GitLab

This selects the BCGS_AMG_SSOR linear solver backend for FV computations
in both the Richards and the Transport model.

Use warning paragraphs in doxygen docs to inform about non-collective
methods of BoundaryConditionManager and TimeStepController.

* Implement matric head grid function adapter transforming water content
  into matric head.
* Add parameterization functions necessary to compute matric head from
  water content.
* Use shared_ptr when using instances of initial condition
  objects in models.

Additionally, it updates documentation, targets, and factories related to the polynomial degree of the transport model.

This introduces an option for a heuristic time step reduction scheme:
If the solution contains negative values, it is discarded, the
time step is reduced, and the computation is repeated.

* Pass water content grid function to CFL condition and calculate
  actual velocity together with flux.
* Read courant number once and throw error if it is negative.
* Print CFL condition result into trace log.

* Define classes of polymorphic boundary conditions and factory
* Only use enum in Operator namespace to discern types
* FlowBoundary now returns an entire boundary condition
* Adapt local operators (Richards only)
* Remove header 'util.hh'. Move functions to 'utility.hh', BaseTraits
  to 'typedefs.hh' and GFS helpers to 'gfs_helper.hh'
* Add setting for `horizontal_projection` of Neumann BC (only used in Richards model) #136 (comment 20384).
* Add setting for `concentration_type` of Dirichlet BC (only used in Transport model) #151.

* Increase encapsulation of TimeController and reduce capabilities
* TimeController now only acts upon being called
* Remove validation routine
* Add unit test loading the default config time values

This commit breaks all models.

ModelFactory currently compiles separately, i.e., without linkage to
the 'impl' library. This vastly reduces the amount of code in the main
function.

Use enable_if instead of run-time exception.

The constructors properly initialize the asserted objects. There is
no point in the code where they can be replaced by empty/null pointers.

* Remove transport model constructor expecting grid functions
* Remove transport local operator constructors expecting grid functions.
  This completely removes any validity checking from the local
  operators!
* Add more verbose error message to TransportSimulation::step() if
  grid functions are not properly set.

This branch was take from !111 which already had the parameterization implemented. For the splitting of issues this branch just remove the non related lines for the parameterization

* In order to ensure const-correctness, we instead copy the whole parameterization into a new shared pointer whenever a state is being exported.

* This commit also fixes a mistake in creation of flux reconstruction, which could in certain cases share a cache set up with other times.
* Total solute adapter is essentially equal to product of grid functions defined in PDELab, however, this class get ownership of grid functions. This is necessary when sharing adapters outside of the main simulation.

* This modification allows that it is not needed to know a priori the prescription space, which reduces as well the number of needed instantiations.
* Moreover, now, the flux reconstruction class does not store any shared pointer or reference can may be easily invalidated.

* Now, the current state of the model can be extracted from the model and used later to generate adaptors out of them.

* Some names were wrong and the inheritance structure was not completely right.
* The yaml now file has more expressive nodes to select the parametrization

Signed-off-by: Santiago Ospina <santiago.ospina@iup.uni-heidelberg.de>

Signed-off-by: Santiago Ospina <santiago.ospina@iup.uni-heidelberg.de>

Signed-off-by: Santiago Ospina <santiago.ospina@iup.uni-heidelberg.de>

Signed-off-by: Santiago Ospina <santiago.ospina@iup.uni-heidelberg.de>

Also, fix header in transport.hh