Re: [Json] Limitations on number size?

[Nico Williams <[e-mail]>]

On Tue, Jul 9, 2013 at 5:46 PM, Peter F. Patel-Schneider
<[e-mail]> wrote:
> But everything that I've read about JSON indicates that JSON is supposed to
> be used for (portably) interchanging data.  If all there is to JSON is the
> grammar then what's the point of JSON?

+1.  In practice agreeing on a subset of typing results in interop, at
least for implementations of specific applications.

Sometimes, when straying past that subset, there's loss of
information, and that mostly works out.  Sometimes parsers reject
inputs, this results in implementors figuring out what happened and
how to improve interop, or else the user doesn't bother reporting the
problem.

We can definitely identify a subset of JSON that is extremely likely
to interop successfully:

 - strings being Unicode character strings, possibly with paired
surrogates (there are still some implementations, no doubt, that can't
handle code points outside the BMP)

 - top-level is an array or object (though many parsers have options
to allow other top-level types, so I don't think this is too
important)

 - no dup names (parsers might or might not reject, or deconflict)

 - numbers as 32-bit signed integers, or maybe even numbers as IEEE
754 64-bit doubles -- TBD from a survey we might never do

Note that the proposed I-JSON need not be that subset of JSON most
likely to interop, just the subset we want for Internet protocols.
And we might be happy enough with I-JSON that we never define any
other subsets of JSON.

RFC4627's underspecification, coupled with host languages with widely
varying attributes led to JSON being a superset than most
implementations of it.

ECMAScript JSON is also a subset of RFC4627, and is neither a subset
nor superset of the proposed I-JSON.

> What is (surprisingly) missing in JSON is a firm notion of the data being
> represented by the syntax.   A pale shadow of this issue for strings appears
> to have been consuming this working for quite some time.

Yes.

Nico
--