GitLab

#1385 Make more generic the documentations about copy/move constructor/affectation operator, and starts fixing Geometry missing comments.

#1385 Complete Doxygen documentation so that latest versions of Doxygen may be used without letting warnings.

#1371 Some remaining InputParameterFile replaced, and InputDataInstances directory is now splitted into InputData/Instances.

#1367 Adapt the InputData.hpp XCode template so that it is filled with values that makes the code compile (but are obviously to modify for your model!).

#1367 XCode: update XCode templates to reflect the current state of the library (done only for the basic ones used in model: Model, VariationalFormulation, (renamed) InputData and main).

Previously, you had to fill each individual type of possibility for a Parameter, regardless of the one finally chosen. For instance, you had something like (skipping the documentation...):

VolumicMass
{
    nature = "constant"
    scalar = 1.
    lua_function = none
    piecewise_constant_domain_id = { }
    piecewise_constant_domain_value = { }
}

and each of these entries should be filled (a shortcut 'none' was however created for lua_function).

This made the Lua files rather heavy, and moreover signaled the non used fields (here VolumicMass.lua_function, VolumicMass.piecewise_constant_domain_id and VolumicMass.piecewise_constant_domain_value) as unused.

Now using up the strength of newly introduced in C++17 std::variant (or boost::variant for Apple Clang currently - see #1305), a more synthetic writing ha been introduced:

VolumicMass
{
    nature = "constant"
    value = 1.
}

where 'value' type depends on the 'nature' choice. For instance, a user might replace it by:

VolumicMass
{
    nature = "piecewise_constant_by_domain"
    value = { [3] = 1., [5] = 0.5 }
}

Please notice the usage of the associative container which was introduced when LuaOptionFile replaced Ops.

NOTE: Rather unusually, I have here merged many commits into one: due to the large scope of the modifications to do, I worked only on a specific test. Consequently, other models and tests stopped to compile, and I really like the code to at least compile at all of its commit not to break the git bisect functionality.

#2000 Model: make the type of MoReFEMData a template argument of the class. This was not done earlier to avoid making base Model a template class... but it has long become since a template class anyway, so it's better to use the possible extra leverage (next step is to store reference to the \a MoReFEMDataType and hence avoiding giving it as parameter in three of the class methods).

#1211 Add tests folder to Doxygen scope and fix warnings. ModelInstances not yet added but some of the warnings have nonetheless been fixed in the same time as their test counterpart (for instance NumberingSubsetIndex legend has been added for both tests and ModelInstances folder).

#1192 Modify in SCons directory the instances of HappyHeart (whatever the case used); Sources have been updated accordingly to keep scons compiling everything.

#1090 Introduce a new template parameter to AssembleImpl so that it's now possible to just call InitLocalComputation.

#1076 Introduce aliases in GlobalVariationalOperator.hpp to both cases currently implemented (namely SameForAllRefGeomElt and DependsOnRefGeomElt).

#1076 GlobalVariationalOperator: generalize the class so that each RefGeomElt may get its own LocalVariationalOperator.

The idea is that now the second template argument is not a LocalVariationalOperatorT, but a tuple which contains pair <GeometricEltEnum, LocalVariationalOperatorForGeometricEltEnum>.
As most operators (all present in HappyHeart so far, with possibly GradientBasedElasticityTensor apart) do not need different implementations for each RefGeomElt, the new generic class is named clumsily GlobalVariationalOperatorSeveralLocal and an alias has been defined to cover the most common case where same type is given for all geometries.

Many safeties remain to be added (the tuple is for instance a pair only for that purpose: a static assert will check all RefGeomElt are properly given and in the right order).

#1076 Rewrite Assemble so that it might be used in a metaprogrammation context. I had issues with variadic syntax; following the difficulty I had I'd rather commit now I have the proof of concept. even if a lot of clean-up remains to be done (warnings to begin with, helper structs not named correctly, many missing comments...)

#1069 Replace all lingering Private namespace in the main part of the library, by either Internal namespace or sometimes an Impl one (for very local instances). I also removed unused hxx files.