Are dump and parse guaranteed to be inverse to each other?

[Mathias-Fuchs]

The following example seems to imply that any json can be dumped and re-parsed, and will stay the same. Is this guaranteed to work on any nlohmann::json object?
Serializing stuff is the point of json. So, it would be great if the README.md was clear about this point. So far, the README only speaks about the other direction, parsing and then dumping a string, which is less useful to know.

       nlohmann::json j;
        j["k"] = {"v1", "v2"};


        std::string jj = j.dump();
        nlohmann::json jjj = nlohmann::json::parse(jj);

        std::cout << j << std::endl; // {"k":["v1","v2"]}
        std::cout << jjj << std::endl; // couts the same 

[gregmarr]

I would say no, it wouldn't be guaranteed, since there can be things in the json object that can't be represented in a JSON file, such as NAN or infinity, or non-utf8 strings.

[nlohmann]

Right - if you take an arbitrary json value, then dump may throw in case of invalid UTF-8 or lose information (e.g., by serializing NaN to null). But even if your json value was created by parsing JSON, the result of dump() may not result in the same JSON text as indentation or exact number formatting is not preserved.