This is not yet tested.

...we have no idea why it was in there. But tests should not ignore warnings.

Due to the elimination of variables some statements appeared redundantly.
I removed them.

valgrind reports that some uninitialized bytes are in a buffer to be written out to cdn.hdf.
The bytes surely are in that part of the buffer that does not contain relevant data. To me
The reported line for the error source is the close operation for the file. Therefore
it is not clear whether this originates from fleur or from HDF5. I can only identify that
this appeared while writing out the stars. Therefore I initialized every associated array
that did not seem to be initialized before setting the relevant elements to certain values.

I had to start valgrind with --num-callers=25 to see at least the fleur context for the
reported missing initializations.

...but Lord Valgrind demands that I correct them. :)

It seems to work now for at least one system

Esepecially jij%nmagn was already needed in rw_noco which is called in
a subroutine of dimens. Therefor it has to be initialized before dimens
is called. I hope there are no side effects of this relocation.

Changed cdntot to use convol to calculate interstitial charge. Should be significantly faster for large systems.
Command line switch "-oldfix" can be used to switch back to old algorithm.

This should not really be relevant. But I do everything my master valgrind tells me.

These initializations are relevant for the old inp file if no enpara file is used.

valgrind complained.

The problem is that this array has size 2 and in hsmt_nonsph the maximum
value of this array is used for some array allocation. If the system under
investigation only has a single spin the 2nd value is undefined.

(found with valgrind, only theoretical error source so far)

The issue was due to the reading of a c string into a fortran character array.
Up to now all 255 elements were read into the array and later the output
was reduced to the part in front of the c string termination null character.

Of course, this is wrong and valgrind complained. Everything after the string
termination character might not belong to the program. Therefore I fixed this.
But to my knowledge this issue never produced a problem.

The cause for this error might also be in fleur. Blaming LAPACK only is unfair.

This commit is the first of probably several similar commits that adds
missing initializations found by valgrind.

Special thanks for this commit go to Aristid Breitkreuz and Christoph Dittmann
who pushed me to trying out valgrind.

distinguished from the version 2016.05.004

- do not use qfix file anymore
- do not run qfix if it is not specified in inp.xml
Fixes #96