mulciber-runtime

Experimental game-loop timing and input coordination for Mulciber.

mulciber-runtime combines a fixed-rate simulation clock, render interpolation, bounded hitch catch-up, input snapshots, rendering suspension, and presentation-cadence pacing. It consumes platform events without owning the native event pump or game state.

Frame flow

Forward platform events to Runtime::handle_window_event, drain presentation feedback with Runtime::record_presented, then call Runtime::begin_frame once per rendered frame. The returned FramePlan specifies the exact number and duration of fixed updates plus the interpolation fraction for rendering between the previous and current simulation states.

Frame deltas follow observed presentation cadence when fresh feedback is available and fall back to wall-clock timing otherwise. Catch-up work and accepted frame time are bounded so a hitch cannot create an unbounded simulation spiral; discarded time remains visible through diagnostics.

Input semantics

InputSnapshot exposes held controls and pressed/released transitions with focus-loss clearing. Transient presses, releases, and scroll samples remain latched through render-only frames and are consumed after a frame schedules fixed simulation work. This prevents one-shot input from being lost when the display rate exceeds the fixed simulation rate.

Transitions belong to the whole simulation-bearing frame rather than individual catch-up ticks. Handle an edge-triggered action once before the fixed-update loop, guarded by FramePlan::fixed_steps() != 0; every fixed update may read the latest held state.

Scope

The crate owns timing, input accumulation, presentation-cadence diagnostics, and rendering suspend/resume coordination. It deliberately does not own the platform event pump, graphics submission, collision, scene state, camera policy, process suspension, jobs, or device recovery.

The complete experimental contract and validation record live in the Mulciber repository.