Files
Neutron/docs/gameplay/jimmy-air-motion.md
T

10 KiB
Raw Blame History

Jimmy ordinary airborne movement audit

2026-11-04. User prioritizes stable Jimmy movement and explicitly preserves the current GamePort feel over JN-Engine's replacement movement. JN-Engine's saved original decompiler output was used only to locate/interpret original routines; its C/Godot movement was not used as a behavioral reference. Authoritative source is the immutable project EXE SHA256 f5281076e1a2492f2d61f87aae3f4544b6d0f535ebe33f9fbb6f78710fe1aea8.

Recovered contracts and concrete mismatches

Addresses below are original EXE VAs; independently mapped file offsets equal VA−400000. tools/binary/jimmy_air_motion.py guards source identity, five complete bounded ranges and decoded absolute relocation operands. Source instruction extracts are retained in artifacts/jimmy-air-audit/{input,apply}.asm.

Original VA / offset Original behavior Prior port mismatch / bounded correction
439900..439A69 /39900..39A69 Mode1 keyboard Left precedes Right, changes stored turn by100dt; no-key return toward0 occurs only when complete6E6 keyboard-turn latch is set. Key turning sets latch; mouse turning clears it. AirSteering always applied return, without storing the latch. JimmyAirMotion.Input owns explicit latch state; caller must retain it and share it with other input modes.
439AE6..439B69 /39AE6..39B69 Up adds30 only when scalar<maximum, then clamps. Existing overspeed is preserved. RecoveredMotion.AddSpeed clamped overspeed down to maximum. New helper preserves it. Ground correction is investigated separately.
439B69..439BA2 /39B69..39BA2 Without throttle, coast only above maximum*float.909090936; compare uses the unrounded x87 product. Coast subtracts17.5 per callback. Float-rounded C# comparison misses a representable threshold case. At max600, speed545.45459 must coast to527.95459. New helper uses double intermediate for the comparison.
439BA2..439BED /39BA2..39BED Independently test body218 support, even with stored air mode1. Supported positive scalar subtracts17.5 with no crossing clamp; nonpositive scalar resets0. Old AirSteering had no support argument. Supported speed300+Up ends312.5; speed5 with no keys ends−12.5. These affect landing/rejump and retained input phases.
439C1D..439CB2 /39C1D..39CB2 Down after throttle/coast/support subtracts35 supported or17.5 airborne, clamps positive scalar to0; negative scalar adds35 toward0. Old helper used17.5 for positive scalar regardless support. No reverse-flight initiation is added.
439D81..439DB7 /39D81..39DB7 Ordinary mode1 turn caps±30, distinct from Bubble's±50. Existing cap is correct and retained.
437FD0..438011 /37FD0..38011 Read incoming live velocity. Exact zero magnitude skips independent yaw AND lateral computation. It still publishes forward scalar afterward. LabSimulation gated yaw but computed side unconditionally. Zero incoming velocity/scalar300/turn30 produces(0,0,300), not a lateral launch.
438011..438062 /38011..38062 For nonzero velocity, call334 AddEuler(0,turndt,0), then side=turnfloat(1/55)*(1−friction)*scalar, without dt on side. Existing side formula is generally right but ungated. Plain yaw modulo omitted full-axis AddEuler normalization. New helper reuses original-leaf-compatible RocketOrientation.Add.
438062..4380E6 /38062..380E6 Rotate the lateral contribution around current yaw using original OMT table values. Retain yaw-only lateral contribution. No invented normalization or clamp.
4380E6..438124 /380E6..38124 game38C rotates local(0,0,scalar) by full Euler. Sum its X/Z with lateral contribution and preserve incoming Vy, discarding rotated forwardY. Old port used flat yaw only. Nonzero pitch must scale horizontal forward; roll does not alter this local-forward vector. Use original-compatible rotate/lookup helper.

The probe complete-object fields are6C8 scalar,6D4 turn,6D8 applied turn, 6BC acceleration step30,6C0 decay35,6C4 maximum600,6E4/6E5 throttle/backward, 6E6 keyboard-turn latch,7C4 movement mode1. Its game interface is complete+C0. Game Euler is224/228/22C; getVelocity2BC/setVelocity2C4, friction248, support218, getEuler328, AddEuler334, local-vector transform38C are separate calls. Synthetic probe binds the last three to original4723F0/472480/472B10. Real class wrapper 334 may add gameplay gates; source publication/controller ownership remain caller responsibilities and are not reconstructed by this isolated component.

Keyboard steering and scalar increments retain current signs, precedence, source per-callback30/17.5/35 values and mode1 turn cap. Existing ordinary flat-air movement with nonzero incoming velocity is preserved. None of these corrections adopts JN-Engine's tuned jump650/gravity1600 or instant tank movement.

