I have nothing to add: I think the undefined/unspecified/implementation-specified trichotomy is better spec language than ‘it is an error’, but as I wrote in my original email, I don’t think SRFI authors should be required to adopt it.

I see Sergei’s point about the implicit ‘should’ in ‘it is an error’ compared to the original C definition of UB, but note the macrological fascicle says that ‘undefined behaviour’ as used there is equivalent to the old Scheme reports’ ‘it is an error’. Ultimately, there is still no restriction on how an implementation is allowed to behave in any case, and an implementation of R5RS, R7RS small, or SRFI 257 which made demons fly out of your nose upon encountering any ‘error’ case would be entirely conformant.

Daphne

On 16 Dec 2025, at 00:13, Arthur A. Gleckler <xxxxxx@speechcode.com> wrote:

> The last call for SRFI 257 ended today.  Daphne hasn't replied to my question below, but her earlier message makes me think that she would like us to switch the language from "it is an error" to "undefined behavior" throughout.  That's a reasonable request.  Sergei, would you mind making that change?
>
> That's a small enough change that we won't need to wait for another review after that.  We can finalize the SRFI if you like.