Hm, first pass worked. From chats it seems repr(C) is not essential there anyway, and the UnknownError is not used (and rather bad practice in error types compared to non_exhaustive); adding two more simple (semver breaking, but so is removal of the numeric values) changes.

I said it on chat but I was wrong: let's keep repr(C). Right now I don't use it but it is good to make EDHOCError available for implementations that need to check the error.

The numeric value can stay omitted, though.

Shouldn't the #[repr(C)] be added if and when the type is exposed to C, given the C side can't make use of the data without copying constants? And I think that the preferred abstraction would be to have an edhoc_error_get_err_id() function depending on how things go with aligning with the EDHOC spec. (I'll update to repr(C) if you persist, just wanted to mention the alternative before.)

Added an accessor as per chat. This is not yet mapped to C, but can be on demand (but so far, the repr(C) on the error type was not used either).