (language, skill; LanguageWords per-language index at 0xB6E5B4 built by 0x4982C0 at startup) only copies plainly at skill >= 300. v15 hooks 0x6DE040 with v6's rule (LookupRace: the morph's race in your byte -> your own race), shared with the 0x6DDF90 hook. Bytes at both places match WoW.exe and WoW.exe.original. Built with winlibs gcc; not installed while the game ran (install_helper.ps1 needs the game closed). Untested in game: chat while race-morphed, and the other 13 callers (likely the same skills/spellbook paths 0x6DDF90 covers).
  • 2.20.0 / DmorphHelper v16 (18 Oct), user: (1) screenshot at Doras (Orgrimmar) during a race morph of the other faction: the flight map shows only the current node. Traced in WoW.exe.original: TaxiNodeGetType (0x4DCCF0 -> 0x4DC4E0) and NumTaxiNodes only read a cached list (0xBB4F10, 32-byte rows, count 0xBB4AA8; +0x1D reachable). The list is built by 0x4DB340 (SMSG_SHOWTAXINODES case, called from 0x48F5C2, i.e. before TAXIMAP_OPENED), which calls 0x4DC010 (no args, plain ret) to work out routes; its step 0x4DBCE0 reads the race byte (+0x110 -> +0x78) -> ChrRaces (0xC0DD3C) +0xC (3 / 5) and skips nodes whose TaxiNodes mounts (+0x38/+0x3C: 2224 / 3574 Horde, 541 / 3837 Alliance) are only the other team's. So 2.4.0's HoldRace("taxi") redraw never helped. v16 hooks 0x4DC010 to run with your own race (same bytes in WoW.exe and WoW.exe.original). (2) "after changing race and set it bugs out", screenshot: own Undead model, "Not morphed", wrong-looking skin, no gear (9 ported pieces morphed), only a full restart fixed it. Taken to be the race byte left on another race: VanillaHelpers (gitea D-C-M-Inc/VanillaHelpers master, src/Morph.cpp ApplyField) stores the byte it finds when a race morph goes on (g_original, only if not stored yet) and writes it back when the morph comes off, never checking it; v15's chat/reputation hooks put the saved race back after the call whatever happened inside (addons' Lua runs in the chat handler). v16: HoldOwnRace/ReleaseOwnRace (chat, reputation, taxi) put it back only if the byte still holds your race and UNIT_FIELD_DISPLAYID is unchanged. Addon: RepairRace in WatchHelper (every second, not while a window holds your race): the byte (DmorphHelper_GetRace) must be DisplayRace(UnitDisplayInfo's current display) or your own race; else SetRace + Rebuild, counted in /dmorph status. Works with v14/v15 too. Gender isn't checked (no getter; VanillaHelpers writes it for racial models). Tests: wrong byte during an Orc morph and after a reset (3 checks, mutation-checked). Untested in game: all of it; the DLL needs the game closed to install (the user was still on v14, "Old helper v14: fix").
