Re: [Json] Canonicalization

["Manger, James H" <[e-mail]>]

> On Wed, Feb 20, 2013 at 11:58 PM, Manger, James H
> <[e-mail]> wrote:
> > Canonicalization (c14n) may be a dirty word, and barred from the
> charter, but I hope it isn't too inappropriate to still discuss it on
> this email list.
> >
> > There is a better c14n format for numbers than I or draft-staykov-hu-
> json-canonical-form have suggested: it is the format produced by
> JSON.stringify, as defined in ECMAScript v5.1 [http://www.ecma-
> international.org/ecma-262/5.1/#sec-15.12.3]. Examples: 0, 1, -1000,
> 0.00002334, 1500000, 6.022e+23, 9.1e-28.
> >
> > In fact, c14n for JSON can (and should) be simply defined as per
> JSON.stringify, with two extra constraints: object name/value pairs are
> sorted; and \uxxxx escapes use lowercase hex digits (a-f). Done.
> Developers even have a readily available reference implementations
> simply by typing JSON.stringify(...) into the JavaScript console of
> their browser.

> Not done. By saying that it should be done as per JSON.stringify(), you
> basically preclude the use of JSON for languages which _do_ have the
> ability to represent numbers outside of IEEE 754, and that includes
> numbers as simple as 0.1.
> 
> It is not because ECMA 262 has made the braindead decision of limiting
> numbers to IEEE 754 that JSON should follow way. RFC 4627 didn't make
> that mistake and I certainly hope any IETF-related work concerning JSON
> will not make such a mistake either.

Done (almost).

The JSON.stringify format for numbers is NOT limited to 64-bit floats. It effectively says format the number without an exponent if it is in the range [1e-6, 1e21); otherwise format it with 1 digit before the decimal point. There is no limit to the number of digits, ie the precision.

--
James Manger