chore(drivers): move the bundled pin to device-drivers@76c968bd - #722
Conversation
Closes #713. The bundled snapshot is what a gateway runs when it boots offline, so a fix that lands upstream does not reach those gateways until the pin moves. Pin moves a4d0205 -> 76c968b. One bundled driver changes: sungrow 1.5.6 -> 1.5.7, which stops rewriting the EMS control block (13049-13051) once the inverter has refused it three times. An SG string inverter rejects every write there, and the driver retried three at startup and three more on every watchdog tick, logging a failed self-consumption reset once per tick for the life of the session. The upstream fix keys on whether the device took the write rather than on the model family, which matters here: a hybrid shipped under a device-type code the driver has not been taught classifies as "string", and gating on the label would leave exactly that inverter forced with nothing writing mode 0 back. A device that refuses the write cannot be holding a forced state, so there is nothing to release. Generated by scripts/sync-bundled-drivers.sh; nothing edited by hand. scripts/sync-bundled-drivers.sh --behind now reports the pin current. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a4c195ab31
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| ], | ||
| "repository": "srcfl/device-drivers", | ||
| "commit": "a4d0205d5f4439f0994b2b708a4d69a7a0c20023", | ||
| "commit": "76c968bd888193cc9362da951ec9bcca7488b256", |
There was a problem hiding this comment.
When a controllable hybrid is left in forced EMS mode and three default-mode attempts encounter a temporary TCP outage or reconnect backoff, the pinned 1.5.7 driver counts those errors as refusals and suppresses subsequent reset writes for the rest of the session. host.modbus_write returns transport errors as well as definitive Modbus exceptions, so if connectivity recovers while dispatch remains stopped—for example, because the site meter is still stale—the watchdog cannot return the inverter to autonomous mode until FTW restarts. Only latch the unsupported-block state for a definitive device rejection, or resume attempts after transport recovery.
AGENTS.md reference: AGENTS.md:L31-L32
Useful? React with 👍 / 👎.
1 participant
Closes #713.
The bundled snapshot is what a gateway runs when it boots offline, so a fix that lands upstream does not reach those gateways until the pin moves.
What actually changed
The issue lists roughly twenty drivers as outdated. That reading is stale — it predates #704 and #712, which moved the pin twice on 2026-07-29. Against the pin as it stands today (
a4d0205),scripts/sync-bundled-drivers.sh --behindreports exactly one:The upstream range
a4d0205..76c968bis two commits and touches one bundled driver. Everything else in it is tests, docs and manifests, which the snapshot does not carry.The Sungrow fix (1.5.6 → 1.5.7)
An SG string inverter refuses every write to the EMS control block (13049–13051). The driver retried three at startup and three more on every watchdog tick, so an SG12RT logged a failed self-consumption reset once per tick for the life of the session — while it was already offline for an unrelated reason.
The fix stops writing the block after three consecutive refusals. It keys on whether the device took the write, not on the model family, and that distinction is the reason to take it rather than a narrower patch: a hybrid shipped under a device-type code the driver has not been taught classifies as
string, and gating the release on that label would leave exactly that inverter forced with nothing left to write mode 0 back. An inverter that refuses the write cannot be holding a forced state, so there is nothing to release. A restart clears the counter, so the startup reset is always attempted at least once — the case it exists for.Verification
scripts/sync-bundled-drivers.sh; nothing edited by hand.scripts/sync-bundled-drivers.sh --behindnow reports the pin current.make verifypasses: Go tests, optimizer suite, Lua driver contract checks, compose migration, vet, build.