[Paraview] h5 reader not avaialable?

[Michael Jackson]

On Apr 7, 2011, at 9:54 AM, Jeff Baumes wrote:

> On Thu, Apr 7, 2011 at 7:51 AM, Biddiscombe, John A. <biddisco at cscs.ch> wrote:
> Dave
> 
> >>>
> One way to go about making an automatic reader that should cut down on development time would be to write a utility to produce xdmf files from the content of arbitrary h5 files.
> <<<
> 
> https://hpcforge.org/projects/xdmfgenerator/
> wiki (with very little help!) here
> https://hpcforge.org/plugins/mediawiki/wiki/xdmfgenerator/index.php/Main_Page
> 
> wiki needs to be improved. building generator requires a paraview build to get xdmf and hdf settings using cmake.
> 
> 
> FWIW, Titan (titan.sandia.gov) has a vtkHDF5TableReader. It can find a single hdf5 dataset containing records and store it in a vtkTable. It is certainly limited, but people are free to enhance it. As Dave said, hdf5 is so complex I'm not sure how generally useful it can become. Right now it reads what I'm sure is a small subset of what could possibly be in hdf5.
> 
> Jeff
> 

Every once in a while I think about this idea and after going through the "thought experiment" I end up reimplementing the "HDFView" program that "The HDF Group" releases which is a generic HDF data viewer. I think the best they can do with regard to dataset dimensionality is 2D. I think ParaView could go to 4D with the Time Series functionality in PV. The tedious part is getting the information from the HDF5 file to create the "correct" vtk data structures, i.e. if the data set is a 16 bit signed integer data set then a 16 bit signed integer data set is created. Those are the first hurdles. Then how many data sets do you allow a user to render at any give time? Just one? A folder full of data sets? And how does one ultimately represent a bunch of datasets with vtk? MultiBlock? Multi-Dimensional?
  I love HDF5. I think it is great. I use it almost exclusively for all my data and analysis programs that I write. Someone said "The best thing about HDF5 is that you can put anything you want in it. The worst thing about HDF5 is that you can put anything you want in it."

  As Dave stated earlier this leads to domain specific readers because there is no way of knowing how the programmer intended the data to be read since there is an almost infinite possibility of organization of the data within the HDF5 file. Xdmf is probably the best solution at the moment.
  
  What I would like to see is the serialization of VTK objects into an HDF5 file. I have a few of these that I have done (vtkPolyData and vtkUnstructuredGrid - OpenSource) so that the power of HDF5 can be used on the writing side (HPC environments) and the speed of loading pure binary data on the reading side (ParaView client). Plus the files are portable and possibly compact (internal Zip compression in HDF5).
  
  Just my 2 cents worth of thinking this morning.

___________________________________________________________
Mike Jackson                      www.bluequartz.net
Principal Software Engineer       mike.jackson at bluequartz.net 
BlueQuartz Software               Dayton, Ohio