  • 2.21.0 (18 Oct), user (screenshots of the Mounts tab, "the preview is zoomed in too much", "all tabs with lists should have this same layout", "look at OctoBG and ShootyEPGP, make it neat like this but more vibrant"; approved a mockup: flat dark panels, purple accent, gold for what you wear; arrows wear previous/next; Models keeps the grid as an option). UI.lua: the look section (C palette, Roboto from media\fonts via the GetFont check, Flat/Hover/FlatButton/EditBox/FlatScroll as in OctoBG's Widgets.lua, seen working in game 17 Oct); window 922x534, PAGE_W 550. Every list tab (Models list view, Mounts, Favourites, Glows, Items, Slutmog, Sets, Wardrobe) is list (LIST_W 330) + preview box (P.NewBox, BOX_W 178, names DmorphOctoBox / BoxModel / BoxTitle / BoxInfo / BoxStatus / Prev / Next / BoxZoomIn / ZoomOut / ZoomReset): what you hover, else what you have on (list.leave); wheel / + / - zoom, drag turns, right-drag moves, Reset; Previous / Next wear the entry above / below yours and scroll to it. The floating hover cards are gone. Creature boxes ("model"): ClassicAPI Model with the portrait camera (SetCamera(0)), framed by FrameView from Data/ModelFrames.lua (tools/build_frames.ps1, run by update_from_game.ps1; GameFiles.Frame in MpqDbc.cs reads the M2 header's bounding box at 0xB4 and the type-0 camera at 0x124: half height seen = distance * tan(fov / 2), taken as a diagonal fov; 13,390 displays from 696 models) and, for ported models, DMorph_Ported\PortedFrames.lua (-Ported; packed and installed with the pack, listed in that addon's toc). Scale = 0.88 * fit * 1.25^zoom, SetPosition(target x, y + pan, target z - scale * middle + pan), re-applied every frame (a loading model forgets them, as the Shenanigans addon notes). Tiles use it too. Item boxes ("dress"): DressUpModel as the old item card (TryOnAll moved). Slot icons: item 0 (hidden) shows the empty slot tinted purple, not a question mark; uncached items are asked for. The weather buttons' kind field (clobbered the mock's widget type) is now weatherKind. Lua 5.0 limits: the new helpers live in table P (the chunk hit 241 locals). tests/layout.lua draws every tab from the mock's anchors (the mock now keeps all points) to tests/layout/*.svg; render with headless Edge to look. Tests: framing (mutation-checked), zoom/reset, Next/Previous wear and scroll (mutation-checked), hidden-slot icon. Untested in game: all of it, above all whether SetModelScale/SetPosition leave the portrait camera in place (if models still show only a head or vanish, that's the first suspect) and the fov guess.
  • 2.21.1-2.21.3 (18 Oct), user screenshots: size slider says "Size in game" + Reset (DmorphOctoSizeReset); the plate's info line is one line ("9 items - mount skin - size 125%"). 2.21.3: grid tiles of ported models blank, cut off or off to one side until zoomed; a big creature morph filled the dressing room. Frame data now measures the vertices (M2Array 0x44, 48-byte records; ported headers had nonsense boxes, a pterrordax 231 units tall) and stores the box middle: "R Tx Ty Tz Cx Cy Cz height width" (old 7-value lines still read). FrameView puts the middle where the camera looks, turned with SetFacing (P.ANGLE: trig units checked once), and for 4 s after a model starts loading (view.fresh) re-sends SetCamera(0) and a scale 0.1% off every other frame (theory: the client skips a value it has and loses it when loading ends; zooming fixed it for the user). Creature morphs with frame data show on the stage in a creature view (DmorphOctoStageModel, P.StageCreature) following the stage's zoom / turn / pan, not on the PlayerModels. Untested in game: all of 2.21.3.
  • 2.21.4 (19 Oct), user: "I shouldn't have to do anything" (re-packing for PortedFrames). The ported frames are only numbers per display ID, so they now ship in DMorph as Data/PortedFrames.lua (build_frames.ps1 -Ported) with the pack's ID (pack = "P8157x48889x915", worked out from Ported.lua's list rows as M.GetPortedPack does); GetModelFrame only uses it when the pack matches. The 2.21.0 pack/install additions were taken out again and DMorph_Ported\PortedFrames.lua removed from this PC. Rerun build_frames.ps1 -Ported and commit after every porting run.
  • 2.21.5 (19 Oct), user's screenshot: after clicking an item, set, glow or outfit the try-on box drew your character white until the next hover. Each click refreshes the window several times and every refresh ran Dress() + TryOn again while the textures loaded. P.BoxShowItems now leaves the box alone when it already shows the same items (box.dressed; cleared when the window closes or the model is set up). The box title is the item's plain name (DescribeItem's "[name] (id)" read oddly), and a ported piece's slot icon is the icon of the item it copies (M.GetPortedIcon). Test + mutation.
