[zigpy-review-bot]

Summary

zigpy-znp has written NWK_CHILD_AGE_ENABLE=False at every start since 18448dc (2020). Koenkk's Z-Stack 3.x.0 builds enable child aging from 20230507 on and skip the NotMyChild leave for Xiaomi devices (prefix 00:15:8d) in firmware, so zigpy-znp's write turned off what those builds ship with. Related: #285

Changes

• Write NWK_CHILD_AGE_ENABLE=True when the Z-Stack build id is 20230507 or newer, False otherwise (older builds, builds without a CodeRevision). The item is still written in both cases, because a coordinator that already ran under zigpy-znp has False in NVRAM.
• Request SYS.Version before the stack settings are written (the build id is needed there); it is no longer requested after start_network.
• Reword the comment on the AssocRemove recovery, which assumed aging is always off.

Testing

• Full test suite passes; new test_child_aging_by_build covers builds before and after the cutoff and a missing CodeRevision.
• Checked against the Koenkk/Z-Stack-firmware changelog and firmware.patch: the 20221226 build set aging off with a Xiaomi comment, 20230507 enabled it with the per-address leave exemption.

Open items

• The cutoff relies on CodeRevision being Koenkk's build date, the same assumption as the existing 20210708 check. A non-Koenkk build reporting a later date without the Xiaomi leave exemption would now run with aging on.
• Not changed here: the recovery in send_packet still drops a child at its first MAC_TRANSACTION_EXPIRED; its interplay with aging is discussed in Child aging is forced off at every start: why, and should zigpy-znp stop writing NWK_CHILD_AGE_ENABLE? #285.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.54%. Comparing base (34098a5) to head (6a6ba9d).

Additional details and impacted files

@@            Coverage Diff             @@
##              dev     #287      +/-   ##
==========================================
- Coverage   98.60%   98.54%   -0.06%     
==========================================
  Files          43       43              
  Lines        3581     3582       +1     
==========================================
- Hits         3531     3530       -1     
- Misses         50       52       +2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[TheJulianJES]

@zigpy-review-bot How does this compare to Z2M?

[zigpy-review-bot]

Short version: Z2M (zigbee-herdsman) never writes this NVRAM item at all, so it just runs with whatever the firmware ships. After this PR, zigpy-znp does the same on Koenkk's 20230507+ builds.

• Z2M: NWK_CHILD_AGE_ENABLE (NVRAM id 62) only exists as an unused enum entry in src/adapter/z-stack/constants/common.ts. A grep over src finds no other use, so the value in NVRAM is never touched.
• Firmware builds that enable aging (20230507+): Z2M users get aging on, together with the firmware's Xiaomi leave exemption described in the PR body. With this PR zigpy-znp writes True on those builds, so the result is the same.
• Older builds: Z2M leaves whatever the firmware or a previous tool left in NVRAM. zigpy-znp still forces False there, which is where the two differ. That is the conservative choice, since those builds have no Xiaomi exemption.
• Recovery after failed sends: Z2M's sendZclFrameToEndpointInternal also retries on MAC_TRANSACTION_EXPIRED, with an assocRemove/assocAdd recovery gated on build 20200805 / 20201026 (supportsAssocRemove / supportsAssocAdd in zStackAdapter.ts). I did not compare it line by line with send_packet here, so the interplay with aging discussed in Child aging is forced off at every start: why, and should zigpy-znp stop writing NWK_CHILD_AGE_ENABLE? #285 is not checked against Z2M.

The one thing the comparison does not settle is the PR's open item: the cutoff trusts CodeRevision as Koenkk's build date. Z2M has the same assumption in its revision checks above.