Add nested object capability to pointers

[clwill]

You currently can create a new object with

j["this"]["that"] = 27;

and if this and/or that don't exist, it's all good, it creates this as an object and puts that in it.

But if you use pointers (e.g. "/this/that") it throws an exception if this doesn't exist.

For example:

j.at("/this/that"_json_pointer) = 27;

would be nice.

[nlohmann]

I can confirm your observation - however, at is supposed to throw when applied to a nonexisting key.

Using j["/this/that"_json_pointer] = 27; still throws - this is a bug. What currently works is j["/this"_json_pointer] = 27;. That is, the operator[] for JSON Pointers shall be extended to work with arbitrary nested pointers even if parts are not existing yet.

[clwill]

Of course, doh!

Thanks, I'm dealing with a complex structure and the ability to not have to parse/check the existence of each level is quite helpful.

[clwill]

Also, don't know if you saw my comment on #278, but there is a parallel issue with value. I think that

j.value("/this/that"_json_pointer, 27);

should return 27 if either this or that are not present, not throw on a missing this.

Thx.

[gregmarr]

Presumably it's "/this/that"_json_pointer that's throwing, not .value() itself.
I think I misread that previous comment as _json_pointer was the part throwing.
Never mind.

[nlohmann]

Thanks for noting. The reason is that both functions use the same function inside the json_pointer class.

[clwill]

Got it. Didn't dive inside, it was late :)

[nlohmann]

The bug was that there was no code to cope when a JSON pointer was used on a null value. I'm on it.

[nlohmann]

I fixed the issue with operator[](const json_pointer&) and now shall have a look at your comment regarding value(const json_pointer&, ValueType).

[nlohmann]

Let's discuss the semantics of value with JSON pointers in #323.

@clwill - Can you please have a look at branch https://github.com/nlohmann/json/tree/feature/issue323 if this fixes the issue?

[clwill]

I've done some playing with the branch above, and it seems to work as expected. I like it!

[clwill]

BTW, just so you can understand my use case... I have a generic configuration module that puts all the other modules' defaults into one master .json file. The value setting function used to be:

template<typename T>
void setValue(const std::string keyName, T val) {
    jsonMaster[moduleName][keyName] = val;
}

Which worked and was OK. But with this update, I can do:

    jsonMaster[json::json_pointer("/" + moduleName + "/" + keyName)] = val;

The modules setting a value can now not only pass in a keyName of, say "this" but they can also make their own nested values of, say "this/that". This is very helpful.

Thank you.

[nlohmann]

Great. I shall merge the branch by the end of the week. However, I have no release date for 2.0.6 yet.

[clwill]

Thanks! Timing's OK, I can live with the [this][that] approach for the time being.

[nlohmann]

Merged the feature branch to develop.