Câble le dispatch renderer RIG via deux subscribers auto-registered (split
MOD bus pour AddLayers, FORGE bus pour RenderLivingEvent.Pre) — pas besoin
de wiring explicite dans TiedUpMod.
onAddLayers (MOD bus) : construit la map entityRendererProvider avec
EntityType.PLAYER → TiedUpPlayerRenderer (Phase 2) ; poste
PatchedRenderersEvent.Add pour extensions tierces.
onRenderLiving (FORGE bus, priority HIGH) : filtre strict instanceof
Player || AbstractTiedUpNpc (protège MCA villagers cf. V3-REW-10) ;
vérifie patch.overrideRender() ; dispatche vers PatchedEntityRenderer et
cancel l'event. Try/catch robuste : log WARN une seule fois par UUID sur
exception, fallback vanilla (event non-canceled).
3 tests unitaires (pure-logic, sans MC runtime) : null-safety du filtre
et idempotence du reset. Le dispatch complet sera validé Phase 2.8
runClient smoke test.
Le biped armature étant identity (Phase 2.4 stub), le hook rendra le
player effondré à l'origine dès qu'il s'active — attendu, warn déjà en
place depuis Phase 2.5.
P0-BUG-001 : LOGGER.warn dans TiedUpArmatures.Holder static init quand les
joints biped sont en identity (Phase 2.4 stub). Sans ça, Phase 2.6 renderer
afficherait un mesh effondré sans signal dev. Warn apparaît une seule fois
(class-init lock JVM).
P0-BUG-002 : Lists.newArrayList(renderer.layers) copie défensive dans
PatchedLivingEntityRenderer.renderLayer. Mods tiers (Wildfire, SkinLayers3D,
Cosmetic Armor) peuvent muter renderer.layers runtime via AddLayers event
-> CME garantie sans copie. TiedUp va au-dela d'EF upstream.
P1-RISK-001 : Pose.orElseEmpty(name) utilise Map.getOrDefault qui retourne
un JointTransform.empty() detache non stocke dans la map -> toute mutation
.frontResult est perdue si le joint est absent. En pratique l'Animator peuple
toujours Head via getComposedLayerPose, mais defensif : on recupere l'instance,
mute, puis pose.putJointData(Head, mutatedTransform). Meme bug latent en EF
upstream.
P1-RISK-003 : PlayerItemInHandLayer extends ItemInHandLayer mais le dispatch
patchedLayers.containsKey(layer.getClass()) est strict -> layer player-specifique
passe au travers. Ajout du mapping explicite dans TiedUpPlayerRenderer
constructor avec le meme stub PatchedItemInHandLayer. Preventif Phase 3.
P2-RISK-004 : @SuppressWarnings scope resserre des 2 methodes aux 2 call sites
dispatch raw PatchedLayer qui en ont reellement besoin. Commentaire pointant
vers l'invariant runtime addPatchedLayer.
Tests : 15/15 rig tests GREEN, compile GREEN.
Doc : docs/plans/rig/PHASE0_DEGRADATIONS.md section Phase 2.5 findings ajoutee
(fichier gitignore, changements locaux pour future session).
P0-BUG-001 — Ordre IDs joints Elbow/Hand/Tool ≠ EF
TiedUpArmatures.buildBiped() assignait Arm_R=11, Elbow_R=12, Hand_R=13,
Tool_R=14 (idem gauche 16/17/18/19) alors que VanillaModelTransformer.RIGHT_ARM
encode upperJoint=11, lowerJoint=12, middleJoint=14 — donc Arm_R=11, Hand_R=12,
Elbow_R=14. Résultat : lowerJoint=17 (SimpleTransformer qui attache les
vertices à Hand) pointait en fait vers Elbow_L → bras tordus au rendu.
Fix : réassigner les IDs dans buildBiped() pour matcher le layout EF
(Arm_R=11, Hand_R=12, Tool_R=13, Elbow_R=14 ; symétrique 16-19).
VanillaModelTransformer non touché (source de vérité EF).
P0-BUG-002 — Singleton BIPED non thread-safe
if (BIPED_INSTANCE == null) BIPED_INSTANCE = buildBiped() est un double-init
race. En SP intégré (client + server threads sur la même JVM), deux threads
peuvent entrer simultanément dans le if null et créer deux HumanoidArmature
distincts — pose matrices incohérentes selon les call sites.
Fix : Holder idiome (static inner class). Le class-init lock JVM garantit
(JLS §12.4.1) qu'une seule init, visible à tous les threads, sans
synchronized ni volatile. Zero overhead après init.
