Creating JSON arrays in Boost using Property Trees

Question

I'm trying to create a JSON array using boost property trees.

The documentation says: "JSON arrays are mapped to nodes. Each element is a child node with an empty name."

So I'd like to create a property tree with empty names, then call write_json(...) to get the array out. However, the documentation doesn't tell me how to create unnamed child nodes. I tried ptree.add_child("", value), but this yields:

Assertion `!p.empty() && "Empty path not allowed for put_child."' failed

The documentation doesn't seem to address this point, at least not in any way I can figure out. Can anyone help?

Answer 1

What you need to do is this piece of fun. This is from memory, but something like this works for me.

boost::property_tree::ptree root;
boost::property_tree::ptree child1;
boost::property_tree::ptree child2;

// .. fill in children here with what you want
// ...

ptree.push_back( std::make_pair("", child1 ) );
ptree.push_back( std::make_pair("", child2 ) );

But watch out there's several bugs in the json parsing and writing. Several of which I've submitted bug reports for - with no response :(

EDIT: to address concern about it serializing incorrectly as {"":"","":""}

This only happens when the array is the root element. The boost ptree writer treats all root elements as objects - never arrays or values. This is caused by the following line in boost/propert_tree/detail/json_parser_writer.hpp

else if (indent > 0 && pt.count(Str()) == pt.size())

Getting rid of the "indent > 0 &&" will allow it to write arrays correctly.

If you don't like how much space is produced you can use the patch I've provided here

Answer 2

In my case I wanted to add an array to a more or less arbitrary location, so, like Michael's answer, create a child tree and populate it with array elements:

using boost::property_tree::ptree;

ptree targetTree;
ptree arrayChild;
ptree arrayElement;

//add array elements as desired, loop, whatever, for example
for(int i = 0; i < 3; i++)
{
  arrayElement.put_value(i);
  arrayChild.push_back(std::make_pair("",arrayElement))
}

When the child has been populated, use the put_child() or add_child() function to add the entire child tree to the target tree, like this...

targetTree.put_child(ptree::path_type("target.path.to.array"),arrayChild)

the put_child function takes a path and a tree for an argument and will "graft" arrayChild into targetTree

Answer 3

Simple Array:

#include <boost/property_tree/ptree.hpp>
using boost::property_tree::ptree;

ptree pt;
ptree children;
ptree child1, child2, child3;

child1.put("", 1);
child2.put("", 2);
child3.put("", 3);

children.push_back(std::make_pair("", child1));
children.push_back(std::make_pair("", child2));
children.push_back(std::make_pair("", child3));

pt.add_child("MyArray", children);

write_json("test1.json", pt);

results in:

{
    "MyArray":
    [
        "1",
        "2",
        "3"
    ]
}

Array over Objects:

ptree pt;
ptree children;
ptree child1, child2, child3;


child1.put("childkeyA", 1);
child1.put("childkeyB", 2);

child2.put("childkeyA", 3);
child2.put("childkeyB", 4);

child3.put("childkeyA", 5);
child3.put("childkeyB", 6);

children.push_back(std::make_pair("", child1));
children.push_back(std::make_pair("", child2));
children.push_back(std::make_pair("", child3));

pt.put("testkey", "testvalue");
pt.add_child("MyArray", children);

write_json("test2.json", pt);

results in:

{
    "testkey": "testvalue",
    "MyArray":
    [
        {
            "childkeyA": "1",
            "childkeyB": "2"
        },
        {
            "childkeyA": "3",
            "childkeyB": "4"
        },
        {
            "childkeyA": "5",
            "childkeyB": "6"
        }
    ]
}

hope this helps

Answer 4

As of boost 1.60.0, problem persists.

Offering a Python 3 workaround (Gist), which can be syscalled just after boost::property_tree::write_json.

#!/usr/bin/env python3


def lex_leaf(lf: str):
    if lf.isdecimal():
        return int(lf)
    elif lf in ['True', 'true']:
        return True
    elif lf in ['False', 'false']:
        return False
    else:
        try:
            return float(lf)
        except ValueError:
            return lf

def lex_tree(j):
    tj = type(j)
    if tj == dict:
        for k, v in j.items():
            j[k] = lex_tree(v)
    elif tj == list:
        j = [lex_tree(l) for l in j]
    elif tj == str:
        j = lex_leaf(j)
    else:
        j = lex_leaf(j)
    return j


def lex_file(fn: str):
    import json
    with open(fn, "r") as fp:
        ji = json.load(fp)
    jo = lex_tree(ji)
    with open(fn, 'w') as fp:
        json.dump(jo, fp)


if __name__ == '__main__':
    import sys
    lex_file(sys.argv[1])

Answer 5

When starting to use Property Tree to represent a JSON structure I encountered similar problems which I did not resolve. Also note that from the documentation, the property tree does not fully support type information:

JSON values are mapped to nodes containing the value. However, all type information is lost; numbers, as well as the literals "null", "true" and "false" are simply mapped to their string form.

After learning this, I switched to the more complete JSON implementation JSON Spirit. This library uses Boost Spirit for the JSON grammar implementation and fully supports JSON including arrays.

I suggest you use an alternative C++ JSON implementation.