The dnssd changes that involves browsing on both local and srp domains has some major bugs that need to be fixed.

[nivi-apple]

line - https://github.com/project-chip/connectedhomeip/pull/32631/files#diff-eba5148bba8ab5cd4e32114e5ddc8315f4f54f588b70a32039c09ba6bec34a8cR148

This should always be the last character of the string though? If it isn't then something else is going wrong.

Member
@bzbarsky-apple bzbarsky-apple 17 hours ago
Yes, exactly. This is making some assumptions, but then acting as if those assumptions are not being made, which means it will fail in bad ways if those assumptions happen to be violated.

Member
Author
@nivi-apple nivi-apple 6 hours ago
added a check that last_pos should be equal to the string length.

Member
Author
@nivi-apple nivi-apple 4 minutes ago
N/A. The API was changed to look for the first occurence of '.' after talking to Ted seems like the hostname with domain has hostname. after the first '.'. I will follow up on this after checking the matter spec on how the hostname could look like in different variations and will make any fixes as required.

Line - https://github.com/project-chip/connectedhomeip/pull/32631/files#diff-eba5148bba8ab5cd4e32114e5ddc8315f4f54f588b70a32039c09ba6bec34a8cL227

Can kDNSServiceFlagsMoreComing be relied upon to work this way when kDNSServiceFlagsShareConnection is in use? It seems like we're assuming here that it means the DNSServiceBrowse and DNSServiceGetAddrInfo calls for the domain of this last result is done for now.

https://developer.apple.com/documentation/dnssd/1823436-anonymous/kdnsserviceflagsshareconnection

Member
Author
@nivi-apple nivi-apple 5 hours ago
When Boris, Ted and I discussed we didn't realize this can be an issue. I need to spend some time on this as I am not an expert in the domain. @bzbarsky-apple do you have an opinion here?

@Abhayakara Abhayakara 5 hours ago
kDNSServiceFlagsMoreComing says that we have any more data on the shared connection. If you are doing multiple queries on the same connection, the response that doesn't set the MoreComing flag for any particular series of results can happen on any of the queries.

Member
Author
@nivi-apple nivi-apple 3 minutes ago
This needs more thought and has to be fixed correctly. I need to check the API documentation and see how best to handle the kDNSServiceFlagsMoreComing for a shared connection. will do as a follow up

Line - https://github.com/project-chip/connectedhomeip/pull/32631/files#diff-eba5148bba8ab5cd4e32114e5ddc8315f4f54f588b70a32039c09ba6bec34a8cR367

bzbarsky-apple 17 hours ago
I don't understand this logic. We want to call GetAddrInfo any time we get OnResolve without kDNSServiceFlagsMoreComing, and only do the GetAddrInfo on the things we didn't do it on yet.

As in, this isResolveRequested flag needs to be per interface/domain pair, not per-context, no?

Member
Author
@nivi-apple nivi-apple 3 minutes ago
I misunderstood this. This needs rework and I am going to do that as a follow up.

Line - https://github.com/project-chip/connectedhomeip/pull/32631/files#diff-eba5148bba8ab5cd4e32114e5ddc8315f4f54f588b70a32039c09ba6bec34a8cR386

This is incorrect. You should resolve the domain that you got in the browse callback. There is no promise that if you do a browse in local, all your responses will be in local. Same for default.service.arpa. There is no need to do two resolves here, and in fact it's incorrect because there might be totally unrelated data that happens to match the hostname you got for local in default.service.arpa, or vice versa.

Member
Author
@nivi-apple nivi-apple 2 hours ago
We discussed this with you in our meeting with Boris. So we get a browse before resolve only for the pairing use cases. for operational reads/writes, we do not get a browse before a resolve and hence we had to do the same thing as browse. I know you re-iterated that what we get in the browse result can be different than what we requested. so what i could do is if the domain in the browse result is not null i.e if a browse preceded a resolve, i will use the domain from the browse result otherwise we will resolve on local domain and srp domain. would that work?

Member
Author
@nivi-apple nivi-apple 3 minutes ago
This needs a discussion with @bzbarsky-apple. The code right now should work for the 2 domains we care about the local and srp domain but definitely has a bug. Need to be re-written as follow up

Line - https://github.com/project-chip/connectedhomeip/pull/32631/files#diff-d3d1d99d691d0aef90e9de721cad391450d0e28aa0b727d812022acc3755f248R722

This would have been a great place to use auto.

Member
Author
@nivi-apple nivi-apple 6 hours ago
filed - 32681