Remarks

[jacob-wg2]

Looks great.

Some minor issues (to be continued):

https://github.com/sebastiw/sock/blob/cbcb23ab699a51c6bfef2798c40ee0ec0d6566dc/src/sock.erl#L62-L63

Binding to all local interfaces by default is probably problematic, especially when local IP addresses are used. Which local IP addresses are announced to the peer in that case? What happens if a new interface is added or an existing goes down/up?

https://github.com/sebastiw/sock/blob/cbcb23ab699a51c6bfef2798c40ee0ec0d6566dc/src/sock_assoc.erl#L62-L64

Wondering what dialyzer will say depending on USE_SOCKET. Why not using -ifdef?

https://github.com/sebastiw/sock/blob/cbcb23ab699a51c6bfef2798c40ee0ec0d6566dc/src/sock_utils.erl#L26-L27

Wouldn't cope with inet6 and shortened addresses (less than 3 dots).

What about having a function that takes two inet:ip_address and a number (mask) instead? Or at least using net:parse_address internally instead of split?

[sebastiw]

1. for testing it's excellent :) I don't hope anyone will use that in prod. all create_ep/0-3 are exported so one can choose to use any.

2. Updated with -ifdef

3. Fixed for ipv6 and shortened addresses, though no function actually use it yet.

[sebastiw]

I did update 1. to loopback instead

[lixen-wg2]

I don't think we should use the supervisor as a lookup. We have seen that it doesn't scale. I also wonder what happens when you have failures and which_children messages filling the message box of the supervisor
https://github.com/sebastiw/sock/blob/16c022414a677310a13340b1d2a2e43ac41ac24e/src/sock_sup.erl#L32

Should we do recv or {active, N} ? I think we greatly increased performance in diameter testing by having active tuned.
https://github.com/sebastiw/sock/blob/16c022414a677310a13340b1d2a2e43ac41ac24e/src/sock_ep.erl#L154

Does it make sense to move this read outside gen_server?
https://github.com/sebastiw/sock/blob/main/src/sock_reg.erl#L84