diff --git a/docs/ISSUES.md b/docs/ISSUES.md index 3728c01c..a795d7ac 100644 --- a/docs/ISSUES.md +++ b/docs/ISSUES.md @@ -24,6 +24,16 @@ What does NOT go here: - Every session: scan OPEN issues at start; promote/close anything we touched during the session before ending. - Promoting to a Phase: mark as `DONE (promoted to Phase X)` + commit SHA where the Phase entry landed. +## #352 — Vendor range-watcher cylinder metric: discriminating unit test deferred + +**Status:** OPEN (filed 2026-08-08). The EnforceRange cylinder-gap fix +(acceptance-band self-close, vendor-verify-gate.log evidence) landed with +the existing 17-range/lifecycle tests green but WITHOUT a unit test that +discriminates cylinder-vs-center (needs an IPhysicsObjHost fake — 38 +members — bound via BindObjectTableHostResolver; center>radius while +cylinder<=radius stays open, radii-sabotage closes). Write it next +session; the live gate covered the behavior. + ## #351 — LandblockBuildOriginTests.FarLoad_StripsEnvCellsAndPhysics flake (Debug, load-sensitive) **Status:** OPEN (filed 2026-08-08). Seen three times today in Debug runs diff --git a/docs/architecture/retail-divergence-register.md b/docs/architecture/retail-divergence-register.md index b59e3e86..f82de966 100644 --- a/docs/architecture/retail-divergence-register.md +++ b/docs/architecture/retail-divergence-register.md @@ -178,7 +178,7 @@ AP-94..AP-112 for the confirmed retail-UI completion gaps. | # | Divergence | Where (file:line) | Why it is safe / justified | Risk if assumption breaks | Retail oracle | |---|---|---|---|---|---| -| AP-160 | **Filed 2026-08-07, Slice 5.3 (vendor browse lifecycle). CORRECTED AND EXTENDED 2026-08-07 at the Slice 5.3 review corrections (fixes 4/5).** **Correction (fix 4):** this row's own Retail-oracle citation originally grouped `WorldObject_Use.cs:50,57` under the SAME citation as `Vendor.CheckClose`/`GetCylinderDistance`, which read as if the `wo.UseRadius ?? 0.6f` fallback lived inside the close watcher. It does not: `WorldObject_Use.cs:50,57` is `WorldObject.IsWithinUseRadiusOf`, the APPROACH check ("how close you need to be to open the shop") — a wholly different method from `Vendor.CheckClose`, which reads `UseRadius` directly with no fallback of its own (`UseRadius` is `float?`; a nullable comparison against a null right operand is always `false`, so `CheckClose` never closes at all on an unauthored radius). `EnforceRange`'s own code comment carried the same mis-attribution and, worse, actually APPLIED that mis-borrowed 0.6f as its fallback; it now passes the raw authored `UseRadius` with no fallback of any kind (0 when absent/unauthored, matching retail's own memset-zero `PublicWeenieDesc::_useRadius` default — a plain `float` field, `acclient.h:37181`, no sentinel). Retail's own behavior for a radius-0 handler is exactly this: close on the very first nonzero-distance check. **Extension (fix 5):** the watcher reads the SERVER-ECHOED ACCEPTED position snapshot (`RuntimeEntityRecord.Snapshot.Position`), sampled once per advanced frame at the post-network-command-phase, not retail's continuous live-pose push (retail's own client simulates and renders every entity's pose every frame; `CPlayerSystem`'s range handler reads that live pose, never a periodically-echoed one). Between accepted-position updates the watcher's distance measurement is therefore up to one update-interval stale. The one BLIND WINDOW this staleness could open into a wrong in/out-of-range verdict — an in-session portal/teleport, where the player's and vendor's position snapshots can briefly sit in DIFFERENT landblock coordinate frames mid-transit — is closed unconditionally by this same review's fix 1b (`RuntimeWorldTransitState.HasPendingTeleportStart`/`IsTeleportActive` short-circuit the whole distance computation before it runs, closing the session instead of measuring across the transit), so the staleness itself never reaches that particular failure mode; it remains recorded here as a standing precision gap for the window fix 1b does NOT cover (ordinary out-of-transit movement between the same-generation position updates a slow network tick can leave briefly stale). **Original text:** The client-local vendor-panel distance watcher closes on PLAIN 3D center-to-center distance instead of retail/ACE's CYLINDER-GAP distance (both objects' own collision