P1 — Nouveau TiedUpArmaturesTest (4 tests)
- bipedHas20Joints : BIPED.get().getJointNumber() == 20
- searchBipedJointByNameReturnsNonNull : vérifie les 20 noms EF
- jointIdsMatchEfLayout : verrou P0-BUG-001 (id=12→Hand_R, id=14→Elbow_R,
etc.) — aurait attrapé le bug en review initiale
- bipedSingleton : BIPED.get() == BIPED.get() (verrou P0-BUG-002)
P2 backlog tracé dans docs/plans/rig/PHASE0_DEGRADATIONS.md :
biped collapsed visuellement jusqu'à Phase 2.7, PlayerPatch.yBodyRot sans
lerp, ridingEntity non géré, isFirstPersonHidden nommage ambigu,
ServerPlayerPatch hérite méthodes client-only sans @OnlyIn.
Tests : 15 GREEN (11 bridge pré-existants + 4 nouveaux biped).
Compile clean.
5 classes ajoutées dans rig/patch/ :
- TiedUpCapabilities.java
Holder du Capability<EntityPatch> CAPABILITY_ENTITY (CapabilityToken
auto-register) + helpers getEntityPatch / getPlayerPatch /
getPlayerPatchAsOptional. Simplifié de EF (pas de ITEM/PROJECTILE/SKILL
caps, combat only).
- EntityPatchProvider.java
ICapabilityProvider + Map<EntityType, Function<Entity, Supplier<EntityPatch<?>>>>.
registerEntityPatches() pour commonSetup (EntityType.PLAYER seul Phase 2),
registerEntityPatchesClient() pour clientSetup (dispatch LocalPlayerPatch vs
ClientPlayerPatch<RemotePlayer> vs ServerPlayerPatch). CUSTOM_CAPABILITIES
pour extensions futures. Pas de GlobalMobPatch combat fallback.
IMPORTANT : n'enregistre PAS EntityType.VILLAGER (MCA conflict V3-REW-10).
- TiedUpCapabilityEvents.java
@Mod.EventBusSubscriber sur AttachCapabilitiesEvent<Entity>. Check oldPatch
pour éviter double-attach, construit provider, appelle onConstructed eager
(D-01 pattern EF), addCapability. Priority NORMAL (order d'attachement
ne matière pas, c'est les runtime cross-cap reads qui importent et ceux-là
sont déjà lazy dans onConstructed).
3 stubs PlayerPatch subclasses (placeholders Phase 2.4) :
- ServerPlayerPatch : overrideRender=false, getArmature=null stub, updateMotion no-op
- ClientPlayerPatch<T extends AbstractClientPlayer> : overrideRender=true, @OnlyIn CLIENT
- LocalPlayerPatch extends ClientPlayerPatch<LocalPlayer> : vide pour l'instant
Ces stubs satisfont le compile de EntityPatchProvider.registerEntityPatchesClient().
Le getArmature() null est non-bloquant Phase 2.3 mais devra être fixé Phase 2.4
pour le vrai rendering (lien avec TiedUpRigRegistry.BIPED à créer Phase 2.7).
Compile BUILD SUCCESSFUL + 11 tests bridge GREEN maintenus.
Résout les 3 items remontés par la review globale pré-Phase 2 :
SMELL-001 — TiedUpRigConstants.ANIMATOR_PROVIDER
Le ternaire retournait ServerAnimator::getAnimator dans les 2 branches
alors que ClientAnimator est maintenant forké (présent dans rig/anim/client/).
Switch vers ClientAnimator::getAnimator côté client (pattern lazy method-ref
préserve la non-chargement sur serveur dédié).
DOC-001 — AnimationManager:211
Commentaire ambigu "SkillManager.reloadAllSkillsAnimations() strippé"
clarifié : préciser que l'appel upstream EF venait de yesman.epicfight.skill.*
et que le combat system est hors scope TiedUp.
TEST-001 — GltfToSkinnedMeshTest coverage gaps
Tests précédents utilisaient [1,0,0,0] → drop trivial, renorm no-op.
Ajoute 3 tests :
- convertDropsLowestWeightAndRenormalizes : poids [0.5, 0.3, 0.15, 0.05]
force le drop du plus faible (0.05) + renorm des 3 restants.
- convertHandlesZeroWeightVertex : weights tous-zéro → fallback Root w=1.
- convertFallsBackToRootForUnknownJointName : joint GLB inconnu ("TentacleJoint42")
→ log WARN + fallback Root id=0 sans crash.
11 tests bridge GREEN (5 alias + 6 convert). Compile BUILD SUCCESSFUL.
