Gledopto GL-SPI-206P (led_controller_spi_tuya) — runtime fix + Athom status (2026-10-10)
Sources (rewritten, credited): zigbee-herdsman-converters src/devices/gledopto.ts (Koen Kanters and contributors) GL-SPI-206P + tzLocal.glspi206p_light; Koenkk/zigbee2mqtt#32754 (bursts freeze the MCU).
Fingerprints (verified, unchanged)
ZHC fingerprint: TS0601 + _TZE204_8fffc3kb, _TZE284_gt5al3bl, _TZE28C1000000_gt5al3bl. Nothing invented. Couples also stay on switch_1gang (reviewed exception in data/native-matrix-reviewed-duals.json).
Runtime fixes (master)
- One EF00 frame per change (
lib/tuya/TuyaMultiDpFrame.js): seq16 +[dp,type,len16,data]..., like ZHCsendDataPoints. Per-DP max fallback only when no single-frame path exists (was: one frame per DP, each through a 6-path cascade = the burst Z2M warns about). - Profile lists all 6 manifest capabilities: base
TuyaDpProfileDeviceused to REMOVElight_temperature/light_modeat each boot (then device re-added them) -> broken Flow cards / insights. - Brightness in colour mode goes inside DP61 (DP3 alone is ignored in colour mode); white mode sends DP4 + DP3 together (as ZHC). dim 0 -> DP1 off (DP3 floor 10). DP2 written only when the mode really changes.
- RX: DP2 work mode ->
light_mode, DP4 ->light_temperature(0 = warm). DP61 stays write-only (h/s in memory + store). - onoff TX through the coalescer too (DP1 in the same frame). Timers are Homey-managed (DpCoalescer).
driver.flow.compose.jsonexplicit empty{triggers:[],conditions:[],actions:[]}(the old fleet-enrich trigger with{brightness}titleFormatted was dropped earlier, e777baa3b6).- energy.approximation (usageOn 15 / usageOff 0.5), no fake measure_power.
Athom status — IMPORTANT
- Master #3480 v9.0.1361 (run 38070570358) WITH the hold active (
[publish-hold] held back 1 driver(s) [led_controller_spi_tuya]) = processing_failed AggregateError. So this driver is NOT the (only) cause on master. Earlier: Test D #3463 (master without the driver) also failed. - J4 #3479 (I + this driver) lacked lib/tuya/DpCoalescer.js, so it is not clean proof against the driver.
- Local payload simulation (Node 22, prepare-publish, hold removed): driver manifest is well formed, 6 mfr forms, no empty driver; no new unresolvable relative require vs 9.0.1330.
- Remaining I -> master payload deltas to bisect: switch_module_whd02 (+2 triggers), action sub_dim_set, energy.approximation on 18 drivers, class socket->light (dimmer_2_gang_tuya, wall_dimmer_1gang_1way).
- Hold stays until a build with this driver reaches Test.