Canonical json support (RFC 8785 / JCS)

[lmussier]

Describe the proposed feature
Hi,

For json document hash use cases, it is mandatory to always have the same json representation for a same set of data.
This is the purpose of https://www.rfc-editor.org/info/rfc8785/

What other libraries (C++ or other) have this feature?
Seen it https://github.com/mirkokiefer/canonical-json and https://github.com/rezn-project/rezn-jcsd - not tested

Include a code fragment with sample data that illustrates the use of this feature
Exemples from here https://connect2id.com/blog/how-to-secure-json-objects-with-hmac (not my production)

Given

{"key":"9cea8d2d","name":"Alice Adams","age":21}

{"key":"9cea8d2d","age":21,"name":"Alice Adams"}

and

{
  "key" : "9cea8d2d",
  "name" : "Alice Adams",
  "age" : 21
}

The canonical version of it is a deterministic string whatever is the order of the keys on the input document.

std::cout << normalize(data) << std::endl;

Would give

{"age":21,"key":"9cea8d2d","name":"Alice Adams"}

and this for every format of the same data set.

[danielaparker]

I've thought about this. The main issues are:

• JSON object keys need to be sorted as arrays of UTF-16 code units.

• Floating point serialization is required to conform to the ECMAScript spec, which is different from the Grisu3 implementation in jsoncons.

We'd need to introduce a new template instantiation for basic_json, say cjson, with a new implementation policy.

It's doable, but would require a significant amount of work, and ECMAScript spec compatible floating point serialization would add a significant code footprint.