diff --git a/docs/architecture/retail-divergence-register.md b/docs/architecture/retail-divergence-register.md index 0fd7b20b..3bf18d91 100644 --- a/docs/architecture/retail-divergence-register.md +++ b/docs/architecture/retail-divergence-register.md @@ -114,7 +114,7 @@ accepted-divergence entries (#96, #49, #50). --- -## 3. Documented approximation (AP) — 92 active rows +## 3. Documented approximation (AP) — 93 active rows Wave-0 UI ledger repair (2026-07-10) retired stale AP-38, resolved the AP-84 collision, restored overwritten paperdoll rows as AP-92/AP-93, and registered @@ -229,8 +229,9 @@ AP-94..AP-112 for the confirmed retail-UI completion gaps. | AP-123 | Item cooldowns use the retained toolkit's existing procedural `UiItemSlot` leaf rather than materializing retail's ten `m_elem_Icon_Cooldown_*` child elements. The group lookup, remaining-time formula, exact DAT sprites, step choice, and topmost ReadOrder-8 outcome are faithful. | `src/AcDream.App/UI/Layout/ItemCooldownUiController.cs`; `src/AcDream.App/UI/UiItemSlot.cs` | `UiItemSlot` already consumes/reproduces UIItem children procedurally under the IA-15 retained-toolkit architecture. Selecting one imported sprite at draw time gives every inventory/equipment/shortcut alias the same output without a parallel widget tree or per-cell timers. | A future feature that observes the individual cooldown child visibility/state rather than the rendered UIItem could see no child elements even though the cell looks and advances correctly. | `CEnchantmentRegistry::OnCooldown @ 0x005943C0`; `UIElement_UIItem::UpdateCooldownDisplay @ 0x004E1E20`; common UIItem prototype `0x1000033E` | | AP-124 | Local ACE omits the new object's CreateObject to the initiating session after `StackableSplitTo3D`, sending only F748 Position for the previously unknown GUID. acdream retains retail's one pending split source/count/time identity for ten seconds and, only for that otherwise-impossible unknown Position, hydrates a canonical clone of the source description with the new GUID and authoritative world placement. | `src/AcDream.App/World/InventoryWorldDropProjectionController.cs`; `src/AcDream.App/UI/ItemInteractionController.cs`; `src/AcDream.App/Physics/LiveEntityNetworkUpdateController.cs` | Nearby/reconnecting clients receive the ordinary CreateObject, while the initiator otherwise cannot render the authoritative object until relog. A server that sends CreateObject never enters this path; the pending identity is consumed before normal hydration and all other unknown Positions remain rejected. | ACE's F748 does not carry WCID or stack size, so an unrelated unknown Position arriving during the exact pending ten-second window could be associated with the split. Retail confirms WCID/count from CreateObject; removing the approximation requires ACE to send that packet to the initiator. | `ACCWeenieObject::UIAttemptSplitTo3D @ 0x0058D850`; `ACCWeenieObject::DeclareValid @ 0x0058E340`; ACE `Player.HandleActionStackableSplitTo3D` / `TryDropItem`; `docs/research/2026-07-26-retail-inventory-placement-and-world-drop-pseudocode.md` | | AP-125 | Transport control packets (the 2.0 s cumulative AckSequence and the 0.6 s RequestRetransmit) are emitted STANDALONE; retail piggybacks optional headers onto queued outbound packets first-fit (`FlowQueue::CoalesceData @ 0x00547740`, invoked at `TransmitNewPackets @ 0x00547A6E`), and `EnqueueNaks` hands the NAK to `PacketController::EnqueueOptionalHeader @ 0x00543C84` rather than emitting directly. | `src/AcDream.Core.Net/Transport/AckNakScheduler.cs` (`EmitCumulativeAck`, `EmitNakRequest`) | ACE honours a RequestRetransmit ONLY when EncryptedChecksum is absent (NetworkSession.cs:283-284) — a retail-style piggyback onto a sequenced packet encrypts the NAK and ACE silently ignores it, making S2C loss unrecoverable; ACE likewise advances its client-sequence watermark on any packet whose flags are not exactly AckSequence (:474-476), so coalesced control content on a borrowed sequence risks skipping a real packet. Standalone exact-flag emission is the only ACE-safe shape; it also keeps reliable packets free of optional