The binary representation is meant to store and provide extensive data sets.
HDF5 seems to be the favoured format. But there is a discussion whether or not to use / allow attributes.
HDF5 (abd other containers) provide attributes for entries. The advantage is that ….
On the down-side, attributes are not forseen in all (relevant) representations. Especially YAML (and related languages) does not know them. It is in principle possible to mark entries in YAML in a way that they are recognised as attributed when converting to HDF5, but these marks and the required type-description contradict our requirement for easy readibility of the ASCII representation.
There are some frightening examples for this on the NXcanSAS pages.
- The attribute-free HDF5 might contradict the orso decision to stay compatible with canSAS and Nexus. This has to be checked.
- Do we insist on / recommend HDF5 for the binary representation?
- If so: with or without attributes?