Respect padding in native HDF5 files

[Francesc Alted]

Hi PyTables devs,

Some months ago, Ken Walker complained about PyTables getting rid of the paddings in compound datatypes of native HDF5 files during copies using PyTables.  After reading this, I thought it would be nice to have this support, and after asking (and getting) for a small grant to NumFOCUS, I decided to have a go at it.  You can see my current solution at:

https://github.com/PyTables/PyTables/pull/720

With this, PyTables can create tables with paddings as long as they come from NumPy arrays with paddings (i.e. paddings in NumPy structured arrays are respected), and the original paddings are respected during copies too.  I believe that this is a change for good (it ensures a smoother interaction with the existing HDF5 ecosystem), and I don't forsee incompatibilities with exisiting PyTables files (compound types without padding are perfectly possible, and still the default).  Still, the changes went quite deep into the type mechanism, so I'd appreciate a review :)

Thanks for your attention!

--

Francesc Alted