NewNet fixes, perfomance tweaks and weapon bugs found. #12
Open
grahamslam wants to merge 25 commits intozenakuten:mainfrom
Open
NewNet fixes, perfomance tweaks and weapon bugs found. #12grahamslam wants to merge 25 commits intozenakuten:mainfrom
grahamslam wants to merge 25 commits intozenakuten:mainfrom
Conversation
Fix 1: Replace naive weighted-average ping with median-of-5 sliding window, 0.5s polling, 150ms cap, and spike rejection. Converges in ~2.5s vs 9-12s. Fix 2: Server sends damage momentum to client via ClientDamageImpulse RPC so client prediction stays in sync — prevents hiccup/snap on taking damage. Fix 3: Smooth rewind delta per PCC (max 10ms change per trace) to prevent visible position jumps when ping fluctuates between shots. Fix 4: 200ms grace period on move error corrections after damage impulse, letting stale in-flight moves expire without triggering rubber-banding. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
… history perf - Fix copy-paste bug in NewNet_LinkAltFire: TimeTravel referenced NewNet_FlakCannon instead of NewNet_LinkGun for mutator lookup - Fix fake projectile timer clobbering on rapid fire: track pending state with bFakeFirePending, fire pending effect before overwriting (NewNet_RocketFire, FlakFire, FlakAltFire, BioFire, LinkAltFire) - Fix forward-iteration removal bug in NewNet_LinkFire and UTComp_LinkFire: LockingPawns cleanup skipped consecutive None entries, now iterates backwards - Fix O(n^2) RemoveOutdatedHistory: count removals first, single Remove call instead of per-element loop shift - Fix AddHistory: use Length++ instead of Insert at end for InterpCurve points - Update build.bat and make.ini for local build layout with C:/UT2004 base asset paths Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Fix Y/Z axis assignment in DoTimedClientFireEffect: all three axes were set to OldXAxis, breaking projectile spawn orientation for delayed fake projectiles on higher ping (rockets, flak, bio, link alt). Add None guard after FindFPM() in CheckForFakeProj to prevent crash when FakeProjectileManager doesn't exist (rocket, link, shock, seeking rocket projectiles). Update make.ini paths for local build layout. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Reset PCC rewind smoothing state in TurnOffCollision so each flak chunk in a salvo traces against the same rewound positions. Previously the smoothing accumulated across 9 chunk traces, causing drift. Add missing FPM None guard in NewNet_FlakChunk.CheckForFakeProj. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…traces The per-PCC rewind smoothing (Fix 3) was clamping dt changes to 10ms per TimeTravelPawn call. This broke: - Shock beam correction loop: searches ±40ms offsets in 20ms steps, but smoothing prevented reaching the target offsets - Flak multi-chunk traces: 9 chunks with independent TimeTravel cycles caused accumulated drift The smoothing was misplaced — PCCs are invisible server-side trace actors, not rendered. The ping estimator median filter (Fix 1) already smooths at the right layer (client ping measurement). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Pre-existing bug: AllowFakeProjectile/RegisterFakeProjectile called without None check after FindFPM in DoClientFireEffect and DoTimedClientFireEffect across multiple weapon fire classes. Fixed in: NewNet_RocketFire, NewNet_FlakFire (2 instances), NewNet_FlakAltFire, NewNet_BioFire, NewNet_LinkAltFire, NewNet_RocketMultiFire. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
High ping players rewinding low ping players 150-200ms+ back caused "shot around corners" — the target had moved well past that position on their screen. Cap rewind at 75ms (half of 150ms RTT target) in TimeTravelPawn. Players at 150ms or below get full compensation. Above that, they need to lead their shots for the excess. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
During lag spikes, the unreliable ClientDamageImpulse RPC could be dropped. For shield jumps, this meant the client never learned about the upward momentum — it predicted staying on the ground, then got corrected to an airborne position, causing the "invisible ceiling" snap. Making it reliable ensures the impulse always arrives, even during packet loss. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When a player fires, switches weapon before the fake timer fires, then switches back and fires again, the bFakeFirePending check was calling DoTimedClientFireEffect with stale aim/position data from the previous shot. This spawned a ghost fake projectile from the old position. Now just clears the flag — the timer delay (max 40ms) is always shorter than any weapon's fire rate, so a valid pending effect would have already fired before the next shot. