12.04.2013 4:37, Michael Caisse пишет:

FWIW, the json example parser in Spirit is based on our json library that we finally pushed to github this week. Links and docs (soon) can be found here: http://cierelabs.org

The goal was to create a json library that allows usage similar to javascript or Python. I believe the parser is fully compliant and we would love to get some feedback from users.

Hopefully it is something useful to the community. I'm glad to hear it but I have to say that for the software development it is important to minimize the number of the external libraries count.

Every external dependence (i.e. library) leads to the additional costs: - the license compatibility must be checked - the version compatibility must be checked - the library must be built and installed - if the library is abandoned, it's a problem - etc. One of the benefits of the Boost is that it is licensed, tested and distributed as one thing. And I think that it would be good for your json parser if it was included into the Boost. PS. For now, I'm using my own implementation of the json parser. I would be happy to switch to the external one if it was included into one of the libraries I'm already using. -- Best regards, Sergey Cheban

Bjorn Reese wrote:

On 04/10/2013 11:46 PM, Stephan Bourgeois wrote:

...

Open Source JSON parsers have already been implemented in C++ and in Java. Examples of Java libraries are: Gson, quick-json. Even if other libraries already exist, developers who are using Boost for their project will appreciate having a JSON parser within Boost.

Agreed.

...

The question is what data structure we should use to represent JSON objects, and how the user can access key/value pairs in those objects. (examples: Boost.PropertyTree, pre-existing C++ object, ...)

I would like to have several different interfaces:

1. Tokenizer API which reads the next token from an input string. This is important for streaming data.

2. Iterator API which iterates to the next token in the input string. Remembers its parent scopes (unlike tokenizer.) This is similar to the XmlTextReader API.

3. Tree API which parses the entire input string into a tree structure. This is a bit like the DOM API, and this is what the Spirit example and Boost.PropertyTree provides.

4. Serialization API which provides a Boost.Serialization input archive without going through an intermediate tree representation.

I would think to add perhaps a fusion API which I think should allow you to have a usage similar to that of JsonFX in C# for any fusion adapted type. http://codepad.org/mu0VD9LG

2013/4/12 Michael Marcin

Arindam Mukherjee wrote:

...

In JSON we typically deal with maps and arrays. The arrays themselves could have arbitrary types (string, object, array, numeric, boolean, null) as elements. The key types in the maps are always strings and the value types in the maps could be anything that can appear in an array, including another map or array.

Due to this, I'd imagine being able to use Boost.Variant or Boost.Any in a list and as a value_type in a map would help.

Probably.

I find the most useful interface is to just provide a datatype you're expecting and let the json parser try its best to do the right thing.

Very like what I experienced with Spirit before, yeah, it just works. I let the user specify the desired types though traits specialization. The old code is here: https://github.com/jamboree/jsume/blob/master/example/pretty_printer/json_co...

For example:

string raw_json = R"({ data:{ a:"hello", b:"world", c:3, widget:3.5

} })";

This doesn't seem like a valid json, the key must be a quoted string.

struct my_type1 { map data; };

struct my_type2 { map> data; };

struct my_type3 { map data; };

struct my_type4 { unordered_map data; };

struct my_type5 { struct my_data { string a; string b; int c; float widget; } data; };

It's impossible for my_type5 without more advanced Fusion adaption.

On 04/12/2013 07:18 PM, Michael Marcin wrote:

In practice you often find json that does not quote keys and I would prefer a useful library to a pedantic one for this task. You could have a strict mode I suppose.

I pretty much just go by what http://jsonviewer.stack.hu/ accepts.

If we are to accept an extended syntax, then it should be absolutely clear what those extensions are. I have no idea what the URL above accepts. Apart from unquoted keys, two other extensions I have seen are C-style comments, and support for the floating-point values of infinity and NaN. Are there other potential extensions?