fix(physics): #265/#166 - stop zeroing grounded residual velocity, wire GroundNormal

Capture bisect (docs/research/2026-07-30-265-capture-bisect.md, mined
from artifacts/matrix-session2-resolve.jsonl records 3415-3434) traced
#265's lost roof slides / permanent landing freeze and #166's missing
downhill sled to a pre-existing (2026-07-20, ten days before Campaign P
- not a regression) mechanism in PlayerMovementController.cs's grounded
quantum block: it hand-zeroed Velocity.X/Y to exactly zero every tick
once OnWalkable whenever animation root motion drives the walk (the
production graphical local-player path), discarding any residual
horizontal momentum a fall left on the body before calc_friction
(AP-7/AD-55, already correctly ported) or PhysicsBody.
UpdatePhysicsInternal's Euler integrator ever got a chance to act on it.

Two changes:

1. PhysicsEngine.cs now syncs body.GroundNormal (the vector
   calc_friction dots velocity against, per retail
   CPhysicsObj::calc_friction 0x0050ee70's `contact_plane.Normal` read)
   from the committed ContactPlane.Normal at the same commit point that
   already publishes ContactPlane. GroundNormal had zero production
   writers before this and silently defaulted to Vector3.UnitZ forever
   - even surviving velocity would have been tested against a fake
   flat-ground normal on any real slope. Core-level, so player, remote,
   ordinary, and projectile movers all benefit uniformly.

2. PlayerMovementController.cs's grounded block no longer reconstructs
   Velocity at all for the animation-root-motion case (only the
   headless/test-controller get_state_velocity fallback still does,
   unchanged). Root motion continues to fully own commanded locomotion;
   this only stops destroying whatever Velocity already holds, letting
   it compose with root motion through the same ResolveWithTransition
   sweep exactly as retail's CPhysicsObj::UpdatePositionInternal
   composes both channels.

Symptom (a), the uphill-jump bounce, traces to a SEPARATE, byte-exact
(re-verified against acclient_2013_pseudo_c.txt:282647-282760),
already-closed retail mechanism (AD-25, PhysicsObjUpdate.
HandleAllCollisions's shouldReflect gate) - confirmed orthogonal to this
fix, not addressed here (see the research doc's as-fixed addendum §9.5).

Issue265SteepSlopeCaptureBisectTests.cs gains a composed harness
(ReplayRealRoofLandingComposed) mirroring PlayerMovementController.cs's
per-tick composition against Core types only, proving: the old model
reproduces the mined freeze exactly; the new model survives the landing
and slides continuously (the real captured geometry glides at constant
velocity per retail's own dot>=0.25 early-return - AP-7); a synthetic
dot<0.25 case shows genuine exponential decay via calc_friction; and a
synthetic uphill-bounce case proves the fix changes nothing about
HandleAllCollisions's reflection decision.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
Erik 2026-07-30 18:45:05 +02:00
parent 61e959169b
commit 06c76009f1
5 changed files with 673 additions and 36 deletions

View file

@ -1865,20 +1865,46 @@ public sealed class PlayerMovementController
* tickDt;
}
if (_body.OnWalkable)
// #265/#166 (2026-07-30, docs/research/2026-07-30-265-capture-bisect.md
// §4): retail CPhysicsObj::UpdatePositionInternal (0x00512C30) composes
// BOTH channels every quantum -- the root-motion Frame just written into
// pmDelta.Origin above (commanded locomotion) AND the integrated physics
// Velocity (residual momentum: jump arcs, landing slides) -- via the SAME
// candidate position that PhysicsBody.UpdatePhysicsInternal's Euler step
// (calc_friction + v*dt, below) and the ResolveWithTransition sweep both
// see. This block used to hand-zero Velocity.X/Y to EXACTLY zero on every
// single grounded tick whenever animation root motion drives the walk (the
// production graphical local-player path since R6) -- regardless of why
// the body was OnWalkable. That discarded any horizontal momentum a fall
// or collision had just left on the body (retail settles it via
// calc_friction over subsequent ticks; PhysicsBody.GroundNormal is now
// synced to the real contact-plane normal by PhysicsEngine so calc_friction
// has real slope data to act on) before the integrator ever ran -- a mover
// that landed on a walkable roof/slope with residual horizontal velocity
// had that velocity vanish the very next tick and never moved again. Root
// motion still fully owns COMMANDED locomotion (walking/running
// displacement comes from pmDelta.Origin above, not from Velocity), so
// this does not reintroduce command- or packet-cadence-derived grounded
// translation -- it only stops DESTROYING whatever Velocity already holds.
// Ordinary walking is unaffected: Velocity is already ~0 while grounded
// with no fall/collision in flight (nothing else writes it), so removing
// this zero is a no-op on that path -- see
// GroundedRootMotion_FrictionThreshold_DoesNotHammerLocomotionTests
// (PhysicsBodyTests.cs) and Update_AnimationRootMotion_WalkSpeedUnaffected
// ByResidualVelocityFix (PlayerMovementControllerTests.cs).
//
// The headless/test-controller fallback below (no animation runtime --
// get_state_velocity's doc comment) is unchanged: it still directly
// writes the commanded state velocity into the body every grounded tick,
// exactly as before -- that model has no separate root-motion channel to
// compose with, so overwriting IS its correct per-tick behavior.
if (_body.OnWalkable && !hasAnimationRootMotion)
{
float savedWorldVz = _body.Velocity.Z;
if (hasAnimationRootMotion)
{
_body.Velocity = new Vector3(0f, 0f, savedWorldVz);
}
else
{
Vector3 stateVelocity = _motion.get_state_velocity();
_body.set_local_velocity(
new Vector3(stateVelocity.X, stateVelocity.Y, savedWorldVz),
autonomous: _body.LastMoveWasAutonomous);
}
Vector3 stateVelocity = _motion.get_state_velocity();
_body.set_local_velocity(
new Vector3(stateVelocity.X, stateVelocity.Y, savedWorldVz),
autonomous: _body.LastMoveWasAutonomous);
}
var preIntegratePos = _body.Position;