Hi Carsten,

Because the serialization draft says so. After the discussion of -00 we
dropped our registry and took "dtrm" as the CBOR WG defines it, and the NaN
rule came with the name. -01 adds no NaN rule of its own: Section 3.1
points to Sections 4.1 and 5.1 of draft-ietf-cbor-serialization.

No implementation problem led to it. The one NaN surprise was in our tests:
struct in Python 3.13 drops the payload of a half-precision NaN, and in
3.14 keeps it, so the reference code decides from the encoded bits.

Best,
Anton

On Tue, Nov 17, 2026 11:07 AM, Carsten Bormann <[email protected]> wrote:

> On 2026 Nov 17, at 09:49, Anton Sokolov <[email protected]>
> wrote:
> >
> > so non-trivial NaNs are excluded.
>
> Is that because the serialization draft says so, or did you have actual
> implementation problems?
>
> Grüße, Carsten
>
>
_______________________________________________
COSE mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to