radius and height subtracted from the center distance before comparing to `UseRadius`). Retail: `gmVendorUI::OpenVendor` registers `CPlayerSystem::RegisterObjectRangeHandler` keyed to the vendor's own `PublicWeenieDesc._useRadius`; ACE's server-side belt-and-suspenders `Vendor.CheckClose` closes on `GetCylinderDistance(lastPlayer) > UseRadius`, i.e. `Position::cylinder_distance`/`Physics.Common.Position.CylinderDistance` with each side's real `GetRadius()`/`GetHeight()`. | `src/AcDream.Runtime/Gameplay/RuntimeVendorRangeQuery.cs` (`EnforceRange`) | `AcDream.Runtime` does not resolve a live per-entity collision radius/height for an arbitrary NPC outside the App-layer's Setup-cylinder resolver (`WorldSelectionQuery`'s `_setupCylinder`, App-only — out of Runtime's reach per the Core-structure rules, and `PhysicsBody`/`RuntimeEntityRecord` carry no radius/height field). Plain center distance is a well-defined, non-degenerate substitute (using `ObjectRangeMath.ObjectsInRange`'s existing `useRadii: false` branch rather than inventing a new metric) for a CLIENT-LOCAL UI convenience that never touches the wire or any authoritative state — closing the panel is not gated by, nor gates, anything server-visible. Reading the accepted-position snapshot rather than a continuously-integrated live pose is the same "Runtime has no live render-side pose, only the last accepted wire snapshot" constraint every other Runtime-side distance query in this codebase already accepts. | The panel can close up to (player radius + vendor radius) sooner than exact retail — typically well under a meter for a two-legged NPC — so a player standing exactly at the boundary of a large-radius vendor's `UseRadius` may see the panel close slightly earlier than retail would. No effect on any transaction, wire message, or authoritative state (Slice 6's buy/sell owns those). Retiring the cylinder-gap half requires a Runtime-owned per-entity collision radius/height source, which does not exist today; retiring the staleness half requires a continuously-updated live-pose source Runtime does not keep either. | `CPlayerSystem::RegisterObjectRangeHandler` pc:203677/0x004C4C34; `gmVendorUI::OnObjectRangeExit` pc:199486/0x004C02F0; ACE `Vendor.CheckClose`/`GetCylinderDistance` (`references/ACE/Source/ACE.Server/WorldObjects/Vendor.cs:322-367`) — a SEPARATE method, `WorldObject.IsWithinUseRadiusOf` (`WorldObject_Use.cs:44-52`), owns the unrelated `?? 0.6f` approach-check fallback; `acclient.h:37181` (`float _useRadius`, plain memset-zero field, no sentinel); `docs/research/2026-08-08-slice5-vendor-browse-research.md` §A.3/§B.1/§B.2 | +| AP-160 | **Filed 2026-08-07, Slice 5.3 (vendor browse lifecycle). CORRECTED AND EXTENDED 2026-08-07 at the Slice 5.3 review corrections (fixes 4/5).** **Correction (fix 4):** this row's own Retail-oracle citation originally grouped `WorldObject_Use.cs:50,57` under the SAME citation as `Vendor.CheckClose`/`GetCylinderDistance`, which read as if the `wo.UseRadius ?? 0.6f` fallback lived inside the close watcher. It does not: `WorldObject_Use.cs:50,57` is `WorldObject.IsWithinUseRadiusOf`, the APPROACH check ("how close you need to be to open the shop") — a wholly different method from `Vendor.CheckClose`, which reads `UseRadius` directly with no fallback of its own (`UseRadius` is `float?`; a nullable comparison against a null right operand is always `false`, so `CheckClose` never closes at all on an unauthored radius). `EnforceRange`'s own code comment carried the same mis-attribution and, worse, actually APPLIED that mis-borrowed 0.6f as its fallback; it now passes the raw authored `UseRadius` with no fallback of any kind (0 when absent/unauthored, matching retail's own memset-zero `PublicWeenieDesc::_useRadius` default — a plain `float` field, `acclient.h:37181`, no sentinel). Retail's own behavior for a radius-0 handler is exactly this: close on the very first nonzero-distance check. **Extension (fix 5):** the watcher reads the SERVER-ECHOED ACCEPTED position snapshot (`RuntimeEntityRecord.Snapshot.Position`), sampled once per advanced frame at the post-network-command-phase, not retail's continuous live-pose push (retail's own client simulates and renders every entity's pose every frame; `CPlayerSystem`'s range handler reads that live pose, never a periodically-echoed one). Between accepted-position