Re: [Json] JSON Schema Language

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

On Tue, May 7, 2019 at 5:14 PM Austin Wright <[e-mail]> wrote:

>
> I would put it slightly differently: While JSON doesn’t define any concept
> of an “integer”, applications that encode integers (e.g. uint32_t or
> similar) might conceivably emit it with the "frac” production (using a
> trailing decimal and zero); this is not ambiguous, and in the interest of
> being liberal in what you accept, parsers shouldn’t produce any error.
>

The problem that arise are:

1) The lexical analyzer may distinguish integers and floats and this may
cause a schema driven parser to reject legitimate number because it is
encoded as a double. (e.g. 1.0)

2) There may be precision loss because the number is presented in
exponential form (1.0E20)

3) There may be precision loss because the integer is larger than the
floating point mantissa can hold.

We should probably stop talking about 'float' values because I don't know
of any JSON tech likely to use them at this point. Pretty much every
programming language out there today will cause all calculations to be
performed in Intel 80 bit floating point until the data has to be stored in
a program data value when the mantissa will be truncated at 52 bits and
stored in a double.

Fact is that a 64 bit processor can do 64 bit mantissa ops more easily than
52 bit. But we are stuck in the choices of the 1980s when memory was scarce
and nobody had 64 bit CPUs yet.