headers, making the resend cache strip provably a no-op. | Slightly higher C2S datagram count than retail (one extra small packet per 2.0 s / per NAK window); marginally more loss exposure for the control packets themselves on a metered path. | `FlowQueue::CoalesceData @ 0x00547740`; `SharedNet::EnqueuePak @ 0x00543B10`; `SharedNet::EnqueueNaks @ 0x00543BD0`; ACE `NetworkSession.cs:283-284,:342-343,:474-476` | +| AP-126 | One monotonic Stopwatch-backed clock (`TransportClock`) drives every transport gate (2.0 s ack, 0.6 s NAK, 0.333 s handshake retry, 0.5 s interval, 5 s assembler sweep); retail splits gates between `Timer::cur_time` (server-adjusted) and `Timer::local_time`. | `src/AcDream.Core.Net/Transport/TransportClock.cs` | The cur/local split only matters for gates that must track server clock adjustments; none of the ported gates semantically depend on server time — they are local cadences. A single injectable source also gives the virtual-clock test seam every conformance suite relies on. | A future port of a genuinely server-clock-relative gate could silently use the wrong clock if it reuses TransportClock without checking this row. | `SharedNet::EnqueuePak @ 0x00543B10` (cur_time); `ClientNet::ProcessConnection @ 0x00545450` (local_time for the 140 s check) | -## 4. Temporary stopgap (TS) — 39 active rows (TS-57 filed 2026-07-29 at Campaign N slice N1 — no outbound RejectRetransmit; TS-27 narrowed same slice to the inbound direction) + TS-37 historical note (TS-20 retired 2026-07-16 — the later named-retail audit disproved the proposed DrawingBSP polygon filter; TS-37 is a retired-row historical note, not an active count; TS-39 retired R5-V3 — sticky seams bound to the ported PositionManager/StickyManager, radii threaded; TS-45 retired 2026-07-07 — hand-rolled `SphereCollision` replaced by the faithful CSphere family port, fixing the player-vs-monster crowd wedge; TS-3 retired 2026-07-07 — `frames_stationary_fall` accounting ported in the #182 verbatim UpdateObjectInternal rebuild, fixing the airborne falling-animation wedge; TS-41 retired 2026-07-07 — SERVERVEL synth-velocity remote body-drive replaced by the retail interp catch-up + unconditional MovementManager::UseTime, the remote-creature de-overlap #184; TS-42 retired 2026-07-19 — semantic animation completion now precedes the ordered Target/Movement/PartArray/Position tail; TS-44 narrowed again 2026-07-19 — complete orientation joined interpolation, only during-stick enqueue suppression remains) +## 4. Temporary stopgap (TS) — 43 active rows (TS-57..TS-61 filed 2026-07-29 during Campaign N — no outbound RejectRetransmit; TS-27 narrowed same slice to the inbound direction) + TS-37 historical note (TS-20 retired 2026-07-16 — the later named-retail audit disproved the proposed DrawingBSP polygon filter; TS-37 is a retired-row historical note, not an active count; TS-39 retired R5-V3 — sticky seams bound to the ported PositionManager/StickyManager, radii threaded; TS-45 retired 2026-07-07 — hand-rolled `SphereCollision` replaced by the faithful CSphere family port, fixing the player-vs-monster crowd wedge; TS-3 retired 2026-07-07 — `frames_stationary_fall` accounting ported in the #182 verbatim UpdateObjectInternal rebuild, fixing the airborne falling-animation wedge; TS-41 retired 2026-07-07 — SERVERVEL synth-velocity remote body-drive replaced by the retail interp catch-up + unconditional MovementManager::UseTime, the remote-creature de-overlap #184; TS-42 retired 2026-07-19 — semantic animation completion now precedes the ordered Target/Movement/PartArray/Position tail; TS-44 narrowed again 2026-07-19 — complete orientation joined interpolation, only during-stick enqueue suppression remains) | # | Divergence | Where (file:line) | Why it is safe / justified | Risk if assumption breaks | Retail oracle | |---|---|---|---|---|---| @@ -276,6 +277,10 @@ AP-94..AP-112 for the confirmed retail-UI completion gaps. | TS-54 | AdminEnvirons sound values `0x65..0x7B` are diagnosed by retail enum name but do not play audio. Retail checks that the local player physics object and UI sound table exist, then calls `SoundManager::PlaySoundFromCenter(Sound_UI_*, table)` for Roar through Thunder6. | `src/AcDream.App/World/WorldEnvironmentController.cs` (`ApplyAdminEnvirons`) | The current audio owner has no typed retail UI-sound-table binding; logging preserves the inbound evidence without inventing wave DIDs or routing the sounds through positional world audio. | Server-authored ambience/thunder packets are silent in acdream while retail plays the centered UI sound. | `CPlayerSystem::Handle_Admin__Environs @ 0x0055DE20` (`0x0055E07F..0x0055E2C7`); `SoundManager::PlaySoundFromCenter @ 0x00550950` | | TS-55 | AdminEnvirons fog values remain a color-only `WeatherSystem.Override` approximation. Retail values 1..5 install authored ambient color/level plus fog color/max; value 6 also forces transition/min/max and blanks radar; Clear restores all override fields and radar; `0x270F` installs a separate authored override. | `src/AcDream.App/World/WorldEnvironmentController.cs` (`ApplyAdminEnvirons`); `src/AcDream.Core/World/WeatherState.cs` (`EnvironOverrideColor`) | Preserves the already accepted enum bridge while Slice 8 moves ownership; porting the complete environment/radar presentation is a separate behavior change requiring focused visual gates. | Forced-fog hue, density, scene ambient, and radar blanking differ from retail; `0x270F` is ignored. | `CPlayerSystem::Handle_Admin__Environs @ 0x0055DE20` (`0x0055DE2B..0x0055E344`) | | TS-57 | No outbound `RejectRetransmit`: a server NAK for an id no longer in the sent-packet cache is dropped silently (counted in `TransportStats.UncachedNakIds`); retail answers `RejectRetransmit @ FlowQueue` so the server abandons the id immediately | `src/AcDream.Core.Net/Transport/OutboundFlowQueue.cs` (`OnRetransmitRequest`) | ACE parses `RejectRetransmit` and no-ops it (NetworkSession.cs — no handler), and the standalone unsequenced form would trip ACE's watermark hole (campaign doc §3 row 3: any cleartext non-ack packet with a live sequence advances the watermark and skips a real packet forever) | Against a server that DOES honor RejectRetransmit, an uncached NAKed id keeps being re-requested until that server's own NAK give-up logic fires — never against ACE, which forgets the id when its next cumulative ack passes it | `RecipientData::ProcessNaks @ 0x00547010`; ACE NetworkSession.cs:299-304 (server-side emit), no client-consume handler | +| TS-58 | No outbound TimeSync/EchoRequest keepalive (retail sends both every 6 half-second intervals, ~3 s). The 2.0 s cumulative AckSequence is the sole idle keepalive; it refreshes ACE's 60 s timeout, which is the only server-side consumer. | `src/AcDream.Core.Net/Transport/TransportClock.cs`; `src/AcDream.Core.Net/Transport/AckNakScheduler.cs` | Standalone unsequenced TimeSync/Echo packets trip ACE's exactly-AckSequence watermark rule (NetworkSession.cs:474-476) and are only ACE-safe piggybacked, which needs retail's CoalesceData (AP-125). The ack keepalive covers the timeout; no transport RTT sample is lost that LinkStatus' app-level ping does not already provide. | No transport-level RTT/latency sample; a future server gating on TimeSync cadence would see silence; ACE's speedhack echo checks never engage. | `ClientFlowQueue::IncrementLocalInterval @ 0x00547F10`; ACE `NetworkSession.cs:474-476`, `Session.cs:101-102` | +| TS-59 | No outbound Flow report (retail emits a 6-byte bytes-received+interval header whenever the inbound remote interval advances). | `src/AcDream.Core.Net/Transport/ReliableTransport.cs` (interval clock present; no Flow emission) | ACE parses the Flow header and has no handler (PacketHeaderOptional.cs:117-124); the standalone unsequenced form would trip the watermark hole. Retail itself never consumes inbound Flow and has no throttle (`WireRoomLeft` is a folded return-1). | A future server that rate-adapts on client Flow reports sees nothing. | `SharedNet::ProcessNewRemoteInterval @ 0x00543A80`; `ClientFlowQueue::WireRoomLeft @ 0x0052C1C0` (folded) | +| TS-60 | No 140 s dead-link declaration or referral auto-reconnect in the transport; a silent server is only visible through `LinkStatusSnapshot.SecondsSinceLastPacket` (presentational). | `src/AcDream.Core.Net/Transport/ReliableTransport.cs`; `src/AcDream.Core.Net/WorldSession.cs` (`BuildLinkStatus`) | The input (seconds since last inbound) is already exposed; session lifecycle/reconnect is Runtime's ownership domain and deserves its own campaign rather than a transport-embedded side effect. Every ACE transport death is silence, so nothing server-side depends on the client reacting at 140 s. | A dead link idles until the user acts; no automatic recall/referral reconnect where retail would attempt one. | `ClientNet::ProcessConnection @ 0x00545450` tail (the two 140.0 literals) | +| TS-61 | A UDP send failure burns the reliable sequence and its ISAAC word (the encode commits before `_net.Send`); retail keeps the sealed packet at the queue head and retries with the same key. | `src/AcDream.Core.Net/Transport/OutboundFlowQueue.cs` (`SendGameMessage`) | A connectionless-socket `SendTo` failure is effectively unreachable in practice (no route/ICMP errors surface on later receives, not sends, on Windows UDP); recovering it faithfully needs a full outbound packet queue. The N1 review accepted the exposure explicitly. | One `SocketException` on send would desync the outbound cipher permanently (session death; observable as `[net-out-EX]` followed by silence). | `FlowQueue::TransmitNewPackets @ 0x00547C2C` (retry-from-head) | | TS-56 | Chase-camera mouse input retains acdream's invented post-filter yaw/pitch scalars (`0.004`/`0.003` radians per count), and held-key pitch/zoom retain their non-retail integration shapes. Retail mouse look passes `FilterMouseInput(delta) × configured sensitivity × 1/15` as the replacement scale to `CameraSet::Rotate`, which then applies the shared 8° angle; retail held pitch uses the same angle and zoom scales the viewer offset multiplicatively. | `src/AcDream.App/Input/CameraPointerInputController.cs`; `src/AcDream.App/Input/MouseLookController.cs`; `src/AcDream.App/Rendering/CameraFrameController.cs` | Slice 8 is behavior-preserving ownership work. The named-retail audit proves the mismatch but has not yet extracted the configured mouse-sensitivity default or the exact caller flags needed for a complete feel port; changing only one scalar here would create a mixed input model. | RMB/MMB orbit, held pitch, and zoom can feel slower, faster, or differently accelerated than retail even though callback ordering and filtering are correct. | `CameraSet::Rotate @ 0x00458310`; `CameraSet::MouseLookHandler` call at `0x00458EF9`; `CameraSet::Raise @ 0x00457B00`; `CameraSet::Closer @ 0x004586D0`; `docs/research/2026-06-11-holistic-map/wf2-camera-viewer.md` | --- diff --git a/docs/plans/2026-07-29-network-transport-campaign.md b/docs/plans/2026-07-29-network-transport-campaign.md index b87e0341..c76c9e74 100644 --- a/docs/plans/2026-07-29-network-transport-campaign.md +++ b/docs/plans/2026-07-29-network-transport-campaign.md @@ -257,4 +257,4 @@ verbatim in each implementer prompt. | N3 | complete | `0265cc42` — Opus review PASS (advisories folded into N4: transitional-state wording, SharedInit citation, pump-order wording, control-packet Time/Iteration rule, stale budget-break comment) | AckNakScheduler + 2.0 s cumulative ack (`Transport/AckNakScheduler`): ONE shared timestamp (`ReceiverData::timeStamp_` @ +0x10) arbitrating NAK-xor-ack per sweep (`ClientNet::ProcessConnection @ 0x00545450`); the ack is one cleartext exact-flags `AckSequence` carrying the tracker's `highestIDReceived_` behind the >= 2.0 s gate (`SharedNet::EnqueuePak @ 0x00543B10` — the binary's only 0x4000 construction site), armed at connection birth (`ReceiverData::Init @ 0x00548EF0`). The Phase 4.9 per-packet reflex ack and `WorldSession.SendAck` are DELETED; the `[net-tick]` acks/s probe now reads `Stats.AcksSent`. Sweep order per `FlowQueue::Empty @ 0x00548A20`: interval clock, NAK/ack arbitration, pending resends, prune. **N3 transitional state (closed by N4):** a non-empty NAK set suppressed the ack and emitted NOTHING — the exposure was real even on loopback, just low-probability: one receive-buffer drop parks an id, every later sweep takes the silent NAK branch, acks