Thanks!

What about moving this paragraph from the "Numerics" section to "Connection registration", so it is more visible? I doubt new client authors will read the details of each numerics, especially as this one looks straightforward

Noice \o/

ping

@slingamn In this repo usually wait 2-4 weeks to give people time for feedback

The server typically responds to rejected NICK messages with ERR_ERRONEUSNICKNAME. It seems this is purely about how to deal with a server assigned nick as a client? Is that new functionality? I'm sceptical about this suggested change because many clients do not currently parse RPL_WELCOME for their actual nick and the change would bring an assumption that isn't backward compatible.

@tommyrot, Hybrid and Plexus both truncate the nick to NICKLEN before validity checking. Likely others do, too. (Code search is tedious on mobile.)

I've already tested (against live servers) this behavior in my bot, just for the convenience of its raw log output. Specifying a longer-than-NICKLEN nickname results in the server sending back that nick truncated to NICKLEN in 001 RPL_WELCOME, as described in the added prose here.

While I can't speak to how long this behavior has been true, it doesn't appear to be new. I'm glad it's been brought up, because it'll let me fix some code that doesn't take this into account.

NICKLEN is a constraint on the validity of a nick and should result in a transparant ERR_ERRONEUSNICKNAME to the client. I think it's a weird choice to silently truncate but can see how it could have been conceived: a client is at fault for requesting a nick longer than NICKLEN but during registration it doesn't know of the NICKLEN value yet. I'm guessing. Still doesn't make it right.

but during registration it doesn't know of the NICKLEN value yet. I'm guessing.

Accurate guess: NICKLEN is part of 005 RPL_ISUPPORT, which the client doesn't receive until after registration.

It's a fair point that this seems like a weird choice. If you have an idea for improving the spec, I'd suggest popping into #ircv3 at Libera, or opening an issue at https://github.com/ircv3/ircv3-ideas. This repo exists to document what IRCds already do; if you want to see the underlying specs changed, you have to go through the working group that ratifies them. 😺

Did some testing as well. Inspircd, ngircd and ergo all respond with ERR_ERRONEUSNICKNAME during registration when using a nick that exceeds NICKLEN. Unrealircd and solanum truncate the nick to NICKLEN in which case RPL_WELCOME is indeed the first message to reflect that. Adding hybrid to the servers that truncate (as @dgw tested) probably makes this about half of all out there.

PR is OK

FWIW, one of the cases that's important to Ergo is that in the default/recommended configuration, if the client authenticates with SASL, we ignore the nickname sent on the NICK line and instead coerce the nickname to be equal to the SASL account name:

https://github.com/ergochat/ergo/blob/v2.8.0/default.yaml#L498-L502

and the primary means of communicating that this happened is RPL_WELCOME.