It's not immediately clear how DataArray instances should be represented in HDF5:
http://fperez.github.io/datarray-doc/0.0.6/design_questions/issues.html#how-do-i-save-a-datarray-in-hdf5-using-h5py
However, there seems to be an official standard for storing axis dimensions and labels in HDF5 that apparently maps quite naturally to DataArray:
http://www.hdfgroup.org/HDF5/doc/HL/H5DS_Spec.pdf
Since the specification proposes to store the metadata in regular HDF5 datasets, this could probably implemented in h5py without resorting to the raw HDF5 API.
It's not immediately clear how DataArray instances should be represented in HDF5:
http://fperez.github.io/datarray-doc/0.0.6/design_questions/issues.html#how-do-i-save-a-datarray-in-hdf5-using-h5py
However, there seems to be an official standard for storing axis dimensions and labels in HDF5 that apparently maps quite naturally to DataArray:
http://www.hdfgroup.org/HDF5/doc/HL/H5DS_Spec.pdf
Since the specification proposes to store the metadata in regular HDF5 datasets, this could probably implemented in h5py without resorting to the raw HDF5 API.