Re: [review][JSON] Review of JSON starts today: Sept 14 - Sept 23

From: Peter Dimov (pdimov_at_[hidden])
Date: 2020-09-16 15:19:56

Vinnie Falco wrote:

> So what is being discussed here is "partial CBOR support?" In other words, 
> only the subset of CBOR that perfectly overlaps with JSON?

Well, it's self-evident that parsing a json::value from, and serializing a 
json::value to, CBOR, only could ever support the subset of CBOR that's 
representable in json::value. Fortunately, that's almost all of CBOR, 
because that's what CBOR is for.

> > (There is actually a fully compatible way to support "binary values" in 
> > json::value, but it will require some hammering out.)
>
> Well, let's hear it!

You may remember my going on and on about arrays of scalars at one point. 
This is what MessagePack gets right - arrays of scalars are an important 
special case and stuffing them into the general value[] representation 
wastes both memory and performance.

If json::array supported internally arrays of scalars, without changing its 
interface at all so it still appeared to clients as a json::array of 
json::values, we could represent a binary value as a json::array (which 
internally uses a scalar type of unsigned char.)

This magically solves everything - value_to<vector<unsigned char>> works, 
value_from same works, it roundtrips.