Native observations and REA

tests/GamePort.JimmyAirTests/original-air-reference.json commits35 copied-original observations and source/generator/shared-map-helper/probe-EXE/output-log hashes: 22 mode1 keyboard input cases and13 velocity-prefix cases. Coverage includes Up/Down combinations, overspeed, exact coast boundary, supported positive/negative crossings, turn cap/precedence/latch/return crossing,60Hz, zero/rising/falling live velocity, slope pitch, roll, Euler seam/negative angle normalization, friction0, reverse scalar, zero scalar and ordinary60Hz output.

Probe executes copied439900 through439DB7 (before orientation/camera work) and complete-entry437F90 through438124 (before jump suffix). It supplies synthetic actor/body/friction/mouse callbacks and uses the copied project's original OMT trig global plus original Euler helpers. Zero mouse/controller input is explicit. 439900's continuation is replaced by a stack-correct return, and437F90's suffix is replaced by its original4387A1 epilogue. No original game entrypoint, physics, collision, camera, animation or audio is executed. Source jump suffix438148 and effect421FB0 are independent parent investigations.

The isolated Wine32 run exits0 and prints JIMMY_AIR_MOTION_COMPLETE; capture.json and wine.log are retained in artifacts/jimmy-air-audit/probe. Thirty-five saved outputs match two managed conformance groups. Two Python groups guard source, relocations, PE offsets, complete fixture counts and generator provenance. This is original component execution, not whole-original or native Windows gameplay.

REA/Ghidra12.1.4 was invoked through the configured local scripts/rea.sh on the generated target copy. The default profile did not recognize439900 as a procedure; the complete explicit mapped-selector failure, export and closed snapshot are retained under artifacts/jimmy-air-audit/rea-input. No pseudocode success or Evidence ID is attributed to that failure, and the unchanged selector was not repeated. Earlier Bubble movement attempts had provider startup failures and do not establish successful decompilation either. Supporting source instructions, whole-source identity and new native observations establish the bounded arithmetic. JN's docs/decomp/evidence/c3dplayer_movement_target3.md supplied an original-code reading lead, independently checked against these project originals.

Integration API and remaining limits

New JimmyAirMotion is an isolated source-backed implementation. Parent owns live integration and existing AirSteering/LabSimulation files:

  1. Pass Input(new(Speed,TurnRate,keyboardTurnLatch),input,Grounded,dt,MaximumSpeed). Retain returned scalar, rate and latch. Complete6E6 is shared with ground and Bubble input: update it when those source paths set it; do not reset each frame.
  2. Call Apply(Speed,TurnRate,new(Pitch,Yaw,Roll),Velocity,dt,friction) in the ordinary air branch, after existing facing/mode selection and before438148 jump suffix. Publish returned full Euler and velocity. The incoming velocity gate must use prior live velocity, not a newly fabricated forward vector.
  3. Keep body-first/fresh-vs-retained order, ordinary landing/reset, jump/jetpack state/effect, raw/effective control and special modes in their existing owners.

This does not implement mouse/controller axes, general camera/pose targets, reverse-run helper states, complete full-Euler/render root behavior, original cadence, dynamic/special contacts or the jump state suffix. Fixed60Hz and source body vs capsule admission remain existing port policies. Ordinary input requires 0<dt<1 because the original divides mouse deltas by dt. Native Windows and whole-original movement/trajectory comparison remain unverified.

Destination-Euler follow-up (2026-11-05)

REA's complete retained objdump Evidence ev_0d64cc58fa8f5573fc8e61e0f2c20e36288099212bd2257d3a3121361fd6c904 adds the next target write. 439DB7..439E33 (offset39DB7; range end39E33) reads accepted forward6E4 before backward6E5, writes pitch7EC or7F0, yaw=currentEulerY and roll=−turn. Neutral pitch is0. Constructor437445/ 43744F (offset37445/3744F) initializes7EC=−40,7F0=10; ordinary437FBC/437FC6 reasserts those values. Bubble43880A/438814 substitutes−10/10. These are retained controller fields, separate from current Euler and destination Euler.

tools/binary/jimmy_air_targets.py extends the previous input probe through 439E33, supplying a synthetic36C destination writer. Twelve native observations are retained in tests/GamePort.JimmyAirTests/original-air-targets-reference.json, including both arrows (Up wins), turn signs, supported zero/overspeed, zero and custom pitch seeds. Isolated capture exits0 with JIMMY_AIR_TARGETS_COMPLETE; source range, generator, executable and output hashes are recorded. No camera, facing interpolation, body or original whole-game entrypoint ran. A Python guard checks source/provenance and saved writes; this is recovery evidence, not a managed/runtime implementation claim.

Live ordinary-air pitch targets are deliberately still the baseline simplification. The source-backed input/velocity helper is integrated, but enabling this target requires the full Euler/actor-root visual bridge and retained mode-field ownership to move together. Otherwise horizontal velocity would gain pitch attenuation while the ordinary rendered Jimmy remains upright. Follow-up must retain source 7EC/7F0 across Bubble→ordinary and fresh/retained callback order, then verify model pivot, takeoff, turning, landing and route returns. This is an identified original mismatch, not a recovered behavior that is already live.