Why
Layer B intentionally remains a developer-machine smoke harness. A future CI gate may be feasible because HonkHonk already runs offscreen wgpu work in CI, but a real windowed wgpu application under kwin_wayland --virtual with lavapipe/llvmpipe is not yet proven reliable enough for required checks.
Spike scope
- Prototype a windowed HonkHonk wgpu launch under isolated
kwin_wayland --virtual on a Linux CI runner.
- Evaluate lavapipe and llvmpipe availability, startup stability, render correctness, runtime, and flake rate.
- Capture a screenshot and reject all-black frames.
- Record required packages and whether GPU/backend differences make the test useful.
- Recommend whether this can become an optional or required CI job.
Constraints
This issue is research-only until the developer-machine Layer B harness lands. Do not weaken or replace the real-wgpu requirement with HonkHonk’s software renderer, and do not make the experiment a required check until its reliability is demonstrated.
Why
Layer B intentionally remains a developer-machine smoke harness. A future CI gate may be feasible because HonkHonk already runs offscreen wgpu work in CI, but a real windowed wgpu application under
kwin_wayland --virtualwith lavapipe/llvmpipe is not yet proven reliable enough for required checks.Spike scope
kwin_wayland --virtualon a Linux CI runner.Constraints
This issue is research-only until the developer-machine Layer B harness lands. Do not weaken or replace the real-wgpu requirement with HonkHonk’s software renderer, and do not make the experiment a required check until its reliability is demonstrated.