src/drv_leia_android/leia_cnsdk.cpp:2817-2830 (82adec6, 2026-09-02) emits a U_LOG_W de-duped on residual_ns/1e6 — a jittery measured latency. Consecutive frames print 35/40/41/45/50 ms, so the de-dupe never holds and the comment "never per frame" is false.
Measured on a K68 today: ~40 lines/s. It rolled the logcat main buffer in ~2.5 minutes and destroyed the startup evidence (HW3D_DBG / HW_DBG_CNSDK mode lines) a live investigation needed — the runtime convention (docs/reference/debug-logging.md) is that U_LOG_W is for one-off init/error/lifecycle events and per-frame WARN is never acceptable.
Fix: a hysteresis band (log only when the value leaves ±N ms of the last logged value) or a rate cap, or demote to U_LOG_I per the convention.
src/drv_leia_android/leia_cnsdk.cpp:2817-2830(82adec6, 2026-09-02) emits aU_LOG_Wde-duped onresidual_ns/1e6— a jittery measured latency. Consecutive frames print 35/40/41/45/50 ms, so the de-dupe never holds and the comment "never per frame" is false.Measured on a K68 today: ~40 lines/s. It rolled the logcat main buffer in ~2.5 minutes and destroyed the startup evidence (
HW3D_DBG/HW_DBG_CNSDKmode lines) a live investigation needed — the runtime convention (docs/reference/debug-logging.md) is thatU_LOG_Wis for one-off init/error/lifecycle events and per-frame WARN is never acceptable.Fix: a hysteresis band (log only when the value leaves ±N ms of the last logged value) or a rate cap, or demote to
U_LOG_Iper the convention.