Cross-check de l'audit Phase 0 contre les sources EF (4 agents) a remonté :
D-08 RÉFUTÉ partiellement :
- EF LivingEntityPatch.getTarget() = getLastHurtMob() — identique à notre stub.
- MAIS MobPatch.getTarget() override avec Mob.getTarget() manquait chez nous.
- Fix : override ajouté, ref commentée à EF MobPatch.java:171-174.
- Sans ça, MoveCoordFunctions.MOB_ATTACK_TARGET_LOOK aurait la mauvaise
sémantique (dernier mob qui m'a frappé vs cible AI courante) → NPC ne
tourne pas vers sa cible pendant attack anim.
D-07 VALIDÉ :
- correctRootJoint zero-out X/Z de la Root en espace monde pour éviter
sliding visuel pendant LinkAnimation vers ActionAnimation.
- Safe Phase 1 (idle/walk = StaticAnimation, pas ActionAnimation).
- Critical Phase 2+ dès qu'une vraie ActionAnimation bondage est jouée.
- Fix : dev assertion LOGGER.warn en IS_DEV_ENV pointant vers
PHASE0_DEGRADATIONS.md D-07. Empêche découverte tardive.
Autres findings post-vérification (traçés en doc gitignorée) :
D-01 getAnimator()=null : fix Phase 2 (pas Phase 1) — field protected EF-style
D-02 sync() stripped : FAUX-POSITIF partiel — BEGINNING_LOCATION non affecté
(IndependentVariableKey non-synced), seul DESTINATION en MP dédié
D-03 InstantiateInvoker throws : swallowed par try/catch, silent no-op
D-04 Patch suppression : doc EXTRACTION.md §3.12/§10 corrigée (Option A)
D-05 reloadAllSkillsAnimations : était déjà dans SkillManager (commentaire OK)
D-06 playAnimationAt : ARCHITECTURE.md §5.5.1 pseudocode (signature fantôme)
→ notes ajoutées pointant vers D-06
D-09 AnimationBegin/EndEvent : listeners EF uniquement skill system interne,
ON_BEGIN/END_EVENTS data-driven continuent de fonctionner
D-10 AT 127 lignes : ~50% utile (GUI TiedUp existant), ne pas fix maintenant
IK stack (S-05 pas dans les docs) : section R12 ajoutée à ARCHITECTURE.md §11.
Compile BUILD SUCCESSFUL maintenu (0 errors).
Drop Commons-Clause and monetization restrictions to enable
incorporating third-party GPLv3 code in upcoming rig system work.
Prior versions (0.1.0–0.5.x) remain under GPL-3.0 WITH Commons-Clause.
From 0.6.0-ALPHA, GPL-3.0-or-later pure.
Architectural debt cleanup on top of the earlier V2 hardening pass.
Minigame:
- LockpickMiniGameState splits the overloaded targetSlot int into a
LockpickTargetKind enum + targetData int. Body-vs-furniture
dispatch is now a simple enum check; the NBT-tag nonce it
previously depended on is gone, along with the AIOOBE risk at
BodyRegionV2.values()[targetSlot].
- PacketLockpickAttempt.handleFurnitureLockpickSuccess takes the
entity and seat id as explicit parameters. Caller pre-validates
both before any side effect, so a corrupted ctx tag can no longer
produce a "Lock picked!" UI with a used lockpick and nothing
unlocked.
Package boundaries:
- client.gltf no longer imports v2.bondage. Render-layer attachment,
DataDrivenItemReloadListener, and GlbValidationReloadListener all
live in v2.client.V2ClientSetup.
- GlbValidationReloadListener moved to v2.bondage.client.diagnostic.
- Reload-listener ordering is preserved via EventPriority (HIGH for
the generic GLB cache clear in GltfClientSetup, LOW for bondage
consumers in V2ClientSetup).
- Removed the unused validateAgainstDefinition stub on GlbValidator.
Extractions from EntityFurniture:
- FurnitureSeatSyncCodec (pipe/semicolon serialization for the
SEAT_ASSIGNMENTS_SYNC entity data field), with 8 unit tests.
- FurnitureClientAnimator (client-only seat-pose kickoff, moved out
of the dual-side entity class).
- EntityFurniture drops ~100 lines with no behavior change.
Interface docs:
- ISeatProvider Javadoc narrowed to reflect that EntityFurniture is
the only implementation; callers that need animation state or
definition reference still downcast.
- FurnitureAuthPredicate.findOccupant uses the interface only.
- AnimationIdBuilder flagged as legacy JSON-era utility (NPC
fallback + MCA mixin).
