I'm a bit unsure how we should handle this nicely. Without Trickle we have to wait for all candidates. But mDNS timeouts are between 3-5 seconds. That's a high startup delay. Especially if hosts are not on the same network. Should we migrate baresip ice to trickle and add only mDNS support there?

There are two ICE stacks, "ice" and "trice".

The latter one has multiple interfaces and supports Trickle ICE, it has been in production
for around 8 years. The long term goal is to merge these 2 stacks into one stack.

We could create a new baresip module trice.so and experiment with how it works.

Waiting for all candidates to finish it not so practical, IMHO. You could also add a new flag
to "ice" which says dont wait for all candidates to finish ...

It is also possible to add optional "Trickle Mode" to the current ICE stack, so that we
don't have to wait for all candidate pairs to finish...

There is also another solution, which is to set the mDNS timeout short, e.g. 100ms.

There is also another solution, which is to set the mDNS timeout short, e.g. 100ms.

Since I had already a 100ms default timeout implemented this workaround was easy. Needs some testing...