Re: [Cbor] Chunks with tags inside indefinite-length string (major type 2 and 3)

Hi Dale,

I completely agree that extensibility (and, more generally, evolvability) is what can make or break a protocol.  One of the big differences between JSON and CBOR is that JSON was designed to be “stable” (fixed, non-extensible), while CBOR has well-defined extension points.

Indefinite length strings are a syntactic feature of CBOR that was not designed to be extensible.  I have a hard time imagining why one would want to extend it.  I don’t think this is proof by lack of imagination, either; this mechanism already has more complexity than one would reasonably like; it just happens to be needed.  There are no data model extensions that would need to extend this mechanism.  If there ever is a need to extend it, this would be much like using additional information values 28..30: A revision of the standard, a new version of the format.  For the present format, indefinite length strings of major type N (2 or 3) are composed of definite length strings of the same major type, only.

Grüße, Carsten

> On Dec 18, 2019, at 04:33, Dale R. Worley <worley@ariadne.com> wrote:
> 
> Carsten Bormann <cabo@tzi.org> writes:
>> Why would we allow tags here?  Indefinite strings are a very limited
>> syntactical structure; there is no need to allow flexibility here (in
>> particular if there is no semantic gain from that).
> 
> The reason I argue is that having worked with SIP, I've become very
> respectful of extensibility.  SIP has taken over the internals of every
> major telecom company, it seems.  And that doesn't seem to be because
> SIP is wonderfully designed; it's sort of a running joke to list the
> things one would change in the next version of SIP if it didn't have to
> be upward-compatible.  But SIP has a number of very general
> extensibility mechanisms that make it possible to implement almost any
> telecomm feature in an upward-compatible way.  And there are dozens if
> not hundreds of SIP extensions.
> 
> And while I cannot guess what use might be made of tagging chunks of
> indefinite-length strings, it seems to me that as long as the
> fundamental structure of CBOR allows such a thing to be constructed
> unambiguously, the wisest course of action is to make it clear that such
> constructs can sensibly be made, but the current standard assigns no
> meaning to them.
> 
> Dale
> 
> _______________________________________________
> CBOR mailing list
> CBOR@ietf.org
> https://www.ietf.org/mailman/listinfo/cbor