Writing to virtual datasets


Is writing to a virtual dataset meant to work? The only document I can find that says anything about this is the outdated RFC document, which mentions using H5Dwrite on a virtual dataset.

We were asked about this in h5py. I found it partly works in practice, but there seems to be a problem with data conversion - when I wrote int64 data to an int32 virtual dataset, it appeared to skip conversion, so every second value was 0 (details on the h5py issue).


Thomas, I believe you are right, but more competent people will have to chime in. On paper and in “straightforward” (TBD) cases, write operations are supported. The write w/ datatype conversion and size change might be one of those undocumented restrictions of the current VDS implementation.

Congratulations on the h5py 3.2 release!