Issue with ordered_json

[amirmasoudabdol]

I'm trying to replace my json object with the ordered_json but some of my to/from_json serializer are not working properly anymore.

So, I have these set of methods to read and write a vector to arma::Row<T> (which is a representation of a vector in Armadillo library).

namespace arma {
template <typename T> void to_json(json &j, const arma::Row<T> &row) {
  j = arma::conv_to<std::vector<T>>::from(row);
}

template <typename T> void from_json(const json &j, arma::Row<T> &row) {
  row = arma::Row<T>(j.get<std::vector<T>>());
}
}

This works fine if I invoke it like this, auto means = j.at("means").get<arma::Row<double>>(); but when I switch to ordered_json I get an error saying: "No member named 'insert' in arma::Mat<double>". If it matters, I actually use this line, auto means = j.at("means").get<arma::Row<double>>(); inside an adl_serializer for another object, like below, and btw the same happens when I want to get the arma::Mat<double>>.

namespace nlohmann {
template <typename T> struct adl_serializer<baaraan::mvnorm_distribution<T>> {

  static void to_json(json &j, const baaraan::mvnorm_distribution<T> &mdist) {

    j["dist"] = "mvnorm_distribution";
    j["means"] = arma::rowvec(mdist.means().as_row());
    j["sigma"] = mdist.sigma();
  }

  static baaraan::mvnorm_distribution<T> from_json(const json &j) {

    auto means = j.at("means").get<arma::Row<double>>();
    arma::Mat<T> sigma;
    auto n_dims = means.n_elem;

    sigma = j.at("sigma").get<arma::Mat<double>>();

    return baaraan::mvnorm_distribution<T>{means.t(), sigma};
  }
};
}

I'm not exactly sure why this is failing, and how I can fix it.

[amirmasoudabdol]

OK, I've tracked this done. I still don't know the reason behind this, but without using ordered_json, I could have defined the serialized to/from_json inside arma namespace and wouldn't have any issue. However, ordered_json doesn't accept this, can cannot find them. So, I've to moved them into the nlohmann namespace, and define a adl_serializer for them.

@nlohmann I'm not sure if this is intended, or whether I've done this incorrectly from the first place, but maybe you like to look into this.