Rely on MPI daemons being started.
mpi start scripts not needed any more

[[Imported from SVN: r4790]]

Use simple for loop instead of sed to insert commas.

[[Imported from SVN: r4789]]

Fixed duplicate dependencies.

Fixes Flyspray 198

[[Imported from SVN: r4788]]

Check added whether any modules were found and whether all modules we
depend on are found.

Fixes Flyspray 195

[[Imported from SVN: r4787]]

[[Imported from SVN: r4786]]

Fixes Flyspray issue 192.


[[Imported from SVN: r4785]]

[[Imported from SVN: r4784]]

Use C++ headers instead of C headers

[[Imported from SVN: r4783]]

[[Imported from SVN: r4782]]

[[Imported from SVN: r4781]]

[[Imported from SVN: r4780]]

[[Imported from SVN: r4779]]

Fixed a serious bug. (At least sum was not O.K. when working with an
inout array. We really have to allocate an out array and copy the
results.)

Made mpicollectivecommunication more generic and reduced code duplication.

There are to new functions (allreduce) than can be given a binary
function (e. g. std::plus) representing the calculation.

[[Imported from SVN: r4778]]

The collective operations now work with simple C arrays.

[[Imported from SVN: r4777]]

[[Imported from SVN: r4776]]

[[Imported from SVN: r4775]]

and therefor function calls lost. 


[[Imported from SVN: r4774]]

[[Imported from SVN: r4773]]

[[Imported from SVN: r4772]]

[[Imported from SVN: r4771]]

you can not rightmultiply a (n,m)-matrix
by a (n,n)-matrix but by (m,m)-matrices


[[Imported from SVN: r4770]]

[[Imported from SVN: r4769]]

[[Imported from SVN: r4768]]

[[Imported from SVN: r4767]]

[[Imported from SVN: r4766]]

[[Imported from SVN: r4765]]

Now configure works even if no fortran compiler is there.
Just how should be!

[[Imported from SVN: r4764]]

[[Imported from SVN: r4763]]

[[Imported from SVN: r4762]]

[[Imported from SVN: r4761]]

[[Imported from SVN: r4760]]

product.  Measuring the impact on efficiency gave the same results
as for the older TMP stuff.  When compiled with -O3 (g++ 4.1), there
is no measurable speed difference between the old and the new code.
That is so even though temporary variables had to be inserted.
When compiling without optimization, the new code is faster,
presumably because a standard for-loop is faster than a cascade
of function calls.  Depending on your vector size, this speedup
can be considerable!  For a six-entry FieldVector repeatly computing
norms gets about 40% faster.

This concludes FlySpray issue 46.

[[Imported from SVN: r4759]]

[[Imported from SVN: r4758]]

NDEBUG is not defined. 


[[Imported from SVN: r4757]]

which can be used for void interface methods. 


[[Imported from SVN: r4756]]

DUNE_DEVEL_MODE defined, because using this macro leads to an extra method
call. 



[[Imported from SVN: r4755]]

[[Imported from SVN: r4754]]

[[Imported from SVN: r4753]]

[[Imported from SVN: r4752]]

disable dune libcheck, because AC_TRY_LINK fails, although everythins is OK. Perhaps a libtool problem

[[Imported from SVN: r4751]]