🔤 making short code optional?

[favonia]

Both LSP and Rust do not demand a short code. Maybe we can make it optional?

API changes:

• The type of short_code is t -> string option instead of t -> string

[mikeshulman]

Under what circumstances would one not want to have a short code? I thought the whole point was that the user can always look up the short code in documentation; it seems to defeat the purpose if not all errors have a short code.

[favonia]

@mikeshulman Perhaps I was thinking too much about how to translate messages between different systems. When diving into the design documents of Rust, I found this comment: rust-lang/rustc-dev-guide#967 (comment). They seem to avoid creating codes when the message contains everything and the code might not be useful for user. I was thinking about "how to translate those messages from Rust to asai" but of course there's no problem in translating messages from asai to Rust-style libraries.

[favonia]

Maybe a better idea is to make it optional only in our diagnostic type so that someone can write a translator if they want to, but keep the reporter API so that code is still mandatory. Otherwise, we should just abandon this idea (which is also fine).

[mikeshulman]

I think personally I would find it simpler to follow a general rule of "always have a short code", even if it sometimes doesn't add anything, than to have to make the extra decision of "do I make a short code for this?" every time I add a new error. Even the author of that comment thinks it would be fine to have generic error codes for the cases that don't have specific details to add, which is precisely what I'm currently doing with parse errors.

[favonia]

Alright I will close this issue as "won't fix".