It seems we use the numpy dtype.metadata to carry (event) metadata over in the nside dictionary.
I find pretty counter-intuitive to have metadata of the data embedded in the dtype, but I understand this may be the only numpy-supported way for storing metadata in npz files.
Since we are moving towards JSON as a primary exchange format, I suggest to migrate all the dtype.metadata logic in to_npz() and read_npz(), so that this "feature" is no longer part of the core class design.
It seems we use the numpy
dtype.metadatato carry (event) metadata over in thensidedictionary.I find pretty counter-intuitive to have metadata of the data embedded in the
dtype, but I understand this may be the only numpy-supported way for storing metadata innpzfiles.Since we are moving towards JSON as a primary exchange format, I suggest to migrate all the
dtype.metadatalogic into_npz()andread_npz(), so that this "feature" is no longer part of the core class design.