test_refused_write_settles.py silently skips HTTP-protocol control drivers #80

Activity

1. frahlg

   added

   testing

   driver-review

   on Aug 11, 2026
2. frahlg commented on Aug 31, 2026

   

   frahlg

   on Aug 31, 2026 – with Claude

   MemberAuthor

   More actions

   Status check, no code change: audited every currently-merged HTTP/websocket-protocol driver's driver_default_mode (the path this gap concerns) to see whether this is live exposure or still forward-looking.

   easee_cloud, tesla_vehicle, myuplink, esphome_dsmr, esphome-dsmr, zap, tibber — every one of their driver_default_mode functions is currently a no-op ("No-op — cloud charger manages itself.", "Read-only: nothing to release.", etc.). None of them write anything there today, so the untested path this issue describes has nothing to hide yet on main. This is still correctly scoped as blocking on #46 (nibe_local's solar-PV write path), which would be the first HTTP-protocol driver with an actual write in driver_default_mode — not an active gap on any shipped driver right now.

   Leaving this open and unclaimed; worth widening test_refused_write_settles.py/refused_write_probe.lua (protocol filter + host mock support for simulating an http_patch/http_post refusal) before or alongside #46, not urgently before it for unrelated drivers.

   ---

   Generated by Claude Code
3. frahlg commented on Sep 16, 2026

   

   frahlg

   on Sep 16, 2026 – with Claude

   MemberAuthor

   More actions

   Status update, no code change: the gap this issue describes just went from forward-looking to live on main.

   #46 (nibe_local Solar PV write path) merged yesterday (f3771d7, 2026-09-15). Its driver_default_mode (drivers/lua/nibe_local.lua:739-751) now does exactly the unbounded-retry thing this issue warns about:

   function driver_default_mode()
       if not write_cfg.requested then return end
       if feed_last_w == nil or feed_last_w == 0 then return end
       local now = host.millis()
       if default_clear_ms and (now - default_clear_ms) < 60000 then return end
       default_clear_ms = now
       local ok, err = write_pv_surplus(0, "default mode")
       if not ok then
           host.log("warn", "NIBE: default-mode solar PV clear failed " ..
               "(will retry): " .. tostring(err))
       end
   end

   No refusal counter — on a pump where the clear write keeps failing (e.g. Local REST API left read-only via installer menu 7.5.15, or register 2107 not accepting the write), this logs a warn and retries every 60s for the life of the session. Same shape as the Sungrow SG12RT incident that motivated sungrow 1.5.7.

   And it's invisible to CI: confirmed drivers/tests/lua_harness/refused_write_probe.lua:87 still hard-codes if PROTOCOL ~= "modbus" then print("SKIP not modbus") return end, so test_refused_write_settles.py skips nibe_local outright rather than measuring it — make check is green with this gap sitting in a merged, live driver.

   This is no longer "worth widening before or alongside #46" — #46 is in, so the exposure is live now. Recommend prioritizing the PROTOCOL == "http" widening (+ a bounded refusal counter in nibe_local.driver_default_mode, mirroring sungrow's count-to-N-then-report-held pattern) rather than leaving it queued behind other work. Flagging for a maintainer — this is a control-path change so it should get the same hardware-evidence treatment as the rest of nibe_local's write path, not a same-day autonomous merge.

   ---

   Generated by Claude Code