Add extended-isupport

[emersion]

Proposal for ircv3/ircv3-ideas#124

[sadiepowell]

Rather than adding more numerics can we just require that it gets batched?

[emersion]

Yeah, that would be a cleaner solution. I wonder if this would be hard for older servers to implement?

[emersion]

Switched to a batch. Does this look good?

[progval]

Could you mention that:

1. as before, ISUPPORT messages are additive (ie. if a new ISUPPORT overwrites or adds values, but the absence of a value does not mean it's not present anymore)
2. specify the - prefix to remove a key? I don't think this existed prior to this spec.

[emersion]

as before, ISUPPORT messages are additive (ie. if a new ISUPPORT overwrites or adds values, but the absence of a value does not mean it's not present anymore)

Not sure how to phrase this. Maybe:

As usual, servers can update or delete existing values by sending additional RPL_ISUPPORT messages in a draft/isupport batch after the initial batch.

But maybe you want something more precise?

specify the - prefix to remove a key? I don't think this existed prior to this spec.

It's in modern: https://modern.ircdocs.horse/#rplisupport-005

[sadiepowell]

specify the - prefix to remove a key? I don't think this existed prior to this spec.

It's in modern: https://modern.ircdocs.horse/#rplisupport-005

Also worth noting that Modern gets it from from the original IETF draft isupport specs.

[jwheare]

-NETWORK isn't a very realistic example. Also, maybe show a key being changed via removal and readdition. Also these examples could do with the full CAP negotiation being shown, including CAP END and maybe the 001 being sent.

[sadiepowell]

I think the intended meaning here is that users will get updates before connection registration?

[jwheare]

Yeah the last part refers to the previous example sorry.

[emersion]

Right. I wasn't sure how far the examples should go, I wouldn't want to list the full registration burst. Will add.

-NETWORK could happen if the network configuration file is reloaded with the network name removed. Do you have another example in mind which might be more realistic?

[emersion]

Also, maybe show a key being changed via removal and readdition.

The ISUPPORT spec says that changing a key is done by advertising it again, rather than removing it and then re-adding it.

[emersion]

I've tried to address the comments. Let me know what you think!

[jwheare]

Changes look good. No need to force push updates, makes it a bit harder to review. I squash commits on merge anyway.

[emersion]

Alternatively, instead of spec'ing that servers must send the ISUPPORT list when the cap is enabled, we could guarantee that VERSION works pre-reg and recommend that clients send it after enabling the cap.

[slingamn]

I would be slightly more comfortable with allowing VERSION pre-registration --- the "edge-triggered" API has always bothered me a little (on the other hand, I haven't been able to identify any concrete problems with it).

VERSION has a wrinkle where you also get RPL_VERSION, so presumably you would get RPL_VERSION outside the batch and then a batch of RPL_ISUPPORT? We could add a command like ISUPPORT that just sends the RPL_ISUPPORT.

[jwheare]

Just specifying an ISUPPORT command that can work pre-reg works for me. Regarding batch, I feel like this was already discussed, but what use case does batch enable? My implementation just incrementally updates the ISUPPORT list whenever an 005 is received, I don't have a need for an end of isupport marker.

[slingamn]

@jwheare there is a note on this here: ircv3/ircv3-ideas#124 (comment)

tl;dr it allows the UI to be updated atomically, reducing "flicker".

[jwheare]

Ok so maybe define the ISUPPORT command to require use of a batch if the cap is enabled? But leave the existing automatic isupport sent post reg unbatched. Would that work?

[emersion]

That wouldn't work if the server wants to send an atomic update of more than 13 tokens.

[emersion]

A new ISUPPORT command has been introduced, and the RPL_ISUPPORT list is no longer sent when the capability is enabled.

[slingamn]

Looks good, I merged support for this in ergo and deployed it on my staging server ircs://files.stronghold.network

[emersion]

I've finally got some time to implement this in Goguma! Here's the patch: https://codeberg.org/emersion/goguma/pulls/2

Ergo's implementation seems to have a small bug: there is an extraneous chathistory param in the batch parameters.

flutter: [3] <- :files.stronghold.network BATCH +1 chathistory draft/extended-isupport
flutter: [3] <- @batch=1 :files.stronghold.network 005 * AWAYLEN=500 BOT=B CASEMAPPING=ascii CHANLIMIT=#:100 CHANMODES=Ibe,k,fl,CEMRUimnstu CHANNELLEN=64 CHANTYPES=# CHATHISTORY=100 ELIST=U EXCEPTS EXTBAN=,m FORWARD=f INVEX :are supported by this server
flutter: [3] <- @batch=1 :files.stronghold.network 005 * KICKLEN=1000 MAXLIST=beI:60 MAXTARGETS=4 MODES MONITOR=100 MSGREFTYPES=msgid,timestamp NETWORK=OraStage NICKLEN=32 PREFIX=(qaohv)~&@%+ STATUSMSG=~&@%+ TARGMAX=NAMES:1,LIST:1,KICK:,WHOIS:1,USERHOST:10,PRIVMSG:4,TAGMSG:4,NOTICE:4,MONITOR:100 TOPICLEN=1000 UTF8MAPPING=rfc8265 :are supported by this server
flutter: [3] <- @batch=1 :files.stronghold.network 005 * UTF8ONLY WHOX draft/CHATHISTORY=100 :are supported by this server
flutter: [3] <- :files.stronghold.network BATCH -1

With a small workaround for the broken batch type, it seems to work fine:

[slingamn]

Sorry about that. I fixed it in ergochat/ergo#2197 and redeployed.

[emersion]

The batch is named "draft/isupport", not "draft/extended-isupport".

[emersion]

Gentle ping :)

[jwheare]

Fine to merge this as is, but would the non draft cap name be more self explanatory as prereg-isupport? The format is not being extended by this afaict.

[emersion]

The name "extended-isupport" was picked because the cap also introduces a new batch and a new command. I'm fine using any other name if you prefer.

[jwheare]

early-isupport was also suggested by @SadieCat to avoid the potentially confusing "registration" term being used. Anyway, no need to change the draft cap, but merged version should be. Hopefully someone remembers when the time comes!