Proposal: Pure HDF5 schema
Writing XML files is somewhat onerous and error-prone for applications that are free to choose their own HDF5 structure, and it limits the workflow because appending to a time series while ensuring the file is never corrupted is difficult. It would be easier to write directly to an HDF5 file with attributes to define the metadata that's normally written to the XML file. Some projects do this and have converter scripts (e.g., petsc_gen_xdmf.py) to write the XML file from an ad-hoc HDF5 schema. But for a new project or one that doesn't need backward compatibility within their legacy format, they'd have a simpler and more robust experience writing pure HDF5 and ditching the bespoke converter.
I'm curious if the community would be amenable to proposals for an HDF5 schema and converter script so that users can opt to write pure HDF5 and it will behave to readers (such as Paraview) as though they had run the XML converter and read that.
Other append-friendly formats suitable for parallel IO could be considered as well. The key is that conversion to the XML representation should be well defined.