[Cbor] Re: Data model versus serialization (was Re: Scrapping the CDE "Application Profile" concept)

[Anders Rundgren <[e-mail]>]

On 2024-08-19 19:51, lgl island-resort.com wrote:
> Trying for a bit of clarity on data model versus serialization...
> 
> == Data model ==
> - Fundamentally, the data types supported — the CBOR generic data model (plus extended data model)
> - Most protocol designs are in terms of data types (not serialization)
> 
> == Serialization ==
> - Says how to encode a particular data type
> - A particular serialization is chosen for message size and implementation constraints
> - Also, specific serialization is needed for deterministic encoding
> - Serialization choices don't narrow or otherwise change the CBOR data model
> - Actual implementations of a protocol must agree on serialization to achieve interoperation
> 
> Seems to me that JSON has no such split at all. ASN.1 has a little one with DER/BER. CBOR is quite different in having this separation. Necessarily so to support very constrained environments.
> 
> Preferred Serialization, CDE and Basic/CIE are primarily serialization specifications (however, they stray into data model with the reduction of big numbers and integers (RFC 8949 3.4.3)).
> 
> dCBOR is primarily about the data model. It specifies no new serialization, just re uses CDE. Its main contribution is reduction of floats and integers and exclusion of a few data types.
> 
> Does this sound right?

I don't bother too much about the terminology, more about the consequences.  Carsten talks about mapping.  Numeric reduction is mapping?

Anders

> 
> LL