Darwin browse+resolve does not properly collect multiple IP addresses

[msandstedt]

Problem

Observed while talking to the Linux all-clusters-app from chip-tool on both Linux and Darwin platforms.

On Linux, chip-tool discovery (browse+resolve) and node ID resolution produce a nodeData record with all IP addresses populated as expected. But on Darwin, discovered and resolved IP addresses are split across multiple nodeData records. For discovery, this means that code like the following can only retain a single IP address, as each new address overwrites the record:

https://github.com/project-chip/connectedhomeip/blob/a04c6d916d26785891751cd804b4b[…]4fc8cc4934f/src/controller/AbstractDnssdDiscoveryController.cpp

And for resolve, we only ever watch for the first received record from the host, so we end up with whichever address happens to arrive first.

Proposed Solution

Fix Darwin mDNS implementation to match the expected behavior and present discovery (browse) and resolve records with all IP address populated.

[bzbarsky-apple]

I believe this is a problem, at least for resolve, for everything using src/lib/dnssd/Discovery_ImplPlatform.cpp. That is, OpenThread SRP, Darwin, Linux using Avahi (instead of minimal mdns), etc. Specifically, for resolve the callback that is used is HandleNodeResolve. It gets handed a DnssdService (which contains a single IP address) and immediately creates a DiscoveredNodeData on the stack, populates it with the one address, and calls OnNodeDiscoveryComplete.

Similar for HandleNodeIdResolve.

Browse, on the other hand, should be producing all the IPs for Darwin based on what I see in the implementation, and the corresponding API in Discovery_ImplPlatform allows passing an array of DnssdService, which Darwin does.

That said, I just checked our GenericThreadStackManagerImpl_OpenThread<ImplClass>::DispatchBrowse implementation and that one always dispatches a single DnssdService.