updates the watcher's distance measurement is therefore up to one update-interval stale. The one BLIND WINDOW this staleness could open into a wrong in/out-of-range verdict — an in-session portal/teleport, where the player's and vendor's position snapshots can briefly sit in DIFFERENT landblock coordinate frames mid-transit — is closed unconditionally by this same review's fix 1b (`RuntimeWorldTransitState.HasPendingTeleportStart`/`IsTeleportActive` short-circuit the whole distance computation before it runs, closing the session instead of measuring across the transit), so the staleness itself never reaches that particular failure mode; it remains recorded here as a standing precision gap for the window fix 1b does NOT cover (ordinary out-of-transit movement between the same-generation position updates a slow network tick can leave briefly stale). **Original text:** The client-local vendor-panel distance watcher closes on PLAIN 3D center-to-center distance instead of retail/ACE's CYLINDER-GAP distance (both objects' own collision radius and height subtracted from the center distance before comparing to `UseRadius`). Retail: `gmVendorUI::OpenVendor` registers `CPlayerSystem::RegisterObjectRangeHandler` keyed to the vendor's own `PublicWeenieDesc._useRadius`; ACE's server-side belt-and-suspenders `Vendor.CheckClose` closes on `GetCylinderDistance(lastPlayer) > UseRadius`, i.e. `Position::cylinder_distance`/`Physics.Common.Position.CylinderDistance` with each side's real `GetRadius()`/`GetHeight()`. **NARROWED 2026-08-08 (vendor-verify gate): the watcher now measures retail's cylinder-gap via the ResolveObjectTableHost radii — the plain-center shortcut was self-closing sessions inside the walk-to-use acceptance band (opened at 4.29 m center vs authored radius 3, closed same frame). Residuals: heights pass 0, unresolvable hosts degrade to center distance (close-early only).** | `src/AcDream.Runtime/Gameplay/RuntimeVendorRangeQuery.cs` (`EnforceRange`) | `AcDream.Runtime` does not resolve a live per-entity collision radius/height for an arbitrary NPC outside the App-layer's Setup-cylinder resolver (`WorldSelectionQuery`'s `_setupCylinder`, App-only — out of Runtime's reach per the Core-structure rules, and `PhysicsBody`/`RuntimeEntityRecord` carry no radius/height field). Plain center distance is a well-defined, non-degenerate substitute (using `ObjectRangeMath.ObjectsInRange`'s existing `useRadii: false` branch rather than inventing a new metric) for a CLIENT-LOCAL UI convenience that never touches the wire or any authoritative state — closing the panel is not gated by, nor gates, anything server-visible. Reading the accepted-position snapshot rather than a continuously-integrated live pose is the same "Runtime has no live render-side pose, only the last accepted wire snapshot" constraint every other Runtime-side distance query in this codebase already accepts. | The panel can close up to (player radius + vendor radius) sooner than exact retail — typically well under a meter for a two-legged NPC — so a player standing exactly at the boundary of a large-radius vendor's `UseRadius` may see the panel close slightly earlier than retail would. No effect on any transaction, wire message, or authoritative state (Slice 6's buy/sell owns those). Retiring the cylinder-gap half requires a Runtime-owned per-entity collision radius/height source, which does not exist today; retiring the staleness half requires a continuously-updated live-pose source Runtime does not keep either. | `CPlayerSystem::RegisterObjectRangeHandler` pc:203677/0x004C4C34; `gmVendorUI::OnObjectRangeExit` pc:199486/0x004C02F0; ACE `Vendor.CheckClose`/`GetCylinderDistance` (`references/ACE/Source/ACE.Server/WorldObjects/Vendor.cs:322-367`) — a SEPARATE method, `WorldObject.IsWithinUseRadiusOf` (`WorldObject_Use.cs:44-52`), owns the unrelated `?? 0.6f` approach-check fallback; `acclient.h:37181` (`float _useRadius`, plain memset-zero field, no sentinel); `docs/research/2026-08-08-slice5-vendor-browse-research.md` §A.3/§B.1/§B.2 | | AP-141 | **Filed 2026-08-04, C4 route 5 (projectile authoritative placement); NARROWED 2026-08-04 at the round-2 delta review (B1/B2) — the far-branch clause was factually wrong for the adopted-body case and is corrected below.