CHANGELOG

Notable Differences Between DuMuX 2.0 and DuMuX 2.1
===================================================

- The thermodynamics framework has been overhauled:
  - The programming interfaces for fluid systems, fluid states and
    components has been formalized and cleaned up.
  - Fluid systems now have the option to cache computationally
    expensive parameters if they are needed for several relations.
  - Fluid systems are not charged with the computation of the
    chemical equilibrium anymore.
  - Fluid states are now centralized infrastructure instead of being
    model-specific.
  - Constraint solvers (which simplify solving thermodynamic
    constraints) have been introduced.
- Outflow boundary conditions have been implemented for the
  fully-implicit models 1p2c, 2p2c(ni) and stokes(2cni).
- The problem and spatial parameter base classes also provide optional
  model-independent interfaces. These methods only get the position in
  global coordinates as argument and are named *AtPos()
  (e.g. boundaryTypesAtPos()). This allows an easy transfer of problem
  definitions between implicit and sequential models.
- The following fully-implicit models have been added:
  - 3p3c, 3p3cni: Isothermal and non-isothermal three-phase,
    three-component models for flow and transport in porous media
    based on primary variable switching.
  - MpNc: A model for arbitrary number of phases M > 0, and components
    (N >= M - 1 >= 1) for flow and transport in porous media. This
    model also comes with an energy and a molecular diffusion module.
  - stokes, stokes2c, stokes2cni: Models for the plain Stokes
    equation as well as isothermal and non-isothermal Stokes models
    for two-component fluids.
- The sequentially-coupled models have been overhauled:
  - A common structure for cell centered standard finite volume
    implementations has been introduced.
  - The data structures where overhauled to avoid large clumps of data
    in large-scale simulations: Each cell stores data in its own
    storage object.
  - The too large assemble() methods have been split into submethods
    getStorage(), getFlux() etc. By this, inheritance of classes has
    been improved and code duplication was reduced.
  - The conceptual seperation of the "VariableClass" (central
    infrastructure), data storage, transport model and pressure model
    has been improved.
  - More of infrastructure is now shared with the implicit models
    (e.g. the BoundaryTypes). This results in significant performance
    improvements and makes debugging easier.
- The 2padaptive sequentially coupled model has been added. This model
  implements a grid-adaptive finite volume scheme for immiscible
  two-phase flow in porous media on non-conforming quadrilateral
  grids.
- The dependencies for the external dune-pdelab and boost packages
  have been removed.
- The build system has received major improvements:
  - There is now much better test coverage of build-time dependencies
    on packages for the default autotools-based build system.
  - Experimental support for building DuMuX using CMake has been much
    improved. In the long run, CMake is projected to become the
    default build system.
- All headers can now be included without any preconditions.
- DuMuX now compiles without warnings if the -pedantic flag used for GCC.
- Specifying run-time parameters is now possible. The mechanism allows
  to use parameter files or to specify parameters directly on the
  command line and fallback parameter input files have been added for
  each test application.  As a consequence, applications can be run
  now without specifying any command line arguments.
- The DuMuX property system has been fine-tuned:
  - Encapsulating property names with the PTAG() is no longer required
    for the GET_PROP* macros (but is still allowed).
  - Setting property defaults has been deprecated.
  - All properties defined for a type tag can now be printed. Also,
    their value and the location in the source where they where
    specified is included in the output.
- Using quadruple precision math has been made possible for GCC 4.6 or newer:
  - To use it, the configure option '--enable-quad' needs to be added
    and the type of scalar values needs to be changed to 'quad'. This
    can be done in the problem file using

        SET_TYPE_PROP(YourProblemTypeTag, Scalar, quad);

    It should be noted, that performance is very poor when using
    quadruple precision arithmetic. This feature is primarily meant as
    a debugging tool to quickly check whether there are machine
    precision related convergence problems.