stop (witness: `[net-tick] acks/s=0`), and ACE disconnects the quiet session at its 60 s timeout. N4 completed the branch. **N1 advisory retired (fold-in):** fresh reliable sends now stamp `Header.Time` = the current interval id (`FlowQueue::TransmitNewPackets @ 0x00547A60`, header build at 0x00547A84); ACE ignores inbound `Header.Time`, so the wire is unaffected. New `WorldSession.TransportClockSource` seam drives the gate on virtual time. 723 Core.Net tests green (keepalive property proven: a quiet session's 2 s acks refresh ACE's 60 s deadline across a 120 s virtual horizon; storm collapse: a 50-packet flood → ONE ack; the model accepts the reused-sequence ack without advancing its watermark). Connected lifecycle + canonical nine-stop gates PASS. | | N4 | complete | `852a59e3` — **Opus review PASS** (reclaim invariant attacked from five angles, held; NAK fidelity verified to the x87 masks; AP-125 filed + F1 false-arithmetic wording + F5 ordinal sentinel fixed in the acceptance commit; F3 Iteration-on-fresh-sends folds into N5) | Client NAK emission + RejectRetransmit consumption. `AckNakScheduler` completes the NAK branch: one cleartext exact-flags `RequestRetransmit` per sweep behind the STRICT 0.6 s gate on the ONE shared timestamp (`SharedNet::EnqueueNaks @ 0x00543BD0` — the 0x41-mask x87 test at 0x00543C03 proceeds only on strictly-greater, contrast the ack's >=), body u32 count + ids ascending capped at 114 (`ReceiverData::GetNaks @ 0x005490C0`, cap 0x72), borrowed sequence, never an ack in a NAK sweep. Control-header rule decided for BOTH emissions: `Time` = interval id, `Iteration` = session iteration, per the shared retail header build (`FlowQueue::TransmitNewPackets @ 0x00547A60` @ 0x00547A84); ACE reads neither. THE design piece: the AD-51 reclaimed-word pool in `InboundSequenceTracker` closes the ACE cleartext-reject keystream hazard the N2 ledger row recorded — ACE's `RejectRetransmit` consumes a fresh cleartext sequence with NO keystream word, so the gap walk mis-parks a word for it and the whole inbound stream runs one word ahead. On a VALIDATED cleartext reject (`WorldSession` calls `OnCleartextRejectSequence` post-checksum), the tracker removes the mis-park, bubble-shifts every later-drawn parked word down one position (per-word draw ordinals; ascending id ⇔ ascending draw order), and pools the excess for the next fresh draws, consumed lowest-draw-order-first — exact for any number of interleaved rejects in any arrival order (a plain FIFO is NOT: reject-after-higher-arrival crosses the parked chain, and dual out-of-order rejects pool out of draw order — both pinned by tests). Reject BODY ids keep N2's discard (word-bearing server-side, consumed-in-place). N3 advisories all folded: honest transitional wording (above), `ReceiverData::SharedInit @ 0x00548EF0` (from `Init @ 0x00548FA0`) citation, `FlowQueue::Empty` pump-order comment (TransmitNaks → TransmitAcks → TransmitNewPackets, interval increment LAST @ 0x00548A9D; our clock-first order is cosmetic vs ACE), the Time/Iteration rule, and the stale `WorldSession` budget-break comment. Gate arithmetic hardened: gate ticks now round (0.6 has no exact double; truncation opened the strict gate AT the boundary). 737 Core.Net tests green, including: strict-gate boundary, shared-timestamp both directions, NAK-suppresses-ack, full wire-shape + 114-cap pins, model-served retransmission round trip, five tracker reclaim proofs, the 130-s virtual prune → fresh-sequence reject system test (victim abandoned, later traffic decodes, pool drains to zero), 10 s long-loss (NAKs on the gate cadence, zero acks, heal inside the window), and the capstone soak: 2% seeded bidirectional loss × 10,000 messages → zero message loss both ways, ACE crypto headroom 256 at convergence with a ≥250 no-erosion floor mid-flight, NAK set / reclaim pool / pending resends / ACE out-of-order buffer all zero, cache at the single watermark entry (retail Flush prunes STRICTLY below the ack). Soak notes: ACE never NAKs a quiet client (§3 row 1), so convergence keeps a C2S trickle flowing — a real idle-client tail loss heals only on the next action, an ACE