Version 3.1.4-beta1

This is a beta release focused on stability fixes that surfaced after 3.1.3 went out. Most users should be safe to install; users on WPU units in particular benefit. Please report back via GitHub issues if you see anything unexpected.

Highlights

• Memory stability: long-running units no longer suffer slow heap erosion that eventually triggers SDIO_RESET after a few days of uptime. Verified on a CVE-Silent test unit running ~14 days continuously without reset since the fixes were applied.
• WPU status reliability: addresses the "status sometimes there, sometimes not" complaint reported with 3.1.3. A boot-time QueryStatusFormat race could leave the device in a state where the status board never populated until the next reboot. Now retried automatically and a partial / corrupt I2C reply no longer wipes a previously-good status list.

Bug fixes

Memory & heap

• Skip WebSocket systemstat broadcasts when no clients are connected — every broadcast otherwise allocated a per-client buffer + deque node, accumulating fragmentation over days even when nobody was listening.
• WebSocket broadcast now bails out when free heap drops below 80 KB, preventing fragmentation spirals during transient memory pressure.
• notifyClients switched to std::nothrow allocation so a failed allocation degrades gracefully instead of throwing.
• MQTT publish buffer is now grow-only — repeated realloc cycles between the default and a larger size were a major source of long-term heap fragmentation.
• HA Discovery publish skipped when free heap is below 100 KB. Fixes a crash path where an already-low-heap device cascaded into SDIO_RESET while building the (large) discovery JSON.
• WebSocket queued-message cap reduced from 64 to 8.
• REST 503 log now includes free heap and largest-block size for diagnosis.

I2C status reliability

• QueryStatusFormat is retried once at boot if the first attempt returned 0 entries. Previously a transient mutex-contention or partial I2C reply at boot would leave the status board permanently empty until the user rebooted manually.
• Status-format parsing now happens into a local list and is only swapped into the live state if the new reply has at least one entry. A 0-length / corrupt reply at runtime no longer wipes the previous (good) data.
• Same defensive 0-length check added to the 31DA and 31D9 paths.
• Mutex acquire timeout raised from 100 ms to 500 ms for the QueryStatusFormat, QueryStatus, Query31DA, and Query31D9 paths. Failure to acquire now logs a warning instead of silently dropping the update.

Cosmetic / diagnostic

• The RF: TX power 0x... for Send remotes log line and the corresponding remote-flash write are now skipped on units without a CC1101 module — the value was meaningless and the log line was confusing on I2C-only setups.

Known issues / not addressed in this release

• CVE ECO 2 SP fan control after upgrade (#364 / #365): if your fan does not respond to the speed slider, please confirm System Settings → Enable/Disable I2C commands → CVE fan control (PWM2I2C) is set to on. We are still investigating whether anything in the firmware needs to change beyond this user setting.

How to test

If you upgrade and notice anything off, please open a GitHub issue with:

• Your /api/v2/deviceinfo output
• The first ~50 lines of the System Log after a fresh boot (so we can see QueryDevicetype, QueryStatusFormat - items:N, etc.)
• A description of what you expected vs. what you saw