Re: [Cbor] Reminder and call for agenda: CBOR WG Virtual Meeting on 2023-05-31

[Emile Cormier <[e-mail]>]

On Fri, May 26, 2023 at 1:35 AM Anders Rundgren <
[e-mail]> wrote:

> Based on recent postings on the mailing list, the CBOR WG seems to rather
> be drifting towards the opposite goal; creating sector- or/and
> application-specific encoding profiles like dCBOR, "fintech", or breaking
> away from the crippling past (C/C++). Since I have no interests in such
> ventures, I won't attend the meeting.
>

I'm curious to know how C/C++ has anything to do with CBOR and the
direction you think it's going. Except perhaps for a few example code
snippets, CBOR seems pretty language-agnostic to me.

While implementing C++ encoders for "important" CBOR tags, I found that
some of them were fighting against the C++ way of doing things. In
particular, the extended time tags want base 10 exponents that are
multiples of 3, whereas C++ std::chrono allows any arbitrary ratio for its
time unit. Another one is CBOR bignums that are not signed magnitude like
how most C++ bignum libraries are implemented.

I've come to accept that CBOR cannot please everyone. It has to contend
with competing pressures: encoding size vs ease of decoding, determinism vs
ease decoding, the ease in which competing programming languages can
generate/consume CBOR, etc.

As much as I've grown attached to CBOR, I would like to see a binary
serialization format that sits somewhere between CBOR and Protocol Buffers,
where bytes could be directly copied to the programming language variables
(as much as possible), yet there is no fixed schema requiring an IDL. This
would be my wish list for such a hypothetical format:

- Little endian!
- Two's complement signed integers instead of separate types for
positive/negative. Rules provided for determinism if an application
requires it (e.g. signed vs unsigned positive integers).
- Tags are purely for semantics and don't encode a portion of the data item
(I'm looking at you, positive/negative bignum!). It should be possible to
strip away the tag while transcoding to JSON, with the receiver being able
to reconstitute the item if they know what type to expect.
- Tags are only for items that have a general purpose and are not
application/protocol/vendor-specific

Cheers,
Emile Cormier