Re: [Json] serializing sequences of JSON values

[Tatu Saloranta <[e-mail]>]

On Fri, Mar 14, 2014 at 3:02 AM, "Martin J. Dürst"
<[e-mail]>wrote:

> Hello Matt,
>
>
> On 2014/03/14 07:23, Matt Miller wrote:
>
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA512
>>
>> On 3/13/14, 3:23 PM, Larry Masinter wrote:
>>
>>> I wonder if there is any lesson from XMPP--  which wanted
>>> sequence-of-XML-values in an indefinite length stream -- that would
>>> apply to JSON. What would XMPP look like if it were redone in JSON
>>> instead of XML?
>>>
>>>
>> /me exchanges JSON chair hat for XMPP enthusiast hat ...
>>
>> If XMPP didn't exist and we instead did JMPP, it might be a series of
>> discrete objects.  However, the direction of the later "XMPP over <x>"
>> adaptations (e.g., [BOSH] and [XMPP-WS]) speak toward wanting to use
>> an existing transport that provides block-oriented framing around
>> complete structures.
>>
>> Dealing with partial XML-like structures in XMPP has ranged from
>> inconvenient to impractical because the vast majority of existing XML
>> software has assumed a whole XML document when parsing starts.  There
>> are few libraries that truly allow one to feed XML in fits and starts
>> and get timely output without some amount of modification or
>> preprocessing.  When the XMPP community took efforts to apply XMPP
>> over other transports, the direction has so far leaned toward the
>> whole documents that most implementations expect; even though the
>> preprocessing required is rather small, the ability to re-use existing
>> implementations without first considering oft-unpublished details was
>> a very powerful motivator.
>>
>
> This is very valuable feedback.
>
> As far as I understand, the problem for XMPP has at least two parts:
> 1) The XML document arrives in pieces.
> 2) Parsing is done in pieces
>
> My understanding is that parsing in pieces is well covered by SAX parsers,
> so that the problems essentially came from 1). But problem 1) doesn't apply
> to JSON text sequences exchanged as MIME types, and it doesn't apply to log
> files as these are rarely read, and when read, that could just be done with
> something like
> > echo ']' > json-array-close.txt
> > cat logfile.json json-array-close.txt | json-parse
> (using [pseudo-]shell syntax).
>
> As for delayed exchange of multiple JSON texts, e.g. a (completely
> hypothetical) JMPP (JSON Messaging and Presence Protocol), one would try to
> use a message-oriented transport (e.g. WebSockets) and then not need a
> concept of JSON text sequences at all.
>
>
Major challenge with sequence of XML documents is really just implementors,
and inertia.
Having written a fully conformat XML parser that optionally supports
"fragment mode" (Woodstox for Java; implements SAX and Stax APIs) I can say
that compared to all other work, supporting multiple root-level documents
is a minor addition. Mode is used/useful for supporting XMPP, as well as
for use for streaming logging, similar to what has been discussed for JSON.
It has to be explicitly enabled (since such usage is non-compliant, having
multiple roots).

There are some minor details (like xml declaration, what to do if it is
repeated) to deal with,
to allow simple concatenation, but all in all it is not a big technical
problem. I mean, compared to writing actual fully compliant parser
(supporting DTD is 80+% of the deal).

-+ Tatu +-



> Regards,   Martin.
>
>
> _______________________________________________
> json mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/json
>