Re: [Cbor] Implementing float->int numeric reduction

[Anders Rundgren <[e-mail]>]

On 2023-08-16 8:02, Wolf McNally wrote:
> Anders,
> 
>> On Aug 15, 2023, at 8:38 PM, Anders Rundgren <[e-mail]> wrote:
>>
>> On the encoding level I don't find it logical or useful handling a probably rather small set of floating point data differently than the rest.
> 
> Am I correct that JCS, of which you are the first author, re-encodes “0.0” as “0”? By your own statement to do so is neither “logical” nor “useful”. So why did you bother? Why did you not just “let floats be floats”?

In this case the goal was to canonicalize IEEE 754 64-bit numbers. There could thus only be a single representation of a specific value.  To make it simple (...), we adopted the serialization method specified by ECMA.  Aided by a public algorithm and some 200 lines of rather intricate Open Source code (both well outside the team's competence), we got a solution.  That number serialization worked right out of the box for JavaScript gave us a head start.  However, this journey took 7 years!

Anyway, this is not directly translatable to CBOR, since CBOR is not built around a single "Number" type, uses binary encoding, and floating point support is even optional.  FWIW, I started my CBOR adventures early 2021 using https://cbor.me as a reference system.  My current encoding scheme is (probably) to 99.9% a copy of the reference.  My only "invention" is targeting the enterprise market with a "JSON killer" :)

Cheers,
Anders
reasonably good engineer, lousy salesman

> 
> https://datatracker.ietf.org/doc/html/rfc8785#name-serialization-of-numbers <https://datatracker.ietf.org/doc/html/rfc8785#name-serialization-of-numbers>
> 
> ~ Wolf
>