constraint outside N4's scope. | | N5 | complete | `4e290f00` — **Opus review PASS** (structural absence + arming gate + un-gameability + teardown all verified; ledger arithmetic reconciled). Acceptance folded in the review's gate strengthenings: per-direction recovery conjunction + the three keystream-health invariants (cksum-fail/sanity-drop/uncached-nak == 0) + the EnterWorldBody unrecoverable-tail caveat. Revert: `git revert 4e290f00d86ebd320d2da609fd7cd15d7175c4f0`. Test-collection hygiene (static NetDiagnostics mutation) folds into N6 | Loss observability + the LossyTransportDecorator + the connected loss gate — the permanent removal of the loopback blindness (§1). `[net-tick]` gains `resend/s nak-out/s nak-in/s rej-in/s dup-drop/s parked/s reclaim/s cache= nakset=` (window deltas mirroring acks/s; `TransportStats` gains `RejectsReceived`; string work probe-gated, counters unconditional) and `WorldSession.Dispose` emits one cumulative `[net-final]` totals line so the gate asserts exact counters, not rounded rates. N4-review F3 folded: fresh reliable sends stamp `Iteration` = the session iteration through the same shared retail header build already cited for `Time` and the N4 control packets (`FlowQueue::TransmitNewPackets @ 0x00547A60`, the build at 0x00547A84/0x00547AA8) — the control-header rule now holds across all three send shapes; ACE reads neither field inbound. `Transport/LossyTransportDecorator`: deterministic seeded per-direction loss (`ACDREAM_NET_DROP_PCT`/`_SEED`/`_DIR` via `NetDiagnostics` typed properties, Rule 5), armed only after the first ENCRYPTED outbound datagram is forwarded (parse-free flags-word check — the cleartext handshake always survives; handshake loss belongs to N6), structurally absent at 0% (`WrapIfConfigured` returns the raw transport; the default factory is the only production seam). The logoff-confirmation wait now runs the transport sweep — retail's pump (`Client::UseTime @ 0x00411C40`) never stops before `LogOffServer`, and the loss gate exposed that a dropped S2C confirmation was gap-detected but never NAKed during `Dispose`. `tools/run-connected-loss-gate.ps1` (default 2%/seed 1) runs the standard lifecycle route through the decorator vs local ACE and FAILS unless `[net-final]` shows resends>0 OR nak-out>0 OR nak-in>0 AND the decorator's own dropped ledger is non-zero — a loss gate that never dropped proves nothing, asserted explicitly. The lifecycle gate defensively clears the drop vars (decorator-absent baseline). **First loss-observing gate evidence (2026-07-29, 2%/seed 1, local ACE):** decorator dropped out=3 in=10 of forwarded out=183 in=496; `[net-final] resends=2 nak-in=2 nak-out=6 rej-in=0 acks-out=114 acks-in=119 dup-drop=0 sanity-drop=0 cksum-fail=0 parked=9 reclaimed=0 uncached-nak=0 cache=1 nakset=0` — both recovery directions fired on a real connected route (ACE NAK → cached resend; client gap-walk park → NAK → ACE retransmit), all six checkpoints validated, graceful logout confirmed, ACE recorded the transport Disconnect, RESULT=PASS. The gate immediately paid for itself: it exposed that the Dispose logoff-confirmation wait processed inbound but never swept the transport, so a lost S2C confirmation could be gap-detected yet never NAKed — fixed by running the sweep in that third blocking pump (retail's `Client::UseTime @ 0x00411C40` pump runs until `LogOffServer`). Known tail caveat recorded in the gate header: a drop landing on the single-shot logoff request or transport Disconnect (~pct each) is unrecoverable by ACE's arrival-driven NAK design (§3 row 1) — rerun with another seed, never widen teardown tolerances. #261 filed for `LinkStatusSnapshot.PacketLossPercentage` (retail `CLinkStatusAverages` formula required; inventing a ratio forbidden). 