* before, fluxes were called for each eqIdx

* non-backward compatible change
* store const ref to fvGridGeometry
* call resize() without argument

* adapt test

* use attachConvergenceWriter() instead
* store shared_ptr of convWriter in NewtonSolver
* deprecate old interface
* fix compiler error when using convWriter in stationary case

Kilian Weishaupt authored Mar 06, 2019 and Ned Coltman committed Mar 11, 2019

* use small indicies for loops
* do not query for dofs using < 0 to check if they exist
* remove superfluous this-> calls
* use efficient types, clean up loops
* move axis and pair data to namespace Detail

* suppress compiler warning

Dennis Gläser authored Feb 09, 2019 and Timo Koch committed Feb 28, 2019

This flux variables cache precomputes the shape values at the tpfa
flux support point necessary on interior boundary scvfs with Neumann
BCs specified on them.

Dennis Gläser authored Feb 09, 2019 and Timo Koch committed Feb 28, 2019

This way we can formulate the flux dependent on both the matrix as
well as the facet permeability. This significantly reduces the condition
number of the resulting matrix and thus improves newton convergence.

Dennis Gläser authored Feb 11, 2019 and Timo Koch committed Feb 28, 2019

On 3d grids, the maximum value of 8-bit integer types can easily be
surpassed for the local indices within interaction volumes. We use the
larger local index type now. This way we should always be on the safe
side.