Re: [Json] Limitations on number size?

["Peter F. Patel-Schneider" <[e-mail]>]

*Any* number?  Not so!  Assuming sane rules, JSON numbers are inadequate for 
representing all rationals, and the same holds, of course, for any superset of 
the rationals, including reals and complex.  (Yes, yes, no usable data format 
can represent all the reals or complex numbers.)

JSON numbers are unlike the numeric type system of any programming language 
that I had been familiar with, even LISP.  That JSON numbers are so close to 
ECMAScript numeric literals is, to me, a very strong indication that JSON 
numbers are supposed to represent ECMAScript numbers.

I do agree that implementations may be range-restricted, and that a data 
interchange format shouldn't always nail the ranges down exactly.   However, 
floats (being rather unusual numbers) are one place where it can be fruitful 
to nail things completely down.

Peter F. Patel-Schneider



On 07/09/2013 04:24 PM, Jorge Chamorro wrote:
> On 10/07/2013, at 00:46, Peter F. Patel-Schneider wrote:
>>
>> OK, then, what I inferred was that a JSON number represents an IEEE floating point double and I was very surprised to find out that this was not the case.
> And given the super-simple rules in the RFC *any* number can be properly represented which is -it seems to me- exactly what Crockford had in mind: that JSON (the data interchange format) does not impose the limits.
>
> That the platforms, languages, libraries, and stringify()er and parse()r implementations may have numeric data type limitations, isn't JSON's (the data interchange format) business.