747 Core.Net tests green (decorator determinism/direction/arming/structural-absence, the 5% seeded WorldSession lossy lifecycle with zero message loss + Headroom 256, `[net-tick]` field pins, Iteration-stamp pins). | -| N6 | complete | SHA recorded at closeout | ConnectResponse handshake retransmit + fragment-assembler eviction — the final implementation slice. **Retransmit:** while unconfirmed, the Connect character-list pump resends the IDENTICAL cleartext ConnectResponse (same sequence 1, same cookie, the one encoded datagram — no new outbound state) on retail's strict 0.333333333 s gate (`ClientNet::ProcessConnection @ 0x00545450` case `cs_ConnectionRequestAcked` at 0x0054547B, the constant at 0x00545481, the mask-0x41 strictly-greater test at 0x0054548C; `ClientNet::SendConnectAck @ 0x005440F0` re-stamps `lastSentHandshake_` at 0x00544102 and rebuilds the same cookie packet). Confirmation = the first checksum-valid post-negotiation packet without the ConnectRequest flag, retail's `cs_ConnectionRequestAcked → cs_Connected` edge (`ClientNet::ProcessPacket @ 0x00545100`: the 0x40000 exclusion at 0x0054514E, `SetConnectionState(..., 5)` at 0x00545160). The cadence rides the TransportClock (virtual-clock testable via `TransportClockSource`); the Connect deadline stays wall-clock. ACE-safety pinned against the N0 model: a duplicate while still `AuthConnectResponse` re-routes idempotently through NetworkManager's pre-route; after acceptance `CheckState` clause 2 drops it pre-CRC at zero keystream cost. Pre-N6 a lost ConnectResponse was a hang to the Connect deadline — routine on a real path, and the N5 decorator deliberately arms after this window, so nothing covered it. **Assembler eviction (AD-52):** partials evict 60 s after their last accepted fragment (re-stamp-on-update per retail `ArrivedEphInfo::UpdateNetBlobID @ 0x0054AE00`), swept from `ReliableTransport.Sweep` on retail's 5 s flush cadence (`Indicator::FlushTimedOutEphInfo @ 0x0054A3D0`, gate at 0x0054A3DC; per-entry `fTimedOut @ 0x0054AE30`) — N4's RejectRetransmit abandonment had made an unrecoverable partial a reachable permanent state. A 64-entry completed-sequence ring drops late duplicates of already-completed messages instead of re-partialing them (the completed-then-duplicate leak). 60 s is a floor, never a tunable to shrink. **Fold-ins:** N5-review LOW-5 — `NetProbeTests` + `LossyTransportDecoratorTests` (the static-`NetDiagnostics`/`Console.SetOut` mutators) share one `DisableParallelization` xunit collection so they never run alongside classes constructing `WorldSession`. 757 Core.Net tests green (drop-first-ConnectResponse-retry-heals with exactly one retry and none after confirmation; clean handshake sends exactly one; server-responses-lost retries drop harmlessly at the model while the N2/N4 gap-walk → NAK → cached-retransmit path heals the handshake responses; model-level duplicate-after-acceptance CheckState pin; assembler TTL floor boundary/refresh-on-update/ring-drop/ring-bound; transport-level sweep eviction). Connected lifecycle gate PASS (capped six-checkpoint route + uncapped reconnect, graceful exits, zero failures). The N5-STRENGTHENED loss gate PASS on its first live run (2%/seed 1, local ACE): decorator `[net-loss] dropped out=3 in=10 forwarded out=181 in=496 armed=True`; `[net-final] resends=1 nak-in=1 nak-out=5 rej-in=0 acks-out=114 acks-in=116 dup-drop=0 sanity-drop=0 cksum-fail=0 parked=8 reclaimed=0 uncached-nak=0 cache=1 nakset=0` — the per-direction recovery conjunction held (C2S: resends+nak-in > 0; S2C: nak-out > 0), all three keystream-health invariants zero, ledger converged at the single watermark cache entry. | +| N6 | complete | `f9c5e47e` — **Opus review PASS** (all six retail address claims verified incl. the three distinct gate strictness masks; packet-identity, latch, TTL, and ring all clean). Acceptance settles LOW-2 (TS-58/TS-59/TS-60/TS-61/AP-126 filed) + INFO-4 (DropAll resets the completed ring) + INFO-5 (the post-acceptance retry drop is NetworkManager's pre-route, not CheckState — same silent pre-CRC outcome). LOW-1 noted: retry cadence dilates to ~0.5 s via the pump's 250 ms receive granularity — mechanism/constant/strictness retail-exact, fewer retries than retail, harmless. Revert: `git revert f9c5e47e`. | ConnectResponse handshake retransmit + fragment-assembler eviction — the final implementation slice. **Retransmit:** while unconfirmed, the Connect character-list pump resends the IDENTICAL cleartext ConnectResponse (same sequence 1, same