[sunzenshen]

Description

Case 1: jump pose ends at the top of the jump.
Review page: exported review page (Dropbox) · live review page (opening it lets you request access with your email or Claude account).

Other players drop out of the jump pose near the top of their jump in neo and come down in the walk pose, with a walk bob of the torso as they land. Classic NEOTOKYO holds the jump pose until the landing and goes straight to the idle legs. Two causes, one commit each:

1. The jump ended on the ground flag as received. HandleJumping ends the jump when FL_ONGROUND is set, 0.2 s or more after the jump event. The flag is networked and applied as soon as it arrives, but other players are drawn at an interpolated position that trails it: logged per frame, the flag arrived 0.19-0.27 s before the drawn body reached the ground, around the top of a standing jump. On the client the jump now lasts until a short hull trace (2 u, the player's own hull) finds ground under the drawn body. If the trace finds nothing while the flag has been set for 0.5 s, the jump ends anyway, so a surface the client cannot trace never leaves a player stuck in the pose. The server's anim state and the local player keep the flag test; for the local player the predicted origin and flag agree, and the trace ends the jump on the same frame.
2. "Moving" used 3D speed. HandleMoving and ComputePoseParam_AimYaw treated any velocity over 0.5 / 1 u/s as moving, so a rising or falling player played the walk legs (also when dropping off a ledge or riding a lift). Both now use horizontal speed, as classic NEOTOKYO and the 2006 SDK do. A player on a ladder keeps the 3D test, so ladder climbing looks as before (classic's ladder look was not compared).

Independent of the other animation PRs from the review page: this touches only hl2mp_playeranimstate.{cpp,h}; the upper-body blend PR (case 2) touches multiplayer_animstate.{cpp,h}. The "standing players stay twisted" case (3) will touch ComputePoseParam_AimYaw too; whichever merges second rebases.

How to test

1. Two clients (or a bot on bot_mimic 1), watch the other player jump in place: the jump pose lasts until the feet touch the ground, then the idle pose, no walk legs or torso bob in the air.
2. Running jumps and jumps onto a higher or lower surface: the pose ends when the drawn body lands there.
3. Drop off a ledge without jumping: idle legs while falling straight down, walk legs only with horizontal speed.
4. Climb a ladder: same walk animation as before (not part of the logged pass).

Results

Same scripted session as the review page: nt_rise_ctg, mirror bot on bot_mimic 1 (NSF Assault, MPN), viewed from a listen-server client at cl_updaterate 20 / cl_interp_ratio 2 (0.1 s lerp, as classic's defaults); classic NEOTOKYO logged with cl_showanimstate, neo with a temporary client-side per-frame logger of the anim layers, ground flag and drawn height (not part of this PR). neo: Release client builds of master (f426c22) and this PR. Standing jumps: 2 in the review pass plus 4 in a jump pass.

	classic	master	this PR
Jump length, standing jumps	0.54, 0.55 s	0.37-0.43 s	0.52-0.61 s
pose ends, relative to the drawn landing	not logged	0.19-0.27 s before	on the landing frame
legs, 20 frames after Jump	Idle_lower 20	walk_lower 6-7, then Idle_lower	Idle_lower 20
torso >= 0.5 walk, 20 frames after Jump	0	6-8	0
running jumps (6), Jump length	0.58 s (1)	0.35-0.44 s	0.50-0.74 s (lands on lower or higher ground)

In the review pass the running jump lands on a 12 u step in neo, so it is short in both neo builds (0.43 / 0.44 s); with this PR it ends where the drawn body meets the step. The local player's own jump (thirdperson) is 0.56-0.57 s in this PR; the bug did not affect it.

Unchanged, checked on the same pass: crouch / stand timing, fire and reload gestures, upper-body weights while walking and peeking, standing breathing. A Debug client with 10 bots on ntre_rise_ctg (Recon crouch jumps of 57-63 u held until landing) logged no new assert (only the known neo_root.cpp (378) localizedWszStr at startup).

Not changed: the jump pose starts 0-0.16 s after the drawn lift-off (the jump event fires when it arrives, not interpolated), the same as master. Ladders were not filmed; the ladder path keeps master's test.

GIF (classic / master / this PR, standing jumps): case1-jump-land.gif

How this was compared

The code came first: NT;RE's player animation code was read against the 2006 Source base class that classic NEOTOKYO is built on (still in this tree as shared/base_playeranimstate.cpp); classic's client.dll was read only for class names, console settings and constants. Then both games logged every frame (classic's built-in cl_showanimstate_log, an equivalent harness log in NT;RE) on the same map and spot, with the same class, weapon and scripted keys, and the same scripts read both logs. The clips only illustrate; the numbers above come from the logs.

[sunzenshen]

Comparison of the classic/master/fixed versions (Click the gif to animate):

[sunzenshen]

#2244 might actually accomplish the same and better than this, so waiting of the result of that review before deciding what to do with this PR.

(This PR edits hl2mp_playeranimstate.*, which #2244 no longer uses and could even remove from the build.)