| Description | Being able to use just the XML part of XDMF to describe data that is in a raw binary file (ParaView is able to use this feature) is a very useful alternative to having to have the heavy data in HDF, especially for codes that currently haven't moved (or won't, legacy for example) over to something like HDF5 and still produce their data dumps in raw binary. However, if I've understood XDMF as it stands, the limitiation is that one large dump file would have to broken down into many, many small binaries to represent each variable on the grid at each time point. This is a lot of post processing I/O. If however the 'binary' type specifier also allowed a byte offset into a raw file (almost analagous as what the path specifier after the : does for the H5) then a single raw file could be used with the offset specifying how many bytes to 'skip' to reach the actual data the XML is referencing. This would also allow a simple post-processing tool to be written to create the XML part of XDMF for any binary file. This then means there is no short term urgency to alter the main hydrocode to produce XDMF and also old/existing archived binaries could also be post-processed.
Given the availability of simple C functions like FSEEK using such an offset I would imagine is fairly trivial.
I realise that XDMF readers themselves would probably need to be updated to recognise such an addition though before it would actually work and its not just a matter of adding something to XDMF. |