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ClientDamageImpulse now sends absolute post-damage velocity instead of a delta. Previously the client added the delta on top of its predicted velocity, causing a brief double-velocity spike on high-momentum hits (shock combo, rockets). The normal server correction then snapped the position back, producing a visible hitch. With absolute velocity, the client sets its velocity to exactly what the server has — no double-up, no delayed correction needed. Low-momentum weapons were unaffected; this fixes shock combo, rocket, flak shell, and shield gun knockback specifically. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…ocity Previous fix sent absolute velocity but the client still had stale position from frames predicted before the impulse arrived. After the grace period, the server snapped the position to correct the accumulated error — causing the hitch on high-momentum weapons. Now sends Location, Velocity, and Physics together. Client syncs all three via SetLocation + SetPhysics in one call. No position error accumulates, no delayed correction needed. The grace period still suppresses redundant corrections during the brief overlap window. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
SetLocation causes a physics reset/collision recalc hitch on the frame it's called. Reverted to velocity-only sync — the position error from pre-impulse prediction frames is small (few units over 2-3 frames) and converges naturally through continued prediction. Extended grace period from 200ms to 500ms so the server doesn't snap- correct during convergence. The velocity is correct immediately, so forward prediction is accurate — the grace period just needs to outlast the stale saved moves flushing from the queue. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Bypass the impulse replication to isolate and test the other netcode improvements (move accum fix, MaxResponseTime, MaxSavedMoves) without interference. TakeDamage just calls super — engine handles corrections natively. The impulse approaches tried so far (delta, absolute vel, full state with SetLocation) all had issues: double-velocity, physics reset hitch, or delayed snap after grace period. Need to revisit with an interpolated correction approach. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The correction pass runs when the client believes it hit but the server's initial rewind trace missed. It searches nearby rewind times to find the hit. At 20ms steps, fast-dodging targets could fall between search points — the collision copy oscillated around the real position without landing on it. 10ms steps halve the maximum position error during the search (~5 units at dodge speed instead of ~10). The extra traces only run during unreg attempts, not normal hits, so no performance impact on regular gameplay. Changed in: ShockBeamFire, SniperFire, SuperShockBeamFire. Other hitscan weapons (minigun, link, assault) don't have the correction search and are unaffected. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace Broadcast() text chat with BroadcastLocalizedMessage() using new UTComp_WinByTwoMessage (extends CriticalEventPlus). Renders centre-screen in yellow at 40% height so players with chat disabled still see it. Updated in both UTComp_GameRules (CTF/team games) and UTComp_ClanArena (round-based). Wipeout inherits via GameRules — no change needed. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…figurable rewind cap Root cause of unregs for 75-100ms players: PingDT calculates full RTT but MAX_REWIND was 75ms, so any player above 75ms ping lost rewind. At 100ms ping + dodge speed (1200 UU/s), the 25ms of lost rewind = 30 UU position error, exceeding the 25 UU collision radius — guaranteed miss on moving targets. Changes: - PawnCollisionCopy: MAX_REWIND 0.075 -> 0.240 (now a var, not const) - MutUTComp: new config var PawnCollisionMaxRewind (default 0.240), exposed in web admin under "UTComp NewNet" - PawnCollisionCopy reads PawnCollisionMaxRewind from MutUTComp in PostBeginPlay - MAX_PROJECTILE_FUDGE 0.075 -> 0.100 across 12 weapon fire classes (FlakCannon both FUDGE and FUDGE_ALT, RocketLauncher stays at 0.275) - CHANGES.md updated with full RTT analysis and V24 bump Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
