Re: [Json] Encoding Schemes

[Nico Williams <[e-mail]>]

On Tue, Jun 18, 2013 at 3:05 PM, John Cowan <[e-mail]> wrote:
> Carsten Bormann scripsit:
>
>> Clearly, the text in section 3 does not work with the
>> BOM-based CESs.  But you might also read the text in 6 as asking for
>> UTF-16 CES and the text in 3 then excluding BOMs so the UTF-16 CES is
>> implicitly big-endian.  So much effort for something so theoretical.)
>
> Unfortunately, it isn't even clear that it means that.  It could simply
> mean "UTF-16, either big-endian or little-endian".

I read it as "UTF-16, big-endian or little-endian, without BOMs".  I
read the RFC as specifying that JSON encodings are (not normative)
UTF-8, UTF-16 (either endianness, no BOM), or UTF-32 (either
endianness).  It may well be that JSON is only ever sent as UTF-8...

...But then, following the thread on what a JSON string is, I now
understand that JSON strings generally contain Unicode character data
but really they may contain any codepoints from the BMP, and, really,
any 16-bit values, with some values having to be escaped.  Of course,
given the whole discussion re: surrogates, JSON uses neither UTF-8 nor
CESU-8, but an amalgam of sorts, so, to wrap it up, JSON is intended
to be encoded in UTF-8 (and maybe -16 and -32), with all the caveats
about surrogates getting smuggled in because bad Unicode character
data in JSON strings bleeds into the encoding.

Nico
--