Added XdmfExodusReader to utils directory
Added exodusii and netcdf sources to xdmf to support XdmfExodusReader.cxx and XdmfExodusConverter
Modified install locations for executables to prefix/bin
Modified install location for python scripts to be correct for all platforms (was previously incorrect for OS X)
Adjusted python install process to work for all Xdmf incantations, irregardless of how much is built (python was previously working only when Utils was built)
Python testing automatically sets correct paths and works automatically
Major improvement to XDMFConfig.cmake file for applications that build against Xdmf
More correct information in XDMFConfig.cmake --- e.g. no need to include hdf5 directories in addition to xdmf, an INCLUDE_DIRECTORIES(XDMF_INCLUDE_DIRS) should be sufficient when building against Xdmf
Fixes to Xdmf.h to include newer Xdmf classes like XdmfSet

ENH: Modified Xdmf configuration so that the Xdmf (libsrc) can be compiled and used directly as a library on Windows by some external projects (For example, NetDMF)

ENH: Modified Xdmf configuration and added some logic to get around python library linking issues when building on Windows so that the Xdmf (libsrc) can be compiled and used directly as a library on Windows by some external projects (For example, NetDMF)

ENH: Add XDMF_BUILD_RPATH and fix some places where tests were being built even without BUILD_TESTING turned on

BUG: These were broken when wrapping python without BUILD_UTILS set on.  XdmfUtil will have its own python file separate from the main Xdmf python so that they no longer interfere.

STYLE: Renamed files to maintain cxx extension like the rest of Xdmf.  Also made a change to XdmfDiff to write out all data to heavy format for more accurate comparisons.

ENH:When the Origin and Spacing are written out, they are done so in the same order, but they are not 'named' so when reading the file one is not sure which is which. By naming them, we can read the xmf file better and the order is not so important. Not really a bug fix, but might help.

BUG:If H5_USE_16_API is specified then don't use the hdf5-1.8.x interface, go back to the 1.6 interface

comes to 2D/3DCoRectMesh and uses vtkImageData instead of vtkUniformGrid.

ENH:Before writing attributes to disk, sort them into alphabetical order. This is important if the attributes were added in some random order and change in two different blocks of a multiblock structure. Any classes which access by index end up with erroneous arrays. Very Rare, but this sort helps prevent segfaults in later filters down the pipeline after reading the data back.

ENH: Add support for reusing static geometry / topology from one timestep to the next. Before calling CreateGeometry or CreateTopology you must find the previous xml node holding ghe data you wish to reuse, when passed in, with appropriate flags, the build heavy is bypassed by setting the XML directly. Not enabled by default, but working in CSCS test code and will gradually be moved into CVS.

ENH:some reworking of the writer2 class to allow subclasses to do modify the way data is written

BUG: Fix new xdmf writing of vector and tensor arrays, while not breaking unstructured data geometry.