Artist guide: corrected the "Monster Seat System (Planned)" section
to match the ISeatProvider single-impl reality.
Broad consolidation of the V2 bondage-item, furniture-entity, and
client-side GLTF pipeline.
Parsing and rendering
- Shared GLB parsing helpers consolidated into GlbParserUtils
(accessor reads, weight normalization, joint-index clamping,
coordinate-space conversion, animation parse, primitive loop).
- Grow-on-demand Matrix4f[] scratch pool in GltfSkinningEngine and
GltfLiveBoneReader — removes per-frame joint-matrix allocation
from the render hot path.
- emitVertex helper dedups three parallel loops in GltfMeshRenderer.
- TintColorResolver.resolve has a zero-alloc path when the item
declares no tint channels.
- itemAnimCache bounded to 256 entries (access-order LRU) with
atomic get-or-compute under the map's monitor.
Animation correctness
- First-in-joint-order wins when body and torso both map to the
same PlayerAnimator slot; duplicate writes log a single WARN.
- Multi-item composites honor the FullX / FullHeadX opt-in that
the single-item path already recognized.
- Seat transforms converted to Minecraft model-def space so
asymmetric furniture renders passengers at the correct offset.
- GlbValidator: IBM count / type / presence, JOINTS_0 presence,
animation channel target validation, multi-skin support.
Furniture correctness and anti-exploit
- Seat assignment synced via SynchedEntityData (server is
authoritative; eliminates client-server divergence on multi-seat).
- Force-mount authorization requires same dimension and a free
seat; cross-dimension distance checks rejected.
- Reconnection on login checks for seat takeover before re-mount
and force-loads the target chunk for cross-dimension cases.
- tiedup_furniture_lockpick_ctx carries a session UUID nonce so
stale context can't misroute a body-item lockpick.
- tiedup_locked_furniture survives death without keepInventory
(Forge 1.20.1 does not auto-copy persistent data on respawn).
Lifecycle and memory
- EntityCleanupHandler fans EntityLeaveLevelEvent out to every
per-entity state map on the client.
- DogPoseRenderHandler re-keyed by UUID (stable across dimension
change; entity int ids are recycled).
- PetBedRenderHandler, PlayerArmHideEventHandler, and
HeldItemHideHandler use receiveCanceled + sentinel sets so
Pre-time mutations are restored even when a downstream handler
cancels the render.
Tests
- JUnit harness with 76+ tests across GlbParserUtils, GltfPoseConverter,
FurnitureSeatGeometry, and FurnitureAuthPredicate.
Two fixes from the animation audit:
1. Variant cache key now includes the resolved animation name (e.g., Struggle.2).
Previously, the cache key only used context+parts, so the first random variant
pick was reused forever. Now each variant gets its own cache entry, and a fresh
random pick happens each time the context changes.
2. FullHead check changed from contains("Head") to startsWith("gltf_FullHead")
to prevent false positives on names like FullOverhead or FullAhead.
The resolver now tries FullHead* before Full* at each fallback step.
Example: FullHeadStruggle → FullStruggle → Struggle → FullHeadIdle → FullIdle → Idle
Previously FullHead* names were dead — the resolver never constructed them,
so animations named FullHeadStruggle were unreachable.
FullStruggle, FullWalk etc. animate body+legs but preserve head tracking.
FullHeadStruggle, FullHeadWalk etc. also animate the head.
The 'Head' keyword in the animation name is the opt-in signal.
Previously, any GLB with keyframes on free bones would animate them,
even for standard animations like Idle. This caused accidental bone
hijacking — e.g., handcuffs freezing the player's head because the
artist keyframed all bones in Blender.
Now the Full prefix (FullIdle, FullStruggle, FullWalk) is enforced:
only Full-prefixed animations can animate free bones. Standard
animations (Idle, Struggle, Walk) only animate owned bones.
This aligns the code with the documented convention in ARTIST_GUIDE.md.
- DataDrivenItemParser: downgrade animation_bones absence log from INFO to
DEBUG (was spamming for every item without the optional field)
- GlbValidator: read 12-byte GLB header first and reject files declaring
totalLength > 50 MB before allocating (prevents OOM on malformed GLBs)
- GlbValidator: WEIGHTS_0 check now uses the same mesh selection logic as
GlbParser (prefer "Item", fallback to last non-Player) instead of
blindly checking the first mesh
Strip pipe-delimited armature prefixes (e.g., "MyRig|body" -> "body")
from bone names in parseSeatSkeleton and remapSeatAnimations, matching
the existing stripping already present in GlbParser. Without this,
isKnownBone checks fail when Blender exports prefixed bone names.
