Agree on handling of JSON value types

[mattheworiordan]

Creating this issue to cover the discussion at #51

Probably better if we continue the discussion here as an issue

[paddybyers]

Copying my comment from there:

OK, you're right, the RFC is explicit: "JSON text is a serialized object or array"

http://www.ietf.org/rfc/rfc4627.txt

By contrast, the ECMA spec permits any text that satisfies the grammar for a JSON value (which does include primitives):

http://www.ecma-international.org/publications/files/ECMA-ST/ECMA-404.pdf

So I agree remove support for primitives.

[mattheworiordan]

Ok, so do we explicitly reject data that is not a supported type, or allow it but then let the developer deal with the possible crash on the other end. I think we should explicitly reject unsupported types so that we can guarantee type portability between all client libraries.

[paddybyers]

Yes, on balance I think it should be rejected.

[paddybyers]

.. but then there is the issue of how we claim to police the supplied values. For example, NaN can't be represented in JSON - so do we check for those as well?

[mattheworiordan]

Nope, I think it is acceptable that we support String, Binary (ByteArray), and JSON Array / Objects. If you have a value in there that the native serialiser accepts, such as NaN becoming null then I think that is acceptable and something the developer should be able to figure out themselves given we say we support JSON.

[paddybyers]

ok, agreed

[mattheworiordan]

Please see ably/ably-ruby#48 where I have implemented this