  • 2.21.6 (19 Oct), user's screenshots ("last try, it's wasting my time"): (1) the glow box never showed a glow (a DressUpModel can't draw enchants): hovering a glow now puts it on your weapon for a look (M.PreviewGlow, glowPreview in WatchHelper, cleared by SetGlow, slot switch, leaving the tab, closing the window) and reloads the box and stage 0.7 s later. (2) The box went back as soon as the mouse left the row, so it couldn't be reached: rows no longer revert on leave (list.leave only runs on refreshes after a change); list tooltips anchor ANCHOR_LEFT, off the box. (3) Mounts still cut off or invisible (Colossal Umbrahide Mawrat 63568 only a claw at the top right; Grinning Reaver 64648 a bit too big). The data is right (the Mawrat's stand animation bounds = the vertex box, display scales all 1), so the fit's assumptions about how the client applies SetModelScale / SetPosition / SetFacing with the portrait camera are wrong somewhere. /dmorph frametest [display] (P.FrameTest, default 63568) shows A-E x two angles (see its comment) for the user to pick from a screenshot; the fit then follows the winner. OnWindowHide moved out of Build (33 upvalues).
  • 2.21.7 (19 Oct), SEEN IN GAME, /dmorph frametest on 63568 (Colossal Umbrahide Mawrat): A (fit, world-unit move), B (move / scale) and E (no turning of the middle) were pixel-identical, and the quarter-turned row matched the first: with a model camera set this client ignores SetPosition and SetFacing; only SetModelScale shows (weaker than predicted). C (portrait, scale 1) = the head close-up; D (no camera, scale 0.1 / height) drew nothing. So every fit of 2.21.0-2.21.6 was moot. The ported mounts carry a "character info" camera (type 1, lookup 0,1), Blizzard's own whole-model framing: GameFiles.Frame now adds a 10th value (1 = has one; 2,918 of 2,984 ported models, 389 of 679 stock), and views use SetCamera(1) for those, else SetCamera(0) (the portrait, as before 2.21), never moved; zoom = scale (P.ViewCamera, FrameView). Boxes and tiles don't turn while a camera is set (SetFacing ignored). The frame test now shows A camera 1, B camera 0, C/D camera 1 at scale 0.7/1.4, E no camera. Unverified: that SetCamera(1) picks the type-1 camera (index or ID: both 1 for these models).
  • 2.21.8 (19 Oct), user: 2.21.7 "all good apart from I can't drag the model around or zoom in myself, use the same controls as the character window" (screenshot: Rusty Mechanocrawler whole and centred in the Your skin box). Views (boxes, tiles, the stage's creature view) are now PlayerModels when ClassicAPI's SetDisplayInfo exists on one (ui.playerViews, probed at Build; else Models as before): they turn with SetRotation (P.TurnView) and FrameView also sets SetPosition(zoom * 0.3, panX * 3, panY * 3) like the dressing room, on top of the zoom scale. Untested in game: SetDisplayInfo on a PlayerModel, and that a PlayerModel with SetCamera(1) honours SetPosition / SetRotation as the dressing room's does.
  • 2.21.9 (19 Oct), user: 2.21.8 still didn't drag or zoom. An explicit SetCamera seems to lock any model frame's transform (the frame test's A/B/E; the 2.21.8 PlayerModels still had SetCamera(1)). The dressing room's PlayerModel never gets SetCamera and moves/turns fine with its own player-sized view; the Arboon morph filled it because a 14-unit creature was drawn at full size in that view. So PlayerModel views (ui.playerViews) now get no SetCamera at all, are scaled to a player's height (P.PLAYER_SIZE 2.2 / max(height, width * 0.75) from the frame data) and zoom / pan by SetPosition like the dressing room; SetRotation turns them. Model views (no ClassicAPI on PlayerModels) keep 2.21.7's camera choice. SEEN IN GAME (user, 19 Oct): "the preview drags, zooms and moves now". Lesson: never SetCamera on a preview you want to move; ClassicAPI's SetDisplayInfo works on PlayerModels.
  • 12 Oct 2026, seen in game: the user and a friend with the same pack (Install Ported.bat on the friend's PC, game at C:\client) see each other's ported looks (2.14.0 sharing works). The friend's list first showed no ported models until a full game restart; a quick check is /run DEFAULT_CHAT_FRAME:AddMessage(tostring(DmorphOcto.GetPortedPack())) (nil = Ported.lua not loaded; the user's pack is L201088x2612617x53695). The user's Config.wtf has scriptMemory 512000 (Ported.lua is 2.8 MB): suspect a low value if Ported.lua fails to load elsewhere.