H5easy Eigen vector shape question

[shivupa]

Reproducer: https://github.com/shivupa/highfiveh5easyquestion

#include <Eigen/Eigen>
#include <highfive/H5Easy.hpp>

int main(int argc, char **argv) {

    Eigen::Matrix<double, 10, 1> col_vec = Eigen::Matrix<double, 10, 1>::Random();
    Eigen::Matrix<double, 1, 10> row_vec = Eigen::Matrix<double, 1, 10>::Random();

    H5Easy::File file("example.h5", H5Easy::File::Overwrite);
    H5Easy::dump(file, "/col_vec", col_vec);
    H5Easy::dump(file, "/row_vec", row_vec);

    return 0;
}

Question: When dumping a eigen column vs row vector, I would expect the shape in the hdf5 file to be (nelem, 1) or (1,nelem) respectively. Instead when I dump using H5Easy::dump, I get the same shape

HDF5 "example.h5" {
GROUP "/" {
   DATASET "col_vec" {
      DATATYPE  H5T_IEEE_F64LE
      DATASPACE  SIMPLE { ( 10 ) / ( 10 ) }
      DATA {
      (0): 0.680375, -0.211234, 0.566198, 0.59688, 0.823295, -0.604897,
      (6): -0.329554, 0.536459, -0.444451, 0.10794
      }
   }
   DATASET "row_vec" {
      DATATYPE  H5T_IEEE_F64LE
      DATASPACE  SIMPLE { ( 10 ) / ( 10 ) }
      DATA {
      (0): -0.0452059, 0.257742, -0.270431, 0.0268018, 0.904459, 0.83239,
      (6): 0.271423, 0.434594, -0.716795, 0.213938
      }
   }
}
}

Is this expected and if so is there a way to get the (nelem, 1) or (1,nelem) shapes if desired?

[1uc]

This seems intentional, yes:
https://github.com/BlueBrain/HighFive/blob/master/include/highfive/h5easy_bits/H5Easy_Eigen.hpp#L48
https://github.com/BlueBrain/HighFive/blob/master/include/highfive/h5easy_bits/H5Easy_Eigen.hpp#L89

For row or column vectors you're likely not affected by the bug in core HighFive. Hence, you could try using something like

H5Easy::File file("example.h5", H5Easy::File::Overwrite);
file.createDataSet("row_vector", row_vector);
file.createDataSet("column_vector", column_vector);

(A H5Easy::File is simply a HighFive::File it seems.) If I read the output of your reproducer with the modification, it actually gets the order of columns and row right. However, I tend to mix them up. So please check it carefully before using it.

However, you should not start using core HighFive with Eigen matrices or anything genuinely 2D since core HighFive isn't aware that Eigen uses FORTRAN style indexing conventions; while everything else uses C-style.

[shivupa]

I see #532 thanks I'll be careful with 2D data. This solved my issue for 1D data. Thank you!