Re: Lua and JSON

["Thomas Harning Jr." <harningt@...>]

On Tue, Feb 9, 2010 at 2:19 PM, Fabio Mascarenhas <mascarenhas@acm.org> wrote:
> On Tue, Feb 9, 2010 at 4:52 PM, Thomas Harning Jr. <harningt@gmail.com>
> wrote:
>>
>> On Tue, Feb 9, 2010 at 1:34 PM, Fabio Mascarenhas <mascarenhas@acm.org>
>> wrote:
>> > On Tue, Feb 9, 2010 at 4:17 PM, Thomas Harning Jr. <harningt@gmail.com>
>> > wrote:
>> >>
>> >> My LuaJSON projects uses LPeg for parsing and the strict mode passes
>> >> all tests provided by the JSON validator project from JSON.org  The
>> >> non-strict mode permits some more loose handling of constructs.
>> >>
>> >
>> > +1 for Thomas' module, it works great. Thomas, have you thought about
>> > having
>> > json.encode honor a "__tojson" metamethod (along with a rawencode
>> > function),
>> > so it is easier to extend the encoder to understand app-specific data
>> > types?
>> Hm... that is a good idea.  I would probably have it be something of the
>> sort:
>>
>>  __tojson(self, rawencode)
>>
>> rawencode passed in so that you can take advantage of some of the
>> infinite recursion avoidance, as well as encoding options.  If I get
>> it right, I'd expect that rawencode would skip the __tojson field of
>> the immediate object and handle inner objects.
>>
>
> My idea was to json.rawencode just what json.encode does right now and
> ignore __tojson metamethods, just like rawget/rawset/raw*. The __json
> metamethod could receive the options the user passed to json.encode.
The problem is that it loses certain encoding state, such as the list
of already encoded tables to prevent infinite recursion.

I should have a preliminary version up on the 'next' branch with an
implementation.
>
> But it is important that __json is a metamethod instead of a regular field,
> just to keep things idiomatic, even though having to work with metatables
> can be a drag sometimes. :-)



-- 
Thomas Harning Jr.