[zigpy-review-bot]

Summary

Fixes ZHA failing to start on ConBee II / RaspBee II with zigpy 2.4.x: No available endpoint slots exist (home-assistant/core#189366).

Closes #280

zigpy 2.4.0 registers the Green Power endpoint 242 next to the HA endpoint 1 and the ZLL endpoint 2, but the ConBee II / RaspBee II firmware only has two endpoint slots. add_endpoint handled one endpoint at a time:

• When both slots already held zigpy's endpoints 1 and 2 (any stick last run with zigpy 2.1–2.3), both were treated as used and 242 raised ValueError, so startup failed on every attempt.
• Otherwise, a slot it had just written was never marked as used, so a later endpoint could overwrite it: 242 overwrote endpoint 2 and slot 1 was rewritten twice (2, then 242) on every start, and with two slots held by another application, every endpoint went into slot 1 and endpoint 1 was never registered at all.

Changes

• register_endpoints now collects all of zigpy's endpoints first, then fits them into the slots in one pass.
• An endpoint with the same ID as an earlier one replaces it, so a configured endpoint can override one of zigpy's, as a standalone add_endpoint call already could.
• If they don't fit, the ZLL endpoint 2 is dropped (a configured ZLL endpoint with another ID is kept). That's the layout deCONZ and Z2M use on these sticks (slot 0 = endpoint 1, slot 1 = endpoint 242), and zigpy-deconz never sends from endpoint 2 anyway: send_packet sends everything except ZDO (endpoint 0) and Green Power (242) from endpoint 1. If it still doesn't fit (extra endpoints configured on a two-slot stick), it raises as before.
• Endpoints already in a slot stay where they are; only missing ones are written, preferring a slot with the same endpoint ID. A configured stick is no longer written on every start, and an upgraded one is written once (slot 1: 2 → 242).
• A standalone add_endpoint call goes through the same path with the endpoints registered so far, so it can't overwrite one of them.

Firmware with more slots (a four-slot ConBee II test build exists) keeps endpoint 2 and puts 242 in a slot not holding one of zigpy's endpoints.

Trade-off on two-slot firmware

puddly's hardware test on a ConBee II (0x26780700) shows that the firmware only delivers incoming frames whose destination endpoint is in a slot with a matching profile, and doesn't ACK the others. With two slots, one of zigpy's three endpoints is always missing:

• without 242, Green Power frames forwarded by proxies are dropped (frames received directly still arrive through the firmware's GP indication);
• without endpoint 2, ZLL-profile (0xC05E) frames sent to the coordinator are dropped.

This PR gives up the ZLL endpoint, the same choice deCONZ and Z2M make. Sending is unaffected either way, since the firmware doesn't check source endpoints.

Testing

• New tests run zigpy's real register_endpoints against mocked slot tables: two slots upgraded from zigpy ≤ 2.3 (ValueError before), two slots already holding 1 + 242 (two writes per start before, none now), two slots from another application, and three/four slots. Each also checks that registering again writes nothing.
• Too many endpoints still raise without writing anything, also when the only other ZLL endpoint is a configured one; a configured endpoint with zigpy's endpoint ID replaces it; a later add_endpoint uses a free slot or the slot with the same endpoint ID.
• 154 passed with zigpy 2.4.1. The four slot-layout cases and the override case fail against dev.
• Not tested on hardware yet. A test on a ConBee II that ran ZHA before 2026.11 would confirm it; ConBee II / RaspBee II fail to start since zigpy 2.4.0: no endpoint slot left for the Green Power endpoint 242 #280 has instructions and a test build for Home Assistant OS.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.17%. Comparing base (1bb30c8) to head (7d9b369).

Additional details and impacted files

@@            Coverage Diff             @@
##              dev     #281      +/-   ##
==========================================
+ Coverage   99.15%   99.17%   +0.02%     
==========================================
  Files           7        7              
  Lines         945      971      +26     
==========================================
+ Hits          937      963      +26     
  Misses          8        8              

☔ 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.

[Copilot]

🔵 Needs a closer look

Maintainer acceptance of the loss of incoming ZLL reception and hardware validation of two-slot startup and Green Power proxy reception are still needed, as slot-table mocks cannot verify firmware behavior.

No open issues 🎉

What changed in this PR

Fixes endpoint registration for ConBee II / RaspBee II firmware with only two slots, addressing the ZHA startup failure introduced by zigpy’s Green Power endpoint registration.

Changes:

• Collects endpoints before assigning slots, omitting ZLL endpoint 2 when needed to fit HA and Green Power.
• Preserves matching descriptors, supports same-ID overrides, and tracks registered endpoints to prevent unintended overwrites.
• Adds regression tests for slot layouts, repeat registration, capacity errors, overrides, and later endpoint additions.

File	Description
zigpy_deconz/​zigbee/​application.py	Batches endpoint registration and allocates available firmware slots while avoiding unnecessary writes.
tests/​test_application.py	Tests slot allocation, registration stability, capacity handling, and endpoint replacement.

_{Review effort: Balanced}

[TheJulianJES]

(note: do not press "Delete branch" yet, as the custom component installs this branch currently)