Re: [apps-discuss] JSON Schema considered harmful

[Phillip Hallam-Baker <[e-mail]>]

On Thu, Sep 20, 2012 at 12:53 PM, Tim Bray <[e-mail]> wrote:

> On Thu, Sep 20, 2012 at 9:26 AM, Erik Wilde <[e-mail]> wrote:
>
> > i have to admit that i am really curious why the seasoned spec writers
> that
> > have voiced general concerns about the usefulness of schema languages are
> > feeling this way, but so far i have not really understood what they are
> > saying. could somebody please try again, i'd really like to understand
> why
> > the general idea of describing protocol syntax in a semi-formal (as
> opposed
> > to non-formal) way is considered a bad idea.
>
> I gave a preso at IETF70 in Vancouver a few years ago which is advertised
> as being XML-centric, but has (I think) useful things to say about
> plain-text/binary/ASN.1/JSON/XML, and discusses about the proper place of
> schemas: http://www.ietf.org/proceedings/70/slides/xmltut-0.pdf
>
> I believe:
> - ASN.1 has been bypassed by history and has truly horrible tooling. Just
> don’t go there.
>

Agree


> - If you’re interchanging documents, use XML; if (much more common) you’re
> interchanging data structures, use JSON.
>

Agree


> - Do not use multiple syntaxes for the same protocol. Pick one of XML or
> JSON and stick with it.
>

Agree but note that this conflicts with the REST notion of using URL
encoded requests and JSON responses. Which I think is yucky but some insist
on.


> - The most important thing in defining a protocol is good clean
> comprehensible human-readable prose.
>

Agree [hence I do not want to see a JSON or an XML syntax]


> - All the protocols I’ve ever seen have important constraints that can’t
> be expressed usefully in declarative schema languages.
>

Agree and furthermore I have found a good use for the constraints they do
provide.


> - Therefore, the next most important thing is an automated
> validator/test-suite.
>

Agree and want to lock that to the schema and the specification (which I
have done in ProtoGen)


> - A schema is a nice-to-have.
> - If you’re in XML and want schema-ware, use RelaxNG. For an example, see
> RFC4287.
> - If your protocol is JSON, investment in a validator/test-suite has much
> higher ROI than chasing schema unicorns.


Unless you have a tool that will produce the test-suite and the spec (and
other schemas if you really must).


-- 
Website: http://hallambaker.com/