Re: [weirds] data formats (was one protocol to rule them all)

[Eric Brunner-Williams <[e-mail]>]

On 2/18/12 1:13 PM, Avri Doria wrote:
> I think that in general data models need to be extensible and that is a default requirement.  

Extensibility is both abstract and concrete. The concreteness is the
(possibly implicit) model, which, in theory, could be useful in the
application problem domain. The abstractness may make implementation
an interesting problem in computer science.

As others have pointed out, life before XML Schema was iffy, and life
after XML Schema has disproportionate complexity.

As others have pointed out in related venues e.g., PROVREG, and
DOMAINREP, implementation of extension of a protocol syntactically
expressed in XML can be difficult, in theory and in practice.

As a choice of working idioms, semantics via a decade old subset of a
text markup language, for which validation tools are recent, and
limited to text markup use, or semantics via a subset of C, for which
syntax validation tools (compilers, editor modes, etc) have existed
for a generation, and common to a vast array of programming idioms and
uses, is a reasonable software architecture and engineering choice.

My view is that the gang of N (Scott, Jordyn, Ross, ... self) erred in
2002 in selecting XML syntax to specify the semantics of a replacement
to the then prevalent provisioning protocol specified in key-value
pairs. See http://tools.ietf.org/html/draft-ietf-provreg-grrp-reqs-06,
where we'd not yet taken the step of committing to a syntax for
command payload.

So my two beads are for JSON, or rather, ease of implementation of a
known application domain, and therefore, ease of extension within the
implied and express constraints, that is, for C programming as the
idiom rather than text and its markup.

Eric