Redesign client command support. about irc

[aatxe]

So, as @retep998 pointed out to me on IRC, there's no way to actually construct such a Command. So, we'll have to look at different design options. This may just involve finding a way to make from_message more maintainable.

[isomorpheme]

My 2 cents as someone stumbling upon this: shouldn't Command really be a part of Message, instead of Command being a separate type that can be converted into a Message? Something like:

struct Message {
    prefix: Option<String>,
    command: Command
}

enum Command {
    NICK(String),
    QUIT(Option<String>))
    // etc...
}

As far as I've understood from the RFC, every IRC message is a command as well - chatting in a channel is really just privmsg-ing that channel. So from the data point of view, having only a Message type containing Commands makes more sense. From the usability point of view though, being able to just send Commands to a server makes more sense. But for that case, perhaps you could impl Into<Message> for Command in such a way that the command just gets copied to the message struct, and the prefix left blank. Seems much simpler IMO.

[aatxe]

The motivation for keeping them separate before was that there are potentially custom or otherwise unknown commands. This is perhaps less of an issue now that we have mostly complete IRCv3 support, but one example would be Twitch chat has a custom command. One way around this would be to add an Unknown command to the enumeration and to just have it default to either a Vec or perhaps a Vec and a String (separating out the suffix).

One of the IRCv3-WG's unofficial policies regarding unknown commands is that they should just be passed to the server straight through—most server implementations these days return numerics for unknown commands, so the client is unlikely to be surprised by a lack of response. Additionally, it makes the need for aliases to work with e.g. services more efficiently entirely obsolete (aside from those servers which don't provide their own aliases, but that's comparatively rare).

A generic 'raw' command, through which unknown-to-the-crate commands flow by default, would probably be most appropriate.

[aatxe]

I'll try to work on this at some point, but I'd accept any help. I'd like to implement @ijks's proposal with the Raw command as the default as per @Aerdan's suggestion.

[ketsuban]

Looks like that commit didn't actually close this issue. Github's magic is a little too magical at times, it seems.

[aatxe]

I believe it's because it's on a separate branch that I haven't merged in yet. I'm waiting a bit to merge because I'm going to do some other major breaking changes, and I don't want to merge it into master until all the changes are ready for release.