[Cbor] Re: Early allocation for packed CBOR (Re: Reminder: CBOR WG Virtual Meeting on 2024-12-11)

[Vadim Goncharov <[e-mail]>]

On Wed, 11 Dec 2024 11:45:16 -0500
Michael Richardson <[e-mail]> wrote:

> Michael Jones <[e-mail]> wrote:
>     > A typical draft allocating tags allocates a number that you can
>     > count on one hand.  Each such tag has a human-readable
>     > description of its purpose in the registry so people know what
>     > to use them for.  
> 
> Sure.
> 
> But, CBOR packed isn't the same as, for instance, RFC9164 (v4/v6

Yes. And more, it's not *justified*.

> tag), or even for RFC9277 which allocated 64K tags to match all
> future (CoAP) Content-Format values.   RFC9277 explicitely encouraged
> use of 4-byte (1+4) tags, mind you.

There is no objection for 4-byte range.

>     > I'd suggest that a reasonable maximum number of tags for a
>     > draft to allocate for its purposes is on the order of ten.  
> 
> a normal draft, sure.
> CBOR Packed is infrastructure for use by other applications.
> It's essentially a significant update to STD94.

Such an update should not grab such big number of tags from
applications, while performing notoriously bad for it's goal. Such
infrastructure update should use unused codes from main specification.
For example, exerpt from my CBAR (CBOR & generic BLOB by-Atom Reduc) idea:

   Internal pass: for CBOR, use unused codes 28..30 in each Major Type. For
   integers, it's shorter one (e.g. 3-byte), for each string type, prefix it
   with atom length as prescribed by CBOR, for tags, put common like
   stringref, and in special, make generic escape mechanism:

   e.g. 0xfc to escape one byte
   0xfd - extended escape, and e.g. for atoms in middle of strings

  FC - copy 2+ bytes literal
  FD - dispense (decompress) atom
  FE - escape a single byte

and all 0xFx codes switch off CBOR going to BLOB mode
[...]
  * for BLOB media type, initialize "remaining_bytes" to entire length, and go
    to INBLOB state
  * in CBOR state every opcode is allowed, and special one for strings emits
    CBOR string prefix, then sets remaining_bytes to string's length
  * in BLOB state, only FC/FD/FE allowed, any other MUST throw error, and
    decrement remaining_bytes for every emitted piece
  * when remaining_bytes becomes zero, go back to CBOR state

so, for strings:

  5C - 2+ byte length (varint) BLOB state inside
  5D - decompress one atom
  5E - single byte length BLOB state inside

and same for UTF-8 text strings (7C/7D/7E)

-- 
WBR, @nuclight