return json objects from functions

[itodirel]

Given a function like json json_add(json& j, std::string name), which given a json object, adds another object to it and returns it, I expect that if I call:

json j;
json_add(j, "1");
json j2 = json_add(j, "2");
json_add(j2, "3");

The last call to json_add would add another object on an existing one created before, to create a hierarchy, but instead nothing happens, is there a way achieve this programatically?

json j = 
{
    { "filename", "file.cpp" }
};
json j2;
j2["prop"] = "1";

j["j2"] = j2;

j2["prop2"] = "2"; // after adding j2 to j, their reference is lost, so further updates to j2 does nothing

Expected:

{
  "filename": "file.cpp",
  "j2": {
    "prop": "1"
    "prop2": "2"
  }
}

[gregmarr]

j["j2"] = j2;

This stores a copy

j2["prop2"] = "2";

This modifies the original.

j["j2"]["prop2"] = "2";

This modifies the version in j.

[itodirel]

@gregmarr that would not work, if you need to do it programatically, N-levels deep, which I need to :)

I understand the value semantics and the copies, but that doesn't mean that internally each "node" can't track each other. Without it, you can't build and patch hierarchies easily, and it's unnatural.

What I had to do is to make a pointer and find the element I want to patch using .at, and return it by reference.

[itodirel]

the other thing, it would be cool if there was an add or append API that specifically returns a json reference of the element that was just added

[gregmarr]

I'm having a hard time visualizing what it is that you're trying to do. Maybe this:

auto &nested = j["j2"];
nested = j2;
nested["prop2"] = 2;

[nlohmann]

I agree with @gregmarr in that I do not understand the issue yet. Could you try to describe your expected outcome and the difference to the actual behavior in more detail?

[itodirel]

Here is an example:

json& add_node(json& node, std::string text)
{
    json new_node;
    new_node["node"]["text"] = text;
    node["nodes"].push_back(new_node);
    return new_node;
}

And the calling code, which creates a hierarchy but doesn't work because of the value semantics and copies:

json j;
add_node(j, "1");
add_node(j, "2");
add_node(j, "3");
json& node = add_node(j, "4");
add_node(node, "1");

I have to write to something like this, which does work, but is hacky:

json& add_node(json& node, std::string text)
{
    json new_node;
    new_node["node"]["text"] = text;
    node["nodes"].push_back(new_node);
    json::json_pointer p(std::string("/nodes/") + std::to_string(node["nodes"].size() - 1) + "/node");
    return node.at(p);
}

But then I eventually wrote it as, which is better:

json& add_node(json& node, std::string text)
{
    json new_node;
    new_node["node"]["text"] = text;
    node["nodes"].push_back(new_node);
    return node["nodes"].back()["node"];
}

I still expect an API that appends and returns a reference directly though, something like:

json& add_node(json& node, std::string text)
{
    json new_node;
    new_node["node"]["text"] = text;
    return node["nodes"].push_back(new_node);
}

[nlohmann]

As explanation: The API is the same as for a std::vector, so push_back is void.

[itodirel]

personally, I don't think that's not sensible, and I think it's inconsistent to have a push_back API, on something that might not be an array, but when used becomes an array, at which some of the other APIs wouldn't work, it's very unintuitive

[nlohmann]

I don't understand: for arrays, push_back behaves just like std::vector::push_back. Only for null values we extended the behavior to allow for an implicit conversion to an arrays. Why is it unintuitive to return void?

[itodirel]

not because it returns void, but because a json object has a vector API and non vector APIs on it, all together

[nlohmann]

I see. That's this library's way to deal with JSON's different value types. When you only see a json type, you don't know which type it is. What would be the alternative? Throwing?