insert(or merge) object should replace same key , not ignore

[itviewer]

json j1 = {{"one", 123}, {"two", 123}};
json j2 = {{"one", 456}};
j1.insert(j2.begin(), j2.end());

As key "one" has exist, now the result j1 is still {{"one", 123}, {"two", 123}}
while I think more commonly used logic is replace the same key，that is the result should be {{"one", 456}, {"two", 123}}

Thanks

[nlohmann]

This method should behave equivalent to std::map::insert: The documentation there states:

Inserts element(s) into the container, if the container doesn't already contain an element with an equivalent key.

[itviewer]

It was a pity that we limited by std::map::insert does not overwrite the exist key.
I'm not familiar with c++ stl, but I often use python and js ,either python's dict or js‘s array are implemented to replace the exist key. the approach of std::map::insert is really counter-intuitive.

So what is the problem with the operator [] ?example as

void merge(json &j1, const json &j2) {
    for (json::const_iterator it = j2.begin(); it != j2.end(); ++it) {
        j1[it.key()] = it.value();
    }
}

[nlohmann]

Interestingly, there will be a std::map::merge function in C++17 which again states:

If there is an element in *this with key equivalent to the key of an element from source, then that element is not extracted from source.

[nlohmann]

From Python 3.6, we could borrow the ideas from update (emphasis added):

Update the dictionary with the key/value pairs from other, overwriting existing keys. Return None.

update() accepts either another dictionary object or an iterable of key/value pairs (as tuples or other iterables of length two). If keyword arguments are specified, the dictionary is then updated with those key/value pairs: d.update(red=1, blue=2).

So this would basically be your code with a different interface:

void merge(const_reference j)
{
    if (not (is_object() and j.is_object()))
    {
        JSON_THROW(type_error::create(305, "cannot use merge with " + type_name()));
    }

    for (auto it = j.begin(); it != j.end(); ++it) {
        m_value.object->operator[](it.key()) = it.value();
    }
}

Any thoughts on this?

[jaredgrubb]

std::map::insert_or_assign basically does this, although it does not operate on a range, just one element. I would say that is the most discoverable name.

[itviewer]

I think @nlohmann 's suggestion ok for now, The interface name can be either merge or insert_or_assign，but the function implementation no need to use c++ 17's std::map::insert_or_assign，until c++ 17 becomes a formal standard and is supported by mainstream compilers.

[jaredgrubb]

Yes, I would not use the C++17 function until this library is C++17 only. It's very easy to implement the functionality. My suggestion was just about the naming.

[nlohmann]

I would implement insert_or_assign once I focus on C++ beyond C++11. I think the name is misleading if it would be used for more than a single value.

I tend to add the merge function I described in #661 (comment). However, I am unsure about the name, because std::map::merge has the semantics we do not want.

What about going the Python way and use update?

[itviewer]

What about going the Python way and use update?

It's an acceptable solution

[nlohmann]

Via Twitter, I got further proposals:

• apply
• combine
• insert_or_update
• extend

[DmitryKuk]

upsert as an option

[nlohmann]

Added a functionupdate. I shall merge the feature branch once the Travis build succeeds.