** Three related projectile-only shapes, all pinned by design (D-P4) rather than ported: (a) the near-`Interpolate` disposition is a NO-OP for a live missile, where retail would lazily build interpolation machinery (`InterpolateTo` @0x005163AF) for it; (b) the post-operation `ConstrainTo` @0x00454272 (`MakePositionManager` @0x00510523 then `PositionManager::ConstrainTo`) is never ARMED for a projectile — retail's single arming site has no kind test, so retail WOULD build a `PositionManager` on demand and arm a missile's leash on any nonzero `MoveOrTeleport` return; acdream never arms it on any disposition, including the adopted-body case (whose PRE-EXISTING leash the teleport/far branches now un-arm or clear queue state for, but never RE-anchor, per retail's post-operation `ConstrainTo`); (c) a null-classified or `Rejected*` accepted Position for a missile is swallowed (write nothing) rather than caught up through any remote-shaped policy. | `src/AcDream.Runtime/Session/RuntimeRemotePlacementDriveController.cs` (`ApplyAcceptedProjectilePosition`) | acdream deliberately does not construct an `EntityPhysicsHost`/`PositionManager`/`InterpolationManager` chain for a ballistic body — the route-5b split the C4 route 5 contract rejected. The context that makes this safe rather than merely convenient: ACE never sends `UpdatePosition` for a missile (`references/ACE/Source/ACE.Server/WorldObjects/WorldObject_Tick.cs:333-334`, `SendUpdatePosition()` commented out inside the `PhysicsState.Missile` branch at `:265`) — every half of this row is deterministic-test-gated only, never exercised against a real server. **The far branch's `StopInterpolating` skip is retail-faithful ONLY for a BARE missile** (no `RemoteMotion` — retail's own `position_manager != 0` guard @0x005163C9 skips it for a never-interpolated object, so acdream's skip is faithful by consequence there). For the ADOPTED-BODY case (`TryBind`'s shared-body branch: an ordinary remote whose Missile bit was set by a later State packet, still carrying its `RemoteMotion`), retail's guard IS satisfied and retail WOULD clear the queue — acdream now ports this (`route.StopInterpolating && record.RemoteMotion is RemoteMotion adopted → adopted.Interp.Clear()`), matching the teleport branch's equivalent `StopInterpolating` action inside `teleport_hook`. What remains divergent for the adopted case is the post-operation `ConstrainTo` re-anchor @0x00454272 — retail re-anchors an existing leash at the just-updated position on every nonzero return; acdream never arms/re-anchors it on any projectile disposition (clause (b)). | A future change that DOES give projectiles a `PositionManager` (or a headless/no-window remote-motion consumer that expects one) must re-decide this row rather than silently building the machinery ad hoc; until then, a live missile never shows an ARMED constraint leash and never catches up via the near/UnroutedCatchUp policy — both unreachable in play. An adopted-body missile's INHERITED leash (armed before it became a missile) is un-armed by the teleport hook, has its queue cleared by both teleport and far, but is never re-anchored at the new position by either — its brake accumulator (`ConstraintPosOffset`) is not reset to zero at each accepted Position the way retail's @0x00454272 re-anchor does. **Correction, round 3 (2026-08-04): the round-2 wording here — that a stale leash "would drag the body toward a stale anchor" — was wrong and is retracted.** `ConstraintManager.ConstraintPos` is write-only in both retail and the port (never read by `AdjustOffset`), and `ConstraintManager::adjust_offset` @0x00556180 only tapers or zeroes an already-composed per-tick offset while `InContact` — a leash brakes motion the interp/sticky chain already produced; it has no mechanism to move anything toward the anchor. The real residual is confined to one tick of un-reset brake accumulator, contact-gated, and it cannot move an airborne far-snapped missile at all (the clamp branch does not run while airborne). | `CPhysicsObj::MoveOrTeleport` 0x00516330 (`InterpolateTo` @0x005163AF, `IsMovingTo` @0x0050EB10 returning 0 without a `MovementManager`; far branch `StopInterpolating` @0x005163C9-@0x005163CB); `SmartBox::HandleReceivedPosition` 0x00453FD0 (`ConstrainTo` arming site @0x00454272); `CPhysicsObj::ConstrainTo` 0x00510520 (`MakePositionManager` @0x00510523); `ConstraintManager::adjust_offset` 0x00556180 (brake-only