cookie, the one encoded datagram — no new outbound state) on retail's strict 0.333333333 s gate (`ClientNet::ProcessConnection @ 0x00545450` case `cs_ConnectionRequestAcked` at 0x0054547B, the constant at 0x00545481, the mask-0x41 strictly-greater test at 0x0054548C; `ClientNet::SendConnectAck @ 0x005440F0` re-stamps `lastSentHandshake_` at 0x00544102 and rebuilds the same cookie packet). Confirmation = the first checksum-valid post-negotiation packet without the ConnectRequest flag, retail's `cs_ConnectionRequestAcked → cs_Connected` edge (`ClientNet::ProcessPacket @ 0x00545100`: the 0x40000 exclusion at 0x0054514E, `SetConnectionState(..., 5)` at 0x00545160). The cadence rides the TransportClock (virtual-clock testable via `TransportClockSource`); the Connect deadline stays wall-clock. ACE-safety pinned against the N0 model: a duplicate while still `AuthConnectResponse` re-routes idempotently through NetworkManager's pre-route; after acceptance `CheckState` clause 2 drops it pre-CRC at zero keystream cost. Pre-N6 a lost ConnectResponse was a hang to the Connect deadline — routine on a real path, and the N5 decorator deliberately arms after this window, so nothing covered it. **Assembler eviction (AD-52):** partials evict 60 s after their last accepted fragment (re-stamp-on-update per retail `ArrivedEphInfo::UpdateNetBlobID @ 0x0054AE00`), swept from `ReliableTransport.Sweep` on retail's 5 s flush cadence (`Indicator::FlushTimedOutEphInfo @ 0x0054A3D0`, gate at 0x0054A3DC; per-entry `fTimedOut @ 0x0054AE30`) — N4's RejectRetransmit abandonment had made an unrecoverable partial a reachable permanent state. A 64-entry completed-sequence ring drops late duplicates of already-completed messages instead of re-partialing them (the completed-then-duplicate leak). 60 s is a floor, never a tunable to shrink. **Fold-ins:** N5-review LOW-5 — `NetProbeTests` + `LossyTransportDecoratorTests` (the static-`NetDiagnostics`/`Console.SetOut` mutators) share one `DisableParallelization` xunit collection so they never run alongside classes constructing `WorldSession`. 757 Core.Net tests green (drop-first-ConnectResponse-retry-heals with exactly one retry and none after confirmation; clean handshake sends exactly one; server-responses-lost retries drop harmlessly at the model while the N2/N4 gap-walk → NAK → cached-retransmit path heals the handshake responses; model-level duplicate-after-acceptance CheckState pin; assembler TTL floor boundary/refresh-on-update/ring-drop/ring-bound; transport-level sweep eviction). Connected lifecycle gate PASS (capped six-checkpoint route + uncapped reconnect, graceful exits, zero failures). The N5-STRENGTHENED loss gate PASS on its first live run (2%/seed 1, local ACE): decorator `[net-loss] dropped out=3 in=10 forwarded out=181 in=496 armed=True`; `[net-final] resends=1 nak-in=1 nak-out=5 rej-in=0 acks-out=114 acks-in=116 dup-drop=0 sanity-drop=0 cksum-fail=0 parked=8 reclaimed=0 uncached-nak=0 cache=1 nakset=0` — the per-direction recovery conjunction held (C2S: resends+nak-in > 0; S2C: nak-out > 0), all three keystream-health invariants zero, ledger converged at the single watermark cache entry. | diff --git a/src/AcDream.Core.Net/Packets/FragmentAssembler.cs b/src/AcDream.Core.Net/Packets/FragmentAssembler.cs index 0041e65c..877ee295 100644 --- a/src/AcDream.Core.Net/Packets/FragmentAssembler.cs +++ b/src/AcDream.Core.Net/Packets/FragmentAssembler.cs @@ -277,8 +277,20 @@ public sealed class FragmentAssembler return evicted; } - /// Discard all in-flight partial messages. - public void DropAll() => _inFlight.Clear(); + /// + /// Discard all in-flight partial messages AND the completed-sequence + /// ring. The ring reset closes a latent trap (N6 review INFO-4): a + /// future session-reset caller would otherwise retain completed marks + /// while ACE restarts its fragment sequence at 0, false-dropping the + /// new session's first messages — exactly the failure class the ring's + /// zero-slot guard exists to prevent. + /// + public void DropAll() + { + _inFlight.Clear(); + _completedNext = 0; + _completedCount = 0; + } private bool WasRecentlyCompleted(uint sequence) {