one true XDMF to rule them all
VTK and ParaView support two versions of XDMF (2 and 3) simultaneously. Supporting both is a maintenance burden, a source of confusion among new users, and splits the attention of developers and potential funding sources.
The feature sets and bug status of 2 and 3 are mostly overlapping, but there are differences that matter. We should drop support for 2 because the code base is more dated. However before we do that we have some work to do on 3 to catch up on areas of the feature set, robustness or performance is currently inferior.
Things that need attention:
- finish removal of boost dependence from 3 now that Visual Studio 2015 is VTK's minimum compiler.
- fix cases where 3 reader fails on files that 2 accepts (default attributes lacking in many cases)
- bring back support for parallel decomposition of structured data
- address advanced array expression differences with code support or xml file translation scripts
- address element reference (reuse) differences between the two versions
- deprecate xdmf2 in a vtk release
- remove xdmf2 in a subsequent release
To do the work well we will need to gather a number of data files. These should be made easily available for users to use as examples and at the same time added to VTK's regression test infrastructure.