taper, write-only anchor); `WorldObject_Tick.cs:333-334`/`:265` (ACE never-sends evidence) | | AP-142 | **Filed 2026-08-04 (C4 route 7, pickup/parent/delete). AMENDED 2026-08-04 at the dual-Opus retail-conformance/architecture review round (R1/A8 MAJOR+LOW; R10 MINOR) — clause (d) added, clause (b) corrected. AMENDED AGAIN 2026-08-04 at the round-3 dual review (N1/N2/N4, B3) — clause (d)'s reasoning corrected and its risk-column scope widened; clause (e) RETIRED — the depth cap it described is deleted outright, replaced by an iterative worklist with no depth concept at all. AMENDED AGAIN 2026-08-05 at the #319 fix — clause (f) added. AMENDED AGAIN 2026-08-05 at the #319 dual-review round (retail PASS, architecture FAIL/6 MAJORs) — clause (f) rewritten: the tripwire moved above the canonical commit and no longer throws (A1), and the deferred late-bind queue A1's fix text originally described was deleted per A6 (both reviews proved it production-unreachable for both producers).** acdream collapses retail's `CPhysicsObj` pair — a `cell` pointer plus a separately-written `objcell_id` — into ONE canonical `RuntimeEntityRecord.FullCellId`, which is also the residency/liveness predicate acdream reads at 45+ sites. Four consequences, all intentional: (a) the removal path propagates ZERO to a subtree's children (withdrawal, delete, `EndGeneration`), where retail's `leave_cell` recursion nulls only each child's `cell` pointer and leaves a STALE non-zero `objcell_id` (`change_cell`'s removal tail @0x005133C1 never touches a child's id) — reproducing that stale-id residue would leave a child "resident" per every acdream predicate while retail's own gating field (`cell == nullptr`) says it is not; (b) retail's same-cell depth-1 per-tick `objcell_id` refresh (`SetPositionInternal` @0x0051539c-@0x005153d8, gated on the parent NOT crossing a cell) is subsumed by the value-idempotent propagation chokepoint (`RuntimeEntityDirectory.SetFullCell`'s "skip a child whose `FullCellId` already equals the target" guard) rather than ported as a separate tick loop — a same-value restamp is unobservable with one field playing both retail roles. **Correction (R10): this is a clean equivalence only on the REMOVAL side.** The skip ALSO prunes the child's whole subtree on a same-value WRITE, which retail's `enter_cell` does not do — it recurses over children unconditionally (@0x00510f03); only `leave_cell` prunes (@0x00510f5b, on `cell != 0`). Currently unreachable-by-construction (after D4 nothing writes a grandchild's cell independently of its own committed parent), but it is an asymmetry, not a proven equivalence; (c) the sustaining propagation itself: retail re-cells children when the parent crosses a cell, recursively, on EVERY `SetPositionInternal`/`change_cell` (@0x00515372/@0x00513390), not only at attach — acdream ports this as a single hook every canonical cell-write funnels through, so an attach-only write (the pre-existing shape) is deliberately NOT what shipped. **(d) retail's `enter_cell` gates its ENTIRE body — the write AND the recursion into children — on `this->part_array != 0` (@0x00510ed8); a child with a null part array receives nothing and its whole subtree is skipped. acdream's propagation has NO analogue and writes unconditionally. CORRECTED reasoning (round-3 review, N1/N2): the original draft of this clause argued acdream's `HasPartArray` means something semantically different from retail's `part_array` (a "renderer built a mesh" flag vs. "this CPhysicsObj has any part array"). That framing is WRONG — retail's `part_array` has exactly ONE assignment site, `CPhysicsObj::makeAnimObject` @0x0050e930 → `CPartArray::CreateSetup`, assigned @0x0050e94d, so retail's flag is ALSO a mesh-construction product; the two are near-synonyms, not different concepts. The REAL reason acdream cannot gate the canonical D1/D2 write on `HasPartArray` is LAYERING, not semantics: Slice J made the Runtime canonical layer presentation-independent by design (`docs/research/2026-07-25-slice-j1-runtime-contract-closeout.md` and the Slice J campaign generally), and `HasPartArray` is populated exclusively by App/graphical code (`EquippedChildRenderController.cs:609`, `DatLiveEntityProjectionMaterializer.cs:203`) — the canonical layer structurally cannot depend on a flag only the presentation layer ever writes, headless or not. CORRECTED scope (round-3 review): this is NOT headless-only. `PrepareAndTryRealize` calls `CommitAcceptedParentCellless` (hence D1's re-cell) BEFORE `TryRealize` sets `HasPartArray = true` at `:609` — so at the exact moment D1 runs, `child.HasPartArray` is FALSE in the GRAPHICAL host too, and gating on it would break attach there as well, not just headless. Retail has no equivalent window at all: `part_array` is assigned once at construction and `enter_cell`'s guard reads that same, already-settled field.