[autolog]

700 and 800 series controllers

Adds backup, restore and verify for 700 and 800 series sticks (Zooz ZST10 700 / ZST39, Aeotec Z-Stick 7 / Z-Stick 10 Pro, UZB-7 and similar), as discussed on the forum thread. The 500-series code paths are untouched; the new generation gets its own functions in controller_image.py and its own restore body in plugin.py.

How it works

Same shape as the 500 path, but through the NVM3 commands these sticks use — a straight port of zwave-js backupNVMRaw700 / restoreNVMRaw700, whose serial frames I captured from Z-Wave JS UI and matched byte for byte:

• SetRFReceiveMode off and StopWatchdog before any NVM access; SoftReset afterwards and wait for the unsolicited SerialAPIStarted. The reset is always sent on the way out (failure paths too), so the stick is never left with its radio off. No unplug on restore: the reset is the documented step and the read-back follows at once.
• NVMOperations 0x2E (16-bit offsets), or ExtendedNVMOperations 0x3D when the stick advertises it in its function bitmask (needed once the NVM passes 64 KB; the ZST39 and Z-Stick 10 Pro do, the 700 stick does not).
• Probe with a 255-byte read, use the chunk that comes back (64 on all three sticks), 48 on an empty probe.
• Node ids parsed by payload length, because zwave-js leaves these sticks in 16-bit node id mode and it survives a soft reset (they read as node 0 otherwise).
• Full SDK version from GetProtocolVersion (7.17.2 where the library string says 7.17), used for the same-software restore guard and shown in Show Plugin Info / the device.

Two things learnt on hardware that shape the restore path

1. The final write is never performed. The write that reaches the NVM size is answered 0xFF (end of file) and dropped; zwave-js behaves the same and never checks. Those bytes hold nothing any file refers to. The plugin reports it and leaves it alone — and deliberately does not do what the 500 path does for the Gen5's last 16 bytes (split and retry): a 1-byte write near the end hung a ZST39 outright, no ACK, soft reset ignored, watchdog already stopped, replug needed. write_nvm_700 writes in the probed chunk and accepts EOF on the last one; a test asserts the fake's "hang" is never triggered.
2. Byte-identical read-back is not achievable on NVM3. The firmware appends housekeeping objects into erased space when it restarts (74 bytes after one restore), and an hour of live running adds a few hundred bytes of route updates the same way; a power cycle can relocate whole pages. So compare_images_700 allows differences only inside the declined tail and where the image holds 0xFF; anything overwritten is "did NOT hold". Restore additionally checks the Home ID and the node table reported after the reset against the sidecar. Verify says how many bytes were added in free space and compares the node table too.

Two small fixes found in live use

• A backup folder pasted with quotes around it was accepted as a relative path and the images landed inside the plugin bundle. clean_folder_path strips quotes, expands ~, and the dialog refuses anything not absolute.
• The Home ID sanity check took the first Home…_Node… key in Indigo's Z-Wave prefs; a real prefs file carried five Home IDs from past networks and the first was not the live one, so it warned about a mismatch that was not there. It now checks against the set.

Tests

FakeStick700 in tests/fake_stick.py, built from the captured ZST39 and 700-stick payloads, covers the declined tail, the hang, 16-bit node ids, the probe fallback, housekeeping on reset, and a stick that says OK to writes and keeps none. 27 new tests; tests/run.sh green (96 tests, syntax, XML, ruff). All existing tests pass unchanged apart from the one that asserted a 700 stick is refused.

Live verification

Stick	Result
Zooz ZST39 LR, 800 series, Z-Wave 7.24, 10 nodes	backup (19 s) → verify → restore → verify, all clean; backups byte-identical to Z-Wave JS UI's on both 0x2E and 0x3D; an older image restored brought back its older node table
Aeotec Z-Stick 10 Pro, 800 series, Z-Wave 7.23 (its Z-Wave port; the Zigbee port is ignored)	backup (13 s) → verify → restore → verify, all clean
Silicon Labs 700 series stick, Z-Wave 7.17, 108 node ids, running Indigo 2025.2's live network	backup from the plugin menu inside Indigo, 16 s, byte-identical to a terminal run; Indigo reconnected in 2 s; restore not exercised on the production stick

Version bumped to 1.1.0; README, SPEC and release notes updated. I don't have a 500-series stick, so the Gen5 path was covered by the existing tests only — worth one run on yours before release.

🤖 Generated with Claude Code