There is no guarantee that it matches the topology of the actual machine
in any way.  Also, it does not appear to be used anywhere in the ug core
at all.  Finally, it contains some small memory leaks.  Admittedly, these
leaks would be easy to fix, but since the code is not used I may as well
rip it out altogether.

[[Imported from SVN: r8530]]

UG implements a small hack to save memory and communication bandwidth:
the NODE data structure contains a VECTOR*, which points to the node data.
If there is no node data then this pointer is not used.  In that case,
instead of allocating sizeof(NODE) bytes for each NODE, the program
allocates sizeof(NODE)-SIZEOF(VECTOR*) bytes instead, and promises to
never touch the missing bytes.

(The same is done for edges).

This 'special' computation is done twice: once in ugm.c, where the memory
for NODEs is allocated, and once in initddd.c, where the DDD communication
layer is told about the NODE type.  The special hack, i.e., the subtraction
of sizeof(VECTOR*) from the object size in case no node data was used,
was massing in initddd.c.  This lead to valgrind giving 'Invalid read' errors
during load balancing (see below).

This patch fixes the size computation in initddd.c, and hence avoid the
valgrind errors.

==1863== Invalid read of size 2
==1863==    at 0x4C2DA60: memcpy@@GLIBC_2.14 (mc_replace_strmem.c:877)
==1863==    by 0x99457C: UG::D3::XferPackSingleMsg(UG::D3::_XFERMSG*) (pack.c:564)
==1863==    by 0x994A36: UG::D3::XferPackMsgs(UG::D3::_XFERMSG*) (pack.c:762)
==1863==    by 0x9842D0: UG::D3::DDD_XferEnd() (cmds.c:611)
==1863==    by 0x9B132C: UG::D3::TransferGridFromLevel(UG::D3::multigrid*, int) (trans.c:835)
==1863==    by 0x9AADC0: UG::D3::lbs(char const*, UG::D3::multigrid*) (lb.c:659)
==1863==    by 0x96D684: UG::D3::LBCommand(int, char**) (commands.c:10658)
==1863==    by 0x4B475F: Dune::UG_NS<3>::LBCommand(int, char const**) (ugwrapper.hh:1019)
==1863==    by 0x4B9CD6: Dune::UGGrid<3>::loadBalance(int, int, int, int, int) (uggrid.cc:534)
==1863==    by 0x42828E: Dune::UGGrid<3>::loadBalance() (uggrid.hh:491)
==1863==    by 0x41D79D: bool Dune::UGGrid<3>::loadBalance<LoadBalance::LBDataHandle<Dune::UGGrid<3>, std::vector<Dune::FieldVector<double, 3>, std::allocator<Dune::FieldVector<double, 3> > >, 3> >(LoadBalance::LBDataHandle<Dune::UGGrid<3>, std::vector<Dune::FieldVector<double, 3>, std::allocator<Dune::FieldVector<double, 3> > >, 3>&) (uggrid.hh:517)
==1863==    by 0x4117B6: void LoadBalance::test<Dune::UGGrid<3> >(Dune::UGGrid<3>&) (test-parallel-ug.cc:473)
==1863==  Address 0xcdf690a is 2 bytes after a block of size 104 alloc'd
==1863==    at 0x4C2935B: malloc (vg_replace_malloc.c:270)
==1863==    by 0x6562F5: UG::GetFreelistMemory(UG::HEAP*, int) (heaps.c:618)
==1863==    by 0x8252A5: UG::D3::GetMemoryForObjectNew(UG::HEAP*, int, int) (ugm.c:326)
==1863==    by 0x825814: CreateNode(UG::D3::grid*, UG::D3::vertex*, UG::D3::geom_object*, int, int) (ugm.c:584)
==1863==    by 0x8350FA: UG::D3::InsertBoundaryNode(UG::D3::grid*, void**) (ugm.c:4731)
==1863==    by 0x845384: UG::D3::InsertMesh(UG::D3::multigrid*, UG::D3::mesh*) (ugm.c:6661)
==1863==    by 0x831ACC: UG::D3::CreateMultiGrid(char*, char*, char*, unsigned long, int, int) (ugm.c:3297)
==1863==    by 0x958EEB: UG::D3::NewCommand(int, char**) (commands.c:1640)
==1863==    by 0x4D7C1E: Dune::UG_NS<3>::NewCommand(int, char**) (ugwrapper.hh:1028)
==1863==    by 0x4DF3EA: Dune::GridFactory<Dune::UGGrid<3> >::createGrid() (uggridfactory.cc:465)
==1863==    by 0x411425: Dune::StructuredGridFactory<Dune::UGGrid<3> >::createCubeGrid(Dune::FieldVector<double, 3> const&, Dune::FieldVector<double, 3> const&, std::array<unsigned int, 3ul> const&) (structuredgridfactory.hh:189)
==1863==    by 0x40ADF2: void testParallelUG<3>(bool) (test-parallel-ug.cc:518)

[[Imported from SVN: r8529]]

[[Imported from SVN: r8528]]

[[Imported from SVN: r8527]]

Because, according to Steffen, MSIZE is already used by some OS X
system header.

[[Imported from SVN: r8526]]

Thanks to Steffen Müthing for the patch.

[[Imported from SVN: r8525]]

UG_PARALLEL and UG_FOR_DUNE can now be checked by CMake.

[[Imported from SVN: r8524]]

Fix problem with user-defined string literals.

Thanks to Steffen Müthing for the patch.

[[Imported from SVN: r8523]]

[[Imported from SVN: r8522]]

[[Imported from SVN: r8521]]

[[Imported from SVN: r8520]]

[[Imported from SVN: r8519]]

[[Imported from SVN: r8518]]

[[Imported from SVN: r8517]]

[[Imported from SVN: r8516]]

[[Imported from SVN: r8515]]

[[Imported from SVN: r8514]]

[[Imported from SVN: r8513]]

[[Imported from SVN: r8512]]

[[Imported from SVN: r8511]]

[[Imported from SVN: r8510]]

[[Imported from SVN: r8509]]

[[Imported from SVN: r8508]]

[[Imported from SVN: r8507]]

Copyright holders are precisely all those people that directly
contributed to the code.  This list is assembled from the commit
history.  Hence we have all author logins, but not all actual
names yet.

[[Imported from SVN: r8506]]

[[Imported from SVN: r8505]]

Still incomplete, more information needed

[[Imported from SVN: r8504]]

last revision.

[[Imported from SVN: r8503]]

The were local paths that only existed on my particular machine in the test.
Now we use the values provided by configure as we should have from the 
beginning.
I am sorry for having made such a newbie mistake previously.

[[Imported from SVN: r8502]]

[[Imported from SVN: r8498]]

This will be used eventually for dynamic load balancing from Dune.
This patch only adds the data fields, so code from Dune that uses
them does at least compile.

[[Imported from SVN: r8497]]

This makes the DimX = ..., DimY = ..., ... output disappear for good.

[[Imported from SVN: r8496]]

[[Imported from SVN: r8495]]

This should fix the build breakage introduced a few patches ago

[[Imported from SVN: r8494]]

[[Imported from SVN: r8493]]

After 15 years of testing, let's simply assume that the code
in question works.

[[Imported from SVN: r8492]]

This Makefile contained a hack from the times when the AutoTools
build system for UG was still incomplete: for header installation,
all header files that 'find' would find in the source tree were
simply copied to the installation directory.  But now the AutoTools
system is in good shape and the headers can be installed in the
standard way.  Therefore the old hack can be removed.

[[Imported from SVN: r8491]]

[[Imported from SVN: r8490]]