** The guard is deliberately NOT reproduced at the canonical layer. **(e) RETIRED 2026-08-04 (round-3 review, N4/B3 — both reviews independently found the same defect).** Previously: recursion depth capped at 64 levels as hostile/buggy-server hardening. The cap's actual failure mode was worse than what it guarded against: a subtree beyond the cap was left at its PRIOR — on the withdraw path, STALE NONZERO — cell PERMANENTLY, logged only under a probe flag nobody runs by default. On the withdraw path that is the #184 shape verbatim: an entity every acdream residency predicate calls resident that retail (and clause (a) above) says is not. Shipping that inside the slice whose headline is fixing exactly this class was unacceptable. Retired by deleting the cap outright and replacing the recursion with an iterative worklist (`RuntimeEntityDirectory._propagationWorklist`), which has no stack-frame-bounded depth at all — the only limit is the number of committed relations actually in the system, matching retail's own genuinely unbounded recursion with no acdream-only cap and therefore no register row for one. **(f) Filed 2026-08-05 (#319 fix).** A CreateObject-carried parent relation (the raw spawn's `Physics.Parent` field, and the same-generation `CreateParentUpdate` envelope) names the parent's GUID and location only — neither wire shape carries a parent instance sequence, matching retail's own GUID-only attach (`PhysicsDesc::get_parent_id` @0x00558a18 → `CObjectMaint::GetObjectA` @0x00558a2d → `CPhysicsObj::set_parent` @0x00558a3e; the reverse `CObjectMaint::SetChildren` @0x00509370 hash-walks by guid with a `GetNullObject` placeholder @0x005093e6 — no instance-sequence field or comparison exists anywhere in either direction). acdream's committed-relation table is nonetheless keyed by (guid, incarnation) (clause (c)'s D1/D2 requirement), so a CreateObject-carried relation must adopt SOME incarnation to file under; it now LATE-BINDS to the parent's LIVE incarnation at accept time (`EquippedChildRenderController.AcceptLateBoundCreateObjectRelation`, both the raw-CreateObject and same-generation `CreateParentUpdate` producers) rather than the previously-hardcoded 0, which silently mis-keyed every player-parented CreateObject relation (a player's `ObjectInstance` is `Character.TotalLogins`, never 0) and defeated D1/D2 for the local player's own login equipment and every remote player's observed equipment (#319). A commit-time tripwire (`ParentAttachmentState.CanCommitIncarnation`, checked BEFORE either half of the commit mutates state — architecture review A1, 2026-08-05, moved it there after the original throw-after-canonical-commit shape was shown to tear the transaction it was built to protect) refuses (logs, returns false, never throws) rather than silently filing a relation under a mismatched incarnation whenever the parent is currently addressable. **A1 also settled A6's design question**: an initial revision queued a relation whose parent was not yet addressable through a deferred/late-bind retry mechanism; both reviews independently proved that queue was structurally unreachable in production for BOTH producers (`RuntimeEntityObjectLifetime.RegisterEntityCore`'s `EnqueueDeferredCreate` gate defers the ENTIRE CreateObject, for both wire shapes, before either producer ever runs) while carrying three latent defects of its own (a missing child POSITION_TS gate, a placeholder-incarnation collision with the generation filters, unbounded accumulation) — it was deleted rather than fixed in place; the unaddressable-parent case now logs and refuses outright, matching the invariant the layer above already enforces. | `src/AcDream.Runtime/Entities/RuntimeEntityDirectory.cs` (`SetFullCell`, `PropagateFullCellToChildren`, `RefreshSnapshot`); `src/AcDream.Runtime/Entities/RuntimeEntityObjectLifetime.cs` (`CommitAcceptedParentCellless`'s