mirror of
https://github.com/jkdansereau/goldeneye-pc-port.git
synced 2026-10-24 00:02:45 +02:00
James Dansereau
e141e59316
Part (b) of D194 ("sensitivity correlates with movement/frame rate"):
input.c converted raw accumulated mouse px directly to stick counts,
treating "one inputComputePad(0) drain" as a fixed-size time unit even
though the real interval between drains drifts with render/scene load.
Same physical hand motion -> different emitted turn rate depending on
how ragged the poll cadence gets.
Fix: measure real elapsed time between drains with a monotonic clock
(SDL_GetPerformanceCounter) and rescale the accumulated px by
(1/60s / dtActual), clamped [0.25, 4.0], before MOUSE_TURN_GAIN/AIM_GAIN
apply -- so "px this poll" always means "px for a nominal 1/60s tick"
regardless of actual poll-to-poll timing. 1/60s matches D193's own
measured steady-state sim tick rate. Applied only to the aim-mode and
hipfire gameplay-look branches; menu-cursor paths (D165/D169's
poll-count-based P-controller) are untouched.
New Input.MouseDtDecouple config knob (default 1, escape hatch to the
prior frame-coupled behaviour) + a GE_INPUTLOG lookdt trace line.
Part (a) (aim-mode response curve) and the CONTROLLER_CONFIG_SOLITARE
controldef switch are deliberately NOT touched here -- both are design
decisions (D238 overlaps significantly with part (a)) that should get
the user's own interactive feel-testing rather than be baked in blind.
Build green (241/241), verify.sh bunker1 PASS (crash-free). Cannot be
feel-verified headlessly -- needs a human playtest (A/B against
Input.MouseDtDecouple=0).
Claude-Session: https://claude.ai/code/session_016TaKy6AAsVUTfjkhrX6CjG