tree 1ad8318647c2104a3fa4de1b3b29744f6a647bbe
parent bdfc185692381cff13d781710ca801a9c4bc2abb
author Boris Zbarsky <bzbarsky@apple.com> 1668702788 -0500
committer GitHub <noreply@github.com> 1668702788 +0000
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsBcBAABCAAQBQJjdmJECRBK7hj4Ov3rIwAAvmgIAEUUZwx3V00K4E8CZhPZo6sT
 8jZ3V/D0Dxvw+MV9rPiaiyXhicARNshhNYyLTw88bczqJXUphLghUg5GvORFp9lz
 qPI2siqyPTUq0f8vj/uUC3Jg6uwRLC/KBlXO80ETu2GbNrmuJU86ykR8WnzdCFnK
 z1GFEtLb2BC/H2pO9UfgnjZu7NFXvqVFheBBQjYGe92CI4SresIUuairafo8KYkH
 5TN3EtqGWyUs39sKPH/3zOetRxgL9n0pGXggmXi/CwZg5bVNJtDQjqsxuRaWcmDv
 Me+GFeGQMCZk964wT3mM7E143Yv2b/Sm1LiNKgSssToE2YpTjFPj5qOcavf6aTM=
 =6ogX
 -----END PGP SIGNATURE-----
 

Add missing endWork calls in MTRClusters. (#23659)

We were missing endWork bits for command invokes when the command has a data
(not status) response.  This would end up permanently blocking the MTRDevice's
work queue once such a command (e.g. UpdateFabricLabel) was issued.

The changes here are as follows:

1) Rearrange the logic in MTRClusters-src.zapt so that the endWork happens
   regardless of whether hasSpecificResponse is true.
2) Add a deprecation message to the timedInvokeTimeoutMs field in _response_
   command params, since that should never be used (ran into this when writing
   the tests for this PR).
3) Set resubscribeIfLost to NO on various subscriptions in MTRDeviceTests, so if
   another test takes a while we don't fail due to the canceled subscriptions
   re-subscribing and over-fullfilling the subscription expectations.
4) Add a test for a mix of passing and failing commands on MTRCluster instances,
   with some of the commands defined as getting status-only responses and some
   defined as getting data responses.  This test fails without the endWork fix.