Redirect latexs stdin from /dev/null so it will actually fail if
   something breaks instead of waiting for input.

doc/istl/istl.tex doc/istl/comm/communication.tex
   Make it build with the new tetex in Debian/etch which uses
   pdfelatex in dvi-mode.

[[Imported from SVN: r523]]

[[Imported from SVN: r522]]

[[Imported from SVN: r521]]

[[Imported from SVN: r520]]

[[Imported from SVN: r519]]

[[Imported from SVN: r518]]

[[Imported from SVN: r517]]

[[Imported from SVN: r516]]

procedure is not correct.

As the pre method of the smoother might (and in parallel actually does) change the
right hand and left hand side, we now copy that information to the outside world.

[[Imported from SVN: r515]]

[[Imported from SVN: r514]]

wrapped sequential preconditioner. This is needed by AMG to automatically
construct the smoother hierarchies.

For Convenience and old code the templates parameter defaults to the abstract class
Preconditioner.

[[Imported from SVN: r513]]

[[Imported from SVN: r512]]

[[Imported from SVN: r511]]

[[Imported from SVN: r510]]

  them into am/latex.
* import the "Praktukumsbericht" about Refinement into cvs.

[[Imported from SVN: r509]]

Move the two images used by doxygen there.

[[Imported from SVN: r508]]

[[Imported from SVN: r507]]

not have any indices to publish.

Code now copes with that information.

[[Imported from SVN: r506]]

[[Imported from SVN: r505]]

not follow the local ones.

[[Imported from SVN: r504]]

[[Imported from SVN: r503]]

[[Imported from SVN: r502]]

[[Imported from SVN: r501]]

There was huge bug in the numbering of the aggregates:

Due to the coarsening of the index set the matrix on the coarse level are ordered like this. First come the row corresponding to vertices owned by the process
(0, 1, ..., n) and the the rows owned by other process (n+1, n+2, ..., N).

Unfortunately during the setup of the sparsity pattern of the matrix this fact was neglected :-(

Fixed this now.

[[Imported from SVN: r500]]

[[Imported from SVN: r499]]

To be consisten with owneroverlapcopycomunication ones has to use the copy flag instead of the overlap flag for marking vertices not owned by the process.

[[Imported from SVN: r498]]

local to global index.

[[Imported from SVN: r497]]

the aggregate numbers.

Needed to invent a marker for for isolated nodes in the communication.
Simply using AggregatesMap::ISOLATED is a bug as there might be a global
index (if sizeof(globalindex)>4) that is the same. We now use -1.

[[Imported from SVN: r496]]

[[Imported from SVN: r495]]

[[Imported from SVN: r494]]

[[Imported from SVN: r493]]

[[Imported from SVN: r492]]

aggregate.

The distance was not calculated correctly before.

[[Imported from SVN: r491]]

constructor that just takes the size argument as this neglects the
structure of the values. I. e. if the values of a block vector are
block vectors itself there was no space allocated and thus the solvers
segfaulted.

In case of CGSolver this fixes Flyspray task #81
<http://hal.iwr.uni-heidelberg.de/flyspray/index.php?id=81&do=details>.

It will not be closed as I want to write some tests for the ISTL
kernels as this bug is likely to occur at other places as well.

[[Imported from SVN: r490]]

[[Imported from SVN: r489]]

[[Imported from SVN: r488]]

off warnings.

[[Imported from SVN: r487]]

[[Imported from SVN: r486]]

[[Imported from SVN: r485]]

[[Imported from SVN: r484]]