D1 half, `WithdrawCommittedChildrenToCellless`); `src/AcDream.Runtime/Entities/ParentAttachmentState.cs` (`TryGetCommittedParent`, `CanCommitIncarnation`, `CommitProjection`); `src/AcDream.Runtime/Entities/RuntimeEntityRecord.cs` (`HasPartArray`); `src/AcDream.App/Rendering/EquippedChildRenderController.cs` (`AcceptLateBoundCreateObjectRelation`, `OnSpawn`, `OnCreateParentAccepted`, `PrepareAndTryRealize`); `src/AcDream.Runtime/Session/RuntimeLiveEntitySessionController.cs` (`ResolveAndCommitChildAttachment`) | Reproducing retail's pointer/id split would require a second field acdream's 45+ liveness call sites would then have to be individually audited for which half they mean — the single-field model is a stated, load-bearing simplification, not an oversight; see `docs/research/2026-08-04-retail-parent-cell-propagation.md` and `docs/research/2026-08-04-c4-route-7-contract.md` D2/D3/D9. Clause (f) is retail-faithful for the identical reason clauses (a)-(d) are: retail's attach has no incarnation gate on this path at all, so adopting the current holder of the guid IS the retail behavior, not an approximation of it. | A future consumer that expects retail's exact stale-`objcell_id`-under-a-null-`cell` shape (none identified) would see a fully cell-less child instead. (d)'s risk: acdream celling a child retail would leave nowhere — none identified in play against a well-behaved ACE, since a server-authored equip always names a real, DAT-resolvable Setup, and the graphical host's own brief pre-`TryRealize` window is bridged by D1 running inside the same synchronous transaction as the rest of the attach commit, not by `HasPartArray` being true. (f)'s risk: none identified against a well-behaved ACE — a CreateObject's parent guid always names the entity that currently holds it by construction. | `CPhysicsObj::change_cell` 0x00513390 (@0x005133C1 removal tail); `CPhysicsObj::enter_cell` 0x00510ed0 (@0x00510ed8 the `part_array` guard); `CPhysicsObj::leave_cell` 0x00510f50; `CPhysicsObj::SetPositionInternal` 0x00515330 (@0x0051536d branch, @0x0051539c-@0x005153d8 same-cell loop, @0x00515372 cell-change branch); `CPhysicsObj::makeAnimObject` 0x0050e930 (`CPartArray::CreateSetup` assignment @0x0050e94d); `PhysicsDesc::get_parent_id` 0x00558a18; `CObjectMaint::GetObjectA` 0x00558a2d; `CPhysicsObj::set_parent` 0x00558a3e; `CObjectMaint::SetChildren` 0x00509370 (`GetNullObject` placeholder @0x005093e6) | | AP-143 | **Filed 2026-08-04 (C4 route 7 D5, headless parent-realize drive). AMENDED 2026-08-04 at the retail-conformance review round (R7 MINOR) — this row originally described only ONE of the three checks the drive skips. Line citations corrected at the round-3 review (N3).** The graphical `EquippedChildRenderController.ValidateParentProjection` performs three retail-anchored checks before accepting a parent-attach request: (1) self-parenting rejection (`relation.ParentGuid == relation.ChildGuid`, `:915-916`); (2) the parent must have a constructed part array (`parent.HasPartArray`, `:920` — the closest acdream analogue to retail's `part_array != 0` guard, AP-142 clause d); (3) `Setup.HoldingLocations` validates the specific holding location (`CSetup::GetHoldingLocation` @0x0050F896, via `PartArray::add_child`). `AcDream.Headless`/`AcDream.Runtime`'s direct-host parent-realize drive (`RuntimeLiveEntitySessionController.ResolveAndCommitChildAttachment`) performs NONE of the three — it commits on the POSITION_TS gate acceptance and relation resolution alone. (1) is inert by construction: D1's re-cell gate reads `parent.FullCellId == 0` (the child was just zeroed by the cell-less edge before D1 runs), and D2's skip-on-equal terminates the resulting one-node cycle — a self-parent headless commits the relation but never observably re-cells through it. (2) has no headless analogue at all (see AP-142 clause d — `HasPartArray` is populated only by the graphical mesh pipeline, never headless, for ANY entity). (3) has no prepared-content surface (repo-wide grep confirms nothing under `src/AcDream.Content`/`AcDream.Bake` carries `Setup.HoldingLocations`). | `src/AcDream.Runtime/Session/RuntimeLiveEntitySessionController.cs` (`ResolveAndCommitChildAttachment`) | Precedent: the content-less host already accepts reduced fidelity elsewhere (`RuntimeLiveEntitySessionController:108-117`'s documented content-less registration). A server-sent self-parent, part-array-less parent, or invalid holding location is unreachable against a well-behaved ACE (ACE only emits `ParentEvent` for a location its own `Player_Inventory`/wield validation already accepted), so this is a defense-in-depth gap, not a live-play one. | A malicious or buggy server could attach a child headless where retail and the graphical host would both reject it — inert against ACE today for all three. Retiring (3) means extending the prepared-content bake format with `Setup.HoldingLocations`, deliberately NOT done in this slice (route 7 contract §4 D5); (2) has no retiring action available until acdream's canonical layer gains its own construction-time part-array concept (a larger architectural question, out of scope here). | `PartArray::add_child` (`CSetup::GetHoldingLocation` 0x0050F896); `CPhysicsObj::enter_cell` 0x00510ed8 (the `part_array` guard); `EquippedChildRenderController.ValidateParentProjection` (graphical port, all three checks) | diff --git a/src/AcDream.Runtime/Gameplay/RuntimeVendorRangeQuery.cs b/src/AcDream.Runtime/Gameplay/RuntimeVendorRangeQuery.cs index cf19c932..27ac07bf 100644 --- a/src/AcDream.Runtime/Gameplay/RuntimeVendorRangeQuery.cs +++ b/src/AcDream.Runtime/Gameplay/RuntimeVendorRangeQuery.cs @@ -87,18 +87,15 @@ public static class RuntimeVendorRangeQuery /// /// /// - /// Distance metric divergence (register AP-160): retail/ACE close - /// on CYLINDER-GAP distance — both objects' own collision radius and - /// height subtracted from the center distance - /// (Position::cylinder_distance/ACE's GetCylinderDistance). - /// Runtime does not resolve a live per-NPC collision radius/height - /// outside the App-layer's Setup-cylinder resolver - /// (WorldSelectionQuery, App-only — out of reach per the - /// Core-structure rules), so this uses plain 3D center-to-center - /// distance via 's - /// useRadii: false branch instead. Effect: the panel can close up - /// to (player radius + vendor radius) sooner than exact retail — - /// typically well under a meter for a two-legged NPC. + /// Distance metric (AP-160 residual narrowed 2026-08-08): the + /// watcher measures retail's CYLINDER-GAP distance + /// (Position::cylinder_distance) with both radii resolved through + /// the same ResolveObjectTableHost seam the movement arrival + /// uses — the plain-center-distance shortcut self-closed sessions the + /// server had just opened (the walk-to-use acceptance band; see the + /// inline comment at the range check). Heights pass 0 (the host surface + /// exposes radius only); an unresolvable host degrades that operand to + /// center distance, which only ever closes EARLY. /// /// public static void EnforceRange(GameRuntime runtime) @@ -148,15 +145,35 @@ public static class RuntimeVendorRangeQuery // an EXACT position match. That is retail's own behavior for a // radius-0 (or unauthored) handler, not a bug to paper over. float useRadius = vendorRecord.Snapshot.UseRadius ?? 0f; + + // #vendor-verify-gate 2026-08-08: the watcher MUST measure with the + // same metric the walk-to-use arrival and ACE's acceptance use — + // cylinder-gap (center distance minus both bodies' radii), retail's + // Position::cylinder_distance. With plain center distance there is + // an acceptance band (bare UseRadius .. UseRadius + both radii, + // observed live: open accepted at 4.29 m center vs radius 3) where + // the server OPENS the session and this watcher instantly + // self-closed it — the panel flashed shut within one frame. The + // radii come from the same GetObjectA seam the movement arrival + // resolves through (ResolveObjectTableHost, lazy-minimal hosts + // included); an unresolvable host contributes 0 and degrades to + // center distance, which can only close EARLY, never hold a session + // ACE has ended. + float playerRadius = + runtime.EntityObjects.Physics.ResolveObjectTableHost(playerGuid) + ?.Radius ?? 0f; + float vendorRadius = + runtime.EntityObjects.Physics.ResolveObjectTableHost(vendorId) + ?.Radius ?? 0f; bool inRange = ObjectRangeMath.ObjectsInRange( AbsolutePosition(playerPosition), - 0f, + playerRadius, 0f, AbsolutePosition(vendorPosition), - 0f, + vendorRadius, 0f, useRadius, - useRadii: false, + useRadii: true, ignoreZDelta: false); if (!inRange)