Re: [Cbor] 7049bis: The concept of "optional tagging" is not really used in practice #126

[Carsten Bormann <[e-mail]>]

> 
> However, I read "schema description" from the /semantic/ point of view:
> a description which explains the meaning of data items.  

Right.  “Schema” probably is one of the most misused terms in this space.
(That’s why CDDL is called a “data definition language”.)

> The IP RFC not
> only tells that the first 4bits are an unsigned int, it also tells that
> this number is the protocol version. 
> CBOR (neither JSON) can't tells this by itself, except if one defines a
> TAG for this.

Tags (we prefer to write simple English words in lower case) tell you how to interpret an enclosed item with different (additional) data semantics.  So a tag with number 1 tells you the enclosed number really is to be interpreted as a POSIX epoch-based date.

> So, the next question is: “is there some guidelines for using TAGs?"
> 
> Well, it's probably too early. One may have to wait that CBOR usages
> grow in maturity.

CBOR has been around for half a decade now, so I think we have a pretty good comprehension now of when to use tags.

> What should I decide for my system design regarding CBOR TAGs?

➔ Use tags when they are useful.

There are no general guidelines like the ones you propose below, because the usefulness depends on the specific context.

> Shall I:
> - prohibit TAGs since this is redundant with other parts of my design
> specifications (which already explicit the meaning of each field); or

If you have a relatively rigid data shape (“schema” in the usual structural sense), you may indeed not need tags, because you can infer an alternative interpretation from structure (e.g., field names in a map used as a struct, position in a record, etc.).  They may still be useful when you want to express a choice, e.g., if you want to support both epoch-based and text-based dates, use Tag 0 or Tag 1.  Another example is integers: If you expect to interchange integers that might not fit into 64 bits, use a choice between a built-in integer (major types 0 and 1) and a tag 2/3:

                  uint = #0
                  nint = #1
                  int = uint / nint
                  biguint = #6.2(bstr)
                  bignint = #6.3(bstr)
                  bigint = biguint / bignint
                  integer = int / bigint


> - put TAGs everywhere for everything because TAGs bring semantic to data; or

"Everywhere” I don’t know.  But if you expect your implementations to rely on generic decoders/encoders doing the work, using tags may be a labor-saving device.  This is particularly useful when CBOR is used for general serialization in a programming environment (where you may not have a hard and fast data definition with your data).

> - add TAGs to some fields and not to others (which ones and why?)

Yes.  Only add them when they are useful.  To express a choice, and/or to have the generic codec do the work.

Grüße, Carsten