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

[Carsten Bormann <[e-mail]>]

On 2024-12-11, at 16:18, Rohan Mahy <[e-mail]> wrote:
> 
> I agree with Joe on this one.
> 
> > What would be the right size then? 250 million?  220 million?
> How many 1+1 tags and 1+2 tags should you be able to consume?  FIVE per RFC seems like a reasonable limit. I have no objection to allocating as many 1+4 or 1+8 tags as you'd like.

What we discussed previously is that we have a budget to spend.

Here is the bank account statement of today:

$ tag-report.rb
range  used     %                 free                total
0 1+0    13 54.17                   11                   24
1 1+1    73 31.47                  159                  232
2 1+2  1087  1.67                64193                65280
3 1+4 65539  0.00           4294836221           4294901760
4 1+8     2  0.00 18446744069414584318 18446744069414584320

Counting RFCs is unlikely to be the best way to manage the budget.
Counting years is more like the “design for decades” Mantra that underlies the CBOR design.

So let’s assume we want to have 30 more years before CBOR 2.0 (or whatever the next big thing will be called) takes over.
As people have said 1+4 and 1+8 seem relatively uncritical, but of course we don’t want to give them away all in one piece.
Let’s focus on the ranges that have usage different from 0.00 %.

1+0 has 11 free code points, this means we should not spend one more than about every three years, and I think we all understand that these will be used for some important purposes only.  RFC 7049 came with 9 code points taken, and we have since spent 4 more (on COSE, which is rather foundational but maybe didn’t really need four 1+0 tags), which is about 12 years’ worth of our budget and thus slightly faster than we should have.  If cbor-packed gets its share, the next Marshmallow will come up in 2028 then.

1+1 has 159 free code points, or about 5 per year.  (We are slightly above that rate with 73 code points used in 11 years, but not dramatically so, and we have slowed down recently.)  Spending 8 years of budget for cbor-packed may seem preposterous, but many of us believe that cbor-packed is worth that.

1+2 has 60000+ code points, or about 2000 per year, except that half of it is FCFS, so we only have about 1000 per year under expert review control.  Spending 5 years of that budget again may sound preposterous and may partially already be in the diminishing returns space, but the 1+1 allocation is really small in the grand scheme of things so we need to have *some* allocation in 1+2.  Saying “your table sizes will never be as large as your 1+2 allocation” is true for constrained applications; I’m not so sure about general purpose computers though.

Grüße, Carsten