Re: [Json] Canonicalization

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

We should avoid the need to canonicalize JSON whenever possible, but there are enough efforts to define a canonical form that it would be worth standardizing one.

Strings:
Escaping is mandatory for controls chars, double quote, and backslash (%x00-1F / %x22 / %x5C).
The simplest string canonicalization rule would be to escape those 34 chars and no others. A rule that might be slightly nicer for people (and hence worth the few extra lines of code) would be to always use the 7 \x escapes for those 7 chars, always use \uxxxx for the rest of %x00-1F, and never use it for any other chars. I would be tempted to drop \/ from the list of seven (only in the canonicalization rules), because / is so common in URIs etc.

Objects:
Sorting string names to canonicalize an object needs a few more words.
Presumably sorting occurs on the logical strings, not the (canonically) encoded versions. So {"a\"b":1,"a#b":2} is in canonical form ("(U+22) < #(U+23) < \(U+5C)).
Presumably sorting uses Unicode code points, not UTF-16 words, or UTF-8 bytes. So {"\uFFE0\uFFE1":3,"\uD834\uDD1E":4} is in the right order (0xFFE0 < 0x01D11E), thought the canonical form would use UTF-8 not \uxxxx for the 3 characters.
[The 21-byte canonical form would be (in hex):
7B 22 EFBFA0 EFBFA1 22 3A 33 2C 22 F09D849E 22 3A 34 7D]

Numbers:
0, 1, 1e3, 2.334e-5, 1.5e6 are the sort of canonical form numbers should have. I don’t like 0.0E0 for zero as per draft-staykov-hu-json-canonical-form-00 -- a person would never write that. That draft also allows any number of trailing 0’s (eg 1.200000e-2), which is a bug. A canonical form should drop the exponent when it is zero, and drop the decimal point when there is nothing after it.
A regex for numbers in canonical form:

0|-?[1-9](\.[0-9]*[1-9])?(e-?[1-9][0-9]*)?


So draft-staykov-hu-json-canonical-form needs a few changes in my mind, but is as good a starting point as any.

--
James Manger