UTComp_ServerReplicationInfo: add bWinByTwo, bEnableCorrectionSmoothing, CorrectionHalfLife, CorrectionJumpThreshold, MaxVisualOffset, bSmoothCameraOffset to replication block + defaultproperties. ModernPawn: smoothed render offset changes (correction smoothing system). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…ace8a8) InitializeClient() calls SetMaxSavedMoves() once at startup, but RepInfo may not have its replicated values yet at that point. Without an ongoing re-sync, clients keep the default MaxSavedMoves=300 even when the server configures a higher value, causing misleading "Exceeded max saved moves" warnings in client logs. Fix: override PostNetReceive() to re-check and re-sync MaxSavedMoves whenever BS_xPlayer's replicated state updates. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…o lookup Multiple fixes targeting movement queue overflow rubberbanding: 1. GetFreeMoveEx: stop nuking the entire SavedMoves list on overflow. Previous behaviour destroyed all in-flight prediction history when the queue exceeded MaxSavedMoves, turning a survivable overflow into a catastrophic prediction reset (massive rubberband). Now drops only the single oldest move and reuses the slot, keeping the rest of the queue. 2. UTComp_ServerReplicationInfo: set NetUpdateFrequency=10.0 Default ReplicationInfo NetUpdateFrequency is 1.0, meaning replicated state changes only push to clients once per second. 10Hz makes initial replication and any settings changes propagate quickly enough. 3. BS_xPlayer.MaxSavedMoves default: 300 -> 500 Insurance for the case where replication fails entirely. High-FPS players (1000+ fps) at 200ms ping can hit 300 with normal jitter. 4. PostNetReceive: proactively look up RepInfo if None instead of just re-syncing from an existing reference. Closes the timing window where PostNetReceive fires before any function has run the DynamicActors lookup that populates RepInfo. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…gated)
Adapts the DarkPlaces approach to UE2 movement: instead of allocating a new
SavedMove every render frame, share a single PendingMove between network
ticks. The PendingMove's Delta is extended and Acceleration is updated each
frame, while local prediction (ProcessMove + Pawn.AutonomousPhysics) still
runs every frame for visual smoothness.
Why: at high client framerate (1000+ fps), UE2 currently allocates ~1000
SavedMoves per second while only sending ~91, causing the queue to grow
~909/sec faster than it drains. This is the structural cause of the
"Exceeded max saved moves" overflow that we previously treated by bumping
MaxSavedMoves and fixing the destructive overflow logic. Coalescing fixes
it at the source — queue size becomes ~1/NetMoveDelta entries regardless
of fps.
Important input changes (jump press, duck toggle, dodge double-click,
acceleration direction change >17 degrees, non-walking physics) bypass
coalescing and force fresh allocation, preserving competitive responsiveness.
Gated behind bCoalesceMoves config (default false). Plumbing:
- MutUTComp.uc: var config bool, FillPlayInfo entry, GetDescriptionText,
push to RepInfo
- UTComp_ServerReplicationInfo.uc: var, replication block, default
- BS_xPlayer.uc: var, LastQueuedMoveTime tracking, coalesce branch in
ReplicateMove (top), SetMaxSavedMoves syncs from RepInfo,
PostNetReceive re-checks both MaxSavedMoves and bCoalesceMoves
Toggle on via web admin under "UTComp Movement" or in WSUTComp.ini under
[WSUTComp.MutUTComp]. Flip back instantly without recompile to revert.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…CK_None enum Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
PlayerController.bDuck is var byte (0/1) while SavedMove.bDuck is var bool. UE2 won't auto-convert across these types in ==. Use (bDuck != 0) to explicitly cast to bool. Restored DoubleClickMove == DCLICK_None since SavedMove.DoubleClickMove is the enum type (the existing combine logic at line 4174 uses the same expression and compiles). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
V22
NewNet ping estimator: median-of-5 sliding window, 0.5s polling, 150ms cap, spike rejection
Damage impulse replication: server sends momentum to client to prevent desync on taking damage
Rewind delta smoothing: cap rewind change per trace to prevent position jumps during ping fluctuation
Move error grace period: suppress corrections briefly after damage impulse
Fix NewNet_LinkAltFire TimeTravel referencing wrong weapon class (copy-paste bug)
Fix fake projectile timer clobbering on rapid fire (rockets, flak, bio, link alt)
Fix LockingPawns cleanup skipping consecutive None entries (forward-iteration removal bug)
Fix O(n^2) RemoveOutdatedHistory — batch removal instead of per-element shift