Multidimensional compound attributes?

I've rarely worked with the HDF5 "array" datatypes, to my knowledge.
Instead I normally use a multidimensional dataspace with a base
datatype. For example, currently with each H5Table I wish to associate
some "recording session metadata". This is a four-field struct, and I
will generally only have one entry in this struct, but sometimes there
may be up to ~5.

Instead of having a non-scalar dataspace with a H5T_COMPOUND type, is it
smarter to create an array data type with the H5T_COMPOUND as the
base-type? In your experience, is tool support better for one approach
vs the other? Is one approach the "right" way?

The (excellent) python package Pytables is by far my preferred way of
interacting with HDF5, but I haveusers who will be using both the Java
HDF5 interface and matlab's built-in HDF5 support, so I'd be curious to
know if anyone on the list thinks there's any virtue in one approach
over the other. Of course, I'm writing my files using the HDF C++

Thanks again,



This mailing list is for HDF software users discussion.
To subscribe to this list, send a message to
To unsubscribe, send a message to