Traceback when adding Xiaomi device

[therve]

I have a few Xiaomi devices, which works great with the new support. But, on join they fail with a traceback:

2018-06-08 15:33:00 ERROR (MainThread) [bellows.ezsp] Exception running handler
Traceback (most recent call last):
File "/usr/lib/python3.5/site-packages/bellows/ezsp.py", line 194, in handle_callback
handler(*args)
File "/usr/lib/python3.5/site-packages/bellows/zigbee/application.py", line 139, in ezsp_callback_handler
self._handle_frame(*args)
File "/usr/lib/python3.5/site-packages/bellows/zigbee/application.py", line 160, in _handle_frame
tsn, command_id, is_reply, args = self.deserialize(device, aps_frame.sourceEndpoint, aps_frame.clusterId, message)
File "/usr/lib/python3.5/site-packages/zigpy/application.py", line 73, in deserialize
return sender.deserialize(endpoint_id, cluster_id, data)
File "/usr/lib/python3.5/site-packages/zigpy/device.py", line 97, in deserialize
return self.endpoints[endpoint_id].deserialize(cluster_id, data)
KeyError: 1

I feel that's due to the fact that quirks only operate after device_initialized, not add_device (though I'm have only a shallow understanding of the problem).

I'd be happy to help testing.

Thanks!

[mrfoxbit]

Same issues. Did you fix it?

[mrfoxbit]

It worked for me. Thank you, @Yoda-x

zigpy/bellows#43 (comment)

[justin8]

Yeah, I'm also seeing this same issue. I haven't tried the fix from zigpy/bellows#43 (comment) yet. I'm running hass.io, so running custom versions of home-assistant seems to be quite painful

[Adminiuga]

Xiaomi are sending attribute updates prior Zigpy finishes device discovery and configure the device to send updates, hence the error.
If you continue getting this error, then:

1. remove the device
2. repair it
3. during the discovery, keep the device awake by pressing the button ever 1.5s
4. you may have to repeat the above a few times, as sometimes Xiaomis are PITA to pair

[therve]

The error only happens when adding the device. But pairing does work. Could we just catch the error?

[Adminiuga]

Hrm, we could take same approach as zigpy-deconz and ignore ZCL traffic until the device is fully initialized or alternatively maybe we should add endpoint/clusters if we receive traffic on those, similar to proposal in #9