[Paraview] Parallel file formats (again)... (UNCLASSIFIED)

[John Biddiscombe]

Further to what everyone else has written on the subject.

Here at CSCS we are convinced that HDF5 is really the only way to 
proceed for parallel IO and we try to encourage all our users to adopt it.
The trouble is that we have a proliferation of 'formats' - they are all 
hdf5, bur as you know there are a myriad way of arranging groups, 
heirarchies, different vector scalar combinations etc. We have something 
like 6 or 7 different hdf5 readers for vtk/paraview. Image/particle/mesh 
etc etc. XDMF allows the possibility of writing a simple xml wrapper 
around each file and then having a single reader which can handle them 
all (in principle at least). XDMF stores the where/form/shape in the xml 
so that its easy to tell the reader to find this group, that dataset and 
put it into this place in the vtk structure.

Of course, the vtk/xdmf support it still quite tatty and needs a lot of 
work, but I am using it whenever possible because every time I fix a bug 
in it, it supports all my hdf formats and not just one. (in principle 
again!).

JB